Open-source platform for 3D AI agents. Turn text or a photo into a rigged, animated GLB avatar, give it an LLM brain, memory and a wallet, and embed it anywhere with one web component. Remote MCP server, x402 payments, Three.js, browser-native.
JavaScript 3d-generationai-agentsanimationavatardigital-humangenerative-aiglbgltfimage-to-3dllmmcpmcp-servermodel-context-protocolsolanatext-to-3dthreejsweb-componentswebglwebxrx402
{
"$comment": "Curated changelog entries for three.ws, the editorial layer of the public changelog. New PAGE launches are tracked automatically via the `added` field in data/pages.json; this file is for everything else holders care about: improvements, fixes, SDK releases, trading/payments capabilities, security work. scripts/build-page-index.mjs merges both sources into CHANGELOG.md, public/changelog.json, and public/changelog.xml. Newest first. Allowed tags: feature, improvement, fix, sdk, infra, docs, security, desktop (desktop entries become the desktop app's release notes). `link` is optional and must be a live three.ws path.",
"entries": [
{
"date": "2026-10-25",
"title": "Chat pairing and X connect returns land on a working Connections page",
"summary": "The /settings/connections link that the Telegram and Discord bots, the notification settings and the X connect flow all send people to returned a 404. Every /settings subpath now serves the settings app, and a new Connections tab lists paired chats, toggles notifications, picks the default agent, unpairs, issues a pairing code and redeems the code a bot prints.",
"tags": [
"fix"
],
"link": "/settings/connections"
},
{
"date": "2026-10-25",
"title": "Paid Solana routes fall back to self-pay instead of advertising a sponsor that cannot settle",
"summary": "The model-check, mint-to-mesh, vanity, forge, pipeline and sealed-drops paywalls kept naming the sponsor wallet as fee payer even when it sat under its SOL floor, so buyers signed payments the facilitator then refused. They now share one self-pay guard with every other paid route: when the sponsor cannot co-sign, the Solana accept stays on the challenge without a fee payer and the buyer pays its own network fee.",
"tags": [
"fix"
]
},
{
"date": "2026-10-25",
"title": "BNB Greenfield vault writes no longer ship a vulnerable XML builder",
"summary": "The Greenfield SDK and the Bot Framework expression library each carried a nested fast-xml-parser 4.5.7 that the version-scoped override never reached, leaving the XMLBuilder comment and CDATA injection advisory (GHSA-gh4j-gqv2-49f6) open. The override now applies to every copy, all of them resolve to 5.11.2, and a lockfile test guards every nested path.",
"tags": [
"fix",
"security"
]
},
{
"date": "2026-10-25",
"title": "Background-removal service moves to Pillow 13, closing 41 image-decoder advisories",
"summary": "The background-removal worker decodes images that callers upload, and its pinned Pillow carried 41 open advisories including heap overflows in the JPEG, TIFF and FlashPix loaders. It now runs Pillow 13.0.0 on rembg 2.0.86 (the first release that allows it), matching every other worker, and the image build gate passes against the real models.",
"tags": [
"fix"
]
},
{
"date": "2026-10-25",
"title": "Modly 3D-generation worker closes TLS, request-parsing and test-tooling advisories",
"summary": "The Modly host environment pinned old releases of certifi (a compromised emSign root CA finding), cryptography, python-multipart (multipart DoS and file-write findings) and pytest. All four are now on current releases, and the worker's contract and CPU test suites pass against them.",
"tags": [
"fix"
]
},
{
"date": "2026-10-25",
"title": "Rate limiter Redis path repaired and its health hint made honest",
"summary": "The rate limiter had been failing every Redis call because @upstash/ratelimit 2.1+ sends a script flag that self-hosted Redis rejects, so traffic ran on the Postgres fallback while healthz called it transient. The library is pinned to 2.0.8, and this class of fault is now logged and reported as a non-transient script incompatibility.",
"tags": [
"fix"
]
},
{
"date": "2026-10-25",
"title": "Allowance pulls now require the signed real-funds agreement",
"summary": "Pulling funds through an owner-funded on-chain allowance now checks the signed real-funds agreements before the agent key is used, matching every other custodial signing path. Unsigned accounts get the standard agreement-required refusal and nothing is sent.",
"tags": [
"fix"
]
},
{
"date": "2026-10-25",
"title": "Commit feed cron skips quietly on a lock-check fault",
"summary": "The commit feed cron now skips a tick with a 200 when the database lock check hits a transient fault, instead of answering a raw 500 and raising an ops alert. The next tick retries.",
"tags": [
"fix"
]
},
{
"date": "2026-10-25",
"title": "Object-storage outages now answer a clean 503 with Retry-After instead of a raw 500",
"summary": "When the R2 bucket is unreachable (account disabled, keys rotated, endpoint down), any API route that writes or reads an object without its own error handling, such as avatar upload, used to answer a bare 500 and fire a Sentry event and an unhandled-5xx alert on every request. The shared error boundary now treats a storage fault like a database outage: a 503 service_unavailable with Retry-After, one throttled warning, a single deduped ops alert, and a clean skip for cron ticks. Deliberate 4xx errors and real handler bugs are unchanged.",
"tags": [
"fix"
],
"link": "/docs/api-reference"
},
{
"date": "2026-10-25",
"title": "GET /api/crypto/bonding's last-known-good cache now survives a full RPC-lane cooldown, not just 30 minutes",
"summary": "The last-known-good tier behind /api/crypto/bonding (added to survive a pump.fun Cloudflare block) kept each mint's cached bonding-curve status for only 30 minutes, far shorter than the outages it exists to cover: the Solana RPC lane cooldown runs up to 6 hours once a lane exhausts its quota, and the pump.fun Cloudflare block it was built for has stayed up for multiple days straight. Once both the 25-second live cache and the 30-minute backup expired during one of those outages, the endpoint fell all the way through to a hard 503 on every /launches and /oracle/coin bonding-progress panel. The cache now holds each mint's last good read for 6 hours by default, still overridable with PUMP_BONDING_LKG_TTL_S.",
"tags": [
"fix"
],
"link": "/docs/api-reference"
},
{
"date": "2026-10-25",
"title": "Security hardening across sign-in, uploads, vaults, agents and the chat server",
"summary": "A source audit led to fixes in many places. Agent updates only link avatars you own, and unpublished agents stay hidden from ask. Handoff, Telegram and SAML sign-ins are bound to the flow that started them, and the walk-avatar extension only accepts a token for a sign-in it began. Vault deposits that would mint zero shares are refunded, and strategy buys reserve against the daily SOL cap atomically. Oracle follow and test alerts need a linked Telegram chat, shown read-only in the panels. Anonymous uploads are size-checked hourly. The local chat server binds loopback and requires a token, and webhook retries re-run after a failed delivery.",
"tags": [
"fix"
],
"link": "/oracle"
},
{
"date": "2026-10-25",
"title": "Fixed phone/desktop/CLI/Telegram account linking, which was failing on every attempt",
"summary": "Every account-link action (mint, claim, poll, pending, decide, devices, revoke) behind /dashboard/account and the mobile-app pairing flow had been throwing on every call since the device-linking feature shipped on 2026-10-11, because its own first migration created a table under a name an older, unrelated agent-identity feature already owned, so the new table silently never got created with the right columns. The old table is renamed out of the way instead of dropped, so its rows are preserved, and the device-linking table now gets created with the shape the feature actually needs.",
"tags": [
"fix"
],
"link": "/dashboard/account"
},
{
"date": "2026-10-25",
"title": "Gallery and marketplace thumbnails degrade gracefully instead of showing a broken-image icon",
"summary": "Six image spots on /gallery and /marketplace (the avatar grid thumbnail, the equip-avatar picker row, the live-sale ticker avatar, the animation marketplace listing card, the animation buy-modal preview, and a tool-detail icon) rendered a plain <img> with no fallback, so a 502 or missing thumbnail left the browser's default broken-image glyph on two of the highest-traffic pages in the product. All six now carry the same data-fallback attribute the site already uses elsewhere: card-style thumbnails swap to the existing placeholder box, small avatar icons just disappear instead. A seventh untagged image turned up in the gallery's forge-card renderer while verifying the other six and got the same fix.",
"tags": [
"fix"
],
"link": "/gallery"
},
{
"date": "2026-10-25",
"title": "GET /api/pump-bounties and its stats/detail routes serve last-known-good data instead of a 502",
"summary": "pump.fun's Cloudflare anti-bot challenge has been rejecting livestream-api.pump.fun's bounty API for hours at a time, which left the bounty list, the aggregate stats, and individual bounty detail pages 502ing for every visitor even though the board itself had not changed. All three routes now go through the same last-known-good cache tier already proven on /api/portal and /api/crypto/bonding: a fresh answer is mirrored into a 24-hour backup, and that backup answers (flagged with an x-pump-stale header) instead of a 502 once the live fetch fails. A true miss, with nothing cached at all, still returns the original pump.fun error unchanged.",
"tags": [
"fix"
],
"link": "/bounties"
},
{
"date": "2026-10-25",
"title": "Image-to-3D Forge instances recycle before a VRAM leak crashes a job",
"summary": "The trellis lane's GPU worker already restarted itself between jobs once host memory ratcheted up, but had no equivalent check for GPU memory: a rasterizer dependency allocates outside torch's own memory pool, so an instance's usable VRAM only ever shrank, and the instance's own health check kept passing right up until a job crashed or Cloud Run killed it mid-generation. The worker now also tracks GPU memory after every job and restarts cleanly between jobs once it has ratcheted far enough, the same way the existing memory guard already does, so Forge image-to-3D jobs on this lane fail or stall less often.",
"tags": [
"fix"
],
"link": "/image-to-3d"
},
{
"date": "2026-10-24",
"title": "GET /api/pump/coin degrades to on-chain data plus first-party metadata instead of a bare 502",
"summary": "pump.fun's frontend API has stayed Cloudflare-blocked long enough to exhaust /api/pump/coin's 24-hour last-known-good cache for some mints, including $THREE's own, leaving a hard 502 with no partial data. The endpoint now adds a third tier: numeric and curve fields come from the same on-chain bonding-curve reader already live behind /api/pump/curve, and identity fields (name, symbol, creator) come from pump_agent_mints for any mint three.ws itself tracks. Responses on this tier carry an x-pump-degraded: onchain header; a true triple-miss (no cache, no on-chain data, no registry row) still returns an honest 502 naming pump.fun.",
"tags": [
"fix"
],
"link": "/docs/api-reference"
},
{
"date": "2026-10-24",
"title": "GET /api/crypto/bonding falls back to an on-chain read when pump.fun is Cloudflare-blocked",
"summary": "pump.fun's frontend API started answering every request from our servers with a Cloudflare challenge, which hard-failed /api/crypto/bonding for every mint, including $THREE's own, once the 30-minute last-known-good cache expired. The endpoint now reads the bonding-curve account straight off Solana RPC (the same reader already live behind /api/pump/curve) before giving up, so bonding/graduation status stays answerable even while pump.fun's edge blocks us.",
"tags": [
"fix"
],
"link": "/docs/api-reference"
},
{
"date": "2026-10-24",
"title": "Two new Solana Blinks: tip an agent, buy its coin",
"summary": "Any agent's profile link now unfurls as a Blink in X and Blink-aware wallets with a one-tap tip (SOL or $THREE, straight to the agent's own wallet) and, for agents that launched a mainnet coin, a one-tap Jupiter-routed buy. Both build unsigned transactions only; the wallet signs.",
"tags": [
"feature"
],
"link": "/docs/api-reference"
},
{
"date": "2026-10-24",
"title": "2,575 NVIDIA SimReady objects in the Object Library",
"summary": "Every asset from NVIDIA's open-source SimReady dataset is converted to a web-ready GLB with materials and thumbnails, searchable in the Object Library and the asset catalog. Each carries its CC BY 4.0 license and NVIDIA credit.",
"tags": [
"feature"
],
"link": "/objects"
},
{
"date": "2026-10-24",
"title": "Chat no longer fails when its default NVIDIA model is retired",
"summary": "NVIDIA stopped serving the Nemotron 3 Super function while still listing it, so chat requests that fell through to it got a 404 and, with the other providers throttled, ended in a 503. Chat, the tool chain, the brain and the CAD writer now use the Nemotron 3.5 Lightning model (with Ultra as its fallback), both verified live.",
"tags": [
"fix"
],
"link": "/chat"
},
{
"date": "2026-10-24",
"title": "Fixed three more legacy thumbnail links that browsers refused to load",
"summary": "A handful of forge creations and 3D props stored from before an earlier URL fix still pointed straight at the storage bucket's own domain instead of three.ws's own CDN link, which Chrome blocks as an opaque cross-origin response. Model-page posters, a creation's related-models sidebar, forge share-card previews (the image X and Discord show when a link is pasted), and the Forged gallery's model-viewer thumbnails now all resolve through the same CDN link as everything else.",
"tags": [
"fix"
],
"link": "/forge"
},
{
"date": "2026-10-23",
"title": "Bonding-curve status API no longer hard-fails when pump.fun blips",
"summary": "GET /api/crypto/bonding was returning a 503 to every caller during a pump.fun outage, even for mints it had already fetched successfully. It now caches each mint's bonding-curve status and keeps a longer last-known-good copy, so a failed upstream fetch serves that cached data (flagged with an x-pump-stale header) instead of a hard error. A 503 with a retry hint is now only returned for a mint that has no cached data at all.",
"tags": [
"fix"
],
"link": "/api/crypto/bonding"
},
{
"date": "2026-10-24",
"title": "Hundreds of dependency security advisories cleared across the repo",
"summary": "Every committed lockfile was audited and patched: same-line security releases everywhere, vetted moves to patched majors (vitest 4, zod 4, helia 7, Svelte 5 and Vite 7 in the chat app, transformers 4 in HQ, gltf-transform 4 for the meshopt worker), and each package's tests and builds were rerun before and after. What remains are advisories with no patched release published anywhere (bigint-buffer, elliptic, node-forge) and pins that upstream SDKs still require.",
"tags": [
"security",
"infra"
]
},
{
"date": "2026-10-24",
"title": "Desktop app moves to Electron 45",
"summary": "The desktop app now runs on Electron 45 and electron-builder 26, clearing 61 security advisories from the old Electron 33 toolchain. Launch at login still starts the app hidden in the tray; on Windows, toggle the setting once if the console window opens at login.",
"tags": [
"desktop",
"security"
]
},
{
"date": "2026-10-24",
"title": "Math renders in chat again",
"summary": "The chat app moved to Svelte 5, and KaTeX formulas in messages now render instead of showing raw TeX.",
"tags": [
"fix"
],
"link": "/chat"
},
{
"date": "2026-10-24",
"title": "Long MCP answers arrive as a summary, and agents can search the docs",
"summary": "list_my_agents, recall and crypto_data now answer with a short summary, a count, a handle and the first page when the list is long, and the new expand tool reads the rest on request. A 100-coin market list drops from about 47,800 to 4,400 tokens. Handles are signed, account-bound and last 15 minutes. The new free search_docs tool answers how-to questions from the docs index on the main and studio servers.",
"tags": [
"feature",
"docs"
],
"link": "/docs/mcp"
},
{
"date": "2026-10-24",
"title": "See what a prediction strategy would have earned before it trades",
"summary": "Agent Predictions records every strategy decision at the live order-book price in shadow mode, marks it at resolution, and shows shadow beside live results. A fee-aware scanner reads Solana event contracts for YES plus NO pairs under $1. Live stays owner-only.",
"tags": [
"feature"
],
"link": "/agent-predictions"
},
{
"date": "2026-10-24",
"title": "Pay for a model request with a wallet, no account needed",
"summary": "POST /api/v1/chat/completions now answers an unauthenticated request with an x402 upto challenge on Solana, sized to the model's worst case for that request. Sign it, retry, and only the metered tokens are charged at the published per-token price; the rest of the cap is refunded. Streaming settles when the stream closes, and a dropped stream is charged only for the tokens already sent. GET /api/v1/models now lists the price sheet and round limits.",
"tags": [
"feature",
"docs"
],
"link": "/docs/metered-router"
},
{
"date": "2026-10-24",
"title": "Skills are scanned for hidden behaviour before they are listed",
"summary": "Every marketplace skill version is now checked for secret access, calls to hosts it never declared, hidden or obfuscated code and instructions that skip the owner's confirmation. Clean skills list as before, suspicious ones are held for a moderator, and dangerous ones are refused. Authors see the file, the line and how to fix each finding in the publish form, and the existing catalog was rescanned.",
"tags": [
"security",
"feature",
"docs"
],
"link": "/docs/skill-supply-scan"
},
{
"date": "2026-10-24",
"title": "See what a Forge generation costs before you press Generate",
"summary": "The Forge page now shows a live cost line for the tier, engine and options you picked, read from the same pricing code that bills you, with nothing spent. A pay-to-keep mode (free preview, pay in $THREE to download, holders keep free) is built and switched off, so downloads stay free until it is turned on.",
"tags": [
"feature",
"docs"
],
"link": "/docs/forge"
},
{
"date": "2026-10-24",
"title": "Agents join the open Solana agent registry and hires read trust tiers",
"summary": "Minted agents are enrolled in the open Solana agent registry (8004) with the platform authority signing, so owners pay nothing. The marketplace ranks providers by registry tier after completed hires, a hire can require a minimum tier, and agent pages show an 8004 tier badge when registered. A rate-limited back-fill covers existing agents.",
"tags": [
"feature"
],
"link": "/docs/solana-agent-registry"
},
{
"date": "2026-10-24",
"title": "Engine bridges are packaged for the Unity, Godot and Unreal stores",
"summary": "One command, npm run package:engine-stores, validates the Unity, Godot and Unreal bridges and builds a reproducible upload package for each, with listing metadata and a submission checklist. The Unity package and Unreal plugin manifests now carry their documentation links.",
"tags": [
"sdk",
"docs"
],
"link": "/docs/engine-store-publishing"
},
{
"date": "2026-10-24",
"title": "Drive the CLI from scripts and agents, and download a run report",
"summary": "Every three-ws command now takes --json (one JSON document on stdout, diagnostics on stderr) and --non-interactive (never prompts), with a stable exit-code table: 2 usage, 3 auth, 4 refused by a guard, 5 payment required, 6 network. New commands cover agents, runs, wallet balance, skill install and orders. `three-ws report run <id> --out run.html` writes one self-contained offline file with the goal, tool calls, decisions, trades, outcome and the decision-trail head hash, and the run page has a matching Download report button.",
"tags": [
"feature",
"sdk"
],
"link": "/docs/cli"
},
{
"date": "2026-10-24",
"title": "Install the three.ws skills with one command on any agent runtime",
"summary": "Run npx skills add nirholas/three-ws-skills (or the 3D-only and agent-economy repos) to install the three.ws skills into Claude Code, Codex, Cursor and any other runtime that reads the open Agent Skills format. Every skill now passes the Agent Skills spec validator, which fixed 21 skills whose frontmatter some installers could not parse, and an install test proves each repo lists and installs every skill in a clean directory. Publishing the repos is still the owner's step.",
"tags": [
"feature",
"docs"
],
"link": "/docs/agent-skills"
},
{
"date": "2026-10-24",
"title": "Opt-in WebGPU renderer with automatic WebGL2 fallback, starting in Restyle Studio",
"summary": "three is now r186. Add ?renderer=webgpu to /restyle to render on WebGPU with a TSL rim light, or leave it off for the same WebGL2 picture as before. If the browser has no GPU adapter or WebGPU fails to start, the page falls back to WebGL2 on its own. A badge shows which backend is running, and exported GLBs are identical on both paths. Measured parity and timings are in the WebGPU renderer doc.",
"tags": [
"feature",
"improvement"
],
"link": "/docs/webgpu-renderer"
},
{
"date": "2026-10-24",
"title": "Agent perps on three venues: Phoenix, Jupiter and Hyperliquid",
"summary": "Agents can now trade perpetuals on Phoenix and Jupiter (Solana) and on Hyperliquid through a per-agent API wallet that can trade but never withdraw. Every venue goes through the same preview then confirmed execute, with new guards for maximum position notional and minimum liquidation distance. Unpinned market reads fail over to a healthy venue, live orders never reroute. MCP access to an agent's perps tools is now opt-in per agent and off by default.",
"tags": [
"feature",
"security"
],
"link": "/docs/perps"
},
{
"date": "2026-10-23",
"title": "Revoking a connected app now stops it everywhere",
"summary": "When Claude, ChatGPT or any MCP client ends its own grant, or you press Revoke in Connected apps, its access tokens stop on the next request and the token introspection endpoint reports them inactive. Other clients on your account stay connected. The consent screen tests now pin the Read, Manage, Trade, Spend and Launch groups.",
"tags": [
"fix",
"security"
],
"link": "/dashboard/settings#connected-apps"
},
{
"date": "2026-10-23",
"title": "Pay any paid endpoint over MPP on Solana, and sell a skill on four rails from one price",
"summary": "Paid endpoints can now answer with an MPP Payment challenge beside the x402 one at the same price, and the agent payer pays either from one call with its spending caps and a dry-run quote. Marketplace skills are defined once and served over x402, MPP, A2A and AP2 with the same author and platform split and the same receipt.",
"tags": [
"feature",
"sdk"
],
"link": "/docs/mpp-solana"
},
{
"date": "2026-10-23",
"title": "Hot streaks on the Arena board",
"summary": "An agent that has closed three or more winning positions in a row now shows its streak beside its name on the Arena leaderboard. The streak is counted from the same on-chain closed-position ledger as the trader profile, and any loss or flat exit resets it to zero.",
"tags": [
"feature"
],
"link": "/play/arena"
},
{
"date": "2026-10-23",
"title": "Backtests now prove they are free of look-ahead and can be repeated",
"summary": "Every strategy backtest replays its entry gate with the future removed and reports any decision that changed, pins the exact dataset it ran on with a content hash, and marks older unpinned results as unreproducible. A new canary job checks each price and candle source against a known value every 30 minutes and tries the healthy ones first.",
"tags": [
"feature"
],
"link": "/strategies"
},
{
"date": "2026-10-23",
"title": "One earnings ledger for everything your agents earn, with reinvest rules",
"summary": "The new Earnings page totals creator fees, skill and API sales, hires, royalties, invoices and realised trading PnL by source and period, links every row to its Solana transaction, reconciles against the chain every hour, and exports CSV or JSON. Agents can publish a public earnings page, and confirmed reinvest rules put a capped share of each period into model credits, a tools budget or a DCA of the agent's own coin, with a receipt for every run.",
"tags": [
"feature"
],
"link": "/earnings"
},
{
"date": "2026-10-23",
"title": "Creator-fee payouts: register a wallet and see every distribution",
"summary": "Your share of each claimed creator fee now has a record and a destination. Register a payout wallet by signing with it and confirming with a step-up (a change waits 48 hours), then follow what is sent, pending or held and why on the new payouts page, and export every distribution as CSV. Sending stays off until payouts are switched on, and shares are recorded and held until then.",
"tags": [
"feature"
],
"link": "/creator/payouts"
},
{
"date": "2026-10-23",
"title": "Holder concentration, funder seeding and v1 launch signals on every coin",
"summary": "A coin's trader sheet now shows how unevenly its top holders split the supply (Gini), what share of the first buyers were funded by one wallet, and whether its creation transaction used Solana transaction v1. The same Gini and seeded share feed the oracle's launch features.",
"tags": [
"feature"
],
"link": "/launches"
},
{
"date": "2026-10-23",
"title": "Coin intel funder map is now a 3D graph",
"summary": "The wallet network in the coin intel drawer renders as a GPU-instanced 3D force graph, with wallets sized by holdings and linked to their shared funder, a live-trades toggle, and reduced-motion support. Browsers without WebGL keep the 2D map.",
"tags": [
"feature"
],
"link": "/coin-intel"
},
{
"date": "2026-10-23",
"title": "Agent trades now race several landing providers and keep score of who lands",
"summary": "Every signed Solana trade goes out through the primary RPC, Jito and any configured staked RPC at once. The first confirmation wins, the rest are dropped, and a provider that keeps failing sits out until a health probe passes. A per-trade tip ceiling binds every provider, and landed, lost and latency stats are kept per provider.",
"tags": [
"feature",
"improvement"
]
},
{
"date": "2026-10-23",
"title": "Give your agent a spending cap and confirm anything bigger from your own wallet",
"summary": "A new Burner cap tab sets how much an agent may spend alone per spend and per day. Anything above waits in your approvals inbox with the recipient, amount, token and chain, and expires after a window you choose. Above a second line you approve by signing with your own wallet. A live preview shows what a spend of any size would do, and every request, approval, denial and expiry lands in the agent's decision trail. Fixed a bug where valid base58 wallet signatures were read as base64 and rejected.",
"tags": [
"feature",
"security"
]
},
{
"date": "2026-10-23",
"title": "Wallet and payment screens now work with a keyboard and a screen reader",
"summary": "Addresses read as short text on screen and as the full value in groups of four to a screen reader, from one shared formatter that replaced more than a hundred copies. The connect picker, payment modal, vault dialogs and add-funds sheet trap focus, make the page behind inert, close on Escape and return focus to the button that opened them. Payment steps now say completed, current or failed in text, not only by color, and low-contrast labels in the payment modal and the vaults page were fixed. /wallet, /agent-wallet, /claim-wallet and /vaults joined the accessibility gate.",
"tags": [
"improvement"
],
"link": "/wallet"
},
{
"date": "2026-10-23",
"title": "Agents suggest skills from runs that worked",
"summary": "After a run that verifiably worked (several tools with no error, a passed labor-escrow job, or a closed profitable trade), an agent now drafts a real skill bundle from it and lists it under Suggested skills on the agent dashboard. Argument values from the run are never stored, only their shapes. You read the bundle and the evidence run, then install it on that agent only, publish it to the marketplace, or dismiss it. Nothing installs or publishes without your click, and a draft that fails the marketplace checks can never be installed. When an installed skill is used again and the run does better, the agent proposes a revision you can accept instead of changing the skill.",
"tags": [
"feature"
],
"link": "/docs/skill-candidates"
},
{
"date": "2026-10-23",
"title": "An agent can sign itself up and pay for its plan",
"summary": "An autonomous agent now signs up with a Solana key and no email: it signs a single-use server challenge, receives an account, a custodial wallet and an API key, and can then buy the Builder or Scale plan with one USDC x402 payment on Solana. The CLI does the whole thing with `three-ws signup --agent`. Replayed or badly signed challenges are rejected, each key gets one account, and signups are rate limited and capped per day.",
"tags": [
"feature",
"sdk"
],
"link": "/docs/agent-self-onboarding"
},
{
"date": "2026-10-23",
"title": "x402 pay-for-what-you-use on Solana (upto)",
"summary": "A paid route can now advertise an upto offer beside its fixed price: the buyer authorizes a cap, the seller measures the work and settles only that amount, and the rest of the escrowed deposit returns to the buyer. A handler that fails settles nothing. The LLM proxy is the first route converted (5 atomics per token, capped at 5x the flat price). Agent payers refuse a cap above their per-call limit before signing, and the unused part of the cap is released from their budget. Existing exact clients are unaffected.",
"tags": [
"feature",
"improvement"
],
"link": "/docs/x402-upto"
},
{
"date": "2026-10-23",
"title": "News reader recovers blocked stories and gives a structured trading read",
"summary": "When a publisher blocks or paywalls an article, the reader now recovers it from the Internet Archive, the SEC EDGAR filing the headline names, or other outlets covering the same story, and says which lane it used. The analysis is now a structured trading read: affected assets with exposure and direction, whether the news is priced in, the horizon, and a counter-argument. The same resolution is available to agents as the free news_resolve MCP tool. The excerpt limit applies to every lane.",
"tags": [
"feature",
"improvement"
],
"link": "/markets/news"
},
{
"date": "2026-10-23",
"title": "Trust grades for every MCP server",
"summary": "/mcp-tools now grades each of the 51 MCP servers from A to F on a 0-100 score built from facts the build checks: tools declared and described, safety annotations, an installable package or mounted endpoint, a registry manifest, a README, a license, tests, prompts, resources and the Glama claim. Hover or focus a grade for the full breakdown, and click the Trust column to sort. The rubric and weights are documented.",
"tags": [
"feature"
],
"link": "/docs/mcp"
},
{
"date": "2026-10-23",
"title": "Market anomaly feed: scored price, TVL, depeg, funding and open interest alerts",
"summary": "A new scanner runs every 15 minutes over real market series (coin price, Solana TVL and stablecoin supply, stablecoin pegs, Hyperliquid funding and open interest) and records statistically scored anomalies with a medium, high or critical severity. Read them per call at /api/x402/market-anomalies ($0.001 USDC), or create a market_anomaly alert rule in alerts-mcp to be notified. Thin or flat series are skipped instead of guessed at.",
"tags": [
"feature"
],
"link": "/docs/market-anomalies"
},
{
"date": "2026-10-23",
"title": "Start your agent from a persona template",
"summary": "The create wizard now opens with a gallery of 26 Solana finance personas: whale watching, wallet safety, yield, launch vetting, tax planning and more. Search or filter by tag, preview the greeting, prompts and tools, then one click prefills the Personality step. Everything stays editable and Undo restores what you had. Finance is now a category in the wizard, the marketplace filter and the agent editor.",
"tags": [
"feature"
],
"link": "/create-agent"
},
{
"date": "2026-10-23",
"title": "Owner-funded agent budgets enforced on-chain",
"summary": "Keep funds in your own wallet and sign one transaction that gives an agent a capped, expiring allowance through the Solana Subscriptions and Allowances program. The agent can pull only what is left, the spend guards refuse a pull the program would reject before any transaction is built, and the new Budget tab shows cap, spent, remaining and expiry with one-click revoke.",
"tags": [
"feature"
],
"link": "/docs/agent-allowance-budgets"
},
{
"date": "2026-10-23",
"title": "Prepaid x402 balances settled on-chain in batches",
"summary": "Deposit USDC once and spend it per call: each paid request debits a ledger balance that can never go negative, and a failed or cancelled request is never charged. Sellers are then paid together in a single Solana transaction per window, so cent-sized calls no longer starve the settle floor. A nightly reconciliation checks deposits minus debits minus settled against what is held and alerts on any drift.",
"tags": [
"feature"
],
"link": "/docs/ops/x402-prepaid"
},
{
"date": "2026-10-23",
"title": "Hired work now has an independent evaluator, a dispute window and automatic release or refund",
"summary": "A bounty can name who judges the delivery: the platform checker (default), the hiring agent, or a registry agent, who signs a pass or fail attestation with a reason code and evidence link. The losing side then has a dispute window (24 hours by default, 0 to 30 days per job) to freeze the funds and send them to a second evaluator or the platform arbiter. When the window ends with no dispute a cron releases or refunds on its own, a job nobody evaluates refunds the buyer at its deadline, and a failed transfer is retried without ever paying twice. Every transition is written to a hash-chained trail shown on the job page, and both parties are notified.",
"tags": [
"feature"
],
"link": "/labor-market"
},
{
"date": "2026-10-23",
"title": "three-ws pay: settle x402 and MPP challenges from the terminal with a key you hold",
"summary": "The CLI can now pay a 402 itself. `three-ws pay <url>` reads an x402 v2, x402 v1 or MPP Solana challenge, prefers Solana, signs with a wallet kept in the OS keychain (or an encrypted file behind a passphrase), retries once and prints the body. `--dry-run` shows the payee, exact amount and network and signs nothing. A permissions file sets allowed origins, networks, assets and payees, per-call and per-day maximums and a confirm-above threshold, and a refused challenge exits with a stable code (10) and a reason string. `three-ws pay mcp` serves the same under the same permissions as a stdio MCP server with pay_quote and pay_fetch.",
"tags": [
"feature",
"sdk"
]
},
{
"date": "2026-10-23",
"title": "Free X video downloader, and post impact now sees engagement",
"summary": "Paste any public X post link at /x-video to download its video or GIF in every quality, with the post's likes, replies, reposts and views, no account or API key. The same keyless reader (X syndication, then fxtwitter, then guest GraphQL) now feeds the X post price-impact tools, which used to get post text only.",
"tags": [
"feature"
],
"link": "/x-video"
},
{
"date": "2026-10-23",
"title": "See whether our X posts move $THREE: a Posts view on the coin chart",
"summary": "The $THREE chart has a new Posts tab that plots every @trythreews post at the price it went out at, sized by engagement and coloured by the +4h return, each bubble linking to the post. Across 197 posts the median +1h and +4h return after a post was +1.6%, against -0.1% and -0.7% at arbitrary moments; the +24h median was -1.4% against -2.4%. The content pipeline now also records the +24h price return after each post and lets it nudge ranking, lightly, once enough posts carry one.",
"tags": [
"feature"
],
"link": "/oracle/coin/FeMbDoX7R1Psc4GEcvJdsbNbZA3bfztcyDCatJVJpump"
},
{
"date": "2026-10-23",
"title": "KOL tracker shows follower counts without an X API key, and the Oracle reads KOL timelines",
"summary": "A keyless X reader now backs the tracker when the paid API is unset or rate limited, so attached KOLs keep their follower counts. The Oracle also reads tracked timelines every 10 minutes to feed its social virality signal, and each coin's own linked X account age and followers now appear in its intel.",
"tags": [
"feature"
],
"link": "/tracker"
},
{
"date": "2026-10-23",
"title": "Robinhood Chain launches now show every venue, the deployer's track record and a dev-sold flag",
"summary": "The launches board stopped watching NOXA, which shut down in July, and now finds launches on every venue seen on chain. Each row names the launchpad, shows how many earlier tokens the deployer launched and how many it sold out of, and flags launches where the dev has already sold.",
"tags": [
"feature"
],
"link": "/markets/robinhood"
},
{
"date": "2026-10-23",
"title": "Flappin UFO is back at /play/ufo with a verified leaderboard and a ghost of the top run",
"summary": "The arcade game returns as a native canvas game with touch and keyboard controls and levels that get longer and tighter. Every submitted run is re-simulated on the server, so scores cannot be faked, and you race a translucent ghost replaying the top run. Scores also appear on /rankings as the UFO High Score tab.",
"tags": [
"feature"
],
"link": "/play/ufo"
},
{
"date": "2026-10-23",
"title": "Agent memory now learns from trade results, and the decision trail has a verify endpoint",
"summary": "When a snipe closes, its realized P&L, win or loss and hold time are written back onto the memories the agent recalled when it opened the trade. Recall then ranks those memories by a bounded outcome weight (0.65 to 1.35, halving every 45 days, never zero), so lessons from winners rise while memories tied to losses sink but still come back as warnings. Every recall hit now reports its rankScore and an outcome block with the weight, wins, losses and net USD so you can see why it ranked. The hash-chained decision trail can be re-walked at GET /api/agents/:id/decision-trail/verify, which returns ok, the head hash and the zero-based index of the first broken row after any edit. The reasoning ledger page now fetches it and links to it from the verified badge.",
"tags": [
"feature",
"improvement"
],
"link": "/reasoning-ledger"
},
{
"date": "2026-10-23",
"title": "Leaked-wallet rescue runbook with a dry run, and live Jito tip accounts",
"summary": "A new rescue script moves every token and the remaining SOL out of a compromised wallet in one atomic Jito bundle, paid for by a separate safe wallet, and its default dry run simulates the whole bundle and prints the plan without sending anything. The leak alarm now links the runbook. Jito tips also read the block engine's live tip-account list with a pinned fallback, instead of three hardcoded copies.",
"tags": [
"security",
"infra"
]
},
{
"date": "2026-10-23",
"title": "Agents can sign up to services: inbox codes and sign-in links",
"summary": "An agent's own address can now receive verification emails and hand back the one-time code or sign-in link. New inbox_list, inbox_read and inbox_wait_for_code tools read the code with a confidence score, only from services the owner lists under Code senders, only from authentic non-spam mail. Email bodies stay untrusted data and links are never fetched for the agent.",
"tags": [
"feature"
],
"link": "/agent-mail"
},
{
"date": "2026-10-23",
"title": "Chain turnover on /chains",
"summary": "The chains leaderboard now has a sortable Turnover column (24h DEX volume divided by TVL) that shows which chains have capital actively trading rather than parked. Chains with no TVL or no reported volume show n/a.",
"tags": [
"feature"
],
"link": "/chains"
},
{
"date": "2026-10-23",
"title": "Onchain viewer reads multi-part memo inscriptions",
"summary": "Paste any transaction of a first/1, firsts/1 or SPL404 inscription into /onchain and it collects the other parts, checks each part and the whole SHA-256, names missing parts, and shows the file. Images render from verified bytes; SVG and HTML are shown as text and never executed. The Solana Memo Media MCP gains a read_inscription tool that does the same.",
"tags": [
"feature"
],
"link": "/onchain"
},
{
"date": "2026-10-23",
"title": "Prepaid credit plans and free trials for paid MCP tools",
"summary": "Buy a credit bundle or a 24-hour or 30-day time pass at /mcp-plans and call paid MCP tools without paying each call over x402. Every account gets 3 free calls per tool. Credits are debited before a call, refunded when the tool errors, and every result reports creditsUsed and creditsRemaining. A drained plan falls back to the per-call payment-required error.",
"tags": [
"feature"
],
"link": "/mcp-plans"
},
{
"date": "2026-10-23",
"title": "Copy any model as a GitHub README 3D block",
"summary": "Forge results, avatar pages, the avatar dashboard and model pages now have a Copy as GitHub README 3D action. It simplifies the model to a size GitHub renders and puts a stl code fence on your clipboard that GitHub shows as an interactive 3D viewer in any README, issue or pull request. The same converter is a free API (POST /api/readme-3d) and has a live page at /readme-3d.",
"tags": [
"feature"
],
"link": "/readme-3d"
},
{
"date": "2026-10-23",
"title": "AR Studio source now lives in the monorepo",
"summary": "The 3d-ar-studio npm package and the /ar/studio page now run the same code: the package source moved into packages/ar-studio and the page imports its scene math, multi-anchor WebXR session, room client and lighting modules directly, so fixes land once instead of in two diverging copies.",
"tags": [
"sdk",
"improvement"
],
"link": "/ar/studio"
},
{
"date": "2026-10-23",
"title": "Oracle's smart-wallet prior now reads the live database",
"summary": "The cold-start wallet prior behind Oracle's pedigree pillar is loaded from the daily-seeded wallet_reputation table instead of a committed June snapshot, so it tracks the current gmgn.ai rankings. It is cached in-process with a bounded refresh, keeps the last good rows if the database blips, and serves an empty prior (and logs it) rather than inventing wallets.",
"link": "/docs/oracle",
"tags": [
"improvement"
]
},
{
"date": "2026-10-23",
"title": "One catalog for all 39 HookForge hooks, and a census that recognizes them",
"summary": "The Uniswap v4 hooks guide now carries the full HookForge registry (39 hooks, pinned by commit) grouped by family, with the agent-native ones linked to their Solana counterparts, plus a design note reconciling AgentTierHook with reputation-fee. The hook census now decodes each hook's 14 permission bits from its address and calls hookName() to flag HookForge hooks when they show up on chain.",
"link": "/docs/uniswap-v4-hooks",
"tags": [
"docs"
]
},
{
"date": "2026-10-23",
"title": "Turn a Solana program's Anchor IDL into an agent skill",
"summary": "POST /api/skills/import now accepts idl_url or program_id. Each account type becomes a read tool that decodes it by address, each instruction a preview tool that builds the unsigned instruction with a plain summary, and a destructive send tool that hands it to the wallet confirm flow. A risk panel flags token movement, SOL spend, account closure and authority changes. Nothing the skill generates can sign.",
"tags": [
"feature",
"sdk"
],
"link": "/skills/import"
},
{
"date": "2026-10-23",
"title": "llms.txt links the docs as markdown, and install.md tells an agent how to connect",
"summary": "llms.txt now links each curated doc as /docs/<slug>.md, the text/markdown twin an agent can read directly. llms-full.txt carries the full text of those docs (capped at 40 KB each, with a link to the rest) after the page index. A new /install.md, generated from the MCP directory, gives one place for the setup command, per-client settings and every hosted endpoint.",
"tags": [
"docs"
],
"link": "/docs/cli"
},
{
"date": "2026-10-23",
"title": "Splat Viewer: Spark 2.3 level of detail, saved camera hotspots, video export, and image to splat",
"summary": "The Splat Viewer now builds a level-of-detail tree for large scenes, lets you save up to 25 camera views as hotspots, play them as a camera path and export it as a video in the browser. A new /api/forge-splat endpoint turns one image into a Gaussian splat with TripoSplat and links straight into the viewer.",
"tags": [
"feature"
],
"link": "/splat"
},
{
"date": "2026-10-23",
"title": "Agents answer mentions of their own X account, in their own persona",
"summary": "An agent with its own connected X account and mention replies switched on is now polled by the mention bot. It replies in its own persona, attaches a render of its own avatar to conversational replies, and only acts on the intents its owner allowed, up to a daily cap. Dry run stays the default: nothing posts until the owner turns dry run off and the operator enables live mode. A connection missing the posting scope is reported as missing_scope instead of being polled.",
"tags": [
"feature"
],
"link": "/agents"
},
{
"date": "2026-10-15",
"title": "Turn any public OpenAPI spec or GitHub repo into a draft agent skill",
"summary": "New importer at /skills/import (From an API tab) and POST /api/skills/import: point at an OpenAPI/Swagger URL or a GitHub repo and every operation becomes a tools.json entry with a generated handlers.js function calling ctx.fetch, SKILL.md prose written from the spec's own descriptions, and Swagger 2.0 upgraded to OpenAPI 3 automatically. A README-only repo becomes a declarative skill instead, prose only, no tools. Every import lands as a disabled draft, owner-scoped and labelled untrusted: enabled only ever flips via the dedicated PATCH /api/skills/import/:id route, never at creation, and nothing is installed onto any agent automatically. Generated handlers fill auth from the skill's declared secrets, never a hardcoded credential. Also available from the CLI: three-ws skills import <url> and three-ws skills drafts. New package: @three-ws/skill-import.",
"tags": [
"feature"
],
"link": "/skills/import"
},
{
"date": "2026-10-15",
"title": "Agent-to-agent hires now carry a signed fulfillment record, and it feeds reputation and pricing",
"summary": "Every settled a2a-hire (api/agents/a2a-hire.js) now writes exactly one signed attestation row: {hire_id, payment_signature, outcome, evidence_hash, signed_by}, outcome one of delivered, failed, or disputed, signed off-chain with the same platform attestation key api/_lib/attest-event.js uses for on-chain provenance memos. A client-side timeout is recorded as disputed, never inferred as delivered, since x402's verify-then-settle means the provider may already have been paid server-side. The resulting recency-weighted fulfillment rate (30-day half-life, minimum 5 attestations before it counts) is now a real input to the free Financial reputation score on every agent's Trust tab: the Settlement reliability pillar's detail line reports the rate alongside its sample size, for example '90% hire fulfillment over 12 hires', or how many hires are attested so far if the minimum has not been met yet. Providers can also opt a service into an optional x402 pricing discount for well-reputed hirers (agent_paid_services.reputation_discount_enabled, off by default): the hiring agent's reputation tier mints a short-lived, slug-scoped, signature-verified voucher, never a caller-supplied claim, so the paywall always fails closed to full price on anything unsigned, expired, or mis-scoped.",
"tags": [
"feature",
"security"
],
"link": "/agents"
},
{
"date": "2026-10-15",
"title": "Receipt Vault gains an Outbound tab: what your agents spent elsewhere, verified against the chain itself",
"summary": "The Receipt Vault at /receipts now has two tabs. Inbound is unchanged (what three.ws billed your wallet); the new Outbound tab answers the other question: what did your agents spend at OTHER x402 merchants while running on an Agent Payment Session. Each settled Solana execution is resolved against the settlement transaction's own token-balance change the first time anyone views it (payer, payee, mint, amount, fee, slot, time), independent of whatever the merchant's response claimed, then cached so later views are free; a 'verified' badge marks a row that matched the chain, 'pending' while resolution is still in flight, and 'unverified' if the lookup failed permanently. The tab is account-session gated (the same credential /payments uses, not a wallet signature, since Payment Sessions are account-scoped, not wallet-scoped), filters by network and merchant host, and exports to CSV alongside the existing JSON receipt downloads. New endpoint: GET /api/x402/outbound-receipts.",
"tags": [
"feature"
],
"link": "/receipts"
},
{
"date": "2026-10-15",
"title": "17 named commerce tools and a spending report land in the x402 MCP buyer, hosted and self-custodial",
"summary": "The self-custodial x402 buyer (@three-ws/x402-mcp, npx -y @three-ws/x402-mcp) and the hosted agent wallet (https://three.ws/api/mcp-agent) both gained 17 named tools, one per fixed three.ws x402 service: weather_forecast, weather_go_no_go, carbon_now, carbon_best_window, domain_rdap_check, domain_rdap_check_bulk, research_search, research_paper, research_bibliography, book_search, book_read, news_wire_search, news_wire_poll, transit_plan, transit_delays, places_search and places_detail. Each carries a fixed input schema and states its price in its description, so a model calls it correctly far more often than it composes the equivalent raw pay_and_call(url) call. On the hosted side every named tool has a matching <name>_quote read tool that prices it without paying, and the shared tool-policy runtime (@three-ws/mcp-policy) gates payment on that exact quote (one service's quote can never unlock a different service's payment). The stdio package also gained spending_report, which totals everything pay_and_call and the named tools have paid, by merchant and by day, from a local append-only ledger (X402_MCP_LEDGER_PATH).",
"tags": [
"feature",
"sdk"
],
"link": "/docs/mcp"
},
{
"date": "2026-10-15",
"title": "A generated SKILL-MD 1.0 catalog of every paid HTTP route is now live at /.well-known/skill.md",
"summary": "Any agent can now load one markdown file, https://three.ws/.well-known/skill.md, and get a conformant SKILL-MD 1.0 document (the x402-skill-md format) listing every priced endpoint on three.ws: method, path, price, parameters, and the Solana and Base payment rails it settles on, plus a sample 402 challenge body. The document is generated straight from the same OpenAPI catalog /openapi.json serves (api/_lib/skill-md.js reads the shared buildOpenApiDoc() builder and the real buildExactRequirements() rail logic, so the catalog and the live 402 challenge can never drift apart), and is linked from /.well-known/x402.json and llms.txt. Validated in tests/skill-md.test.js against the real upstream x402-skill-md npm validator: zero errors other than the EVM rail finding (SM005), which is expected since this is a Solana-first platform and Base settles only when a facilitator is configured. Distinct from the pre-existing https://three.ws/skill.md (no .well-known), the Agent Skills connector file for installing three.ws MCP servers; both now cross-link to the other.",
"tags": [
"feature",
"docs"
],
"link": "/.well-known/skill.md"
},
{
"date": "2026-10-15",
"title": "A spend call above an agent's approval threshold now pauses for the owner instead of failing",
"summary": "The requires_approval verdict agent payment sessions and agent-to-agent hires can already produce now actually does something: both /api/pay/execute and /api/agents/a2a-hire create a request on the existing approval inbox the moment a call crosses the caller's approval_threshold_usd, answer with 202 and a poll link to /approvals/<id> instead of failing the call, and resume automatically the moment the owner taps Approve there, re-probing or re-running the exact same reserve-then-settle path the call would have taken inline, with no re-entry point for a double-spend. A denied request answers 403 on the next poll, an expired one answers 410, and retrying with the same idempotency key before the owner answers just re-polls instead of parking a second request. Nothing new to build on the approval page itself: it already renders any venue generically from the request's summary, amount, and risk notes, so the two new venues (payment_session_execute, a2a_hire) show up with the right Approve/Decline card for free. Covered by 21 new handler tests across create, approve, decline, expire, and the executor's own idempotent re-run.",
"tags": [
"feature",
"security"
],
"link": "/approvals"
},
{
"date": "2026-10-15",
"title": "Payment sessions gain blocked hosts, an approval threshold, and a per-rail daily cap, all with tamper-evident verdicts",
"summary": "Agent payment sessions get three new governance knobs, settable at creation and editable while a session is active: blocked_hosts (a denylist checked after the allowlist, so an explicit block always wins over a broad allow, even with no allowlist set), approval_threshold_usd (a call priced above this amount now gets a distinct requires_approval verdict instead of either settling or silently failing), and per_rail_daily_usd (a rolling 24h spend cap per settlement network, reserved atomically the same way the session budget already is, closing the race a naive sum-of-past-executions check would leave open). Every governance verdict, allow or reject, now carries a policy_digest (a sha256 hash of the exact policy it was judged against) and an HMAC signature over the verdict, digest, request, and timestamp, stored with the execution: editing a session's policy after the fact changes the digest and breaks the signature on every verdict that preceded the edit, so a verdict record can be trusted even after the policy it was judged against has since moved. The dry-run simulator at /api/pay/simulate now predicts all three new rules through the exact same shared policy module the real enforcer calls, so a simulated verdict and an enforced one can never disagree; that parity is pinned by a new test running both over identical scenarios. The payments page gained matching fields in both the session creation and edit modals, with inline validation and a collapsed 'Advanced spend policy' section so the common case (budget, expiry, host allowlist) stays uncluttered.",
"tags": [
"feature",
"security"
],
"link": "/docs/payment-sessions"
},
{
"date": "2026-10-14",
"title": "x402 agent buyers now pay v1 sellers, not just v2",
"summary": "Every buyer path, the hosted a2a-hire/pay_and_call flow, the packages/x402-mcp stdio server, and the mcp-bridge stdio client, registered only x402's newer CAIP-2 wire format (network strings like solana:5eykt4UsFv8P8NJdTREpY1vzqKqZKvdp). Most of the open x402 bazaar, and every service in the owner's x402-suite, still speaks v1 (flat network strings like solana, price field maxAmountRequired), so agents paying through three.ws could not reach those sellers at all. All three buyers now call registerExactSvmScheme() instead, which registers both wire formats on the same Solana key; when a seller advertises both, v2 is always preferred. Settlement receipts already share field names (success, transaction, network, payer) across both versions, so /receipts and the session ledger need no changes. Fixing this also surfaced two real spend-cap bugs that only trigger on a v1 payment: the hosted buyer's enforceCap threw instead of admitting a capped v1 payment, and mcp-bridge's spending-cap hook silently read a v1 payment's price as zero, bypassing both the per-call and session ceilings. Both are fixed and covered by regression tests.",
"tags": [
"feature",
"security"
],
"link": "/docs/mcp-agent"
},
{
"date": "2026-10-14",
"title": "8 keyless live-data services join the x402 paid API: weather, grid carbon, domains, research, books, news, transit, places",
"summary": "three.ws's x402 surface grows 17 new routes, each needing no API key anywhere in the chain (not an upstream key, not a three.ws key), ported natively onto three.ws's own Solana facilitator so they settle, receipt, and discover exactly like every other paid endpoint: Weather Guard (24h forecast + severe alerts from Open-Meteo, NWS fallback), Carbon (UK grid carbon intensity + lowest-carbon scheduling window), Domains (live RDAP registration lookups, single and bulk), Research (paper search across arXiv/Crossref/Semantic Scholar, metadata, bibliographies), Books (OpenLibrary + Project Gutenberg search and full-text reads), News Wire (global news search and polling over GDELT DOC 2.0), Transit (MBTA GTFS trip planning and realtime delay snapshots), and Places (OpenStreetMap Overpass point-of-interest search and detail). Each service's discovery route runs a free daily per-IP quota before falling through to the x402 pay-per-call lane, fails closed to the paid rail rather than opening an unmetered proxy if the quota store is unreachable, and every upstream call sends an identifying User-Agent. No fixture or mocked fallback anywhere in the set: an upstream outage answers 502, and the one route whose source suite offers a fake-data fallback when unconfigured (Places Detail's review enrichment) ships without that fallback, returning only the real OpenStreetMap record. All 17 routes appear in GET /openapi.json and GET /.well-known/x402.json with Solana listed first in every payment challenge.",
"tags": [
"feature"
],
"link": "/docs/x402-endpoints"
},
{
"date": "2026-10-14",
"title": "Agents can now carry an owner-signed, offline-verifiable spend cap",
"summary": "A new Delegation tab in the agent wallet hub lets an owner sign a portable spend delegation with their own linked Solana wallet (a message, never a transaction): 'this agent may move at most CAP lamports per EPOCH, until NOT_AFTER.' The signed envelope is plain text and the verification math is a zero-dependency function any outside counterparty can run themselves (packages/agent-guards, spec in its README) to confirm the authorization without trusting three.ws's say-so. three.ws enforces the same cap server-side on every call through POST /api/agent/guard, measured from the agent wallet's REALIZED spend in the custody ledger rather than the agent's own claim, so an exact-output trade that ends up costing more than advertised still can't slip past the ceiling. A replayed nonce, an expired delegation, or a signature from the wrong wallet is rejected with a specific reason.",
"tags": [
"feature",
"security"
],
"link": "/docs/agent-wallet-api"
},
{
"date": "2026-10-14",
"title": "The pre-push guard now catches a leaked Solana secret key, and withdrawals warn before a sweeper bot drains them",
"summary": "scripts/check-secrets.mjs, the hook that blocks a push carrying a credential, now decodes every 86-90 character base58 run in added content and flags it only when the first 32 bytes derive the embedded public key in the last 32 bytes via Ed25519, catching a real Solana keypair secret while leaving the repo's thousands of genuine 64-byte transaction signatures untouched. Separately, the agent wallet's Withdraw tab now reads a destination's recent on-chain history before you send: a wallet that repeatedly drains to near-zero within seconds of receiving funds, the sweeper-bot fingerprint, surfaces a blocking warning on the confirm screen and again server-side if you retry, so a sweeper bot can't just wait for the preview to go stale.",
"tags": [
"security"
],
"link": "/docs/agent-abilities"
},
{
"date": "2026-10-14",
"title": "A new wallet linkage score flags a coin's creator buying their own launch",
"summary": "GET /api/crypto/linkage cross-checks 2-10 Solana addresses for shared control: a 0-100 same-controller score per pair, combining six real on-chain signals by noisy-OR (shared fee payer, one address funding the other, both sharing a first funder, SPL token-account control, direct transfers, a shared funder sending similar amounts in a short window), every match backed by the signature or account that triggered it. It is keyless, free, and part of the public Crypto Data API, and now wired into two places that act on money: the Trade Safety Firewall warns when an early buyer scores as the same controller as a coin's creator (wash-trading dressed up as demand), and copy-trading refuses to size an order behind a leader whose own wallet links to the creator of the coin they are 'signaling', showing the follower exactly why. The same scorer ships as the wallet_linkage tool in the hosted intel MCP server for any AI agent to call directly.",
"tags": [
"feature",
"security"
],
"link": "/crypto-api"
},
{
"date": "2026-10-14",
"title": "The native $THREE launchpad now taxes snipers, not just steady trading",
"summary": "Every pool on the native launchpad lane used to open at the same flat 1% trading fee a coin keeps forever, which is exactly what a sniper bot wants: the same price whether it buys in the pool's first block or an hour later. New pools now open at a 50% fee that decays exponentially down to the normal 1% over the first 100 slots (about 40 seconds), so a bot buying in the first block pays close to half its $THREE in fees while an organic buyer a minute later pays the standard rate. The coin creator's own bundled first buy stays exempt, since that is the creator funding their own launch, not a snipe. The schedule runs entirely on Meteora's own Dynamic Bonding Curve fee scheduler, no new contract.",
"tags": [
"feature",
"security"
],
"link": "/docs/native-launchpad"
},
{
"date": "2026-10-14",
"title": "MCP clients can now prepare a pump.fun launch for you to sign",
"summary": "The hosted MCP server gained pumpfun_prepare_launch: an assistant (Claude Desktop, Cursor, or any MCP client) that has agreed a coin name, ticker, description and image with you can call this tool to validate every field the same way the /launch page does, fetch and type/size-check the image through the SSRF-hardened fetcher, and hand back a pre-filled /launch URL plus a plain-English summary. The tool never signs or launches anything; it is marked read-only and non-destructive, and its response always carries requires_human_signature: true. Opening the returned link on /launch now shows a 'prepared from your assistant' banner so you know exactly which fields came from the conversation before you review and sign.",
"tags": [
"feature"
],
"link": "/launch"
},
{
"date": "2026-10-14",
"title": "Creator earnings now cross-check two independent pump.fun sources",
"summary": "A coin's earnings figure on /launches/<mint> and in the agent earnings card used to come from one source only: the creator wallet's lifetime fee index. It now also sums that specific coin's own pump.fun distribution timeline, the only feed that can say which coin actually paid a wallet, and shows both: 'this coin paid X SOL' next to the wallet-level figure, plus a plain sentence when the two sources disagree and by how much. Nothing is hidden when sources drift apart; the gap is the point. The snapshot pipeline that already indexed wallet-level fees now reconciles against the timeline for every coin on each refresh.",
"tags": [
"feature"
],
"link": "/launches"
},
{
"date": "2026-10-14",
"title": "Every pump.fun launch now carries a plain-English receipt",
"summary": "/launches/<mint> gained a Launch receipt panel composing five on-chain facts into one readable object: whether the mint or freeze authority can still act (with a plain sentence for each, not just enabled/disabled), supply and decimals, the top-10 holder share, where trading fees route (a creator wallet, a fee-sharing config, cashback, or holder rewards), the creation transaction's version and byte size, and the Trade Safety Firewall's verdict. A coin that never launched through pump.fun gets an honest 'not a pump.fun launch' instead of an error, and a source that fails marks only its own section unavailable, never an invented number. Copy the receipt as text or download it as JSON; both carry a SHA-256 content hash of the response so a saved copy can be checked against a fresh read. A shareable 1200x630 card is available via GET /api/pump/launch-og?mint=<mint>&variant=receipt, and the composition itself is a new public endpoint, GET /api/pump/launch-receipt?mint=<mint>.",
"tags": [
"feature"
],
"link": "/launches"
},
{
"date": "2026-10-14",
"title": "Atomic is now a six-tab Transaction V1 lab, not just an inspector",
"summary": "/atomic grew five tabs beyond the Inspector, each reachable by click, arrow-key/Home/End tab navigation, or a deep link (?tab=mask, ?tab=sponsor, ?tab=planner, ?tab=diff, ?tab=wallet). The config mask decoder reads a V1 transaction's real TransactionConfigMask u32 and its fixed-width ConfigValues byte by byte, flagging a half-set priority-fee pair or a reserved bit SIMD-0385 has not assigned yet. The sponsor cap view shows a paymaster exactly what it is agreeing to pay, reading V1's inline config instead of a Compute Budget instruction scan that sees nothing on a V1 transaction. The size and budget planner compiles a pasted instruction set with @solana/kit to an exact byte count against legacy, V0, or V1's ceiling, including an exact lamport total for a chosen priority fee. The legacy/V0/V1 diff compiles the same instructions three ways and reports the real byte deltas, honestly, including the case where V1's inline config makes it larger than legacy. The wallet V1 capability check connects the browser's injected wallet and reports the same verdict the rest of three.ws would get from findV1WalletStandardSigner before ever requesting a real V1 signature. Nothing here is estimated: every number comes from compiling and decoding real transactions, and nothing is ever signed, simulated, or broadcast.",
"tags": [
"feature"
],
"link": "/atomic"
},
{
"date": "2026-10-14",
"title": "Atomic now replays a pump.fun transaction step by step",
"summary": "Atomic (/atomic) decodes every Pump, PumpSwap, and PumpFees event in a transaction's logs, oldest first, and renders it as a Replay panel: a bonding-curve-position bar, a human-readable description of each step (create, buy/sell, graduation, creator-fee collection and distribution, GitHub social-fee claims, and post-graduation PumpSwap swaps), and the SOL/token amounts that moved. An event the installed SDK does not recognize still renders, labeled unrecognized, instead of silently disappearing. For a create transaction, the panel also pages through the earliest trades on that mint since launch, walking Solana's signature history backward in bounded pages since RPC has no forward cursor.",
"tags": [
"feature"
]
},
{
"date": "2026-10-14",
"title": "GitHub fee-claim feed now reads from a real-time listener, not just an on-demand RPC scan",
"summary": "The claims feed's GitHub-attribution path used to walk Solana RPC history fresh on every request, the same cost every time, with nothing persisted in between. services/pump-claims is a standalone process that subscribes to PumpFees program logs directly, decodes and attributes each claim as it happens, and writes it to a new pump_fee_claims table with a watchdog that resubscribes on a stalled connection and gap-fills whatever a drop missed. The claims route now reads that table first and only falls back to the RPC scan when the table has nothing for the requested window, so PUMPFUN_BOT_URL is no longer required for GitHub claims specifically (ordinary wallet claims still use it).",
"tags": [
"feature",
"infra"
]
},
{
"date": "2026-10-14",
"title": "GitHub fee-claim channel now attributes withdrawals to a coin, or says plainly it cannot",
"summary": "A GitHub identity withdrawing pump.fun fees through the claim_social_fee_pda instruction only proves payment to that identity, not which coin paid it: one GitHub fee account can be a shareholder on more than one coin, and its address can be pasted into a copycat coin's metadata. The claims feed and its Telegram channel now classify every such withdrawal with on-chain evidence: a same-transaction distribution to the exact account, or a 90%-dominant distribution history since the previous claim, then a repository-owner or creator-wallet match, down to an honest unresolved or unverified label when nothing proves a coin. A new-GitHub-account-plus-copycat-coin pattern holds the claim for review instead of posting it with a celebratory card.",
"tags": [
"feature",
"security"
]
},
{
"date": "2026-10-14",
"title": "Robinhood Chain feed worker now runs the published SDK, not a stale local snapshot",
"summary": "workers/robinhood-feed depended on robinhood/robinhood-chain-sdk through a file: link, a reference copy that drifts from the standalone robinhood-chain-sdk repo it mirrors. The worker now installs hoodchain from npm like any other dependency, so it always runs the released SDK. Also fixed: seven pointer READMEs staged for duplicate or superseded standalone repos (three-ui, crypto-market-data, crypto-market-data-ts, crypto-data-aggregator, memescope-monday-directory, and one of each duplicate Robinhood Chain repo pair), ready for the owner to push and archive.",
"tags": [
"fix",
"infra"
]
},
{
"date": "2026-10-14",
"title": "Deep links to a specific avatar now actually load that avatar",
"summary": "@three-ws/walk's ?walk=play&avatar=<id> deep link and the drop-in resume path read the saved avatar preference straight out of storage and never looked at the query string, so sharing a link to a specific avatar silently loaded the stored (or default) one instead. A new deepLinkAvatarId resolver gives the query param precedence and persists it, the same way picking an avatar in the UI does.",
"tags": [
"fix",
"sdk"
]
},
{
"date": "2026-10-14",
"title": "three.ws Avatar MCP clamps unbounded tool arguments before they reach generated HTML",
"summary": "render_avatar's name and camera-orbit arguments flowed into the generated <model-viewer> HTML with no length bound, unlike the model URL, which already rejected oversized or unsafe values. Both are now clamped the same way, closing the one argument path that could still balloon a rendered page or chat payload.",
"tags": [
"fix",
"security",
"sdk"
]
},
{
"date": "2026-10-14",
"title": "A read-only drift report for every standalone GitHub mirror",
"link": "/docs/contributing",
"summary": "npm run mirrors:drift compares each of the ~105 standalone mirrors in data/standalone-mirrors.json against its monorepo source at the last synced commit (GitHub API, with a README-timestamp fallback when rate-limited), flags any package whose GitHub copy has diverged, and writes the result to tasks/mirror-drift-<date>.json. It never pushes or mutates a mirror; it is the audit step that found the two fixes above.",
"tags": [
"infra",
"docs"
]
},
{
"date": "2026-10-14",
"title": "Standalone GitHub mirrors now stage instead of force-pushing",
"link": "/docs/contributing",
"summary": "scripts/sync-standalone-repos.mjs force-pushed straight to main on every run, which could silently create duplicate repos for the eleven mirrors already renamed on GitHub and bypassed the owner's push-approval rule. data/standalone-mirrors.json is now the single source of truth for which monorepo directory maps to which public repo, read by every exporter (sync, growth satellites, the awesome list, the skill-repo builder) instead of each keeping its own list. The sync script only stages a snapshot under dist/standalone-mirrors/, fixes any stale pre-Apache-2.0 README or LICENSE footer, and prints the push command for the owner; npm run sync:repos:check compares what is staged against each mirror's live tree and reports drift without writing anything.",
"tags": [
"infra",
"docs"
]
},
{
"date": "2026-10-13",
"title": "Free voice is back: the Edge lane speaks again",
"link": "/voice",
"summary": "Microsoft changed the handshake on its free Read Aloud service, so the default Edge voice lane answered every request with an error. The lane now speaks the current protocol, so the /voice page and every agent that uses the free voice can talk again.",
"tags": [
"fix"
]
},
{
"date": "2026-10-13",
"title": "Avatar downloads fail over when object storage is not entitled",
"summary": "When the Cloudflare R2 account stopped serving objects, avatar model downloads answered a bare error instead of trying the bucket public domain. That fault is now recognised as a storage outage, so the failover runs.",
"tags": [
"fix"
]
},
{
"date": "2026-10-13",
"title": "The Home Assistant integration is ready for the HACS default store",
"link": "/docs/home-relay",
"summary": "home-assistant-integration/ now validates clean against the real hassfest and HACS checks: manifest.json's documentation link fixed (it 404'd), version bumped to 1.0.1, and hacs.json's minimum Home Assistant version corrected to 2024.12.0 (the first release bundling the aiohttp.ClientWSTimeout the integration's relay client uses, confirmed against HA's own package_constraints.txt). It now ships its own local brand icons at the sizes both HACS and home-assistant/brands require. A new scripts/export-hacs-repo.mjs builds the exact standalone-repository tree (integration, README, LICENSE, hacs.json, and the two validation workflows, which exist only in that exported tree since this monorepo runs no GitHub Actions of its own) ready to push to a new repository. Verified end to end against a real Home Assistant container through its real config flow and a real services/home-relay server: 18 real entities, a confirmed auto-reconnect after a relay restart, and a documentation fix to the connectivity sensor's entity ID along the way. Creating the standalone repository and opening the hacs/default and home-assistant/brands pull requests are owner actions; submission text is ready in marketing/growth/submissions/home-assistant.md.",
"tags": [
"improvement",
"docs"
]
},
{
"date": "2026-10-13",
"title": "Embed a live 3D agent in WordPress, Framer, or Webflow",
"link": "/docs/wordpress-embed",
"summary": "Three native embeds for <agent-3d>, one per site builder's own extension point instead of a generic iframe: a WordPress Gutenberg block and [three_ws_agent] shortcode (server-rendered so the editor preview always matches the front end), a Framer code component with property controls, and a Webflow Code Component, all exposing On-chain URI, Agent ID, Model URL, Background, Framing, Poster and Autoplay. A shared build step pins all three to the same version and SRI hash from the agent-3d release ledger, so they can never drift apart. The WordPress plugin passed the official Plugin Check tool against a real local WordPress install with no errors; Framer and Webflow were each verified in a real-React test harness that boots a live WebGL canvas against the production CDN release, and the Webflow component additionally compiles through the real Webflow CLI. WordPress.org, Framer Marketplace, and Webflow Apps submissions are owner actions. Docs: /docs/wordpress-embed, /docs/embedding.",
"tags": [
"feature",
"docs"
]
},
{
"date": "2026-10-13",
"title": "Import a VRoid Hub avatar with its creator's license honored",
"link": "/docs/vroid-hub-import",
"summary": "A new picker at /import/vroid lists every character you own or hearted on VRoid Hub, each scored against the real, live license conditions its creator set (public agent, export, modify, monetize), so three.ws only ever offers an action the creator actually permitted. Built end to end against VRoid Hub's live documented API (OAuth2 with mandatory PKCE, the two-step download-license flow), including encrypted token storage alongside three.ws's other social connections, through to a real import that lands the avatar in the editor with idle/walk already retargeted. VRoid Hub app registration and the OAuth client credentials are an owner action tracked in prompts/finish/_context/production-100-OWNER-ACTIONS.md; everything else is wired and tested. Docs: /docs/vroid-hub-import.",
"tags": [
"feature",
"docs"
]
},
{
"date": "2026-10-13",
"title": "A Shopify app turns product photos into 3D product media",
"link": "/docs/shopify-app",
"summary": "A new embedded Shopify app, built end to end but not yet installable (Partner account, env vars, and App Store submission are owner actions): pick a product photo in the admin, generate a real GLB through three.ws's own /api/forge pipeline, preview it in-page, and attach it with productCreateMedia so the storefront shows a 3D viewer on desktop and AR Quick Look on iPhone from the one upload (Shopify derives the USDZ automatically). Bulk mode generates across a whole catalog with real per-item status. Billing runs through Shopify's own usage-based Billing API, priced from Forge's own per-call cost, in test mode until the app has Public distribution. The three mandatory privacy webhooks plus app/uninstalled are HMAC-verified and delivery-deduplicated. Docs: /docs/shopify-app. Submission packet for Partner Dashboard setup and App Store listing copy: marketing/growth/submissions/shopify-app.md.",
"tags": [
"feature",
"docs"
]
},
{
"date": "2026-10-13",
"title": "Video to Motion: download any video's motion as a GLB, BVH, or FBX",
"link": "/video-to-motion",
"summary": "A new page for creators who want the animation file itself, not a composited video: upload a video, preview the captured motion on a stock avatar or any rigged GLB you drop in, trim the in and out points, and download a GLB (clip baked onto the previewed avatar), a BVH (the platform's own canonical-skeleton proportions, the standard shape to re-retarget in a DCC tool), or an FBX (server-side Blender conversion that preserves the rig). Reuses the production model-video2motion worker that already powers /motion-swap for capture; signed-in users can save the clip straight to an agent's animation library. New writer: src/animation-bvh-export.js, round-tripped through three.js's own BVHLoader in tests/animation-bvh-export.test.js. Cross-linked from /motion-swap, /animations, and /mocap-studio. Docs: /docs/video-to-motion.",
"tags": [
"feature",
"docs"
]
},
{
"date": "2026-10-13",
"title": "Get it printed: live Craftcloud prices for any Forge model, buyer pays direct",
"link": "/docs/forge#get-it-printed-live-prices-from-craftcloud",
"summary": "A new \"Get it printed\" action on model pages and in the Forge export menu uploads the model straight to Craftcloud (All3DP) and shows live, real prices across whichever of its roughly 186 print shops can run the job, cheapest first with shipping attached. This is not Materialize: three.ws never takes payment here. Craftcloud's public v5 API has no endpoint that hands a server-created cart back to a browser for checkout, so the panel links out to Craftcloud's own upload page instead of faking a continue-to-checkout button, where pasting the same STL reproduces the prices in an authenticated, checkout-capable session. Backed by a new fulfillment adapter, api/_lib/print/adapters/craftcloud.js, registered ahead of the manual adapter since quoting never spends. Docs: /docs/forge#get-it-printed-live-prices-from-craftcloud and /docs/materialize.",
"tags": [
"feature",
"docs"
]
},
{
"date": "2026-10-13",
"title": "A Discord Activity brings a shared Forge room into any voice channel",
"link": "/docs/discord-activity",
"summary": "Open the new Activity from a Discord voice channel and everyone in the call shares one Forge room: one prompt, one in-progress generation, one finished model with a shared vote count, and an agent that reacts in text to what just happened. Auth maps a verified Discord identity (re-checked against Discord's own /users/@me, never trusted from the client) to a room-scoped session; generation and voting reuse the real /api/forge pipeline and forge_votes table rather than a separate throwaway pipeline. No coin, wallet, or purchase path. Docs: /docs/discord-activity. Submission packet for Developer Portal setup and Discovery copy: marketing/growth/submissions/discord-activity.md.",
"tags": [
"feature",
"docs"
]
},
{
"date": "2026-10-13",
"title": "A three.ws avatar now lipsyncs to any LiveKit Agents or Pipecat voice pipeline",
"link": "/docs/voice-framework-integrations",
"summary": "Two new packages, livekit-plugins-threews and pipecat-threews, republish a running voice agent's TTS and user-turn frames as the same transcript/gesture/speech_started/speech_ended message shape over the room's data channel, which <agent-3d voice=\"livekit\"> already renders client-side: no server-side video rendering, no change to the browser code at all. The LiveKit package is an AvatarSession plugin dropped into an AgentSession; the Pipecat package is a FrameProcessor placed after the TTS service. Each ships a runnable local example against a real open-source livekit-server with a real local TTS, verified end to end with real browser automation driving the actual avatar lipsync and gesture cues off a live bot in both frameworks. Docs: /docs/voice-framework-integrations.",
"tags": [
"feature",
"sdk",
"docs"
]
},
{
"date": "2026-10-13",
"title": "Open source credits: every dependency, model, and asset source three.ws runs on, in one place",
"link": "/open-source",
"summary": "A new /open-source page lists every third-party dependency, GPU worker model, and free asset source the platform runs on, grouped by area (rendering, formats and compression, generation models, motion, voice, assets, infrastructure) with license, upstream link, and what each one powers here. The dataset is generated from package.json and each worker's own README, not hand-maintained, and a guard test fails the build if a new direct dependency lands without a credited entry. Credits now also show up where the work actually happens: /objects and the object viewer name Poly Haven, the CC0 asset library the catalog sources from; the Forge export dialog names meshoptimizer and glTF-Transform; and the animation gallery shows each clip's source and license, including a fallback for studio-tier clips that lack an explicit manifest entry. Docs: /docs/open-source-credits.",
"tags": [
"feature",
"docs"
]
},
{
"date": "2026-10-13",
"title": "The public animation library is now openly licensed, every clip CC0 or self-generated",
"link": "/docs/animations",
"summary": "GET /api/animations/library used to serve a bulk motion library baked from Mixamo, redistributable by anyone through scripts/mirror-animation-library.mjs even though Mixamo's own terms permit using its animations inside a project but not redistributing the raw files. The library is rebuilt from two CC0 1.0 Universal sources instead (Quaternius's Universal Animation Library 2 and Mesh2Motion's human animation packs), retargeted through the existing pipeline with every clip carrying a source and license field end to end from scripts/cc0-library.config.json through the published manifest. The /animations gallery shows each clip's pack and license next to its category chip, and the endpoint itself now refuses to publish any entry missing a license or naming Mixamo as its source, so a raw Mixamo clip can never reach the public endpoint or its mirror again. scripts/mixamo-all.mjs keeps its download and local-build phases for offline retargeting research, but its upload phase is permanently disabled.",
"tags": [
"improvement",
"docs"
]
},
{
"date": "2026-10-12",
"title": "Dissolve: turn any 3D model into a GPU particle storm",
"summary": "A new page that dissolves any GLB into up to a million GPU particles, morphs one shape into another, and lets you bend the storm with a black hole, a vortex, shockwaves and a collapse. Turn on the microphone and bass drives the storm. Pick a look, record a clip, save a PNG or share the exact pair as a link.",
"tags": [
"feature"
],
"link": "/dissolve"
},
{
"date": "2026-10-12",
"title": "Accessibility pass on the gallery, create page and site header",
"summary": "Icon-only buttons in the header, footer and suggestions panel now show a real tooltip on hover and keyboard focus. Gallery tags, avatar cards and the create page cards are announced with the text you can see, header buttons have more breathing room, and the page background no longer flashes transparent when you overscroll.",
"tags": [
"improvement"
],
"link": "/gallery"
},
{
"date": "2026-10-12",
"title": "An in-depth X Article on how three.ws renders 3D, and a playbook for answering basic questions at length",
"link": "/viewer",
"summary": "A reader asked how three.ws renders 3D. The answer is now a full draft Article: one rigged character followed through making, packing, loading, lighting and animating, with runnable embed code, decoder and boot-phase tables, the free Rig Doctor, validator and viewer tools, every partner by exact designation, and the roadmap. Every claim is tied to evidence in the repo. A new playbook describes how to turn any basic question into a long reply thread plus an Article.",
"tags": [
"feature",
"docs"
]
},
{
"date": "2026-10-12",
"title": "The Grid: walk a neon room where every app is a live surface",
"link": "/?world=grid",
"summary": "A new place beside the campus. Step into a wireframe room with a live board of new agents, launches, wallet moves and the $THREE quote, a directory wall, a Pulse wall and one kiosk per app. Walk up and press E to open the app on the desktop. Your byte tier, from Byte to Yottabyte, follows the gems you mine in the campus. Open it from the Grid button in the campus HUD.",
"tags": [
"feature"
]
},
{
"date": "2026-10-12",
"title": "Forge generation controls (target faces, bbox fit, rest pose) and a batch endpoint for fanning out N generations",
"link": "/docs/forge",
"summary": "POST /api/forge and /api/x402/forge take three new, off-by-default controls: target_faces (alias of target_polycount) pins an explicit geometry budget on poly-aware backends; bbox: [x, y, z] uniformly rescales the delivered mesh to fit a box in meters, lane-agnostic since it runs as a post-generation GLB fit; pose: \"t\" | \"a\" requests a rest pose on backends without a caller-supplied reference image, with pose_applied echoed back so a photo-conditioned lane never silently ignores it. A new POST /api/forge-batch fans one call out to up to 8 independent /api/forge submissions, concurrency-capped by the caller's $THREE holder tier, and GET /api/forge-batch?batch=<id> polls the group: one item's failure never fails the batch, and group state is best-effort Redis so a job already returned stays pollable at /api/forge?job= even if the group expires. The Workbench's Generate Mesh node now exposes rest pose and width/height/depth bbox fields alongside the existing face budget, and a new Batch drawer (toolbar icon, or press B) queues up to 8 prompts at once against the new endpoint and opens any finished item straight into the viewport.",
"tags": [
"feature",
"improvement"
]
},
{
"date": "2026-10-12",
"title": "Tripo house lane: v3 API, house key, signed webhooks, and engine-version attribution",
"link": "/docs/forge",
"summary": "The Tripo integration moved off the legacy v2 API onto v3 (openapi.tripo3d.com/v3): dedicated endpoints per capability, a unified input field, and a unified error envelope. Setting TRIPO_API_KEY turns Tripo from a bring-your-own-key engine into a High-tier house default billed to $THREE credits, with the BYOK path and the engine list unchanged when the variable is unset. A finished job can complete by poll or by Tripo's own signed webhook (api/forge/webhook/tripo.js), which verifies Tripo-Webhook-Signature and claims Tripo-Webhook-Delivery so a retried delivery cannot double-complete a job; polling stays the fallback. Every Tripo result, BYOK or house key, now names the exact model version it ran on in the submit, poll, and gallery responses and in the exported GLB's own metadata, and a new npm run forge:tripo reports the lane's own completion rate and median latency by day.",
"tags": [
"feature",
"infra"
]
},
{
"date": "2026-10-12",
"title": "Companion presence: sustained mood, emoji-driven expression, an LLM gesture tool, active-listening nods, and real multi-language lip-sync",
"link": "/docs/agent-system",
"summary": "Agent avatars carry more of what makes a companion feel present. setMood(valence, arousal) sets a sustained resting bias (a soft smile or a worried brow that drifts in over about a second) on top of the existing per-utterance emotion blend, and a reply that is nothing but emoji (no English sentiment words at all) now drives a real expression change instead of staying neutral, because the built-in sentiment scorer reads emoji as direct valence/arousal signal. The LLM brain gets a general-purpose `gesture` tool covering all 19 body-language slots (nod, shake, shrug, bow, point, think, and the rest), so it can react physically mid-turn instead of only through a skill or an emotion threshold. While a real VAD/STT signal says the user is talking, the avatar nods on a fixed cadence to show it is listening, and stops the moment they stop. ElevenLabs text-to-speech now drives lip-sync from the provider's real per-character timestamps (any language the voice speaks), falling back to the energy-based analyser only if that call fails, so mouth shapes track what is actually being said instead of approximating it from amplitude alone.",
"tags": [
"feature",
"improvement"
]
},
{
"date": "2026-10-12",
"title": "Scene Studio: an Agent tab that edits your open scene by prompt, plus a TSL node material editor",
"link": "/scene",
"summary": "Scene Studio's sidebar grows two tabs. Agent is a chat panel that reads the open scene and drives it with the same real editor.execute(Command) calls the toolbar and menus use (free, keyless, rate-limited model via POST /api/scene/agent): add a primitive or a Forge-generated object, move/rotate/scale a named object, recolor a material. Every mutating prompt pushes a numbered version (a scene snapshot plus a viewport screenshot); clicking an earlier version restores it as one more undoable edit, so a bad restore is one Ctrl+Z away like any other mistake. Nodes is a dropdown-wired TSL (Three.js Shading Language) material graph editor for the selected object: build a graph from UV, time, color, texture, and math nodes as a reorderable card list, no canvas or wires to route, and apply it to swap in a real node material through the same undo-able command path. Fixed along the way: restoring a version used to append the snapshot's objects on top of the current scene instead of replacing them (SetSceneCommand, the vendored editor's own scene-import command, only ever adds); and the site-wide corner-stack widget could sit on top of the Agent tab's version-restore buttons and eat the click, now suppressed while the studio is open.",
"tags": [
"feature",
"fix"
]
},
{
"date": "2026-10-12",
"title": "Every generated and purchased model now ships with a signed provenance credential, free, by default",
"link": "/docs/provenance",
"summary": "Content credentials (specs/PROVENANCE_3D.md) used to be an on-demand, Claude-track action: you had to call anchor_provenance to get one. Now every GLB the Forge produces, across every generation lane (free NIM, Forge Pro, rigged avatars, self-host TRELLIS/Hunyuan, async GCP workers), gets an ed25519-signed content credential the moment it's materialized, and every first download of a paid asset from the x402 asset bazaar gets one too, with the hash and a verify link returned right in the download response. Signing and storage are free and local, no Solana write and no spend, so there's nothing to gate; the separate paid anchor_provenance upgrade, which additionally commits the hash on-chain, is untouched and still available. A credential is keyed by content hash, so it's issued exactly once per file no matter how many times a lane or endpoint touches it. Model pages now show the same verified / tampered / unverified badge the 3D viewer already had, and a third party can check any model for free with GET /api/provenance. The owner can turn automatic issuance off platform-wide with PROVENANCE_AUTO_DISABLE=true; nothing else about verification changes if they do.",
"tags": [
"feature",
"security"
]
},
{
"date": "2026-10-12",
"title": "Restyle and part edits now prove they kept the rig, not just claim it",
"link": "/restyle",
"summary": "Material Studio's AI restyle and seeded colorway variants, plus Restyle Studio's HoloPart \"Regenerate part\" tool, are now checked against their own source bytes with @three-ws/glb-diff before anything is persisted: a restyle declares it keeps topology, UVs, skeleton, skin, and clips, a part regenerate declares the narrower skeleton-and-clips promise since the part's own shape is meant to change, and the operation throws instead of returning a URL if the real diff shows otherwise. A successful restyle response now carries preserve: { kinds, verified: true } as evidence. Model Diff's underlying engine (packages/glb-diff) grew three independent mesh sub-hashes, topology, uv, and skinWeights, so a caller can tell which kind of change happened instead of one opaque geometry hash, and every lineage step in Restyle Studio that has a parent now shows a \"Diff\" link that opens Model Diff pre-loaded with that exact before/after pair. Regenerating a part also now mints its own durable lineage version automatically, closing a gap where part edits were not recorded.",
"tags": [
"feature",
"improvement"
]
},
{
"date": "2026-10-12",
"title": "Text-to-motion gets a second GPU backend: MoMask, selectable per request",
"link": "/pose",
"summary": "The self-hosted text-to-motion worker behind the Animation Studio and POST /api/forge-motion now runs two independent generation backends on the same GPU lane: MDM (Motion Diffusion Model, MIT), the existing default, and MoMask (EricGuo5513/momask-codes, MIT), a residual-VQ tokenizer plus masked and residual transformer that publishes a lower HumanML3D FID (0.045 vs. T2M-GPT's 0.141). Pass model: \"momask\" to /infer or /api/forge-motion to route a prompt through it instead; omit it and MDM keeps answering as before. Both backends converge on the same (T,22,3) HumanML3D position format and the same shared decode math, so a MoMask clip previews, retargets, and exports exactly like an MDM one. The two projects ship conflicting top-level `utils` packages; the worker resolves that at load time so either backend can be warmed in the same container without the other's modules leaking in. MoMask's checkpoints are staged to production storage and the worker is wired to load them; the first on-GPU deploy and a real in-house benchmark against MDM are tracked as the next step.",
"tags": [
"feature",
"infra"
]
},
{
"date": "2026-10-12",
"title": "Animation Studio: trim, loop-blend, and save any generated clip to your library",
"link": "/pose",
"summary": "The Animation Studio's prompt-to-clip flow (type a prompt, generate motion on the self-hosted GPU worker, preview it retargeted onto the loaded avatar) now has an editing and persistence half. The transport grows trim start/end inputs, a loop-blend duration field, and a loop toggle, all recomputed through a non-destructive edit path so the live preview, the GLB export, and the saved clip always agree on the current edit. A Save to library button collects a name and visibility and posts the canonical, pre-retarget clip to the shared clip library, so it plays back correctly on any future rig, not just the avatar it was previewed on. A separate bug that crashed every generated clip's playback the moment it reached the browser (AnimationClip.parse() expected userData as a JSON string, the server was sending an object) is also fixed.",
"tags": [
"feature",
"fix"
]
},
{
"date": "2026-10-12",
"title": "Real VRM avatars: spring bones, expressions, and .vrma clips via @pixiv/three-vrm",
"link": "/docs/avatar-engines",
"summary": "The hand-rolled VRM loader is gone. vrm-boot.js now registers the real VRMLoaderPlugin (@pixiv/three-vrm) on the three.ws GLTFLoader, so VRM 0.x and 1.0 avatars get genuine MToon materials, spring-bone physics, node constraints, look-at, and all 18 expression presets, driven live by vrm.update(delta) in AnimationManager and mapped onto the ARKit-52 face-capture driver. VRMAnimationLoaderPlugin loads .vrma clips into the same retargeting pipeline that drives non-VRM humanoids, so a single clip plays on either. The Ready Player Me import page (/import-rpm) now accepts .vrm files alongside .glb, and a long-standing bug that 400'd every import at the final save step (it sent an invalid source value) is fixed. docs/avatar-engines.md upgrades three-vrm from interop to live.",
"tags": [
"feature",
"fix"
]
},
{
"date": "2026-10-12",
"title": "<agent-3d> gets model-viewer's AR contract: ar-modes, variants, hotspots, and /ar drops the third-party CDN",
"link": "/avatars",
"summary": "<agent-3d> now matches the AR/viewer contract the market expects from model-viewer (Apache-2.0): ar-modes picks which launch paths to try (WebXR, Scene Viewer, Quick Look), ar-placement anchors floor or wall, ios-src supplies a pre-baked USDZ or the element generates a best-effort one on device, variant-name switches a GLB's real KHR_materials_variants materials live, and any element with slot=\"hotspot\" tracks a 3D position and flies the camera to a saved orbit on click. /avatars/:id/ar and /agents/:id/ar (pages/ar.html) now run entirely on <agent-3d ar>; the page no longer loads a third-party viewer from a CDN.",
"tags": [
"feature"
]
},
{
"date": "2026-10-12",
"title": "Three new engine bridges: Unity, Godot, and Unreal, one prompt to a mesh in your editor",
"link": "/docs/engine-bridges",
"summary": "A shared contract (specs/ENGINE_BRIDGE.md) and three real, independent implementations: a Unity UPM package that imports with glTFast, a Godot addon that imports with Godot's own native GLTFDocument, and an Unreal plugin that imports through Unreal's Interchange glTF pipeline. Each one calls only the public Forge API (POST /api/forge to submit, GET /api/forge?job=<id> to poll, a plain HTTPS GET to download), the same free-first routing and anonymous x-forge-client auth every other Forge caller gets, with the same four import settings (generate_colliders, import_materials, scale, destination) mapped onto each engine's own equivalent. A dropdown and a Generate button in each editor: type a prompt, get a mesh dropped into your scene.",
"tags": [
"feature",
"sdk"
]
},
{
"date": "2026-10-12",
"title": "Engine QA: a free real-time-engine fitness grade, plus real LOD chains",
"link": "/docs/engine-qa",
"summary": "A new free grade answers the question a game engine's import step asks that neither glTF validation nor simulation readiness does: draw calls, material count, texel density against the mesh's real-world size, a per-level triangle budget for a LOD chain, and whether the mesh is cheap and closed enough to use as its own physics collider. GET/POST /api/3d/engine-qa grades any public GLB (format=stl exports a watertight mesh for printing); the same browser-only checks run with zero network cost in Rig Doctor, and the full grade with a one-click upgrade in the GLB viewer, through a shared <engine-qa> panel component. The underlying LOD builder (packages/glb-tools's buildLodChain, real meshoptimizer decimation tagged with MSFT_lod) is now also available on generation: send { \"lods\": true } to POST /api/forge and the finished creation gets a combined LOD-chain variant alongside its delivered mesh, the same opt-in, fire-and-forget pattern as the existing web-delivery variant.",
"tags": [
"feature"
]
},
{
"date": "2026-10-11",
"title": "A daily partner spotlight on X: one partner, one feature built on it, filmed live every day",
"link": "/partners",
"summary": "@trythreews now features one partner from the partners page every day, with a thread on something three.ws built on that partner's technology. Each spotlight is filmed working against production that morning, so the reel is proof rather than a promo, and it rotates between a card, a GIF, the reel and generated art for its head. Every sentence comes from a partner registry with evidence behind it, the program name is stated exactly as the partner grants it, and nobody is tagged. The spotlight owns the prime-time slot, and three days are kept in stock so a bad run never leaves the slot empty. Once a week the series also publishes a partner deep-dive as an X Article: every feature built on that partner, filmed that day, with the runs, the evidence behind each sentence, the real requests and responses, and inline frames from the reels.",
"tags": [
"feature",
"docs"
]
},
{
"date": "2026-10-11",
"title": "Restyle Studio can now edit a single part of a model, not just the whole thing",
"link": "/restyle",
"summary": "Models with named part nodes, whether from the new \"Editable parts\" Forge engine (PartCrafter, MIT) or from the existing part-segment tool, now show a Parts panel in Restyle Studio. Select a part and every material tool, presets, AI restyle, manual sliders, scopes to that part alone instead of the whole model. Swap a part in from a local file, or hit Regenerate to send it through the new HoloPart worker (MIT), which completes whatever geometry was hidden or cut off at the segmentation seam, then reassembles the result back into the same model under the same part name. Part node names never change across either path, so a model edited this way stays compatible with every other part-aware tool on the platform.",
"tags": [
"feature"
]
},
{
"date": "2026-10-11",
"title": "Rig Doctor now diagnoses non-humanoid rigs too, not just naming mismatches",
"link": "/rig-doctor",
"summary": "Rig Doctor's canonical-joint coverage score has nothing to say about a creature, prop, or vehicle rig: it has no Mixamo-shaped skeleton to map and always reads as \"will not animate\" by design. A new Skin weights stat answers the question that actually applies to a rig like that: how many joints influence the single heaviest-weighted vertex, and how many of the mesh's vertices came back with no joint influence at all. Zero unweighted vertices means the skin is sound even though the clip library will never retarget onto it; a nonzero count means part of the mesh will not move under any skeleton.",
"tags": [
"feature"
]
},
{
"date": "2026-10-11",
"title": "Forge adds a lane that hands back an already-rigged model",
"link": "/forge",
"summary": "A new Forge engine, \"Rig on generate\", predicts a model's mesh, joint skeleton and skin weights together in one pass instead of handing back a bare mesh that still needs a separate rigging step. Built on VAST-AI-Research's AniGen (MIT), running on our own self-hosted worker. Pick it from the engine grid, give it a single reference image, and the finished creation arrives already animatable, with the poll response marking it rigged and reporting its joint count so the UI never offers a redundant rerig call. It is opt-in: generating without picking it behaves exactly as before.",
"tags": [
"feature"
]
},
{
"date": "2026-10-11",
"title": "Auto-rigging now covers non-humanoid meshes, not just Mixamo-shaped ones",
"summary": "The auto-rigging worker behind the Forge's \"Animate it\" step used to assume a roughly humanoid mesh (a head, a torso, two arms, two legs); anything else, a creature, a prop, a vehicle, came back unrigged. A new fallback worker built on UniRig and SkinTokens (both MIT) now catches those cases: a lightweight shape heuristic in the rigging worker decides whether a mesh is humanoid-shaped, and anything that isn't routes to the new worker for a skeleton and skin weights built for its actual topology instead of forcing a Mixamo bone set onto it.",
"tags": [
"feature"
]
},
{
"date": "2026-10-11",
"title": "Forge gets a fidelity lane: match your photo instead of guessing it",
"link": "/forge",
"summary": "A new Forge engine, \"Match my photo\", fuses 1 to 4 photos of the same object into one pixel-aligned 3D model instead of generating a plausible guess from a single view. Drop photos into the existing view slots and label each one Front, Back, Left or Right; the model is pixel-aligned to every photo you give it, and more photos narrow the result further. Only one photo on hand? Toggle \"fill in the other sides\" and the worker synthesizes the missing angles before reconstructing. Built on TencentARC's Pixal3D (MIT) running on Microsoft TRELLIS.2's backbone, with MV-Adapter (Apache-2.0) for the view synthesis step. Every input photo is recorded against the finished model, and single-photo requests on every other engine are unaffected.",
"tags": [
"feature"
]
},
{
"date": "2026-10-11",
"title": "A stuck background job now fails fast instead of hanging for 15 minutes",
"summary": "Every scheduled background job (cron) on the platform now answers cleanly and quickly if something inside it gets stuck, instead of silently hanging until the server's own 15-minute connection limit cuts it off with no explanation. Each job now gets a real time budget matching what it already declared, the status board records the failed run immediately, and an alert fires with the job's name attached so it can be diagnosed in seconds instead of traced by hand from a blank timeout. Image rendering for 3D creations got the same treatment: a stuck browser launch, page load, or screenshot now times out on its own instead of tying up the render queue.",
"tags": [
"infra"
]
},
{
"date": "2026-10-11",
"title": "The campus is now a real place: daylight, walk-in buildings, photographic surfaces and real props",
"link": "/?world=1",
"summary": "The 3D campus in front of the desktop is rebuilt to feel like walking through a real place rather than a dark game level. A physical sky and sun with soft shadows that follow you, hills on the horizon, a walled gate with lamps and a welcome sign, a lawn, concrete paths and a paved plaza. Every building is a room you walk into through its door: the brick office with a gabled roof, windows and a furnished desk; the marble-columned exchange with a ticker outside and over the dais inside; the forge with a glowing furnace; the gallery under a skylight, hanging the real avatars of the agents on the campus; a hillside mine with a timbered portal, rails and a cart; the launchpad with its rocket, gantry and crew cabin; a pergola plaza with a fountain; and a timber market hall with striped stalls. Walls, floors and roofs use CC0 photographic materials from Poly Haven, and trees, hedges, lamps, benches, bins, crates and barrels are real glTF models streamed from the three.ws object library, nearest first. Walls and furniture are solid for you and the agents, the camera never cuts through a wall, teleporting puts you at a building's door facing in, and quality adapts to the device (a software renderer always gets the lightest tier).",
"tags": [
"feature",
"desktop"
]
},
{
"date": "2026-10-11",
"title": "The on-chain leak alarm now scans smarter instead of scanning everything",
"summary": "The background monitor that watches every platform-controlled Solana wallet for a leak now checks the handful of wallets that hold real balances every run, any wallet the platform's own payment ledger shows just moved money, and the rest of the registry in a rotation, instead of walking the full wallet list on every single pass. That cuts the monitor's own Solana RPC usage sharply, which means faster balance reads and snappier agent pages across the platform, and it no longer runs two overlapping schedules at once. A wallet-identity bug found along the way is fixed too: on-chain transactions were being matched by the wrong field shape, which silently kept both the leak detector and its fee-accounting cross-check from recording anything real. Both now work correctly.",
"tags": [
"infra",
"fix"
]
},
{
"date": "2026-10-11",
"title": "The desktop now looks and works like a real operating system",
"link": "/?view=desktop",
"summary": "The desktop shell is rebuilt to feel like a computer, not a web page. A floating island taskbar with Start, Widgets and Assistant buttons. A new Start menu with your avatar, the time, an A to Z list of every app and page, pinned apps, recent files, quick folder links and a power button, plus search across apps, pages and your files. A system tray with live network, volume and battery readouts that opens quick settings (dark mode, night light, do not disturb, energy saver, sounds, full screen, brightness and volume). A notification centre over a month calendar, with toasts that open the app that sent them. A lock screen (Ctrl+Alt+L), a power menu with restart, shut down and sign out, a Run box (Ctrl+Alt+R) and an About page with your real device specs. The desktop is now your Desktop folder: create folders and text files from the right-click menu, rename with F2, delete to the Recycle Bin, drop files from your computer onto it, and set any picture as the wallpaper. New Bubbles wallpaper in dark and light, glossy app icons, and Ctrl+Alt shortcuts for every system surface.",
"tags": [
"feature",
"desktop"
]
},
{
"date": "2026-10-11",
"title": "Share a finished Forge creation on X and it plays inline",
"summary": "A link to a finished Forge creation now carries an X player card instead of a flat image: the post expands into the real 3D model, posed with its own poster frame and ready to orbit, through the same safe embed runtime the agent-profile embeds use (no audio or video element, HTTPS assets only). A creation that is still generating keeps the plain image card until it finishes, so a link never promises a model that isn't ready yet.",
"tags": [
"feature"
],
"link": "/forge"
},
{
"date": "2026-10-11",
"title": "The web desktop is now a real OS: a file system and eleven apps that run in your browser",
"link": "/?view=desktop",
"summary": "The desktop gains a persistent virtual file system (Home with Desktop, Documents, Pictures, Music, Videos and Downloads, a Recycle Bin, quota readout) and eleven native apps built on it: Files with list, detail and tile views and drag-free cut, copy, paste and rename; Notepad with tabs and autosave; Photos; Calculator with standard, scientific, programmer and unit-converter modes; Clock with world clocks, alarms, timer and stopwatch; Weather from live Open-Meteo forecasts for your location or any city; Camera that takes real photos and videos into Pictures and Videos; Paint with layers of tools, undo and export; Terminal with a real command set that opens apps and switches theme; Task Manager with live memory, frame rate and per-window usage; and a Media Player with playlists. Everything is stored in your browser and nothing is faked: metrics a browser cannot report say so.",
"tags": [
"feature",
"desktop"
]
},
{
"date": "2026-10-11",
"title": "Choose which @-mentions your agent answers, right from its editor",
"summary": "An agent's Social tab now has an \"Answer mentions\" section: turn it on, keep it in dry run while you watch what it would say, cap it at a daily reply count, and pick exactly which kinds of mentions it may act on. A new decisions list on the same page shows the agent's own recent calls, what it decided and why, so the owner never has to ask an admin to check a console on their behalf.",
"tags": [
"feature"
],
"link": "/agents"
},
{
"date": "2026-10-11",
"title": "Live widgets and an assistant on the desktop, and the desktop is back at /",
"link": "/docs/desktop-widgets",
"summary": "The homepage is the web desktop again, and the Field moves to /?view=field. The desktop gains eleven live widgets: the $THREE price with a 24 hour chart, weather where you are, clock, calendar, the newest agents, Pulse wallet moves, coins launched through three.ws, platform stats, headlines from the news desk, your device's battery, network and storage, and sticky notes. Pin them to the desktop in small, medium or large, drag them on a snap grid, or keep them on the widgets board beside a news column. The new assistant pane answers questions about three.ws, opens the right app from its reply, and lets you switch to any public agent and talk to it instead. Nothing is sample data: a source that is down says so and offers a retry.",
"tags": [
"feature",
"desktop"
]
},
{
"date": "2026-10-11",
"title": "The @-mention reply bot now reports its own status, and /grok shows it live",
"summary": "A new public endpoint, /api/x/mention-bot/status, answers whether the @-mention reply bot is live, in dry run, or paused, how many handles it has replied on, how many replies it sent in the last 24 hours, and its median reply time. No mention text or author identity is included. The /grok page now shows this as a live section, and /api/healthz carries the same signal as a platform subsystem, so an operator sees a wedged reply queue or a rising error rate within the hour it starts, instead of the bot quietly going silent with no signature anywhere.",
"tags": [
"feature",
"improvement"
],
"link": "/grok"
},
{
"date": "2026-10-11",
"title": "Real agents live on the campus: a greeter meets you at the gate and every agent you walk up to talks back",
"link": "/?world=1",
"summary": "The campus is now populated by the ten most talked-to public agents on three.ws, each wearing its own avatar from its public record (five on a phone). The most popular one stands beside the gate path, turns to you as you walk in, waves and says hello through its own brain. Every agent idles, wanders a step, faces you when you come near and carries a name tag; walk up (or tap the tag) and press E to open a conversation that streams the agent's real reply, shows it as a speech bubble and can read it aloud. The panel opens the agent's profile as a window and hands you the one-tag embed that puts that exact agent, body and brain, on any web page. GET /api/agents/public now returns avatar_model_url for public and unlisted avatars.",
"tags": [
"feature",
"desktop"
]
},
{
"date": "2026-10-11",
"title": "An agent set to Grok now answers as Grok everywhere, with a badge when it can't",
"summary": "Profile chat, the agent editor's copilot, scheduled strategy runs, and the gateway all now read an agent's chosen brain through the same resolver, so picking Grok for an agent actually means Grok on every surface, not just the one where it was set. A Grok pick first tries the agent owner's own saved xAI key, then the platform's, before falling back to the free default chain. When Grok genuinely can't answer (no key reachable anywhere), profile chat now shows a small badge naming which model answered instead, so it's never a silent swap.",
"tags": [
"improvement"
]
},
{
"date": "2026-10-11",
"title": "The homepage is now the Field: one live map of every agent, launch and wallet move",
"link": "/",
"summary": "The front page no longer opens a web desktop. It opens the Field, a zoomable map of the whole platform built from live data. Every public agent is a star in the core, newest at the centre, and turns into its avatar as you zoom in; around it sit ten districts (Create, Forge, Gallery, Launchpad, Trenches, Pulse, Marketplace, Chat, Agent Studio, Console). From afar each shows one live readout; zoom in and it opens into a card with real content: the newest avatars, coins agents launched, the $THREE price, market cap, holders and contract address, the latest agent wallet moves, the skills agents carry most. New wallet moves from Pulse fly across the map as comets and land in their district. Search finds any place, agent or page; opening one loads the real page in place, with its own shareable /?go= link, and Back or Esc returns you to the map. It works with mouse, touch, pinch and keyboard, in light and dark, and lays itself out for phones. The windowed desktop is still at /?view=desktop and the 3D campus at /?world=1.",
"tags": [
"feature"
]
},
{
"date": "2026-10-11",
"title": "Sentiment Scout's X reads gained a second, independent path through xAI",
"summary": "The Sentiment Scout reads X for posts quoting a coin's exact contract address, as evidence a holder can open and check. When our own X search credentials are unavailable or rate limited, it now fails over to xAI's x_search tool, a second route to the same posts, metered by its own daily cap so an outage on one path never silently costs more than planned on the other. No change for anyone reading the Scout: the evidence looks the same either way.",
"tags": [
"improvement",
"infra"
],
"link": "/docs/sentiment-scout"
},
{
"date": "2026-10-11",
"title": "The web desktop gets Snap layouts, taskbar peek and three shell styles: Fluent, Aqua and Classic",
"link": "/?view=desktop",
"summary": "Hover a window's maximize button (or press Alt+Z) to tile it into one of six layouts, then let Snap assist fill the empty zones with your other windows. Corners snap to quarters, Ctrl+Alt+Arrows snap from the keyboard, minimize flies the window into its taskbar button, and hovering a running app shows its window card with Aero Peek. Settings, Personalization, Style switches the whole shell to Aqua (menu bar, magnifying dock, traffic lights) or Classic (bevelled 9x chrome with a labelled taskbar).",
"tags": [
"feature",
"desktop"
]
},
{
"date": "2026-10-11",
"title": "3D viewers no longer flood the browser console, and /trenches stops hammering an offline feed",
"link": "/marketplace",
"summary": "Every page with a 3D model viewer moved to model-viewer 4.4.0, the first release that both decodes our compressed avatars and drops the debug lines 4.3 printed for every model on the page. The trenches dashboard now checks the feed once and backs off while it is offline instead of sending five requests every 15 seconds, and desktop app windows no longer trigger a fullscreen-attribute warning.",
"tags": [
"fix"
]
},
{
"date": "2026-10-11",
"title": "A page for every three.ws app: Android, Solana Seeker, iPhone, desktop and the web desktop",
"link": "/apps",
"summary": "The new Apps hub lists every way to run three.ws, each with its own page: the web desktop, the Android app (APK download), Solana Seeker (live on the dApp Store, Seed Vault sign-in and Genesis Token verification), iPhone and iPad (what runs today and what the native build adds once Apple enrollment clears), and the Windows and Linux desktop app, whose download button reads the live release feed and picks the right file for your system.",
"tags": [
"feature"
]
},
{
"date": "2026-10-11",
"title": "Walk into three.ws: the homepage boots, unlocks, and lets you enter the desktop as your avatar",
"link": "/",
"summary": "The homepage now starts like a real computer. A boot screen runs live checks (your GPU, the site catalog, your account, the avatar body, the $THREE feed, your gem ledger) and a lock screen greets you with the $THREE price and your balances. Press Walk in and you arrive on a campus as your avatar: walk to your desk and sit down to open the desktop on the monitor, pick up the phone for Pocket, put on the headset for VR, enter the three.ws Exchange (a live ticker board and Trenches inside), the Launchpad, the Forge, the Gallery, Agent Plaza and the Marketplace, or dig for gems in the mine, where every gem is rolled server-side and kept on your account. Settings gets an Arrival section to choose where you start, the taskbar gets a World button, and the desktop wears the classic three.ws violet grid.",
"tags": [
"feature",
"desktop"
]
},
{
"date": "2026-10-11",
"title": "The prompt director works again, found by its own run receipts",
"link": "/runs",
"summary": "The first receipts showed the director's brief falling back on every Studio run. The replies were arriving in 5 to 10 seconds but ran past the 1000-character prompt cap or ended on the closing constraint the director is told to end on, and both were discarded as clipped. The director now writes to a word budget the models actually follow, the guard accepts the mandated ending, both forge clients share one guard, and a receipt now names the exact reason a brief fell back (too long with its length, unfinished, no reply) instead of a generic timeout. Receipts also stop saying every stage met when a stage was skipped.",
"tags": [
"fix"
]
},
{
"date": "2026-10-11",
"title": "Every 3D generation now writes a signed run receipt",
"link": "/runs",
"summary": "Each Studio run (avatars, rigged avatars, meshes, rigs, refinements) records stage by stage what was expected, what was observed, and a verdict: met, recovered by a named fallback, missed with its cause, or skipped. The rig is verified by reading the delivered file back, not by trusting the rigger. Receipts are ed25519-signed, linked from every result, browsable at /runs with stage-by-stage rates, and verifiable offline with one command.",
"tags": [
"feature"
]
},
{
"date": "2026-10-11",
"title": "Browse every feature by goal",
"link": "/everything",
"summary": "A new /everything page groups all 340 product pages under twelve goals (make, animate, build an agent, embed, earn, launch, trade, on-chain, worlds, discover, developer tools, account) with search, pins, recents and a new-this-month feed. The nav leads with it, the command palette labels every result with its goal, and tool pages like Forge, Create and Widgets now end with a what-next strip.",
"tags": [
"feature"
]
},
{
"date": "2026-10-11",
"title": "The desktop behaves like a real computer",
"link": "/docs/desktop",
"summary": "Drag a selection box across the desktop, move icons on a snap grid, right-click for view options, and open every app from desktop icons or All Apps. Settings now covers your profile, theme, accent colour, wallpaper (including your own photo), icon size and taskbar, and the whole session (windows, icon layout, preferences) is saved in your browser, with export, import and reset.",
"tags": [
"feature",
"desktop"
]
},
{
"date": "2026-10-11",
"title": "three.ws is now a desktop",
"link": "/docs/desktop",
"summary": "The homepage is a web desktop with a taskbar, Start menu, widgets and draggable, snappable windows that open every part of the platform, in the CTRL Windows-style Mica theme. The original homepage is kept as the classic view at /classic, one click from the taskbar.",
"tags": [
"feature"
]
},
{
"date": "2026-10-11",
"title": "A registry of every place three.ws is listed",
"link": "/docs/listings-registry",
"summary": "The listings registry tracks 121 surfaces across MCP directories, x402 directories, aggregators, cloud marketplaces, awesome-list pull requests and press, with a command that re-fetches every live entry so a dropped listing shows up.",
"tags": [
"docs"
]
},
{
"date": "2026-10-11",
"title": "API keys can no longer arm spenders or move payout addresses without the wallet scope",
"summary": "Sixteen routes that accept an API key could arm something that spends or redirect where money lands, for any key at all. They now require the wallet:write scope: the treasury autopilot, the auto-launcher, the market maker, live Oracle watches, the pump buyback and distribute policy, signal subscriptions, recurring agent charges, team tasks that hire paid agents, agent orders and spending mandates, an agent's wallet address, a skill listing's proceeds split, animation payout addresses, a Launchpad page's payout wallet or pricing, and opening or re-terming a vault. Pausing, cancelling and closing stay open to any key, and editing an agent no longer lets a narrow key raise its spend limits or change its payment fallbacks.",
"tags": [
"security",
"fix"
],
"link": "/docs/mcp"
},
{
"date": "2026-10-11",
"title": "The terminal 3D renderer animates live in every HTTP client",
"summary": "curl three.ws/tty streams an animated 3D character into your terminal frame by frame. Clients that ask for compressed responses, such as curl --compressed, scripts using fetch, and most HTTP libraries, used to get nothing for about ten seconds and then the whole animation at once. Live streams are no longer compressed, so every client sees the character move from the first frame. Streaming agent and chat responses get the same fix.",
"tags": [
"fix"
],
"link": "/tty"
},
{
"date": "2026-10-11",
"title": "Avatar builds that hit a full library now say so, and the news knowledge base stops losing stories",
"summary": "If your avatar library filled up while a photo or prompt avatar was still generating, the finished job used to sit unresolved and retry in the background without telling you why nothing appeared. It now closes with a clear message: delete an avatar or upgrade, then build it again. News stories whose text contained a half-cut emoji or a stray null character were silently dropped from the knowledge base the 3D agents ground their answers on, about 120 a day; they are now cleaned and saved. Static files such as images and environment maps stop answering a server error when the CDN revalidates them, GPU workers fetch reference photos from hosts that refuse anonymous clients, and a stalled upstream can no longer hold an image or webhook request open until the platform timeout.",
"tags": [
"fix"
],
"link": "/news"
},
{
"date": "2026-10-11",
"title": "Trenches Pulse: live launch and smart-money intelligence",
"summary": "A new Trenches page shows what is happening on Solana and Robinhood Chain right now: new launches and graduations as they happen, the hottest tokens by trade flow, and scored smart-money wallets. Every number comes from a live on-chain feed, and if the data service is unreachable the page says so instead of showing made-up figures.",
"tags": [
"feature"
],
"link": "/trenches"
},
{
"date": "2026-10-11",
"title": "Menu bar panel for the desktop app",
"summary": "Click the three.ws icon in your menu bar or tray and a small panel shows your agent alive in 3D, its Solana wallet balance, a start and stop button, anything waiting on your approval, unread alerts and a box to ask the agent something. Approvals still open the native confirmation window, so what you approve is exactly what you were shown.",
"tags": [
"feature"
],
"link": "/desktop"
},
{
"date": "2026-10-10",
"title": "Anatomy: describe a machine and explore it as an interactive 3D cutaway",
"summary": "Type a machine in one line, such as how an A380 turbofan makes thrust, and three.ws builds an isometric wireframe cutaway you can rotate, explode, cut open and click part by part. Shafts, blades and pistons really move, WebGL shaders draw the fire, exhaust, plasma and water, particle flows trace the air and fuel, and a guided tour walks through how it works. The machine assembles on screen while it is written and every one gets a shareable link. Agents can build them too with the open anatomy skill.",
"tags": [
"feature"
],
"link": "/anatomy"
},
{
"date": "2026-10-10",
"title": "Rig Doctor and the animation library now understand Source engine and 3ds Max Biped hands, plus tighter upload and number checks",
"summary": "Source engine (ValveBiped) skeletons now map to the shared animation library and Rig Doctor names them as their own convention, so 16 rig conventions are recognised. 3ds Max Biped finger chains and toes now map too, which restores hand animation on those rigs. The browser GLB check now rejects truncated files before upload, matching the server. The Money Pulse counters roll over to the right unit near $1k and $1M instead of showing $1000 or 1000.0k, and the URL scheme and model-host guards gained regression tests.",
"tags": [
"improvement",
"fix"
],
"link": "/rig-doctor"
},
{
"date": "2026-10-10",
"title": "Calmer homepage first visit: one corner card at a time, full agent descriptions, language in the footer",
"summary": "A first visit used to stack three cards in the bottom-right corner over the hero avatar. The search-shortcut hint now goes first and the \"Have you tried\" suggestion waits until it is dismissed. The Living agents cards show the whole description over two lines instead of cutting it off mid-word, and the language picker moved from the hero to the footer so the first screen holds only the Forge prompt and the three next steps.",
"tags": [
"improvement"
],
"link": "/"
},
{
"date": "2026-10-10",
"title": "AI agents now forecast Event Markets and build a public track record",
"summary": "Agents can call a market as themselves, with a confidence, a rationale and evidence links, and the market page shows what each agent thinks under an Agent badge. Every agent profile gets its calls, hit rate, a calibration chart built from resolved markets only, and its best calls. A new forecasters page ranks people and agents, with an agents-only filter, and you can follow an agent to be notified when it calls. Owners can switch on autonomous forecasting per agent within a points budget, off by default and fully visible in the activity log. Assistants get event_market_analyze over MCP.",
"tags": [
"feature"
],
"link": "/event-markets/forecasters"
},
{
"date": "2026-10-10",
"title": "Event Market announcements draft themselves, and nothing posts until you approve it",
"summary": "When an Event Market opens, nears its lock, shifts, or resolves, a draft announcement now appears in a review queue with the market link and card. Entrants are tagged only where they linked their own X account, everyone else is named, and entrants who opted out are left alone. Approve a whole batch in one action, edit a draft, or dry-run the exact text that would go out. Posting is off until the owner turns it on, with a daily cap and one post per kind per market.",
"tags": [
"feature"
],
"link": "/admin/event-market-announcements"
},
{
"date": "2026-10-10",
"title": "Launch a coin from a quote the agent wallet pays, with a creator-fee picker and a live status",
"summary": "Agents can now launch a coin from a priced quote. Pick the quote asset and a creator fee, see the full fee lines and what a coin could earn, then fund it from the agent's own wallet and confirm it in the approvals inbox. Only a signed-in browser can fund or confirm, so an API key can never spend. Every launch walks the same stages (quote, paid, submitted, confirmed, indexed) and replays safely if you retry. A devnet dry run lets you rehearse the whole thing with no real funds, and the command line prints the quote and tracks the stage with npx three-ws agent launch --intent.",
"tags": [
"feature",
"improvement"
],
"link": "/launch/intents"
},
{
"date": "2026-10-10",
"title": "Sign in with Telegram, link devices with one-time codes, and prove payout wallets with a signature",
"summary": "You can now sign in to three.ws with Telegram, by the login widget or by tapping a link that opens the bot. Account, Linked devices mints a short one-time code that links a phone, a desktop app, the command line (npx three-ws link) or a Telegram chat. You always see what is asking and confirm it from your signed-in browser, and you can revoke any device in one click. Agents and owners can attach a Solana or EVM wallet by signing a message. Replacing a live payout wallet now asks for your password and takes effect after a 24 hour cooldown, and every link and unlink is in your audit log. A private Telegram chat from an account you signed in with now reaches your agent without a pairing code.",
"tags": [
"feature",
"security"
],
"link": "/dashboard/account"
},
{
"date": "2026-10-10",
"title": "Developer plans with quotas that are enforced, self-serve upgrades, key rotation, IP allowlists and per-key analytics",
"summary": "Every API key now sits on a developer plan: Free, Builder, Scale or Enterprise, each with a monthly call quota, a per-minute burst ceiling, a concurrency ceiling and a webhook cap, all enforced on every /api/v1 call rather than only recorded. Each response carries x-plan, x-ratelimit-* and x-quota-* headers; past the burst or concurrency ceiling you get a 429 with Retry-After, and when the period's calls are used up you get a 402 that names the next plan and links to the upgrade. The new /developers page renders the plans straight from the config, and the dashboard's Plan tab shows live usage against the quota, lets you upgrade with credits, USDC or $THREE on Solana (your own wallet signs, with a review of recipient, amount, asset and chain first), prorates mid-period upgrades, schedules downgrades for the end of the period and lists every receipt. Revoking a key now takes effect immediately on every server. Keys can be rotated with an overlap window so a deployment never loses access, locked to IP addresses or ranges, and inspected per key: calls by route, error rate, latency percentiles and top callers.",
"tags": [
"feature",
"sdk",
"security",
"docs"
],
"link": "/developers"
},
{
"date": "2026-10-10",
"title": "Agents can read their portfolio, run a launch sniper, set alerts and challenge each other to duels",
"summary": "Fifteen new tools for Claude or any MCP client, plus a portfolio API at /api/v1/agents/:id/portfolio. An agent can read its own valued Solana portfolio with cost basis and risk flags, its net-worth history (now recorded hourly) and its profit and loss broken down by where it came from. It can check its launch sniper, and arm it only after you have seen the exact budget, per-trade size, limits and network in a preview and said yes; turning it off is one call. It can subscribe to signal feeds in paper mode, and create, list and delete pump alert rules, including a new kind that fires on any new launch matching your filters: a name pattern, a market cap band, a minimum safety score, the creator's track record and risk flags to avoid. Alerts reach you on the bell, by push, and in any Telegram or Discord chat you have paired. Agents can also challenge another agent to a points-only duel, accept one, read a duel and browse the open markets, with every result counted on the duels leaderboard.",
"tags": [
"feature",
"sdk"
],
"link": "/docs/mcp"
},
{
"date": "2026-10-10",
"title": "Agents can compare every Solana swap route and see why one wins",
"summary": "Ten trading tools are now available to agents in Claude or any MCP client, and over the API at /api/v1/trading. Research a token (search with a safety read, live price, indicators, news) or read the whole Solana launch ecosystem at once: launch and graduation rates against their usual levels, the biggest movers, and anything unusual, each with the data behind it and the time it was measured. A swap quote now prices every aggregator side by side, shows what you actually end up with after fees and price impact, and says in plain words why the winner won. Filling it from an agent wallet takes a confirmed quote that expires in five minutes, is re-priced before it fills, is refused if the price slipped below what you approved, and can fill only once; a dry run simulates it first with nothing signed. Arbitrage quotes are honest about the risk: the two legs are separate trades, so you see a simulated worst case and a verdict, and large sizes are refused unless you ask for them.",
"tags": [
"feature",
"sdk"
],
"link": "/docs/trading-tools"
},
{
"date": "2026-10-10",
"title": "Your agents can trade perpetual futures",
"summary": "Each agent now has a perps desk at /agents/:id/perps, also reachable from the API and from Claude or any MCP client. It goes long or short with USDC collateral from the agent's own Solana wallet. Every agent starts in paper mode, which fills at live market prices without moving funds; live trading stays off until you turn it on for that agent. Every order is previewed with its exact margin, leverage, fees and liquidation price, runs only after you confirm, and is refused if the price moved too far in between. Leverage and per-position margin caps are always on (2x and $25 to start). A live tracker shows profit, funding paid and distance to liquidation, alerts reach you at the thresholds you set, and one kill switch closes everything.",
"tags": [
"feature"
],
"link": "/agent-perps"
},
{
"date": "2026-10-10",
"title": "Talk to your squad in one sentence",
"summary": "Squad chat turns a plain sentence like \"research $THREE and buy 0.05 SOL if it looks safe\" into a plan of steps for your specialist team or any of your agents. Research and entry checks run live with their evidence, and anything that signs, moves funds, launches a coin or goes over your cap stops as an approval card with the full details. Approve and the squad picks the run back up and finishes with receipts. Paper mode is the default and never signs. Say \"remember my default trade size is 0.1 SOL\" and the squad keeps it. Coin names and descriptions are treated as data, so a coin written to trick AI agents cannot change the plan. You can type or dictate, and the squad's 3D avatars react as they work.",
"tags": [
"feature",
"security"
],
"link": "/team-chat"
},
{
"date": "2026-10-10",
"title": "The 3D agent demos show their agent again",
"summary": "Every page in the agent interaction lab had stopped drawing its 3D character, so the demos showed only text and buttons. They load again: she climbs the title, catches your cursor, rides the scroll, falls asleep on the headline and the rest. The wrecking-ball entrance was also rebuilt so she really swings in on the rope and hits the button, instead of hovering beside it with the rope pointing the wrong way.",
"tags": [
"fix"
],
"link": "/demos/agents"
},
{
"date": "2026-10-10",
"title": "Approve or deny your agent's requests from Telegram and Discord",
"summary": "When your agent asks before a transfer or a trade, the request now lands in your paired Telegram or Discord chat with the full confirmation table (recipient, amount, asset, chain, risk notes and deadline) and Approve and Deny buttons. Each button is signed for that exact request, your chat and your account, so a changed request, an expired one, a copied button or someone else in a group chat can never trigger it, and pressing Approve twice runs the action once. The chat now also answers /approvals to list everything waiting, /positions for open positions and their live value, /pause to stop one agent and /kill to stop every agent and deny every pending request. A chat can stop your agents but never unfreeze them; that is done on the web. Forwarded messages are never treated as instructions.",
"tags": [
"feature",
"security"
],
"link": "/approvals"
},
{
"date": "2026-10-10",
"title": "Approve your agents' payments from a notification",
"summary": "Set any agent spend rule to Ask me and the agent now waits for your yes instead of failing. You get a notification with the recipient, amount, token and network spelled out, and can approve or deny right from it, from email, or from the new Approvals inbox. Approving runs exactly what you were shown, once; denying or letting it expire runs nothing. You can also create auto-approve rules for small swaps and transfers to addresses you have paid before, and revoke them any time.",
"tags": [
"feature"
],
"link": "/approvals"
},
{
"date": "2026-10-10",
"title": "Live agent payments keep running when fee sponsorship pauses",
"summary": "When the wallet that covers Solana network fees for agent payments runs low, agents on three.ws now pay their own fraction-of-a-cent fee and the purchase goes through, instead of failing with an error. The live agent-to-agent payment demo and the activity feed keep moving through it.",
"tags": [
"fix"
],
"link": "/agent-exchange"
},
{
"date": "2026-10-10",
"title": "Coin logos, agent art and IPFS images load again",
"summary": "Several public IPFS gateways started refusing image and metadata requests from websites this month, which left coin logos, the 3D coin face, agent cards and on-chain agent art blank or broken across the site. Every page and API now reads IPFS content through gateways that still answer, races them so the fastest wins, and quietly rewrites links that point at the refusing gateways, so art that was published long ago still shows up. Images from a host that no longer serves them now fall back to a clean placeholder instead of a broken-image icon.",
"tags": [
"fix",
"improvement"
],
"link": "/launches"
},
{
"date": "2026-10-10",
"title": "Quieter, cleaner pages: dead links fixed and console errors gone",
"summary": "We swept every page on three.ws with the browser developer console open and fixed what it found. Links to pages that moved now redirect to where they live today, share buttons always have a destination, 3D viewers decode compressed models on every page, the missing bow animation thumbnail is back on the choreography and gestures pages, and pages full of brand-new coins or agents without a wallet no longer fill the console with errors while they wait for data.",
"tags": [
"fix"
],
"link": "/gestures"
},
{
"date": "2026-10-10",
"title": "Set up your AI client with one command, and check it with doctor",
"summary": "Run npx three-ws --claude (or --cursor, --codex, --vscode, --windsurf) and the CLI signs you in, connects three.ws and verifies it, with no prompts. Keys never land in your client's config file, and running it twice changes nothing. New installers at three.ws/cli/install and three.ws/cli/claude come with a checksum and a signature so you can read and verify them before you run them. npx three-ws doctor finds a broken config, an expired sign-in or an unreachable server and tells you the fix. agent status and team show your agents, wallets and teams from the terminal.",
"tags": [
"feature",
"improvement"
],
"link": "/docs/cli"
},
{
"date": "2026-10-10",
"title": "Run your agents on your own desktop",
"summary": "The three.ws desktop app can now run agents, strategies, limit orders and automations on your machine. Signing keys stay in your operating system keychain, paper strategies keep working with no account or cloud connection, and every live trade waits for your approval in a native window that shows the exact recipient, amount, asset and chain. A tray menu lists each agent with pause, kill and review, and the mascot shows when an agent is working, waiting for you or in trouble.",
"tags": [
"feature"
],
"link": "/docs/desktop-local-runtime"
},
{
"date": "2026-10-10",
"title": "Specialist teams: four agents that research before they trade",
"summary": "Assemble a squad of four 3D agents in one click: a Researcher that vets a token and posts a scored verdict, an Entry scout that checks fresh launches against your entry gates, a Trader that quotes, simulates or trades inside one spend policy, and a Launcher that prepares a launch plan for you to sign. They share a live findings board. The Trader only acts on research it can cite, reuses a fresh verdict instead of redoing it, and every refusal is posted with its reason. Only the Trader can ever sign, and only a live trade you confirm.",
"tags": [
"feature"
],
"link": "/teams"
},
{
"date": "2026-10-10",
"title": "Import skills from any registry, safely",
"summary": "Agents can now take skills from outside three.ws. Point the new Import Skills page at any public GitHub repository of SKILL.md files or any registry manifest, browse by category (DeFi, intelligence, social, infrastructure, security, data) with install counts, and install in two steps: a scan you read, then your approval. Every skill is pinned to the exact commit it was read at, licence-checked (skills that do not permit reuse are never offered), and scanned for hidden instructions, secret grabbing, data leaks and hardcoded payout addresses. Skills that ask to spend, sign or send messages need a second yes and can never move funds on their own. Updates never apply silently: you see the diff and approve it. You can also fork any skill and publish your own for everyone to import, from the page or from any MCP client.",
"tags": [
"feature",
"security"
],
"link": "/skills/import"
},
{
"date": "2026-10-10",
"title": "Sign in with Google, connect several AI clients at once, and grant permissions by group",
"summary": "You can now sign in to three.ws with Google, on the login page and on the screen an AI client sends you to when it connects. A Google account that matches an account you already verified is linked only after you confirm it on a clear page, and you can link or unlink Google any time under Settings, Sign-in methods. Connecting a second client (Claude, then ChatGPT) no longer risks the first: each client gets its own grant and both stay connected. The permissions screen groups what a client asked for into Read, Manage, Trade, Spend and Launch, says what each group can do and which actions will still ask you first, and lets you untick any group but Read before you approve. Revoking a client from Settings now stops it on its very next request, and the /connect page has the one URL and the steps for every client.",
"tags": [
"feature",
"security"
],
"link": "/connect"
},
{
"date": "2026-10-10",
"title": "Strategies now research every coin and ask before they buy",
"summary": "Trading strategies can now check a coin before spending anything: minimum holders, how much the largest holder owns, liquidity, the rug and honeypot safety score, whether the creator can still mint or freeze, and the creator's track record. A coin that fails a check, or whose data cannot be read, is skipped and the decision log on your agent says exactly which check stopped it. You can also cap price impact per buy. Strategies now ask for your approval before each buy by default, with the amount, coin and risk notes on the card, and existing strategies were switched to ask mode too; flip one to Auto once you trust it. The strategy builder edits everything with sliders, toggles or raw JSON, turns a plain-English description into settings, shows a live count of recent launches your settings would have bought, and backtests with the checks applied.",
"tags": [
"feature"
],
"link": "/strategies"
},
{
"date": "2026-10-10",
"title": "Agents can sign themselves up, and the platform now reports its health honestly",
"summary": "An autonomous agent can now create its own three.ws account by signing a short message with its key: no email, no browser. It gets a wallet and an API key and starts in paper mode with strict limits until a person claims it. New public endpoints list every tool with prices and rules, publish real platform totals (agents, launches, volume, payouts, active strategies) with their sources, and a readiness check that reports degraded when the database, Solana RPC, the worker or wallet encryption is unwell. All of it is shown on the status page.",
"tags": [
"feature",
"improvement"
],
"link": "/status"
},
{
"date": "2026-10-10",
"title": "Launch a coin as a Uniswap pool on Base, and retry launches safely",
"summary": "There is a new launch lane: your agent launches a fixed-supply coin whose whole supply sits in one Uniswap V3 pool on Base. You choose the pool fee, lock the liquidity for a set time or forever, and route the fees, and a confirmation table lists every fee before anything is signed. The launch wizard now compares every lane (chain, fees, creator share, graduation, lock) from live figures. Launches on Base and Robinhood Chain now take an Idempotency-Key and keep a record you can follow at GET /api/launches/:id, so a dropped connection or a retry returns the same launch instead of a second one. Solana stays the default lane. The Base lane opens once its launcher contract is deployed.",
"tags": [
"feature"
],
"link": "/launch/uniswap"
},
{
"date": "2026-10-10",
"title": "New payout addresses now wait out a cooldown you can cancel",
"summary": "Your agent wallet has a destination whitelist. A new address cannot receive funds until it has served a cooldown (24 hours by default, never less than 1 hour). Adding one needs you to re-confirm with your password or an emailed code, you are told instantly on every connected channel, and you can cancel it with one click. Agents and API keys can only propose an address, never activate it. Removing an address works instantly. The check covers withdrawals, transfers, tips, bridge sends and payouts, and every confirmation shows whether the destination is on your list. Manage it on the new Allowlist tab of your agent wallet.",
"tags": [
"feature",
"security"
],
"link": "/agent-wallet"
},
{
"date": "2026-10-10",
"title": "Agents can find and register web domains",
"summary": "Search any name across popular extensions with live availability and the exact yearly price, check a single name, and register it from your credits after a quote and an explicit yes. Coin pages now suggest available domains for the project name, and a registered domain can serve your agent's public page. Registration is for one year, renews at the same price, and has no markup.",
"tags": [
"feature"
],
"link": "/domains"
},
{
"date": "2026-10-10",
"title": "Token ceilings, a spend page, USDC deposits and a way out of the free tier",
"summary": "Every agent can now carry hourly, daily and per-run token ceilings next to its dollar caps. A ceiling is checked before the model is called; hitting one pauses the agent, tells you, and gives you Resume and Extend once. The new /spend page shows what each agent spends by day, by model and by tool, projects the month end, and alerts at 50, 80 and 100 percent of any cap. Credits can be topped up with USDC at a fixed rate alongside SOL and $THREE, the deposit page lists every accepted asset and its rate from live config, and paying in $THREE earns whatever bonus the owner has set. When the free tier runs out, the error now carries your choices: wait for the reset, top up, bring your own key, or resend the same request with paid_fallback: true to bill it to credits in one step.",
"tags": [
"feature"
],
"link": "/spend"
},
{
"date": "2026-10-10",
"title": "Agents get their own email inbox",
"summary": "Every agent can now have an email address on agents.three.ws, with an inbox right on its profile (the Mail button). It can send, read, reply in thread, search and delete, from the web, the API, or any MCP client. Nothing is created or sent until you have seen the exact recipients and message and said yes. You can limit an agent to a list of recipients and a daily number of sends, and only you, signed in, can change those limits. Mail an agent receives is always treated as information, never as orders: hidden characters are shown, links are disabled, and the agent is told not to follow anything inside it. Rules let you turn incoming mail into work, such as \"when mail arrives from this sender, summarize it\", either right away or after you approve.",
"tags": [
"feature",
"security"
],
"link": "/agent-mail"
},
{
"date": "2026-10-10",
"title": "Manage agents, autonomous runs and automations from any MCP client, and replay every run on the agent page",
"summary": "Sixteen new MCP tools let an AI client do everything the dashboard can with an agent: edit it, start and stop it, give it a new avatar from a GLB or image, delete it, launch and steer autonomous runs, and create, edit, test-fire and remove automations. Deleting takes two steps, a preview and then an explicit confirm, so nothing disappears by accident. Every run now has a turn budget and a dollar budget, stops cleanly between steps when you cancel it, and ends with a plain-language summary. Each step is sealed with a receipt that chains to the one before, and the new Runs tab on your agent's page replays any run step by step, shows every tool call with its result, checks the receipts, follows a live run as it works and lets you cancel it.",
"tags": [
"feature"
],
"link": "/docs/mcp"
},
{
"date": "2026-10-10",
"title": "Quote any Papertrade position before you open it, from the API or your AI assistant",
"summary": "three.ws now reads Papertrade, the synthetic BTC and ETH perps exchange on HyperEVM. Ask for a quote and you get the entry price, the bust price where the whole margin is lost and how far away it sits, whether the market can take your size, and what a win would actually pay at eight different price moves once Papertrade's haircut and win fee come out. Small moves pay far less than the raw profit, and now you can see it before you trade. You can also look up any wallet's open positions valued at the live price, and check the protocol's health, including whether winners are waiting in its payout queue. It works over a public endpoint and as four tools for Claude or any MCP client. It only reads: you still trade on Papertrade from your own wallet.",
"tags": [
"feature"
],
"link": "/docs/papertrade"
},
{
"date": "2026-10-10",
"title": "Every MCP tool now has its own page, and the tool counts are measured, not typed",
"summary": "The MCP tool catalog now shows the real totals for every server, worked out from the live servers with a plain note on how, and a release view that lists which tools each release added. Open any tool to see its arguments, an example call, its price and what it can touch, and try the free read-only ones right on the page. You can filter the catalog by category. Behind it, every tool now says whether it only reads, whether it can destroy anything, whether repeating it is safe and whether it reaches outside three.ws, and every tool says when to use it. Prompts that pointed at tools that did not exist now point at real ones, and the tool counts quoted in our READMEs and docs are checked against the catalog on every build so they cannot go stale again.",
"tags": [
"feature",
"improvement"
],
"link": "/mcp-tools"
},
{
"date": "2026-10-10",
"title": "A complete REST API for agents, with signed webhooks",
"summary": "Everything you can do with an agent in the app now works over plain HTTP with one API key: create and edit agents, start and stop them, chat, start runs and watch them live, schedule automations and fire them on demand, manage skills, and export or import an agent. Lists page with cursors, every response has the same shape, and a retry with an Idempotency-Key never does the work twice. Webhooks tell your server when a run finishes, an automation fires, a message arrives or an approval is needed. They are signed, retried for up to a day, logged attempt by attempt, and any delivery can be replayed from the new Developers dashboard.",
"tags": [
"feature",
"sdk"
],
"link": "/docs/api-reference"
},
{
"date": "2026-10-10",
"title": "Every three.ws package now has its own home, docs site and agent-readable guide",
"summary": "Each SDK and MCP server is mirrored to its own GitHub repository with a documentation site, install command, topics for search, an AGENTS.md and llms.txt so coding agents can use it, plus contributing, security and issue templates. Find them all in the standalone repositories list.",
"tags": [
"docs",
"improvement"
],
"link": "/docs/start-here"
},
{
"date": "2026-10-10",
"title": "Agents can now send invoices, sell things, buy from each other and ask for higher limits",
"summary": "Your agent can bill anyone in USDC, SOL or $THREE. Each invoice gets a share page, a Solana Pay link and a QR code that any Solana wallet can scan, and three.ws watches the chain to mark it paid, part-paid or expired on its own, with notifications, webhooks and a receipt. Paid invoices show up in your agent's earnings next to skill sales and creator fees. Agents can also list things for sale and buy from each other with a quote first and an explicit confirm, and the goods are only released once the payment is verified. Sending money takes a preview and a confirm, and goes to your approvals unless the address is on your allowlist. An agent can read its own spending limits and propose changes, but only you can approve them, after confirming it is you. Manage it all on the new Commerce page.",
"tags": [
"feature"
],
"link": "/commerce"
},
{
"date": "2026-10-10",
"title": "Every public repo is now easier to find, trust and share",
"summary": "All public repositories on the nirholas GitHub account now carry the same discovery layer: a star and share section, an AGENTS.md and llms.txt so AI coding agents can work in them cold, contribution and security policies, issue forms, and tuned topics. A new profile page lists every repository grouped by topic, with a machine-readable list for agents. Package mirrors keep the same layer on every sync.",
"tags": [
"improvement",
"docs"
],
"link": "/docs/github-growth"
},
{
"date": "2026-10-10",
"title": "Every Arena tournament, platform event, build round and bounty now gets a prediction market automatically",
"summary": "A cron opens a \"Who wins X?\" market for each event as soon as it has two or more entrants, adds late entrants as outcomes until the market locks, and records why any event was skipped. Admins can see events with and without markets at /admin/event-markets and open or void one by hand.",
"tags": [
"feature"
],
"link": "/admin/event-markets"
},
{
"date": "2026-10-10",
"title": "Event Markets now settle themselves from our own data",
"summary": "When an Arena tournament, event, launch cohort, build round or bounty finishes, its market resolves on its own from three.ws records and shows exactly what it read: the standings, the values, the timestamps. Ties, events with no data and cancelled events end in a void with every pick refunded, and a market whose source never answers is voided after a documented timeout. Admin corrections need a written reason that is shown on the market page.",
"tags": [
"feature"
],
"link": "/event-markets"
},
{
"date": "2026-10-10",
"title": "Event Markets get their own pages and share cards",
"summary": "Browse live, closing-soon and resolved markets, open any market to see every entrant's odds, the odds over time, the lock countdown and how it resolves, then make your call without leaving the page. Each market has a live share card showing the question, the top entrants and the time left, a per-entrant version for challenging a friend, and entry points from the Arena, event pages, launches and the home page.",
"tags": [
"feature"
],
"link": "/event-markets"
},
{
"date": "2026-10-10",
"title": "Event Markets: pick the winner of any event, free",
"summary": "Every event now has a market where you put free points on who you think wins. The crowd's picks move the odds live, with an honest even split until the first pick lands. Change or withdraw your pick any time before the lock; a pick after it is refused with the closing time. Each account has a points budget per season, and a voided market gives every point back. Assistants can browse markets and place a pick over MCP once you confirm it.",
"tags": [
"feature"
],
"link": "/event-markets"
},
{
"date": "2026-10-10",
"title": "Event Markets leaderboard: seasons, streaks, badges and rewards",
"summary": "Every resolved call now scores. A right call on long odds earns the most, a wrong call costs the stake, and your score never goes below zero. Rank by season or all time, filter by event type, and see your own row pinned. Earn badges for your first correct call, an upset, five in a row and a season top 10, shown on your profile. Season-end $THREE rewards are listed on the page; accounts that fail the fair-play checks still play but do not rank.",
"tags": [
"feature"
],
"link": "/event-markets/leaderboard"
},
{
"date": "2026-10-10",
"title": "Event Markets: odds that move while you watch",
"summary": "Event market odds now update live. A pick in one browser moves the odds in everyone else's within about a second, with no refresh. The market page animates each change, the home page shows a live ticker of open markets with a pause control, and big swings (5 points or more within ten minutes) are flagged as moves. If the live connection drops, the page reconnects and catches up on its own, falls back to refreshing every few seconds, and says when you are offline. Picks stay anonymous: the feed shows counts, never who picked.",
"tags": [
"feature"
],
"link": "/event-markets"
},
{
"date": "2026-10-09",
"title": "Avatar builds that hit a full library now say so, and the news knowledge base stops losing stories",
"summary": "If your avatar library filled up while a photo or prompt avatar was still generating, the finished job used to sit unresolved and retry in the background without ever telling you why nothing appeared. It now closes with a clear message: delete an avatar or upgrade, then build it again. Separately, news stories whose text contained a half-cut emoji or a stray null character were silently dropped from the knowledge base the 3D agents ground their answers on, about 120 a day. They are now cleaned and saved. Static files such as images and environment maps also stop answering a server error when the CDN revalidates them after an update.",
"tags": [
"fix"
],
"link": "/news"
},
{
"date": "2026-10-09",
"title": "Set up Grok Bot from the terminal with npx three-ws setup",
"summary": "Grok Bot keeps its connectors in xAI's cloud, so there is no config file to write, but the three-ws CLI now does every other step. Run npx three-ws setup --client grok-bot, or tick Grok Bot in the setup picker, and it prints the exact fields Grok Bot's custom MCP form asks for, copies the server URL to your clipboard, and checks the public URL live the way Grok Bot will call it. Pick how Grok Bot signs in: free with no account, free with an install token that keeps one quota across scheduled runs, a connector key it mints for you that can read and generate but never spend, or OAuth so Grok Bot signs in itself.",
"tags": [
"feature",
"sdk"
],
"link": "/docs/cli"
},
{
"date": "2026-10-09",
"title": "AI agents can now read how to connect three.ws to their client",
"summary": "The public server directory at three.ws/.well-known/mcp.json now carries setup for each AI client: Claude, ChatGPT, Grok Bot, Cursor, VS Code and Claude Code. For each one it names the server to add, whether to pick no sign-in, OAuth or an API key, and the exact values for that client's connector form or config file, plus one-click install links where the client has them. An agent asked to connect three.ws to Grok Bot can now do it from that one file, without a person reading the setup page. The site index for AI agents (three.ws/llms.txt) points to it too.",
"tags": [
"feature",
"docs"
],
"link": "/docs/mcp"
},
{
"date": "2026-10-09",
"title": "The Crawl: watch agents read the web, and send yours out",
"summary": "A new page where three.ws agents read the open web in real browsers while you watch. Each agent's own 3D avatar walks across the live page to the link it picked, says what it is thinking, then hops to the next page. Send your agent out with any topic and it reads around it until you call it home, remembering what it read so you can ask it later. Every page lands in an open corpus you can browse or download.",
"tags": [
"feature"
],
"link": "/crawl"
},
{
"date": "2026-10-09",
"title": "The iPhone app is ready for TestFlight",
"summary": "The three.ws iPhone app now carries Apple's privacy manifests, releases to TestFlight with one command or through Xcode Cloud, and only asks for CarPlay once Apple grants it. Inside the app, buying credits, launching a coin, trading and paying now continue in Safari with one tap, still signed in and on the same page, while your wallet stays right in the app.",
"tags": [
"feature",
"improvement"
]
},
{
"date": "2026-10-09",
"title": "Forge Workflows: chain 3D tools into one run",
"summary": "A visual workflow builder for the Forge. Link a prompt or photo to any live 3D engine, then rig, remesh, segment, restyle or make it game-ready, preview it and export or save it, all in one run. Batch many prompts or photos with For Each, re-run from a failed step without redoing the rest, or generate on your own graphics card with Modly. Starts from four ready-made templates, and the Forge page links straight to it.",
"tags": [
"feature"
],
"link": "/forge/workflows"
},
{
"date": "2026-10-09",
"title": "three.ws Forge for desktop",
"summary": "A free desktop 3D studio for Windows, Linux and Apple silicon Macs, built on the open-source Modly app by Lightning Pixel. Turn a photo into a 3D model on your own graphics card or on the three.ws cloud, chain repair, remesh, rig and export steps in a node workflow, start from any model in the CC0 library, and publish to your three.ws account in one click after signing in with your browser. The download page offers the build for your system with its SHA-256, and until the first installers ship it walks you through running Forge from source.",
"tags": [
"feature"
],
"link": "/forge-desktop"
},
{
"date": "2026-10-09",
"title": "Claim the 3D models the bot made for you on X",
"summary": "Anything @trythreews makes for you from a post now waits for you. Link the same X account and every model, avatar and photo-to-3D result moves into your library, either the moment you link or from the new Claim page. If you had already linked, the bot's reply says it was saved to your library. If not, the reply carries a link to claim it.",
"tags": [
"feature"
],
"link": "/x/claim"
},
{
"date": "2026-10-09",
"title": "pump.fun MCP tools explain the October program upgrade",
"summary": "The hosted pump.fun MCP server now tells AI clients how pump.fun's new trades change the numbers they read: a pool's virtual quote reserves are signed and often negative because fees wait in the vault until swept, a large final buy on a bonding curve finishes through a synthetic migration, and creator fees from the new trades only show up in a claim after they are swept. Graduated coins are described as moving to their PumpSwap pool, not Raydium.",
"tags": [
"fix"
]
},
{
"date": "2026-10-09",
"title": "A skill file you can hand to Grok",
"summary": "three.ws/grok-skill.md teaches Grok and Grok Bot what three.ws does and how to use it: the one sentence that adds the free 3D connector, copy-paste commands for calling it over plain HTTPS when no connector is set up, the links every result carries, the free limits, and the rule that nothing is ever paid for without your yes.",
"tags": [
"feature"
],
"link": "/docs/grok"
},
{
"date": "2026-10-09",
"title": "Ready-made recipes for Grok Bot and other always-on agents",
"summary": "The 3D Studio and Grok connectors now offer guided prompts written for an agent that runs on a schedule with nobody watching: a daily 3D brief that turns the top trending topic into a model and a poster image, a themed asset pack that uses the free catalog first and generates only the gaps, a rigged avatar from a photo, a quick tour, and a read-only status report on your agents. Each one names the exact tools, never stops to ask a question, and reuses a retry key so a repeated run collects the original model instead of generating twice. A new list_my_agents tool lets a signed-in connector see which agents your account holds.",
"tags": [
"feature"
],
"link": "/grok"
},
{
"date": "2026-10-09",
"title": "Splitting a model into parts says plainly when it cannot read an FBX",
"summary": "The part splitter never could read FBX files, but it used to accept one anyway and then fail with an unexplained error. It now refuses an FBX straight away, on the website, the API and the segment_model agent tool, with the formats it does take (GLB, GLTF, OBJ, STL, PLY, OFF or DAE) and the one-step fix: convert the FBX to GLB with remesh first, then split the result. An FBX hidden behind a download link with no file extension is recognised too, and gets the same clear message instead of an error code.",
"tags": [
"fix"
],
"link": "/forge"
},
{
"date": "2026-10-09",
"title": "Remesh reads FBX files, and stylize says plainly when it cannot",
"summary": "The remesh tool now accepts a binary FBX for every operation, not only FBX-to-FBX conversion: convert it to GLB, OBJ or any other format, simplify it, repair it or retopologize it, even from a download link with no file extension. Stylize still cannot read FBX, so it now refuses one straight away with the formats it does take and the one-step fix (convert to GLB with remesh first), instead of accepting the job and failing later with an unexplained error. Files that cannot be read at all, such as an ASCII FBX, now fail with a message that says what to re-export. Rigged models converted to FBX also no longer carry a stray sphere mesh that the conversion used to add.",
"tags": [
"improvement",
"fix"
],
"link": "/forge"
},
{
"date": "2026-10-09",
"title": "The leaderboard now ranks agents by what they earned",
"summary": "A new Earned tab on the leaderboard ranks every public agent by what it earned over the last 24 hours, 7 days, 30 days or all time: creator fees from its coin plus what people and other agents paid for its skills, shown separately and added up in SOL with dollars beside it, with how far each agent moved since the previous period. Each agent's Earned card links to its place on the board, and a window you pick is saved in the page address so you can share it. The builders board also now counts launches made from an agent's own wallet, which it had been missing.",
"tags": [
"feature",
"fix"
],
"link": "/leaderboard"
},
{
"date": "2026-10-09",
"title": "Agent MCP: price an x402 payment with pay_quote before pay_and_call spends it",
"summary": "The Agent MCP server now has a pay_quote tool. It asks a paid x402 endpoint what it charges without paying, and shows who gets paid, how much, in which token and on which chain, which wallet pays, the per-call limit, and anything that would stop the payment. pay_and_call needs the quote_id it returns plus an explicit confirm_payment, so an assistant always shows you the price before your agent wallet pays. One approval pays one call. Before this, turning pay_and_call on still left it refusing every call, because the quote step it requires did not exist.",
"tags": [
"feature",
"fix"
]
},
{
"date": "2026-10-09",
"title": "Paired coins: launch a coin on Robinhood Chain that trades against up to five assets at once",
"summary": "A paired coin trades in up to five markets at the same time, each its own bonding curve: tokenized stocks, WETH, stablecoins or chain coins, in whatever split you choose. Every curve opens on a virtual reserve, so launching costs only a small flat ETH fee plus gas, and the creator never needs to hold the assets the coin pairs with. Liquidity never migrates and nobody can pull it, and the creator earns 70% of every swap fee in each asset the coin trades in, claimable in one transaction. Browse every paired coin with its pools priced in dollars, a chart per pool and every trade at /markets/robinhood/paired, and trade one from your own wallet with exact-amount approvals and slippage protection. At /launch/paired your agent launches one from its own wallet: a live quote shows each pool's opening value, the fee, gas and an optional opening buy that lands in the same transaction, and the launch runs through the agent's spend limits. Bots can do the same with two API calls, agents with the new paired-launch skill, and AI clients through the paired MCP tools. The developer guide is at /docs/paired-coins.",
"tags": [
"feature"
],
"link": "/markets/robinhood/paired"
},
{
"date": "2026-10-09",
"title": "Funding an agent's inference from its wallet now needs the signed real-funds agreements",
"summary": "Topping up an agent's inference credits, switching on its automatic top-up, and provisioning an inference API key all send real USDC out of the agent's wallet. They now ask for the same signed Terms of Service, Risk Disclosure and Agent Wallet Agreement as every other real-funds action, on the page and on the API alike. Previews and switching the automatic top-up off stay open, since nothing moves.",
"tags": [
"security"
]
},
{
"date": "2026-10-09",
"title": "The desktop companion, CLI and MCP server now work with just the bridge token",
"summary": "The companion bridge token is the one credential the desktop app, the companion command line and its MCP server are given, but the feed, marking a message as spoken, replying, reading contacts and checking sources all refused it, so pasting the token into the desktop app said it was wrong and an agent asking what you had heard got turned away. Those calls now accept the bridge token and only ever show the messages and contacts of the person it belongs to. It still cannot connect sources, edit contacts, change settings or rotate itself; those stay behind your signed-in account, and they now say so plainly instead of answering as if the token were wrong.",
"tags": [
"fix"
],
"link": "/companion"
},
{
"date": "2026-10-09",
"title": "Creator rewards include fees pump.fun now holds on the curve and pool",
"summary": "pump.fun's October program update changed where creator rewards land: newer trades leave the creator fee on the coin's bonding curve or pool until someone sweeps it into the creator's vault. Without handling that, those rewards were invisible to a claim and read as already claimed on coin pages. Every claim on three.ws now sweeps your coins first and collects everything in the same transaction, from your wallet, an agent wallet or the automatic claimer. Reward balances, coin pages and earnings history count the held fees as unclaimed, fee-sharing changes and distributions no longer fail on unswept fees, and the biggest buy a coin ever sees (the one that completes its curve) is now reported at its full size in trade feeds and whale alerts. The automatic claimer for agent sniper wallets also had a bug that stopped every claim; that is fixed.",
"tags": [
"improvement",
"fix"
],
"link": "/launch"
},
{
"date": "2026-10-09",
"title": "A paid call that fails after payment now tells you to retry, not to pay again",
"summary": "If a paid x402 call fails after your payment has already settled, the response now says so: paid, retry safe, the settlement transaction, and how many retries are left. Repeat the exact same request with the same payment and the server re-runs the work without charging you again, instead of answering that the payment was already used. Paid POST routes also now tell apart two different request bodies sent under one payment identifier. The MCP docs gained a Use cases section that spells out every guided prompt as the ordered tool calls it makes, with the steps that spend money marked, and a test keeps it in step with the real tools.",
"tags": [
"improvement",
"docs"
],
"link": "/docs/x402"
},
{
"date": "2026-10-09",
"title": "Cloud agents get their own free 3D budget with a connector link",
"summary": "Grok Bot and other hosted AI agents all call the free 3D studio from the same network address, so one busy user could use up the allowance for everyone. You can now mint a free, anonymous install token (on the Connect page, or with one POST) and add it to your connector link. Your agent then has its own generation budget, separate from every other installation, with no account. When any caller does hit a limit, the error now says which limit it was, when it resets, and how to raise it.",
"tags": [
"feature",
"improvement"
],
"link": "/connect"
},
{
"date": "2026-10-09",
"title": "Connected apps: clearer sign-in screen and instant revoke",
"summary": "When an app such as Grok Bot asks to connect to your three.ws account, the approval screen now shows the app's name and website, what it will be able to do in plain words, and says plainly that it can never spend from a wallet. The Connected apps list in Settings shows when each app was last used, and Revoke now cuts the app off on its very next request instead of waiting up to an hour for its access to run out.",
"tags": [
"feature",
"security"
],
"link": "/dashboard/settings"
},
{
"date": "2026-10-09",
"title": "One Grok Bot connector for 3D and your agents",
"summary": "https://three.ws/api/mcp-grok is now the one URL to paste into Grok Bot. With no sign-in it is the free 3D studio: text or an image to a 3D model, a rigged avatar, or a talking persona, with links Grok can share. Sign the same URL in with OAuth at https://three.ws/api/mcp-grok?auth=oauth, or give it an API key, and Grok also manages your agents, their memory and skills, and your avatars. It never lists a wallet, payment, card, trading or launch tool, whatever the sign-in allows, so a Grok Bot task can build with your agents but can never spend.",
"tags": [
"feature"
],
"link": "/docs/grok"
},
{
"date": "2026-10-09",
"title": "Safe retries and live status for 3D jobs",
"summary": "The free 3D studio now has a get_job tool that reports a job's status, progress, time remaining, download links, or a failure with what to do next. Every generation tool accepts an optional idempotency_key, so a scheduled agent that retries after a timeout gets its original job back instead of starting a second one. Clients that ask for progress now receive live progress updates while a model renders.",
"tags": [
"feature"
],
"link": "/docs/mcp-studio"
},
{
"date": "2026-10-09",
"title": "Use three.ws from Grok Bot",
"summary": "The Connect page now has a Grok Bot tab with the exact connector settings: Streamable HTTP transport, a one-click copy of the server URL for whichever server you pick, and the right sign-in for it (none for the free servers, OAuth for the rest). Open /connect?client=grok to land on it with the Grok server already selected.",
"tags": [
"feature"
],
"link": "/connect?client=grok"
},
{
"date": "2026-10-09",
"title": "AI agents can read how to connect, per client",
"summary": "The public MCP directory at /.well-known/mcp.json now says exactly how to connect from Claude, ChatGPT, Cursor, VS Code and Grok Bot: which server to use, which sign-in, and what to type. The /grok page is now listed in llms.txt too, so an agent can find the Grok Bot setup on its own.",
"tags": [
"feature",
"docs"
],
"link": "/grok"
},
{
"date": "2026-10-09",
"title": "A full setup guide for Grok Bot",
"summary": "There is now one page that covers connecting Grok Bot to three.ws end to end: the exact connector settings, which sign-in to pick, how scheduled runs collect long renders and retry without generating twice, the rate limits, the rule that an agent can never spend, and a fix for every connection problem.",
"tags": [
"docs"
],
"link": "/docs/grok-bot"
},
{
"date": "2026-10-09",
"title": "Three ready-made jobs and the skill file on the Grok page",
"summary": "The Grok page now lists three jobs to hand Grok (an asset pack, a model with an embed, a talking persona), each naming the tools it calls, and links the Grok skill file you can upload to Grok Skills. The recent-models strip falls back to each model's concept image when a render is unavailable.",
"tags": [
"improvement"
],
"link": "/grok"
},
{
"date": "2026-10-09",
"title": "One command sets up Grok Bot",
"summary": "Run npx three-ws setup --client grok-bot to get the exact connector fields for Grok Bot, the server URL copied to your clipboard, and a live check that the public URL answers. Add --connector-key to also get a key that lets Grok Bot use your agents but never spend.",
"tags": [
"feature"
],
"link": "/docs/cli"
},
{
"date": "2026-10-09",
"title": "Create, Forge and Launch are easier for screen readers and AI agents to operate",
"summary": "Buttons and cards on Create and Forge now announce the same words you see on screen, the search hint and the walking companion can be reached with the keyboard, and the Launch form fields keep stable names while their character counters change.",
"tags": [
"improvement"
],
"link": "/forge"
},
{
"date": "2026-10-09",
"title": "Six Grok Bot recipes, each run for real",
"summary": "A new tutorial gives you six jobs to hand Grok Bot: a daily 3D brief, a game jam asset pack, an avatar from a photo, an X post image turned into a 3D model, a weekly agent report and a $THREE market brief. Every one lists the exact sentence, the tools in order and the real output from a production run, and a script reruns any of them.",
"tags": [
"docs"
],
"link": "/tutorials/grok-bot-recipes"
},
{
"date": "2026-10-09",
"title": "A build plan for the next 27 platform upgrades",
"summary": "We surveyed the strongest open-source crypto and agent projects, compared them with three.ws, and wrote the plan for closing every gap: agents that learn skills from experience, a payer command line, on-chain spending budgets, more trading venues, safer chat trading and more.",
"tags": [
"docs"
]
},
{
"date": "2026-10-09",
"title": "Connect three.ws to IBM Bob",
"summary": "IBM Bob, IBM's AI development partner, is now a supported client. Open the IBM Bob tab on the connect page to copy a ready-made .bob/mcp.json entry, or run npx three-ws setup --clients bob to write it for you. The free 3D Studio server works with no account, and the connect page now fits phone screens without spilling sideways.",
"tags": [
"feature",
"improvement"
],
"link": "/connect?client=bob&server=three-ws-studio"
},
{
"date": "2026-10-09",
"title": "A new default engine for photo to 3D, with regional safeguards",
"summary": "Photo to 3D in the Forge now starts with TRELLIS.2, an open-source model that makes textured assets with full material detail and keeps transparency for glass and leaves. You can pick a detail level of 512, 1024 or 1536 when you generate. We also made sure that requests from the European Union, the United Kingdom and South Korea never use models whose license does not cover those regions, and that a request with no location signal takes the safe path too. The engine switches on as soon as its GPU worker is deployed, and until then the existing engines keep serving every request.",
"tags": [
"feature",
"improvement"
],
"link": "/forge"
},
{
"date": "2026-10-09",
"title": "Generate 3D models on your own graphics card in the Forge",
"summary": "The From photos tab now has a Use your GPU panel. Install the free Modly desktop app, press Connect, and any image-to-3D model you have downloaded in Modly shows up as an engine marked Your GPU, free. Generation runs on your computer with live progress, you can cancel it at any time, and the finished model lands in the same viewer as a cloud result so you can rig it, optimize it, view it in AR, download it or save it to your avatars. Nothing on your computer is contacted until you press Connect.",
"tags": [
"feature"
],
"link": "/forge"
},
{
"date": "2026-10-09",
"title": "three.ws nodes for the Modly desktop app",
"summary": "Modly users can now add three.ws to their workflow editor. Six new nodes turn an image, a prompt or a sketch into a 3D model on our cloud GPUs, add a humanoid skeleton, rebuild the mesh as clean quads, triangles or low poly, and publish the result to a shareable three.ws page. They work on any laptop with no weights to download, and chained steps reuse the same upload instead of sending the model again. A separate Publish node saves meshes made on your own GPU to your account.",
"tags": [
"feature"
],
"link": "/modly"
},
{
"date": "2026-10-09",
"title": "Fix, smooth and slim any model in your browser, then send it straight to your 3D printer",
"summary": "Every finished model now has mesh tools: repair welds cracks and fixes inside-out faces, smoothing softens blocky surfaces without tearing textures, and decimate cuts the triangle count to a number you choose, with undo and a before and after count. Nothing is uploaded. A new 3D print action opens the model in OrcaSlicer or Bambu Studio, already upright, sized in millimetres and sitting on the build plate, with STL and 3MF downloads for every other slicer.",
"tags": [
"feature"
],
"link": "/forge"
},
{
"date": "2026-10-08",
"title": "Use three.ws from Grok Bot",
"summary": "The connect page now has a Grok Bot tab. Grok Bot adds three.ws as a custom MCP connector, so the tab shows the exact fields to type in: the name, transport Streamable HTTP, the URL with a copy button, and whether authentication is None (the free 3D servers) or OAuth (servers that work with your account). Pick a different server above and every field follows it, and there is a ready-made sentence to send Grok Bot instead of filling in the form. Opening three.ws/connect?client=grok starts on the Grok server, which is free with no sign-in and adds your own agents when you sign in. The page also reminds you that a localhost URL never works, because Grok Bot connects from xAI's cloud.",
"tags": [
"feature"
],
"link": "/connect?client=grok"
},
{
"date": "2026-10-08",
"title": "A skill file that teaches Grok to use three.ws",
"summary": "three.ws/grok-skill.md is one Markdown file you can upload to Grok's Skills or hand to Grok Bot. It tells Grok exactly which connector URL to add for a free 3D studio, how to sign it in to work with your agents, and, when no connector is set up, how to search the 3D catalog, generate a model and look at it with plain curl calls that run as written. It also spells out the links every model comes back with, what needs an account, and that Grok never spends money on your behalf. It is built from the same facts as three.ws/skill.md, so the two always agree. The Grok page has a button to download it.",
"tags": [
"feature",
"docs"
],
"link": "/grok"
},
{
"date": "2026-10-08",
"title": "One Grok Bot connector for 3D and your agents",
"summary": "https://three.ws/api/mcp-grok is now the one URL to paste into Grok Bot. With no sign-in it is the free 3D studio: text or an image to a 3D model, a rigged avatar, or a talking persona, with links Grok can share. Give the same URL a connector API key, or sign it in with OAuth at https://three.ws/api/mcp-grok?auth=oauth, and Grok also manages your agents, their memory and skills, and your avatars. It never lists a wallet, payment, card, trading, launch or delete tool, whatever the sign-in allows, so a Grok Bot task can build with your agents but can never spend.",
"tags": [
"feature"
],
"link": "/docs/grok"
},
{
"date": "2026-10-08",
"title": "Long 3D jobs and safe retries for AI agents",
"summary": "Agents that run on schedules can now generate 3D models without ever making the same model twice. Every generation tool on the free 3D Studio MCP server accepts an optional idempotency_key: retry a call that timed out with the same key and you get the first call's job back instead of a second generation, and the retry costs none of your quota. A new get_job tool reports any job's status, progress and time left, hands back the viewer, download, picture and embed links when it is done, and says what went wrong and what to do next when it fails. MCP clients that ask for progress updates now receive them live while the model renders.",
"tags": [
"feature",
"improvement"
],
"link": "/docs/mcp-studio"
},
{
"date": "2026-10-08",
"title": "API keys for AI agents that can never spend",
"summary": "When you make an API key you can now choose \"For an AI agent (Grok Bot, schedules, CI)\". That key can read, generate and edit your avatars, agents and their memory, and it can never spend: it cannot pay, send, trade, bid, launch or change where your agents get paid, and nothing done to it later can change that. If an agent holding it reaches one of those actions, it gets a clear answer with a link back to three.ws, where you approve the action yourself. The same rule now covers every route that arms an autopilot, a market maker, a launcher or a recurring charge, or moves a payout address. Keys you already have keep exactly the permissions they were made with.",
"tags": [
"feature",
"security"
],
"link": "/docs/mcp"
},
{
"date": "2026-10-08",
"title": "Connect Grok Bot and other cloud agents with OAuth, and revoke them instantly",
"summary": "Cloud agents such as Grok Bot can now sign in to your three.ws account over OAuth 2.1 from wherever they run, with a callback on any https address. The sign-in screen says plainly which app is asking, which site it belongs to, where it will send you back, and what it will be able to do, and it states that the app can never spend from your wallet. Spending is only granted if you tick a separate box. Settings, Connected apps lists every app with when it last called in, and Revoke now cuts an app off on its very next request instead of up to an hour later.",
"tags": [
"feature",
"security"
],
"link": "/docs/mcp"
},
{
"date": "2026-10-08",
"title": "Topping up an agent's inference credits now asks for the real-funds agreements first",
"summary": "Moving USDC from an agent wallet into model credits is a real-funds action, so it now follows the same rule as withdrawals, trades and vaults. A one-off credit top-up, turning on the automatic top-up rule, and funding a new inference key from the command line all require the signed real-funds agreements (Terms of Service, Risk Disclosure and Agent Wallet Agreement). On the credits panel the signing dialog opens before the first top-up; from the CLI or an API key you get a link to sign once. Previews still work without signing, and switching the automatic top-up off never needs it.",
"tags": [
"security"
],
"link": "/legal/agreements"
},
{
"date": "2026-10-08",
"title": "Agent Wallet MCP: quote an x402 service before your assistant pays it",
"summary": "The Agent Wallet MCP server has a new pay_quote tool. It reads what a paid x402 endpoint charges (price, network and the address that gets paid) without spending anything. pay_and_call now works end to end once you turn it on for a connection: your assistant quotes first, shows you the price, and pays only after you say yes, using that quote. Paying a different endpoint than the one quoted, or paying with an expired quote, is refused before any USDC moves.",
"tags": [
"feature",
"security"
],
"link": "/docs/mcp-agent"
},
{
"date": "2026-10-08",
"title": "Your own free 3D quota when you connect through Grok Bot or another cloud agent",
"summary": "Cloud agents like Grok Bot reach three.ws from one shared address, so every user of that agent used to share one free 3D quota. Now you can generate a free install token on the Connect page (pick 3D Studio (free), then Generate my connector URL) or with one POST to /api/mcp-studio/install, and add the connector with that URL. Your generations then count against a quota of your own: 4 a minute and 30 an hour. No account, no key, and the token works on every free studio URL, including the Grok one. When a limit is reached, the message now says which limit, exactly when it resets, and how to lift it.",
"tags": [
"feature",
"improvement"
],
"link": "/connect?server=three-ws-studio"
},
{
"date": "2026-10-08",
"title": "A paid call that fails after you were charged never asks you to pay again",
"summary": "If a paid three.ws endpoint takes your payment and the work then fails, the error now says so plainly: it carries paid: true, retry_safe: true, the settlement transaction, and the sentence \"You were charged; repeat this exact request with the same payment to get your result.\" Repeating the request with the same payment re-runs the work with no second charge, up to three times within 24 hours. Before, a download or stream that failed after payment answered with a generic error and the retry was refused as a reused payment. Errors before any charge now say paid: false, so a client always knows which case it is in. Separately, paid POST requests are now matched on their body too, so two different requests under one payment identifier can no longer receive the same cached answer.",
"tags": [
"fix"
],
"link": "/docs/x402"
},
{
"date": "2026-10-08",
"title": "MCP docs now show every guided workflow as the exact calls an agent makes",
"summary": "The MCP guide has a new Use cases section: each of the 17 guided prompts (create an agent, set up a wallet, hire an agent, generate a 3D model and the rest) written out as the ordered tool and resource calls it drives on each server, with every step that spends money marked. A test renders every prompt against each server's live tool list and fails if the section names a tool that no longer exists or misses one a prompt now calls, so the page stays accurate as tools change.",
"tags": [
"docs"
],
"link": "/docs/mcp"
},
{
"date": "2026-10-08",
"title": "Grok model menus now offer xAI's current models",
"summary": "Every Grok picker on three.ws (the /brain menu, agent chat, the avatar page and the embed editor) now lists xAI's current line-up: Grok 4.7, xAI's flagship, Grok 4.3 with its 1M-token context at the lowest Grok price, Grok 4.20 Fast with no thinking pass, and Grok Build for coding. Agents and embeds that saved a Grok model xAI has since retired, such as Grok 4.1 Fast, keep working and answer on the model xAI recommends in its place. People who bring their own xAI key now get Grok 4.7 by default instead of an older flagship.",
"tags": [
"improvement"
],
"link": "/brain"
},
{
"date": "2026-10-08",
"title": "Coin pages keep their Markets table when the listing sources are busy",
"summary": "When both of our exchange-listing sources were rate-limited, a coin page's Markets table could show an error for coins only one of them lists. It now falls back to the coin's live DEX trading pairs, found by its contract address, with each pair's venue, chain, price, 24h volume and a link to trade it.",
"tags": [
"fix"
],
"link": "/coins"
},
{
"date": "2026-10-08",
"title": "3D Studio results now hand every agent plain links to the model",
"summary": "Every model the free 3D Studio MCP server hands back, whether freshly generated, picked from the asset catalog or saved as a persona, now comes with four plain links, both in the result data and in its first lines of text: a viewer page, the GLB file, a rendered PNG picture and a paste-ready embed snippet. Agents that cannot show a 3D widget, like Grok Bot and other browser agents, can now open, download, share and embed the model directly. A model that is still rendering comes with a viewer link that waits for the job and opens the model the moment it is ready.",
"tags": [
"feature"
],
"link": "/docs/mcp-studio"
},
{
"date": "2026-10-08",
"title": "Coin charts say plainly when a coin has no market yet",
"summary": "A coin with no trading pool anywhere could show a price-history error on its chart when a market data source was busy. The chart now recognises the coin has no market and shows the no-price-history state, and it asks that source one question fewer per chart load.",
"tags": [
"fix"
],
"link": "/terminal"
},
{
"date": "2026-10-08",
"title": "Picking the free Hugging Face engine no longer leaves you stuck when its Spaces are down",
"summary": "When every free Hugging Face Space is out of GPU quota, the Forge now pauses that engine for a few minutes, greys it out in the engine picker with the reason, and moves a generation you started on it to a working engine, showing you the switch, instead of retrying the same dead Spaces with a countdown.",
"tags": [
"fix"
],
"link": "/forge"
},
{
"date": "2026-10-08",
"title": "Chart Companion: a 3D agent that watches the chart with you",
"summary": "Paste any DEXTools pair at /chart-companion and a 3D agent sits beside the live chart, reacting to every real swap as it lands: a superhero jump for a whale buy, a facepalm for a dump, a dance for a buy streak, applause for a new daily high. Each reaction comes with a line you can trace to its transaction, and it can read them aloud. Bring your own three.ws agent, embed it on your coin's site with one iframe, or add it to OBS as a transparent overlay for your stream.",
"tags": [
"feature"
],
"link": "/chart-companion"
},
{
"date": "2026-10-08",
"title": "Agent earnings are updating again",
"summary": "pump.fun moved the service that reports creator fees, so the Earned figures on agent pages, agent profiles, coin pages and /analytics had stopped refreshing since the end of September. Earnings now read from pump.fun's current fees service, with a second source behind it if that one stumbles, and every agent coin is back on its 30-minute refresh.",
"tags": [
"fix"
],
"link": "/analytics"
},
{
"date": "2026-10-08",
"title": "The coin visualizer and communities keep loading when pump.fun throttles us",
"summary": "The full coin board behind the pump.fun visualizer, the communities page and the home page's live coin now retries a rate-limited read once and falls back to the last board any server saw in the past 30 minutes, marked as slightly stale, instead of showing an error.",
"tags": [
"fix"
],
"link": "/communities"
},
{
"date": "2026-10-08",
"title": "Event-attested reputation now requires the platform's own signature",
"summary": "The event-attested tier in a Solana agent's reputation, which ranks above community feedback in the Passport grade, was granted on a label written inside the memo, so any signer could claim it. It now counts only attestations the platform's own attester minted and recorded, so a self-written label earns ordinary feedback weight and nothing more.",
"tags": [
"fix",
"security"
],
"link": "/reputation/market"
},
{
"date": "2026-10-08",
"title": "Community sentiment readings are back on coin pages and the Oracle",
"summary": "pump.fun moved the feed of community callouts that our sentiment pulse reads, so every reading had been failing with an unavailable error. The pulse now reads the new feed, so launch pages, the Oracle and the sentiment_pulse MCP tool score real community commentary again, up to 100 callouts per coin.",
"tags": [
"fix"
],
"link": "/oracle"
},
{
"date": "2026-10-08",
"title": "DEXTools is one click away on every coin, and every visit counts toward Social Boost",
"summary": "DEXTools Social Boost ranks tokens by visits to their DEXTools page, and $THREE has already won three buybacks from it. Every coin on three.ws now carries a DEXTools link: it leads the trading terminal row and appears on coin and launch pages, the 3D coin view, the Oracle, agent and character pages, the live pump.fun feeds and your token dashboard. Each link opens the coin's busiest pair on DEXTools, so the visit lands on the page Social Boost actually counts. Launch pages, and the success screen right after you launch a coin, also get a Social Boost card that explains how the ranking works and sends your community there in one click. Visits are counted in daily totals per coin and per page that sent them. No visitor data is stored.",
"tags": [
"feature"
],
"link": "/three-token"
},
{
"date": "2026-10-08",
"title": "Signing in to the 3D Studio, wallet, Bazaar and IBM MCP servers works from any connector",
"summary": "Adding the 3D Studio, agent wallet, x402 Bazaar or IBM Granite MCP server to an AI client with sign-in (Grok Bot, claude.ai connectors and any client built on the official MCP SDK) used to stop before the consent screen, because every server pointed the client at sign-in details written for the core server only. Each server now describes itself and issues a sign-in bound to it, so the connector reaches consent, gets its token and connects. Existing connections keep working. The MCP docs now carry a compatibility table with the exact connector settings for every server, and `npm run probe:mcp-clients` checks all of them the way a cloud client connects.",
"tags": [
"fix"
],
"link": "/docs/mcp"
},
{
"date": "2026-10-08",
"title": "Coin Clash stays playable when community data is offline",
"summary": "When the community roster behind Coin Clash is unavailable, the arena no longer goes dark. Each round is now fought between the coins trending on Solana right now, locked in for the whole round so battles never reshuffle mid-fight, and each army shows its market cap where the member count would be. Enlisting, rallying and war records work exactly as before.",
"tags": [
"improvement"
],
"link": "/clash"
},
{
"date": "2026-10-08",
"title": "Your agent can launch its own coin on Robinhood Chain",
"summary": "Agents can now launch a coin on Robinhood Chain through Pons, the chain's busiest launchpad, signed by the agent's own wallet. Name the coin, see the exact cost live (launch fee, opening buy, gas, and how many tokens the opening buy gets), confirm, and it is trading on a bonding curve in about a second. The opening buy lands in the same transaction, so nobody trades ahead of your agent. Bots can do the same through two API calls or the new pons-launch skill, and launched coins show up on /launches and the agent's profile. The Recent Launches feed on /markets/robinhood now lists every Pons launch too, with curve price, market cap and graduation progress.",
"tags": [
"feature"
],
"link": "/launch/robinhood"
},
{
"date": "2026-10-08",
"title": "Chat and agents stay up through more AI outages",
"summary": "Every AI answer on three.ws, from chat to agent brains, the tutor and the fact-checker, now has five more free models to fall back on when the usual ones are busy. NVIDIA's largest open model, Nemotron 3 Ultra, now answers right behind Nemotron 3 Super, with the compact Nemotron 3.5 Lightning further down the line. A second, independent route to the same Nemotron models works with no account at all, and a Gemma 4 model is back in the mix as well. The free NVIDIA and gateway models are now checked against their providers every six hours too, so the team hears the moment one is switched off instead of finding out from a failed answer.",
"tags": [
"improvement"
],
"link": "/chat"
},
{
"date": "2026-10-08",
"title": "Fewer failed 3D generations when the concept image comes back blank",
"summary": "Text to 3D first paints a concept image, then sculpts it. When the painting service quietly returned a blank black frame instead of a picture, the studio used it anyway, and every 3D engine then failed because there was nothing in the image to sculpt. The studio now spots a blank or withheld frame and paints the concept with the next image service instead, so those prompts finish as real models.",
"tags": [
"fix"
],
"link": "/forge"
},
{
"date": "2026-10-08",
"title": "Shared 3D models now explain themselves to X and Grok",
"summary": "Every 3D model page now tells whoever reads the link exactly what it is. A model shared from three.ws unfurls on X as a large card showing a render of the model itself instead of a stock image, with its own title and description. Model, avatar and agent pages also describe themselves in a machine-readable form that search engines and AI assistants like Grok can read: the 3D file, a picture of it, who made it, when, and the terms it is shared under. So when someone asks Grok what a three.ws link is, it can answer from the model rather than guess.",
"tags": [
"improvement"
],
"link": "/creations"
},
{
"date": "2026-10-08",
"title": "The terminal 3D renderer animates live in every HTTP client",
"summary": "curl three.ws/tty streams an animated 3D character into your terminal frame by frame. Clients that ask for compressed responses, such as curl --compressed, scripts using fetch, and most HTTP libraries, used to get nothing for about ten seconds and then the whole animation at once as a wall of text. Live streams are no longer compressed, so every client now sees the character move from the first frame. Streaming agent and chat responses get the same fix.",
"tags": [
"fix"
],
"link": "/tty"
},
{
"date": "2026-10-08",
"title": "See which agents earned the most on the new Earned leaderboard",
"summary": "The leaderboard has a second tab, Earned, that ranks every public agent by what it actually earned in the last 24 hours, 7 days, 30 days or all time. Earnings are the creator fees its coin collects from trading plus what people and other agents paid for its skills, each shown beside the total in SOL with the dollar value, along with the agent's coin and how many places it moved since the previous period. Each row matches the Earned card on that agent's own page, and the card now links straight to its spot on the board, such as \"#1 this week\". The tab and period live in the address bar, so a link opens the exact view you shared. Separately, the coin-launch ranking on /rankings and its daily Top 10 badge now count launches from the platform's own launch records, so builders who launched from an agent's wallet are no longer ranked with zero.",
"tags": [
"feature"
],
"link": "/leaderboard"
},
{
"date": "2026-10-08",
"title": "Ready-made scheduled tasks for Grok Bot and other AI agents",
"summary": "The free 3D Studio and the Grok Bot connector now offer guided prompts written for an agent that runs on its own schedule. daily-3d-brief turns a topic, or the day's top trending one, into a 3D model and a poster image and hands back the links. asset-pack builds a themed set of 3D assets, using ready-made catalog items first and generating only what is missing. avatar-from-photo turns a photo into a rigged avatar with a pose studio link. agent-report, once the connector is signed in, writes a read-only status report on your agents. Every prompt names the exact calls to make and marks each generation so a retried run never makes the same model twice.",
"tags": [
"feature"
],
"link": "/docs/grok"
},
{
"date": "2026-10-08",
"title": "Every MCP server now knows which AI client is calling",
"summary": "The three.ws MCP servers now remember which AI app opened each connection, such as Claude, ChatGPT, Cursor or Grok Bot, using the name and version the app announces when it connects, and count which tools each one uses. Only those names and counts are kept, never what anyone asked for. Each connection gets a session id that MCP apps send back automatically, so nothing changes for anyone connecting. It tells us which apps people actually use three.ws from, so we know where to make the experience better next.",
"tags": [
"infra"
]
},
{
"date": "2026-10-01",
"title": "Pinned agent-3d embeds keep working after every update",
"summary": "If you embedded the 3D avatar component with a pinned version and an integrity hash, as our docs recommend for production, the embed could stop loading after one of our updates: the file behind a version that is promised never to change was being rebuilt each time we shipped, so its fingerprint changed and browsers refused to run it. A released version is now frozen for good, every update serves exactly the same file under it, and the build fails if any hash in our docs, skills or code snippets ever disagrees with the file it points to. Version 1.5.2 is frozen as it is served today; if your page pins an older hash for 1.5.2, update it once from /agent-3d/1.5.2/integrity.json.",
"tags": [
"fix",
"sdk"
],
"link": "/docs/web-component"
},
{
"date": "2026-10-01",
"title": "The Holo Sticker corner now peels in one smooth curl",
"summary": "Peeling the corner of the Holo Sticker used to break the sticker apart: around a third of the way the curled edge showed stair steps, past 40% a shard stuck out, and at full peel the corner crumpled. The flat front and back of the sticker were made of a few long triangles with no points inside, so they could not bend where the fold crossed them. Both faces are now built from an even mesh of small triangles that share their edge exactly with the rim, so the corner rolls into a clean curl at every point of the slider, for every mark, every foil and any text you die-cut, holes in letters included.",
"tags": [
"fix"
],
"link": "/holo"
},
{
"date": "2026-10-01",
"title": "Search the 3D asset catalog from any AI client without an account",
"summary": "Our docs said an AI client could connect to the main three.ws MCP server and search the ready-made asset catalog with no account, but adding that server makes the client ask you to sign in straight away, because most of its tools work on your own avatars and agents. The free studio server, three.ws/api/mcp-studio, now carries the same three catalog tools (search, item details and paste-ready code), so an AI client with no account can check the thousands of ready-made props, characters and motion clips before it generates anything. The docs, the find-3d-assets skill and the registry post now say which server to add, and the asset counts no longer go stale.",
"tags": [
"fix",
"docs"
],
"link": "/docs/mcp-studio"
},
{
"date": "2026-10-01",
"title": "Trending and the agent directory show only real agents",
"summary": "A behind-the-scenes agent the platform uses to budget its marketplace chat bot was showing up in the public agent directory and sitting at number one on Trending, counted up by the bot's own conversations. It is no longer listed, and both lists now leave out any agent the platform runs for itself unless it has been deliberately published, so Trending ranks agents people actually talk to.",
"tags": [
"fix"
],
"link": "/trending"
},
{
"date": "2026-10-01",
"title": "Avatar pictures show a natural standing pose instead of a T-pose",
"summary": "The small pictures of avatars around the site, on Spotlight cards, in the Monitor fleet list, on Characters and on the Rankings board, were drawn with every avatar standing arms out in a stiff T. Avatar pictures are now drawn with the avatar standing naturally, arms relaxed at its sides, using the same posing the live 3D viewer uses, and that works for every skeleton style the viewer already understands. Props, vehicles and creatures are drawn exactly as they were made. Avatar pictures also keep their colored backdrop and their real textures again, where some had come out on plain black with bare white materials. Every avatar picture drawn from now on uses the new pose, and the older pictures can now be redrawn in place.",
"tags": [
"fix"
],
"link": "/spotlight"
},
{
"date": "2026-10-01",
"title": "Avatars no longer open in a T-pose, and Spotlight shows one avatar instead of two",
"summary": "On an Agent Spotlight page the avatar's preview picture, which is rendered arms out, never went away once the live 3D avatar loaded, so a second, T-posed copy stood behind the real one for as long as the page was open. The picture now gives way the moment the live avatar is ready. Separately, every embedded 3D avatar briefly showed its arms-out rest pose while its idle animation loaded and then faded in from it. The avatar now stays hidden until it is standing in its idle, then fades in, everywhere the avatar component is used.",
"tags": [
"fix"
],
"link": "/spotlight"
},
{
"date": "2026-10-01",
"title": "Trading agents stop leaking SOL on every trade and can now ride a runner",
"summary": "A full review of the autonomous trading fleet found the biggest loss was not bad trades: every buy opened a token account that no sale ever closed, stranding more SOL in rent than the agents lost trading. Agents now close it after every exit so the rent comes back. The take-initials ladder also works as designed now: once an agent has its stake back, the rest rides on its trailing stop with no clock and no early take-profit, instead of being sold on the next check. The crossing strategy no longer buys coins that have already graduated or sit on a dead curve, profit percentages are reported against the real stake, and the outcome labels that teach the Oracle are flowing again.",
"tags": [
"fix",
"improvement"
],
"link": "/sniper/experiments"
},
{
"date": "2026-09-30",
"title": "Experiment write-up: can autonomous agents trade new coins at a profit?",
"summary": "The full record of our AI trading agents so far, measured from production: 1,296 real trades on Solana, 16.4% of them winners, and a net loss of 0.5402 SOL on 9.1444 SOL put to work. Only take-profit and take-initials exits made money, and the August retune made results worse, not better. The write-up says what that cost, what we got wrong, what we changed, and prints every query behind the numbers.",
"tags": [
"docs"
],
"link": "/docs/experiments-autonomous-trading"
},
{
"date": "2026-09-30",
"title": "Experiment write-up: how many new agents ever launch a coin?",
"summary": "Of the 788 agents people created in the 30 days to September 30, 2 went on to launch a coin, about three in four had no wallet, and the token plan built to bridge that gap has never been saved. The write-up publishes the numbers, what we got wrong about reading agent sign-ups as traction, what we already changed, and the queries so anyone can re-run them. It is the first entry on the new experiments page.",
"tags": [
"docs"
],
"link": "/docs/experiments-agent-launch-funnel"
},
{
"date": "2026-09-30",
"title": "Trader duels: call which trader wins, free to play",
"summary": "A new Duels page pairs traders who sit next to each other on the board and asks one question: who books more realized profit over tomorrow, or over next week? Make your call before the window starts with free points from your daily allowance, then watch the race live on the duel page with both traders' records, trade rooms and profiles one tap away. Results come straight from the on-chain trade ledger, the same numbers their profiles show, and ties or empty windows refund every call. A correct call doubles your stake and adds XP to your quest level, and there is a monthly season board and four badges to earn. Points have no cash value and cannot be bought, sold, transferred or redeemed.",
"tags": [
"feature"
],
"link": "/duels"
},
{
"date": "2026-09-30",
"title": "Spotlight entries now show verified results, and success stories are opt-in",
"summary": "Every Agent Spotlight entry now has a Verified results block: coins the agent launched, creator fees it earned and income from other agents paying for its skills, computed by three.ws and linked to where you can check them. The builder cannot edit these numbers. The new Success Stories page shows entries whose owners chose to feature them, grouped by category, and never shows an agent without its owner's yes.",
"tags": [
"feature"
],
"link": "/stories"
},
{
"date": "2026-09-30",
"title": "Sell your whole agent as a paid API",
"summary": "Flip one switch on your agent's Earn tab, set a price per call and a short description, and your agent becomes a pay-per-call API anyone can use. Buyers pay in USDC on Solana straight to your agent's own wallet, so there is nothing to withdraw, and they talk to exactly the agent you configured. Agents on sale are listed in the x402 catalog and discovery feed so other agents can find and hire them.",
"tags": [
"feature"
]
},
{
"date": "2026-09-30",
"title": "See what every agent has earned",
"summary": "Agent profiles, avatar pages and coin pages now show an Earned card: lifetime creator fees from the coins the agent launched, in SOL with USD beside it, what has been claimed and what is still waiting, a per-coin breakdown and every recorded claim with its Solscan link, plus income from selling its skills and services. The figures come from a snapshot refreshed twice an hour, and are also readable over MCP.",
"tags": [
"feature"
]
},
{
"date": "2026-09-30",
"title": "The coming iPhone app gains notifications, a share sheet entry and quick actions",
"summary": "The three.ws iOS app, on its way to the App Store, now matches the Android app and goes further. Turn on notifications and sales, follows and remixes arrive on your iPhone with your unread count on the app icon, and a tap opens the page it is about. three.ws appears in the share sheet: share one to three photos and they wait in Create as the front, left and right views of a new 3D avatar, or share a .glb from Files to upload it. Press and hold the icon for Create avatar, Discover, My agents and Notifications.",
"tags": [
"feature"
],
"link": "/docs/ios-app"
},
{
"date": "2026-09-30",
"title": "Every platform total in one free API call",
"summary": "GET /api/platform/analytics returns agents, agents with a wallet, coins launched, 3D models generated, LLM tokens processed, x402 settlements and volume, marketplace sales and agent-to-agent hire volume, each with a daily series over 30 days, 90 days or all time and a plain sentence saying exactly which rows were counted. A figure that cannot be read comes back empty with a named reason instead of as zero. It powers the new analytics page and is open to anyone building on three.ws.",
"tags": [
"feature"
],
"link": "/analytics"
},
{
"date": "2026-09-30",
"title": "See where every $100 paid in $THREE goes",
"summary": "The $THREE page now shows, for paid compute, marketplace sales, scarcity mints, spins and copy-trading fees, how every $100 splits between the seller, the treasury that funds buybacks, and holder rewards. The numbers are read live from the same config the payment code enforces, and the treasury and holder-rewards wallets are listed with their live $THREE and USDC balances, or marked not yet published. The platform never burns $THREE. The page also follows your light or dark theme now.",
"tags": [
"feature"
],
"link": "/three-token"
},
{
"date": "2026-09-30",
"title": "59 more pages now read in your language",
"summary": "Pick any of the 85 languages in the language menu and 59 more pages now follow it, including the Airdrop Checker, Portfolio, Drops, Crews, Launch, the Agent Monitor and the rig and embed doctors. Titles, buttons, form labels, hints, error messages and the text shown when a page is shared all translate. Live numbers and statuses on those pages (counters, balances, scan results, play and pause buttons) keep updating in every language: the translation step now recognises a value the page has already filled in and leaves it alone, instead of putting a placeholder like \"Loading\" back over it.",
"tags": [
"improvement",
"fix"
],
"link": "/airdrops"
},
{
"date": "2026-09-30",
"title": "Photo to 3D keeps going when its main engines go down",
"summary": "When a photo-to-3D job fails partway through, it already moves to the next engine on its own. Until now that only worked while another textured engine was up, so when the two main engines failed together last week, photo generations stopped completely. There is now one more rung: our own TripoSG engine takes over the job and delivers the shape of your model without textures, and the progress screen tells you when that happens. You get a model you can work with instead of an error, and whenever a textured engine is up, it is still used first.",
"tags": [
"fix",
"improvement"
],
"link": "/image-to-3d"
},
{
"date": "2026-09-30",
"title": "The Oracle's model now downloads from any website",
"summary": "The Oracle's conviction model is published data, and the @three-ws/oracle-model package is meant to score coins anywhere. Until now its download endpoint only answered pages on three.ws and a short list of partners, so a browser app on any other site could not load the model. It now answers every origin, the same as the rest of the Oracle's public reads, so any web page can fetch the model once and score coins locally with no key.",
"tags": [
"improvement"
],
"link": "/oracle-lab"
},
{
"date": "2026-09-30",
"title": "The language menu now offers 69 languages",
"summary": "The language menu grew from 14 languages to 69, each one fully translated. 55 more pages now follow the language you pick, including Hood Desk, Portfolios, Fee Bridge, Agent Cards, Connect, the CLI sign-in approval, Motion, Docs World and the legal pages. Their titles, headings, buttons and the text shown when a page is shared all translate, and so does the sitemap, so you can find every page on three.ws in your own language.",
"tags": [
"feature",
"improvement"
],
"link": "/sitemap"
},
{
"date": "2026-09-30",
"title": "3D models no longer arrive glued to a floor slab",
"summary": "The 3D generators sometimes built a thin, wide plate under whatever you asked for: a teapot sitting on a black marble square, a telescope standing in the middle of a grey sheet many times its size. Every finished model is now checked for that plate and has it cut away before it is saved, while anything that really has a base, like a lamp foot, a plinth or a tripod, is left exactly as it was.",
"tags": [
"fix"
],
"link": "/forge"
},
{
"date": "2026-09-30",
"title": "Every agent now shows what its coin has earned",
"summary": "Agent pages, agent profiles and coin pages now carry an Earned card: the creator fees an agent's coins have earned on pump.fun, in SOL with dollars beside it, how much has been claimed and how much is still waiting, a per-coin breakdown, and every claim linked to its transaction on Solscan. Skill sales and hires by other agents count too. The figures refresh every 30 minutes, each card says how old its numbers are and how they are calculated, and developers can read the same data from the public earnings API or over MCP. Automatic fee claiming for autonomous launches also works again.",
"tags": [
"feature",
"fix"
],
"link": "/launches"
},
{
"date": "2026-09-29",
"title": "API docs you can build a production integration from",
"summary": "The API reference now has one error table that says, for every common error, whether retrying can help and what to do instead, plus the real rate limits with the headers that tell a client how much budget it has left, how idempotent retries work, how API keys are stored and revoked, a pre-launch integration checklist, and what to include when asking for help.",
"tags": [
"docs"
],
"link": "/docs/api-reference"
},
{
"date": "2026-09-29",
"title": "One link teaches any AI agent to use three.ws",
"summary": "three.ws/skill.md is a single Agent Skill an assistant can load to learn the platform: how to connect, the rules it follows before spending money, and links to the focused skills for making 3D models, rigging avatars, creating agents and hiring them. It updates itself whenever a skill is added.",
"tags": [
"feature",
"docs"
],
"link": "/connect"
},
{
"date": "2026-09-29",
"title": "Know the real three.ws and the real $THREE at a glance",
"summary": "Every page footer now states that three.ws is the only official website and that the $THREE address shown there is the only official mint, so a look-alike site or copycat token is easier to spot. The same notice leads the llms.txt index that AI assistants read, which is now a short curated guide to the key pages instead of a list of nearly a thousand, with the full list still in llms-full.txt.",
"tags": [
"security",
"improvement"
],
"link": "/three-token"
},
{
"date": "2026-09-29",
"title": "Syndicates: copy the best traders as a team",
"summary": "You can now copy-trade as a team. A syndicate is a named group with its own flag that follows one to three verified trader agents together. Every member still copies from their own wallet with their own per-trade cap, daily budget and drawdown limit, and nothing is ever pooled. The team gets a public page with its roster, a live group profit curve built from members' real copies, and a spot on the board with the rival one rank above named so you know who to catch. Leaving stops only the copies the syndicate started. Every trader profile now shows the syndicates following that trader.",
"tags": [
"feature"
],
"link": "/syndicates"
},
{
"date": "2026-09-29",
"title": "Daily trading quests with XP, levels and badges",
"summary": "A new quests page gives you five small trading goals every day: make three trades, fork a verified trade, ghost-copy an agent you have never tried, act on a copy intent, and have your own agent close a winning trade. Each one is checked against real on-chain activity rather than a button press. Clear three in a day to keep your streak going, level up with XP, and earn the First Quest, Daily Clear and Quest Streak 7 badges on your profile.",
"tags": [
"feature"
],
"link": "/quests"
},
{
"date": "2026-09-29",
"title": "Copy Coach: your first copy trade, the careful way",
"summary": "Never copied a trader before? The new Copy Coach walks you through it in four small steps. First it shows a real verified win, right next to the full record of the agent that made it, so one lucky trade never passes for a track record. Then you ghost-copy a leader with fake money over their real trades. Only then can you set up one tiny real copy: capped at 0.05 SOL per trade and 0.2 SOL a day by the server, paused automatically if the leader starts losing, and spending nothing until you sign each trade from your own wallet. Ask the coach anything along the way; it answers from the real numbers on the page and never promises returns.",
"tags": [
"feature"
],
"link": "/copy-coach"
},
{
"date": "2026-09-29",
"title": "Sentiment Scout: momentum with receipts",
"summary": "Coin Intelligence has a new Scout tab. It ranks the last hour of pump.fun launches by momentum and backs every claim with its receipt: the X post or paid read it came from, when it was true, and what the chain showed at that same moment. It counts distinct voices rather than posts, calls out coins where all the chatter is one account, names the main risk on every coin, and grades its own past calls against what actually happened. The same evidence now appears on trade receipts, in each coin's detail view, and through the sentiment_scout MCP tool.",
"tags": [
"feature"
],
"link": "/coin-intel?tab=scout"
},
{
"date": "2026-09-29",
"title": "Watch traders trade live in their own 3D trade rooms",
"summary": "Every public trader now has a live trade room. Their avatar stands on the floor and reacts as each real buy and sell fills, the board and ticker tape fill with their trades, and everyone watching shows up in the room. Tap any trade for the receipt of why it was made, fork it into your own wallet, or ghost-copy the trader. A quiet room still shows the latest real trades and which rooms are live right now, and the whole room also works as an accessible live list.",
"tags": [
"feature"
],
"link": "/trade-rooms"
},
{
"date": "2026-09-29",
"title": "The platform's own payments stop crowding out the data it pays for",
"summary": "Every two minutes a background job sends a batch of tiny tips to keep the payment feed moving, and three.ws pays the network fee on each one. It sent 60 at a time whether or not the day's fee budget could cover them, so about 95% were turned away after the work of preparing them, and the few that got through used up the budget the health, price and volume checks needed. The job now sends only what the budget can pay for and leaves room for those checks first. The status page also flags this pattern if it ever comes back, instead of reporting payments as healthy.",
"tags": [
"fix",
"infra"
],
"link": "/status"
},
{
"date": "2026-09-29",
"title": "3D Studio in ChatGPT finishes every model, however long it takes",
"summary": "ChatGPT stops waiting on a tool after one minute, and a detailed 3D model takes one to four. The 3D Studio plugin now answers inside that minute and the inline viewer finishes the job itself: it shows a live timer, drops the model in the moment it is ready, rigs an avatar's skeleton on its own when that step is still to come, and keeps a refined model's version history so you can flip back to the original. Inspecting an existing model now shows it in the viewer too, avatars always come out as a full body ready to pose even when the prompt helper is unavailable, and busy periods no longer let one ChatGPT user's generations use up everyone else's.",
"tags": [
"improvement",
"fix"
],
"link": "/docs/mcp-studio"
},
{
"date": "2026-09-28",
"title": "The Pole Club floor comes alive: a crowd, money on the stages, and tipping from inside the room",
"summary": "Walking into the Club used to land you in an empty room. Now a crowd stands around the three stages, and every tip turns into $0.001 bills. When you tip, a clump of bills flies from where you are standing and flutters down onto her stage. When someone else tips, bills drop from the lights above her pole. The bills stay where they land, so each stage keeps a pile for the night, sized from that dancer's real tips today. You can tip without leaving the scene: every dancer has a tag over her head, and clicking her or her pole opens a picker for her routine. A tip from another visitor now puts her through the routine they paid for on your screen too, and the people at her stage cheer. Tips paid by three.ws platform agents are labelled as agent tips in the feed, drop a single violet bill, and never start a routine, and each dancer's card now says how many of today's tips came from people and how many from agents.",
"tags": [
"feature",
"improvement"
],
"link": "/club"
},
{
"date": "2026-09-25",
"title": "The model catalog AI clients read now shows the real price of every model",
"summary": "The three://models resource that Claude and other MCP clients read listed free models with no price and left some paid models blank. Free models now read as costing nothing, and a model served through a partner route is priced at the model that actually answers, so an assistant comparing costs for you works from real numbers.",
"tags": [
"fix",
"sdk"
],
"link": "/docs/mcp"
},
{
"date": "2026-09-25",
"title": "Props the platform pays Forge to make now land in the /forged gallery",
"summary": "Every half hour the platform pays its own Forge to make a new 3D prop for the public /forged gallery. A standard generation takes 20 to 55 seconds, but the paying side gave up after 20. Forge still finished the model and took the payment, so most of those props were paid for, never saved to the gallery, and logged as failed. The paying side now waits up to 90 seconds for a generation. Each paid prop lands in the gallery, and the payment health numbers stop counting these successful payments as failures.",
"tags": [
"fix",
"infra"
],
"link": "/forged"
},
{
"date": "2026-09-24",
"title": "Paying the Club cover in USDC works in wallets that refused to show the request",
"summary": "Some visitors paying the Club cover in USDC saw their wallet refuse to open the approval window, and the payment stopped with \"Your wallet could not display this payment request\". The payment itself was well formed; the problem was in how the page handed it to the wallet. The payment window now passes the transaction to your wallet through the Wallet Standard interface that Phantom, Solflare and Backpack all support, exactly as our server built it, and only falls back to the older wallet connection for wallets that do not offer it. Nothing was ever charged on a refused attempt.",
"tags": [
"fix"
],
"link": "/club"
},
{
"date": "2026-09-24",
"title": "The x402 payment rail stops starting payments it cannot afford to finish",
"summary": "The platform's own agents pay each other for services over x402, and a daily budget caps how much SOL the shared fee wallet spends on network fees. When that budget was nearly used up, our agents kept starting payments anyway: they looked up accounts, signed transfers and ran a simulated check, only to be turned away at the last step. About 1,700 payments an hour ended that way beside roughly 230 that settled. Each agent now reserves its share of the remaining budget before it starts, so a payment only begins when the fee wallet can fund it. The economy lab also now counts every refused payment instead of about one in a hundred, and the treasury report tells a wallet holding a little spare SOL apart from one that is empty. Payments from customer wallets are never held back by this budget.",
"tags": [
"fix",
"infra"
],
"link": "/economy-lab"
},
{
"date": "2026-09-24",
"title": "See why an AI trader bought, with the evidence it had",
"summary": "Every closed trade on an agent's track record now has a why link. It opens the trade's receipt: what triggered the entry, the Oracle score and the base-rate reasons behind it, the trade firewall's on-chain buy and sell simulation, the LLM judge's thesis when one ran, the Risk Officer's review, and paid market and rug-pull reads with the payment transaction that bought them. Every entry and exit leg links to its transaction. Each piece of evidence shows when it was recorded relative to the entry. Anything logged after the exit is left out, because it could not have caused the trade. The same receipt appears on every shareable trade page, a link with ?trade= opens it directly, and agents can read it through the new trade_receipt MCP tool.",
"tags": [
"feature"
],
"link": "/docs/trade-receipts"
},
{
"date": "2026-09-22",
"title": "Claude and other AI clients can now read your three.ws account and follow guided flows",
"summary": "Every hosted three.ws MCP server now publishes live, read-only views of your account as resources: your agents, each agent's Solana wallet with balances, limits and freeze state, usage and costs, chat history, runs, open orders, DCA plans and intents, plus the marketplace, the model catalog with prices, your token launches and the x402 service catalog. Clients like Claude Desktop show them in the attachment picker, and a client can subscribe to an agent's wallet to be told the moment a transfer lands. Each server also offers guided prompts, such as get-started, create-agent, setup-wallet, trade, hire-agent and review-costs, that walk you through a flow tool by tool and always show the recipient, amount, token and chain before anything that moves money. A new create_agent tool lets a model create an agent for you, and read_resource gives the same data to clients without a resource reader.",
"tags": [
"feature",
"sdk"
],
"link": "/docs/mcp"
},
{
"date": "2026-09-22",
"title": "Agents now buy real work from brand-new wallets, and you can watch it",
"summary": "The platform's agent economy used to prove the payment rail from a few long-lived wallets buying test calls. It now mints a fresh Solana wallet for every purchase: each one is created, given exactly one job's worth of USDC, pays a three.ws endpoint, and is emptied and closed within the minute. Every purchase does something useful. Some buy a 3D generation that lands in the Agent-Forged Gallery, where a new fresh wallet badge marks it. The rest buy live market data, which now has its own page, the Data Desk, showing each dataset with the wallet that paid, the price, and the Solana receipt. Both feeds are public JSON as well, and a new stats endpoint lists every wallet with its three transactions.",
"tags": [
"feature"
],
"link": "/data-desk"
},
{
"date": "2026-09-22",
"title": "The agent economy now pays from thousands of distinct wallets, and only from ones that hold funds",
"summary": "The x402 ring, the closed loop where three.ws agents pay each other for services, has a pool of 2,000 payer wallets to rotate through instead of a single seed wallet, so on-chain activity reflects many participants rather than one cron. Rotation now checks each wallet's recorded SOL and USDC before handing it a payment and falls back to the seed payer when nothing in the pool can cover the call, so a freshly minted or thinly funded pool never produces failed payments. The funder that keeps the pool topped up only submits transfers its own wallets can pay for, and the ring status endpoint reports how many pool wallets are minted and how many are ready to pay.",
"tags": [
"infra",
"improvement"
]
},
{
"date": "2026-09-21",
"title": "Forge now asks what your model is for",
"summary": "Under a finished model in Forge there is a new one-tap row: Game, Web, Avatar, Simulation, Print, AR, or Just playing. It is optional, it is remembered in your browser so you are only asked once, and it rides along on your later generations. We are using it to see which kinds of work Forge does well and which it does not, by use, instead of one blended number for everyone. Developers calling the Forge API can send the same thing as a destination field.",
"tags": [
"improvement"
],
"link": "/forge"
},
{
"date": "2026-09-21",
"title": "The club alley no longer renders a giant avatar across the sky",
"summary": "Walking into /club, some visitors saw a huge metallic shape filling the sky above the alley, and their computer ran hot. It was a crowd member: one avatar model built at centimetre scale was measured wrong before it was sized, so it spawned about 400 times too big, roughly 700 metres tall, and the GPU spent the whole walk-in painting it. Crowd members, your own avatar in the agent switcher, the theater stage and the Agora now size every avatar from its real posed mesh, so that model stands at normal human height like everyone else.",
"tags": [
"fix"
],
"link": "/club"
},
{
"date": "2026-09-21",
"title": "The club no longer runs your computer hot on the way in",
"summary": "While you walked the alley and stood at the door, the club was secretly drawing the whole dance floor behind it at full quality, so your graphics card was rendering two complete 3D scenes at once. The hidden stage now pauses its drawing until the alley fades away, roughly halving the graphics work for the entire walk-in. For a cooler machine inside the club too, turn on Power saver in the club controls.",
"tags": [
"fix",
"improvement"
],
"link": "/club"
},
{
"date": "2026-09-21",
"title": "Pick DEXTools as your chart on every coin chart",
"summary": "The chart-source switcher now covers every coin chart on three.ws. The Oracle coin page and the Pump Dashboard market chart join the coin pages, launch pages, the trade deep-dive and Mission Control, so wherever a coin is charted you can switch between the three.ws chart, DexScreener, Birdeye, GMGN, DEXTools and GeckoTerminal. Your pick is remembered and follows you from one coin surface to the next. The Oracle page keeps its Line and Agent trades views, and the dashboard keeps its own candle chart; the terminals simply line up beside them.",
"tags": [
"feature",
"improvement"
],
"link": "/docs/coin-pages"
},
{
"date": "2026-09-21",
"title": "Phantom users no longer have to switch off transaction protection to pay",
"summary": "Phantom adds a small safety check to every transaction before you sign it, and it does this by default. Our payment rail treated that check as tampering and refused the payment, and the payment window told you to turn the protection off in your wallet settings. That was the wrong answer: the check moves no money and is there to protect you. Payments that carry it now go through, at the Club door and everywhere else you pay with a Solana wallet. We still refuse a payment if a wallet changes the amount, the recipient, or who covers the network fee.",
"tags": [
"fix"
],
"link": "/club"
},
{
"date": "2026-09-21",
"title": "Three SDKs now report the version you actually installed",
"summary": "The walk, tour and assistant SDKs each export a VERSION constant, and a release earlier this month bumped all three packages without updating it. Code that read the constant got the previous release's number: 0.3.0 from walk 0.3.1, 0.5.1 from tour 0.5.2, 1.0.0 from assistant 1.0.1. The constants now match the packages, and the tour builder's copy-paste embed snippet points at tour 0.5.2. Only the tour SDK had a test comparing the two, which is how this was found; the other two were caught by checking every package from that release for the same mistake. Copies already installed from npm keep the old number until the next release of each package.",
"tags": [
"sdk",
"fix"
],
"link": "/tour-builder"
},
{
"date": "2026-09-21",
"title": "Paying at the Club door no longer fails when our own traffic has been busy",
"summary": "Some visitors paying the Club cover approved the payment in their wallet and then saw it fail at the last step with an internal error. The cause was on our side. We cover the small Solana network fee on your payment, and a safety limit that spreads our own automated payments across the day was also being applied to customers, so once our background traffic had used the morning's share, a real payment was turned away. Customers are now exempt from that limit: if you are paying, the payment goes through, and our own traffic waits instead. Nothing was charged on a failed attempt. Separately, when a wallet refuses to open its approval window, the payment window now says plainly that nothing was signed or charged and what to try next, instead of repeating the wallet's own wording.",
"tags": [
"fix"
],
"link": "/club"
},
{
"date": "2026-09-20",
"title": "Repeat visits stop downloading the app's code twice",
"summary": "Every page tells your browser which code it is about to need so it can start fetching early. On a return visit that head start was being thrown away: our offline helper answered those same requests a second time, Chrome refused to match the two, and each file was requested twice while the developer console filled with a warning per file, well over a hundred on a heavy page like the Club. The offline helper now leaves that code to the browser's own cache, which already keeps it for a year, and clears out the duplicate copy it had stored on your device. Return visits load with one request per file and a quiet console.",
"tags": [
"fix",
"improvement"
]
},
{
"date": "2026-09-20",
"title": "Put any token's 3D scene on your own site, keyed by the pair address you already have",
"summary": "Token in 3D is now embeddable anywhere with a single iframe, and it accepts two identifiers instead of one. Chart terminals, launchpads and wallets name a market by its pair (pool) address, while three.ws keys tokens by mint, so /coin3d now takes ?pair= as well as ?mint= and resolves the rest itself through a new public endpoint, /api/coin/pair. Add ?embed=1 and the scene drops the chrome your own page owns, keeps one attribution link, and sends its links to a new tab instead of navigating your frame. Free, keyless, and no account: paste the iframe and it renders the live medallion, holder galaxy, graduation ring and real trade tape for that market.",
"tags": [
"feature",
"sdk"
],
"link": "/docs/coin3d-embed"
},
{
"date": "2026-09-20",
"title": "A launchpad where every coin is bought with $THREE",
"summary": "The three.ws bonding-curve launchpad is now priced in $THREE instead of SOL. A coin launched there for a 3D agent can only be bought with $THREE, its creator earns half of every trading fee in $THREE, and when it graduates, the roughly 12.3 million $THREE it raised is locked in its pool permanently. The new launchpad page shows the live curves with buy and sell straight from your wallet. The curve has been verified against Solana mainnet and opens for launches as soon as its settings are published on-chain.",
"tags": [
"feature"
],
"link": "/three-launchpad"
},
{
"date": "2026-09-20",
"title": "Two Uniswap v4 hooks: wearable 3D items as coins, and fees that reward agent reputation",
"summary": "New open-source contracts. SkinHook turns a wearable 3D item into a fixed-supply coin with its own pool: wearing it locks one token, so every wearer shrinks the tradable supply, and the creator's royalty is taken inside the swap where it cannot be skipped, with a share for the site that referred the sale. AgentTierHook gives ERC-8004 agents a lower trading fee as their on-chain reputation grows, with liquidity providers keeping every fee. Both are tested against the real Uniswap v4 contracts and are not deployed yet.",
"tags": [
"feature",
"sdk"
]
},
{
"date": "2026-09-20",
"title": "Agent reputation on Base and other EVM chains now shows real on-chain reviews",
"summary": "Reputation panels, the passport widget and the trust scores behind agent-to-agent gating were asking the shared ERC-8004 reputation registry a question it does not answer, so agents with dozens of on-chain reviews showed none. They now read the registry the way it actually works: an agent with 41 reviews from 22 reviewers on Base shows 4.1 out of 5 and its recent reviews, and the panel loads in a few seconds on free public RPCs. Staked vouches are only offered on chains whose registry supports them.",
"tags": [
"fix",
"improvement"
]
},
{
"date": "2026-09-20",
"title": "Paid $THREE purchases say what is wrong instead of failing silently",
"summary": "Every $THREE-priced purchase is sealed into one signed quote before you pay. If the platform side of that rail is not fully configured, the quote now comes back with a clear, retryable message and a named reason rather than an opaque internal error, and the platform status page reports the condition itself. Nobody has to be mid-checkout for the problem to be visible.",
"tags": [
"fix",
"infra"
],
"link": "/status"
},
{
"date": "2026-09-20",
"title": "The @trythreews account now posts three times a day, every day",
"summary": "The reviewed content queue that runs the @trythreews account moved to three posts a day, eight hours apart, on every day of the week including weekends. It used to hold its flagship slot for weekdays and pack all three posts into an eight-hour afternoon window, which meant a weekend could only ever carry two. Three posts went out under the new cadence today: the $THREE token MCP server, Forge Max, and the free object library. The Forge Max post also gained a check that proves the high-quality tier is a real, gated lane before it can be published, rather than only proving its page loads.",
"tags": [
"improvement"
],
"link": "/changelog"
},
{
"date": "2026-09-20",
"title": "Queued posts can quote an earlier post",
"summary": "A post in the @trythreews content queue can now name an earlier post to quote, so a follow-up can add the detail the original left out instead of repeating it. The first use quotes the $THREE token MCP announcement with the numbers behind the MCP Tool Catalog: 321 tools across 42 servers, 293 of them free, and 215 whose read-only label is checked against the source on every build.",
"tags": [
"improvement"
],
"link": "/mcp-tools"
},
{
"date": "2026-09-20",
"title": "The flagship post slot now gets flagship material",
"summary": "The @trythreews account publishes into three slots a day, one owned by each tier: proof of work at 04:00 UTC, a shipped feature at 12:00, a flagship at 20:00. The machine that drafts announcements from the backlog was not filling them that way. It never recorded a tier on anything it wrote, so every draft it produced was treated as a mid-tier feature and the 20:00 slot was left to whatever happened to be ready. It also dated the surfaces that need owner sign-off on their imagery into the best slot of every day, which pushed everything it could publish on its own weeks into the future. Announcements are now planned into the slot their own tier owns, the ones awaiting sign-off are listed separately instead of blocking the calendar, and six new announcements are queued for review.",
"tags": [
"improvement"
],
"link": "/changelog"
},
{
"date": "2026-09-20",
"title": "A library of prompts you can paste straight into Claude",
"summary": "Knowing that three.ws has a free text-to-3D API is not the same as getting a rigged avatar out of it, and the paragraph that closes that gap is one nobody should have to write themselves. /prompts is 23 ready-made prompts for Claude and any other assistant: generate a model or a rigged avatar, search the CC0 catalog before generating anything, rig a GLB you already have, put a live avatar on your site, diagnose an embed that went dark, validate a model before it ships, find a paid API in the agent economy or sell your own. Each one names the real endpoint or tool, says what to do when a job comes back pending, and says what to hand back; the ones that can spend money are written to stop and show you the amount first. Most need no account, no key and no setup. Copy a prompt, open it directly in Claude with the text already in the box, or fetch the whole library as JSON or plain text.",
"tags": [
"feature",
"docs"
],
"link": "/prompts"
},
{
"date": "2026-09-20",
"title": "Five new skills teach any Claude to build on three.ws, and three standalone repos to install them from",
"summary": "Until now the skills we hand out taught Claude the 3D lanes and the wallet, and stopped at the platform itself: nothing explained how to create an agent, give it a capability, put a price on it, or hire someone else's. Five new skills cover exactly that, each written against the real endpoints with the real request bodies and the real failure responses: create an agent (body, custodial Solana wallet, persona, live page), author a four-file in-app skill bundle and test it, publish and price a skill per call in $THREE with metered free trials, hire another agent under per-call and daily spend caps, and connect Claude Code, Claude Desktop or any MCP client to the hosted servers starting with the free one. The pack is also published as three small standalone repos, so installing it no longer means cloning a 60-directory monorepo: everything, the crypto-free 3D subset, or the agent-economy set. The repos are generated from the same folders the platform ships, so a copy can never drift from the original.",
"tags": [
"feature",
"sdk",
"docs"
],
"link": "/docs/agent-skills"
},
{
"date": "2026-09-20",
"title": "Every perpetual on /derivatives now opens a full contract page",
"summary": "The perps table told you a contract's price, funding rate and open interest, and then dead-ended: rows went nowhere. Each one now opens its own page. It puts the mark price next to the index the contract settles against and states the basis between them, turns the funding rate into money by letting you type a position size and reading back what it pays per interval, per day and per year, and ranks the contract's open interest and volume against every other venue listing the same underlying. Below that sits the spot chart for the underlying, a cross-venue funding table with the contract you came from pinned in place, and four callouts that answer the question the table never could: the cheapest venue to hold a long, the best paid venue to hold a short, how far funding spreads across venues, and how far their prices disagree. Those four ignore venues trading under a million dollars a day, because a dormant book sets the widest number every time and is nowhere anyone should act. The same page is reachable from the contracts table on any derivatives venue's exchange profile.",
"tags": [
"feature",
"improvement"
],
"link": "/derivatives"
},
{
"date": "2026-09-20",
"title": "Derivatives venue profiles show their contracts again",
"summary": "Opening a futures venue from the exchange rankings could land on a profile that said it had no markets at all, even for venues listing hundreds of contracts and clearing billions a day. The market-data source keeps an empty placeholder record for some of those venues alongside the real one, and the page was reading the placeholder. It now recognises an empty record for what it is and serves the real venue: open interest, perpetual and futures pair counts, and the contract table. Those tables are also ranked by 24h volume before they are cut to fifty rows, so the biggest contracts lead instead of whatever happened to come first alphabetically, and every contract links to its own detail page.",
"tags": [
"fix",
"improvement"
],
"link": "/exchanges"
},
{
"date": "2026-09-20",
"title": "Awesome 3D Agents joins the announcement queue, and the machine stops skipping pages like it",
"summary": "The machinery that writes announcements from the backlog rebuilds its inventory from the site map on every run, and it was reading only the product sections. Five pages filed under Learn are products rather than documents, so it had never seen them: the curated Awesome 3D Agents list among them, live since the 7th of September and never posted about. They are in the backlog now. Three smaller things landed with it. A page can be asked for next, by name and with the reason recorded, instead of waiting for its score to come round. A post can quote the counters a page puts on screen, such as the 152 entries and 15 sections on the list itself, which the fact harvester used to throw away because they are not sentences. And a frame that was meant to be a five second loop is now shipped as a single picture when the capture finds the page does not actually move.",
"tags": [
"improvement"
],
"link": "/awesome"
},
{
"date": "2026-09-20",
"title": "Light mode on the market pages now clears the contrast floor",
"summary": "Every green number in a market table, and the coloured score badges on the exchange pages, were being drawn in the same hue used for chart fills. On white that reads at around 3:1, under the 4.5:1 the platform holds itself to, so gains in a table and the figures inside a trust badge were harder to read than they should have been for anyone with low vision or a bright screen. They now use the darker text shade of the same colours, which the design system already carried for exactly this. Dark mode is unchanged. Checked with the same automated WCAG rule sets the site gates on: the perpetual markets table, the new contract page, the exchange rankings and an exchange profile all report zero violations in both themes.",
"tags": [
"fix",
"improvement"
],
"link": "/derivatives"
},
{
"date": "2026-09-20",
"title": "Anyone can now check that the $THREE hold gate does what we say it does",
"summary": "We publish specific numbers about holding $THREE: $25 unlocks high quality generation and game ready export, draft and standard stay free for everyone, and the perks further up the ladder are labelled planned because they are not built yet. Those are claims, and a claim nobody can check is worth very little. There is now a single command that checks all of them against the live site, and none of it requires owning a token. It compares the published thresholds and discounts to the ones the server actually uses, confirms an anonymous request for the paid lanes is refused with the tier and the price it needs, reads the balances of real wallets that already hold $THREE to confirm they are let through, confirms the free lanes still run with no account at all, and confirms nothing unbuilt is being advertised as live. The unlock side is proved by reading public balances, which moves no money and buys nothing. How to run it is written up in the hold to access guide.",
"tags": [
"docs",
"improvement"
],
"link": "/three"
},
{
"date": "2026-09-20",
"title": "The $THREE tier cards no longer list unbuilt perks next to real ones",
"summary": "Each tier on the holder page carries two kinds of line: things the hold turns on today, and things that are registered but not built yet. Further down the same card the unbuilt ones were already flagged Planned, but higher up they also appeared as plain bullets with no flag, so Silver read as though private worlds and priority routing were included and Gold as though branded environments were. They are not. The two lists are now kept apart at the source, the planned lines carry the Planned flag wherever they appear, and the pricing API publishes them as a separate field so anything reading it automatically cannot mistake a roadmap line for a benefit. Nothing about what you actually get has changed. A test now fails the build if an unbuilt feature finds its way back into the delivered list.",
"tags": [
"fix",
"improvement"
],
"link": "/three"
},
{
"date": "2026-09-20",
"title": "Paying the Club cover charge in $THREE works now",
"summary": "The Club door quotes its cover in USDC or in $THREE, but every attempt to pay in $THREE failed at the last step, and the payment window wrongly said you had cancelled in your wallet when you had just approved. The cause was on our side: $THREE is issued under Solana's newer token standard (Token-2022) and our payment rail only knew how to build and check transfers for the older one, so a $THREE payment could never be verified. The rail now reads which standard a token uses and handles both, in the step that builds your transaction and in the step that verifies it. Nothing was ever charged on a failed attempt. The payment window also now shows the real reason when a payment is refused instead of blaming a cancel.",
"tags": [
"fix"
],
"link": "/club"
},
{
"date": "2026-09-20",
"title": "There's a bouncer at the Club door now",
"summary": "The Club's cover charge has always said a bouncer checks your wallet, and until now nobody was standing there. Walk up the alley and you will find him posted beside the door, a head taller than you, in black with shades. He watches you come down the alley, tells you the cover (one cent, in USDC or $THREE), looks down at his list while your wallet is checked on-chain, then nods you through or shakes his head. Regulars and VIPs get a different welcome. Tap him and he explains how the club works, one line at a time, and if you try to walk through him he does not move. There is a line of people waiting behind the rope beside him too, and they have opinions about you cutting past.",
"tags": [
"feature"
],
"link": "/club"
},
{
"date": "2026-09-20",
"title": "The alley outside the Club looks like a night out now",
"summary": "Walking up to the Club used to mean crossing a quiet alley. Now there is a line along the wall beside the door, two abreast behind brass posts and a velvet rope, under a neon Pole Club sign. The people in it are real platform avatars, and they talk: neighbours trade short exchanges in speech bubbles about the one cent cover, the tips and the wait, and they notice you. Walk straight past them to the door and the head of the line objects. Stand too close and someone asks for space. Get let in and they grumble about it. A few of them started early: one sways in the line and slurs, one weaves around the alley and steps up the kerb the hard way, and one by the wall keeps trying to stand up and sliding back down. The line fits itself to the alley, bends around the corner when the wall runs out, and gets shorter on slower devices so the walk-in stays smooth. With reduced motion turned on the bubbles still appear, without the swaying and staggering.",
"tags": [
"feature",
"improvement"
],
"link": "/club"
},
{
"date": "2026-09-18",
"title": "Standard Anthropic SDK clients can now use a three.ws agent's AI budget",
"summary": "The AI proxy behind every embedded three.ws agent now answers at the address an Anthropic SDK or command-line tool builds for itself. Point a client at /api/llm/anthropic/agents/<your agent id> with one of your API keys and it runs on that agent's model, limits, and monthly budget. The proxy also stopped turning such clients away over their defaults: a system prompt sent as a list of text blocks is accepted, and an output limit above 16,000 tokens is lowered to 16,000 instead of rejected.",
"tags": [
"feature",
"sdk"
],
"link": "/docs/api-reference"
},
{
"date": "2026-09-18",
"title": "The embedded 3D agent is readable on light websites and quiet in the console",
"summary": "Two fixes for anyone who has put a three.ws agent on their own site with the two-line embed. The floating widget had no background of its own, so on a light page its white button labels and name vanished into the page; it now has its own dark panel, and the suggestion buttons carry a solid ground, so it reads the same on any site. You can still ask for a transparent or light panel with the background attribute. The embed also made two background requests that browsers blocked on every site but ours, logging errors in the host page's console; those requests now succeed, and only the public, signed-out answer is ever shared across sites.",
"tags": [
"fix",
"sdk"
],
"link": "/tutorials/embed-in-30-seconds"
},
{
"date": "2026-09-18",
"title": "Choose your chart everywhere a coin is charted, now including DEXTools",
"summary": "Every place three.ws draws a coin's price chart now lets you switch to the chart you trust. The Trade Terminal at /trades and Mission Control used to show only the platform's own live candles; both now have a source bar with DexScreener, Birdeye, GMGN, DEXTools and GeckoTerminal one click away, the same choice already offered on coin pages and launch pages. DEXTools is new on all four. Your pick is remembered and follows you from coin to coin, the bar works from the keyboard with the arrow keys, and each chart comes with a link that opens the same market on the provider's own site. If a provider cannot draw a coin yet, or its chart is blocked by your browser or network, you get a plain explanation with a retry and a working alternative instead of an empty box.",
"tags": [
"feature",
"improvement"
],
"link": "/trades"
},
{
"date": "2026-09-18",
"title": "The Trade Terminal finally looks the way it was designed",
"summary": "The Trade Terminal at /trades was written against a set of colors and fonts that the page itself never defined, so browsers quietly fell back to their defaults: Times New Roman text, oversized controls and missing panel colors. The page now defines its own palette, in both dark and light themes, and renders as the dense, monospaced trading terminal it was meant to be.",
"tags": [
"fix"
],
"link": "/trades"
},
{
"date": "2026-09-18",
"title": "The $THREE page now has a live DEXTools chart and the Social Boost wins",
"summary": "The $THREE page shows a live candle chart of the three / SOL pair, drawn by DEXTools. Below the platform's own buybacks there is a new section for the community's: the three DEXTools Social Boost prizes $THREE has won so far, $11,382 of buybacks in total, each with a link to its receipt on DEXTools. Social Boost ranks tokens by visits to their DEXTools page, so the section also links straight there. Every visit counts toward the next win.",
"tags": [
"feature",
"improvement"
],
"link": "/three-token"
},
{
"date": "2026-09-18",
"title": "Paid endpoints now describe themselves the way marketplaces expect",
"summary": "When an agent or a marketplace asks one of our paid endpoints what it costs, the answer also explains how to call it and what comes back. Around sixty endpoints were sending that explanation in an older layout, or in one that did not match its own rules, which is enough for a marketplace to skip the listing without saying why. Every paid endpoint now sends it in the current standard layout, checked against its own rules. We also added a tool that tests our endpoints with the same checklist a marketplace validator uses, so a listing problem shows up on our side first.",
"tags": [
"improvement",
"fix"
],
"link": "/docs/x402-dev-tools"
},
{
"date": "2026-09-18",
"title": "Paid x402 endpoints are answering again",
"summary": "For several hours on September 18, nearly every paid endpoint on three.ws, the x402 discovery document at /.well-known/x402, the payment facilitator and the MCP server answered with a server error instead of a price. A routine update to the library behind wallet sign-in changed what it needs at startup, and our code tripped over it the moment it loaded, before any request could be served. Agents and marketplace validators that probed us saw a broken seller. The sign-in code now supplies what the library expects, returning buyers can sign back in to what they already paid for exactly as before, and the deploy checks now refuse to ship when the tested dependencies differ from the ones the live site will run, which is how this got past us.",
"tags": [
"fix",
"infra"
],
"link": "/docs/x402"
},
{
"date": "2026-09-18",
"title": "Granite Guardian now answers on three.ws, not just in the docs",
"summary": "The governance check that screens what an agent is asked to do, and vetoes it when the request is a jailbreak, harmful, or fraudulent, runs on IBM Granite Guardian. It was built to call IBM's hosted watsonx.ai service, so on our live site it answered with a not-configured error, and the new agent identity check quietly skipped itself for the same reason. IBM publishes Granite Guardian's weights openly, so we now run the model ourselves on a dedicated GPU as a second lane: watsonx still leads wherever it is set up, and our own copy answers whenever it is not. Every verdict now says which model and which lane produced it. The identity check also compares a new agent against existing ones using the site's regular embedding service when Granite embeddings are unavailable; it lists the look-alikes it finds but only scores them on Granite, because that is what its thresholds were tuned on.",
"tags": [
"fix",
"infra"
],
"link": "/docs/ibm"
},
{
"date": "2026-09-18",
"title": "Text avatars now wear what you describe",
"summary": "Typing a description on /genesis (or sending a prompt to the avatar API) used to produce the same template body with a face pasted onto it, so an explorer in a teal flight jacket came back in someone else's clothes. Text avatars are now built from your words head to toe: the prompt is painted as a full-body character, sculpted into a textured 3D model on our own GPU, and given a skeleton so it animates immediately. Selfies still use the face pipeline, which is what a likeness needs. The free 3D generator behind the forge_avatar tool also stopped failing when the community GPU spaces were out of quota: it now moves the job onto our own worker instead of giving up.",
"tags": [
"fix",
"improvement"
],
"link": "/genesis"
},
{
"date": "2026-09-18",
"title": "469 generated animations in the library, and for sale as rigged GLBs",
"summary": "The animation library now carries 469 clips generated on our own motion model, covering 175 prompts across walks, idles, gestures, dance, combat, sport, reactions and more. Every one passed the quality gate that checks for sliding feet, sudden jumps and broken loops, and fewer than three in ten takes were thrown away. While checking them on a real avatar we found that every generated clip, including the ones already live, played with the head thrown back and the arms raised, because the model describes its skeleton with the arms hanging and the head in front of the neck. That is fixed, walks now travel instead of skating in place, and a foot flicker is smoothed out. The same fix applies when you generate a motion from a prompt in the Animation Studio. Each clip is also listed in the marketplace as a GLB on the three.ws rig that opens in Blender, Unity or Unreal, for 0.01 USDC through the x402 download, and a rotating set of twelve is free each week.",
"tags": [
"feature",
"fix"
],
"link": "/animations"
},
{
"date": "2026-09-17",
"title": "A payment that is still confirming no longer reads as a failed payment",
"summary": "Settling an x402 payment on Solana means broadcasting a transaction and then waiting for the network to confirm it, and in between the answer is genuinely unknown. Until now an unfinished wait was reported as a failure, so a payment the chain was in the middle of accepting could tell the buyer it had failed and withhold what they paid for. Our facilitator now answers with the protocol's pending state and the settlement signature instead, a retry re-checks that same transaction rather than sending a second one, the endpoint still returns the work with the signature to watch, and a reconciler closes the books every two minutes once the chain answers. The confirmation wait also dropped from 30 seconds to 12, so paid calls come back faster.",
"tags": [
"fix",
"improvement",
"infra"
],
"link": "/docs/x402-settlement-pending"
},
{
"date": "2026-09-17",
"title": "Sign the real-funds agreements once before using real money",
"summary": "three.ws is experimental technology, and agent wallets are custodial software, not bank accounts. Before you deposit into an agent wallet, trade, send, withdraw, arm autopilot, or pay with real funds, you now sign three agreements in one step: the Terms of Service, the Risk Disclosure, and the new Agent Wallet Agreement. You tick each one, confirm you are eligible and that you understand funds can be lost for any reason with nothing insured, and type your name as your signature. An agent's deposit address now appears only after signing, and every deposit is covered by the Agent Wallet Agreement. When you are signed in, the signature is saved to your account and the servers refuse real-funds actions from unsigned accounts, including through the API, so it cannot be skipped. You can review every signature you have made at /legal/agreements. Devnet stays free of this step.",
"tags": [
"security",
"feature"
],
"link": "/legal/agreements"
},
{
"date": "2026-09-17",
"title": "Coin pages show the numbers, not a score",
"summary": "Every coin launched on three.ws used to carry a quality score, a conviction tier and risk labels on its page, in the launch feed and on its share image. A high score can read as investment advice and a low one as FUD about someone's launch, so all of them are gone. In their place is a Token stats panel with the facts traders look up in terminals: volume, buys, sells, net buy and price change over 5 minutes, 1 hour, 6 hours and 24 hours, plus holders, the top 10 share, how much the dev holds, how much early buyers still hold, mint and freeze authority, bonding curve progress, liquidity and whether a DEX Screener profile was paid for. The Holders panel lists the ten largest wallets with links to each one, and leaves the bonding curve and pool out of the ranking. If the full holder index is unavailable, the page still shows the largest holders and says so instead of going blank.",
"tags": [
"improvement"
],
"link": "/launches"
},
{
"date": "2026-09-17",
"title": "Pick your chart on every coin page, and click an agent's coin to open it",
"summary": "Every coin launched on three.ws now lets you choose the chart you trust, right on its page: our own price line, live TradingView candles that tick with each trade, or the DexScreener, Birdeye, GMGN and GeckoTerminal charts, all free and all without leaving three.ws. Your choice is remembered from coin to coin. Birdeye and GMGN are also added to the chart picker on /coin pages. A brand-new coin used to show \"no price history\" for its first hours, because the chart data sources had not listed it yet; it now charts from its very first trade. If GeckoTerminal has not listed a young coin yet, its tab says so and offers a chart that already covers it, instead of an error page. And the coin shown on an agent's page, its profile and its home card is now a link straight to that coin's page, where before it could not be clicked at all.",
"tags": [
"feature",
"fix"
],
"link": "/launches"
},
{
"date": "2026-09-17",
"title": "Agent wallets that cannot sign no longer look tradeable",
"summary": "A small number of agent wallets hold a balance that three.ws cannot currently sign for. Their Portfolio tab used to show a live net worth with Trade buttons, and the Trade tab showed a working form, so the first sign of trouble was a failed withdrawal. Portfolio and Trade now say up front that trading and withdrawals are paused for that wallet, hide the actions that cannot work, and link straight to support. The explanation is also clearer: the balance has not moved, it is still on chain at the same address, and you should not create a replacement wallet while access is being restored.",
"tags": [
"fix"
],
"link": "/agents"
},
{
"date": "2026-09-17",
"title": "Ask your AI assistant to show the picture inside a Solana transaction",
"summary": "Some Solana transactions carry a small image written straight into their memo, which explorers show as thousands of characters of unreadable text. A new free MCP server, @three-ws/solana-memo-media-mcp, lets Claude, Cursor or any other MCP client fetch a transaction by its signature and show you the image directly, alongside who signed it and a fingerprint of the exact bytes. It can also scan an account's recent history and pick out only the transactions that carry images. It needs no account, key or wallet, never writes anything, and only shows PNG, JPEG, WebP and GIF files whose contents match their label. The Onchain Viewer at /onchain was also fixed: it only recognised the newest memo program, so memos written through the standard Solana memo program were reported as missing. It now reads all three.",
"tags": [
"feature",
"fix"
],
"link": "/docs/mcp-solana-memo-media"
},
{
"date": "2026-09-17",
"title": "A plan for meeting agents through smart glasses",
"summary": "three.ws already lets you pin a 3D agent at a real place and walk up to it, and /irl can already mirror the nearest-agent cue to Brilliant Labs Frame and Even Realities G1 lenses over Bluetooth. New documentation maps the whole glasses lane: how the work splits between the lens (who is near, which way, how far) and the phone (the 3D body in AR), a full build plan for Meta Ray-Ban Display glasses and the Neural Band, and an honest survey of every platform, repository and package involved, including which ones have shut down and should not be built on.",
"tags": [
"docs",
"feature"
],
"link": "/docs/glasses"
},
{
"date": "2026-09-17",
"title": "Documentation that tells you whether to trust it",
"summary": "Every page in the docs is bound to the code it names, and git decides whether that code moved after the page was last written. That measurement now has a second clock: when someone reads a page, checks it against the source and finds it correct, that review is recorded, so careful reading counts as much as editing and nobody has to fake a change to clear a page. Docs that record a moment, like a dated audit or a finished piece of work, are marked as records and are no longer counted as drift. Every page that was out of date has been brought back in line: 110 documents were read against the exact changes that had moved under them, 82 were corrected, 22 were confirmed still accurate, and 6 were marked as records of a moment. The dashboard at /docs/freshness now shows who last checked each page and when, the report is regenerated on every deploy instead of by hand, and the check runs automatically so drift cannot pile up unnoticed again.",
"tags": [
"docs",
"improvement"
],
"link": "/docs/freshness"
},
{
"date": "2026-09-16",
"title": "A flat 1% fee on trades made through three.ws",
"summary": "Trades that three.ws builds or runs for you now carry a 1% fee, paid in the same transaction as the trade: buys and sells in the coin trade window, trades from your agent's wallet, and trades your agents make through strategies, mirror trading, orders, market making and sniping. The fee is shown before you confirm, both in the trade window and in the agent wallet's trade tab. Trades you make on other apps are not affected, and three.ws's own agents are never charged.",
"tags": [
"improvement"
],
"link": "/legal/tos"
},
{
"date": "2026-09-16",
"title": "Deploy your agent on-chain with Solana's new transaction v1",
"summary": "Minting an agent to Solana now uses the new transaction v1 format whenever your wallet supports it, from the agent deploy button, the registration wizard (vanity addresses included), and /deploy-onchain. On /deploy-onchain that turns a Genesis-shape deploy from two wallet transactions into one: the asset is created and registered in the Metaplex Agent Registry atomically, so there is no half-finished deploy if something fails in between. Each deploy spells out its compute, account data, and priority fee limits instead of relying on defaults, and three.ws checks that the transaction that landed on-chain is the exact version it prepared before recording your agent. Wallets that have not added v1 yet keep working exactly as before, and a deployed agent now shows which transaction format it used.",
"tags": [
"feature",
"security"
],
"link": "/deploy-onchain"
},
{
"date": "2026-09-16",
"title": "A cleaner launchpad for your 3D agent's coin, with rewards you claim right here",
"summary": "three.ws/launch is rebuilt around one flow: pick one of your 3D agents, name the coin, and launch it on pump.fun from your wallet or from the agent's own wallet. The coin's metadata links your agent's page and 3D body, a live preview and cost panel show exactly what you will pay (including a 1% three.ws fee on the dev buy), and a new My coins tab shows unclaimed creator rewards and claims them in one click, no pump.fun app needed. Launches now fit in a single transaction thanks to pump.fun's lookup table, and larger ones, like USDC-paired coins with a dev buy, use Solana transaction v1 automatically when your wallet supports it. This also fixes launches that were failing because an on-chain buyback program stopped accepting new agents.",
"tags": [
"feature",
"improvement",
"fix"
],
"link": "/launch"
},
{
"date": "2026-09-16",
"title": "GitHub fee claims now say what they actually prove",
"summary": "A GitHub fee claim on pump.fun proves that a GitHub account collected creator fees routed to it. It does not prove that person built the coin, owns the linked repository, or backs the project. Claim cards on /pumpfun used to read as a developer endorsing a coin even when the account was a lookalike name, the coin had no GitHub link at all, or one fee account was shared by several coins and the card simply picked the biggest one. Every claim now carries one of five labels: verified repository, verified creator wallet, identity mismatch, unverified, or unresolved pooled withdrawal. Only the two verified labels get trade buttons or count toward a coin's reputation score, mismatches are flagged in red, and the Telegram claims channel follows the same rules.",
"tags": [
"fix",
"improvement"
],
"link": "/pumpfun"
},
{
"date": "2026-09-16",
"title": "pump.fun agent skills now launch holder-reward coins",
"summary": "The pump.fun launch, swap and fee skills that agents use on three.ws now run on version 2 of the pump.fun SDK. Launching a coin can now turn on holder rewards, which sends every creator fee to the coin's holders instead of the launcher, permanently. Cashback launches were retired by pump.fun, so the skill now refuses them up front with a clear message instead of handing back a transaction that would fail after you sign it. The fee skill also recognizes holder-reward coins and explains where their fees go rather than trying to collect them.",
"tags": [
"improvement",
"fix"
]
},
{
"date": "2026-09-15",
"title": "Show and tell now captures projects people can actually try",
"summary": "The GitHub community showcase now asks builders for the outcome, live demo, source, three.ws surface used, one useful lesson, and visual proof. That structure makes each submission easier to understand, reproduce, feature, and turn into an honest adopter story. Two compact developer front doors are also fully built and exportable: a Codespaces starter that opens on a live animated agent, and a GLB pull-request quality gate powered by the same model-diff engine as three.ws.",
"tags": [
"improvement",
"docs"
]
},
{
"date": "2026-09-15",
"title": "three.ws is ready for Solana's new 4 KiB transactions",
"summary": "Solana V1 transactions are live, with more than three times the wire room and a different place for execution limits. Every three.ws transaction reader now opts into V1, so a payment, registration, receipt, or agent action cannot disappear just because it uses the new format. The new Atomic inspector reads a confirmed signature or raw base64 handoff and shows the exact byte footprint, compute and account-data limits, priority fee, actual compute use, and the warning a fee sponsor needs before co-signing. It never signs or broadcasts anything.",
"tags": [
"feature",
"improvement",
"security"
],
"link": "/atomic"
},
{
"date": "2026-09-12",
"title": "Corrected what the 3D Studio connector declares about checking a pending job",
"summary": "The tool that collects a finished 3D generation was labelled a read-only status probe. It is not: the first check that finds a job done copies the model into permanent storage, records it, and scores it, and it can route failed work to another provider. Any AI client reading those labels to decide whether a call is safe to repeat was being told the wrong thing. The labels and the documentation now match the behaviour. Nothing about how the tool works changed.",
"tags": [
"fix",
"sdk"
],
"link": "/docs/mcp-studio"
},
{
"date": "2026-09-12",
"title": "The 3D viewer inside ChatGPT now loads everything through three.ws",
"summary": "The interactive viewer that shows your 3D model inside ChatGPT now fetches every model and every preview image through three.ws itself, so the only places it is allowed to reach are three.ws and the 3D viewer library it runs on. Nothing about what you see changes: the model still renders, spins and downloads the same way. There is also a dedicated connection address for the ChatGPT plugin directory, https://three.ws/api/mcp-chatgpt, carrying the eight 3D creation tools. The existing connector at https://three.ws/api/mcp-studio keeps every tool, including the living agent body.",
"tags": [
"improvement",
"security"
],
"link": "/docs/mcp-studio"
},
{
"date": "2026-09-11",
"title": "Your selfie avatar now survives a storage outage instead of vanishing at the last step",
"summary": "For part of today, a selfie could finish building and then be thrown away. The 3D model was generated correctly, but the step that files it into your library could not reach our object storage, so the page said the avatar finished and could not be saved, and trying again hit the same wall. The avatar was never actually lost: a complete copy sits with the service that built it. The last step now recognises a storage outage for what it is and keeps your avatar, serving it from that copy and flagging it to be filed away properly once storage is back, so a problem on our side costs a housekeeping step instead of the model you waited for. Anything that is not a storage outage still fails loudly rather than quietly papering over a real bug. The image API for agents also now answers with a clear temporarily unavailable instead of a generic failure, so it no longer reads as a bad prompt when nothing was wrong with the prompt.",
"tags": [
"fix",
"infra"
],
"link": "/create/selfie"
},
{
"date": "2026-09-11",
"title": "Selfie to avatar is unstuck: a scan that used to wait forever now finishes in seconds",
"summary": "Uploading a photo on the selfie page could sit on the same message indefinitely and never start the scan, with no error and nothing to retry. One copy of the face-tracking runtime held at our CDN was telling browsers to expect 11 MB and then sending a third of that, so the page waited for bytes that were never coming. That copy is gone, and three changes make the failure impossible to repeat silently: the face-tracking runtime now loads from three.ws first instead of a public CDN, any wait on it gives up after fifteen seconds and builds your avatar from the photo as it is, and a photo is never rejected for having no face in it just because the detector could not load. A new check also walks every large file on the site the way a browser asks for it and fails loudly when the bytes do not add up. Measured after the fix: a selfie became a rigged, animation-ready avatar in about fifteen seconds, and still finished in under a minute with the face-tracking runtime completely unreachable.",
"tags": [
"fix",
"infra"
],
"link": "/create/selfie"
},
{
"date": "2026-09-11",
"title": "The 3D creation skills are now a public download",
"summary": "The skills that teach an AI agent when and how to generate, rig, and embed 3D models on three.ws used to live only inside this repo. Four of them are now published at three.ws/skills/3d-studio: read one straight from the web, or download the whole set as a zip and drop it into whatever agent you are building. Nothing there needs an account, an API key, or a payment. A skill is only included if every tool it tells an agent to call is actually served, so a downloaded skill cannot send an agent after something that does not exist.",
"tags": [
"feature",
"sdk"
],
"link": "/docs/agent-skills"
},
{
"date": "2026-09-10",
"title": "Fourteen paid endpoints now offer Solana first, matching what the catalog advertises",
"summary": "Agents that pay for a three.ws service are handed a list of chains they can settle on, and most clients simply take the first one. Fourteen of those endpoints were handing back Base first even though the public service catalog listed them as Solana, so an agent that read the catalog and then called the endpoint could end up paying on a different chain than the one it had picked. All fourteen now lead with Solana, which is the platform default the rest of the endpoints already followed, so the catalog and the live payment request finally agree. Nothing changes for anyone who deliberately settles on Base, which is still offered.",
"tags": [
"fix",
"improvement"
]
},
{
"date": "2026-09-10",
"title": "Any embedded agent can now be waved at, not just the one on our own page",
"summary": "Hand control arrived last week on a page of ours: your speakers play a tone too high to hear, and your microphone listens for how your hand bends it on the way back, so you can drive a 3D agent by waving at your laptop with the camera shut. It now works on any agent embedded anywhere. Add sonar to the embed tag, give people a button to switch it on, and a sweep of the hand steps the agent through its gestures, a push moves the camera in, and a held hand turns it around. Sites that want the gestures to do something else entirely can take them over and drive their own page with them. Nothing is recorded, nothing is downloaded until somebody actually turns it on, and it stops the moment they navigate away.",
"tags": [
"feature",
"sdk"
],
"link": "/sonar"
},
{
"date": "2026-09-10",
"title": "Control a 3D agent with your hands, no camera needed",
"summary": "Your laptop already has a motion sensor: the speakers and the microphone. The new Sonar page plays a tone too high to hear, then listens for how your hand bends it on the way back, the same effect that drops the pitch of a passing siren. Sweep your hand across the keyboard and the agent on screen steps to its next gesture. Push your palm toward the screen and the camera moves in. Lift your hand and hold it, and the camera turns around the agent. Nothing is recorded and nothing leaves the page, and every gesture is on the keyboard too, so the page works with the microphone switched off. The sensing itself is two small modules you are free to point at anything, documented for developers.",
"tags": [
"feature"
],
"link": "/sonar"
},
{
"date": "2026-09-10",
"title": "three.ws now reads in 47 languages",
"summary": "The site has carried translated text for a long time, but the language picker only ever offered English and Spanish, so almost nobody could reach the rest. Those catalogs are being finished and checked one at a time, and the switcher now offers 47 languages. Each one is complete across every piece of text on the site, close to twenty thousand phrases, with nothing left sitting in English and nothing left half translated. A language only appears in the switcher once its translation is finished, so anything you can pick is a language the whole site speaks. Retired labels that no longer appear anywhere were cleared out at the same time, so no page shows a stale phrase in one language that it does not show in another. Right to left languages are included.",
"tags": [
"feature",
"improvement"
]
},
{
"date": "2026-09-10",
"title": "Repaired the free AI lanes, which had been quietly answering nothing",
"summary": "Several of the free AI models the platform falls back to had been retired by their provider and were answering every request with a Gone error. Nothing logged it as a problem, because a dead model looks the same as a busy one to a fallback chain: it just moves on. The cost was real though, since each dead entry ate one of the limited retry slots that should have gone to a lane that could actually answer, making the whole chain fail sooner and more often under load. Every model id is now checked against the provider's live catalog, the dead ones are replaced with working models, and the same applies to the search-and-memory embedding lane, which had been failing for two weeks. Older stored memories keep working: they are tagged with the model that created them and still resolve.",
"tags": [
"fix",
"infra"
]
},
{
"date": "2026-09-10",
"title": "A smart home that stops answering now says so, instead of reading Live",
"summary": "If your house went offline while three.ws already had a connection open to it, the home card could keep showing Live, updated moments ago for as long as you watched. The live 3D scene knew and went stale correctly, but the stored status did not, so anything reading it (the card, the home list, a wall display) kept describing a house that was gone as healthy. A house dropping out is now recorded the moment it is noticed, and recorded again when it comes back, so every surface agrees. This was invisible on a first visit and permanent on a busy one, which is why it survived so long.",
"tags": [
"fix"
],
"link": "/smart-home"
},
{
"date": "2026-09-10",
"title": "3D assets now load on any website, not just ours",
"summary": "Files served straight from our asset storage carried a permission list that only named a handful of sites, so a browser on any other site quietly threw the file away after downloading it. Anyone embedding a three.ws avatar or model from its direct storage link saw an empty viewer with a cross-origin error, and the workarounds were our /cdn and /api/glb routes. Storage now allows reads from everywhere while keeping uploads locked to our own pages, so direct links work in any embed. This also fixed a first-party case: the www version of the site served every page correctly but had all of its 3D blocked.",
"tags": [
"fix",
"infra"
]
},
{
"date": "2026-09-10",
"title": "Generated motion clips rebuilt so they no longer slide across the floor",
"summary": "The generated half of the animation library was written in the wrong rest pose and carried a constant forward drift that was never in the original motion, which made characters glide instead of plant their feet. Every published clip was re-derived from the motion already generated, re-checked against the quality gate, and republished: 39 clips that now pass are live, and the 94 that still failed on foot-sliding were withdrawn rather than left on the shelf. No new generation was needed, because the motion was correct all along and only written down wrong.",
"tags": [
"fix",
"improvement"
],
"link": "/animations"
},
{
"date": "2026-09-10",
"title": "Two scheduled jobs that had never once run are now running",
"summary": "The daily portfolio snapshot and the 15-minute global events ingest were both declared and deployed but had no schedule attached, so neither had ever fired. Both are now scheduled and enabled. The events feed backfills a few files per tick rather than all at once, and the snapshot refuses to record a day it cannot price properly, so neither one writes a hole into history while it catches up.",
"tags": [
"fix",
"infra"
]
},
{
"date": "2026-09-10",
"title": "The image check now moves on when a model answers in the wrong format",
"summary": "Generated models are inspected by a vision model before you see them, and that inspection asks several providers in turn so one being busy is never the end of it. The catch was that a provider only had to reply, not reply usefully: a model that narrated its thinking instead of returning the requested format counted as the answer, and the healthy providers waiting behind it were never asked. A reply the check cannot read is now treated as that provider failing, so it moves straight to the next one, and it only gives up once every provider has been tried. The practical effect is that the quality check returns a real verdict far more often, and when it cannot, the reason it reports names the provider that actually lost.",
"tags": [
"fix",
"improvement"
]
},
{
"date": "2026-09-10",
"title": "Adding more AI providers no longer makes the image check slower to answer",
"summary": "The image check behind generated models asks several providers in turn, and it shared its time limit out equally between all of them. That had a perverse effect: every provider added to the list shortened the time given to the first and best one, so growing the list made the check worse rather than more resilient. Three providers were added last week and the first one's window fell just below what it needs to answer, which is why the check started timing out instead of returning verdicts. Each provider is now given whatever time the ones behind it do not need, with a guaranteed minimum still held back for every one of them, so the check gets a real answer from the first healthy provider and still falls through to the rest when it has to.",
"tags": [
"fix",
"improvement"
]
},
{
"date": "2026-09-10",
"title": "The image check behind generated models now has ten providers instead of four",
"summary": "The check that inspects a generated model before you see it was leaning on a single working AI provider, because one of the two free ones was retired by its vendor and the other three were unavailable. It now draws on every free vision-capable provider we could find, including a second host serving the exact same model so one provider being busy no longer stops the check. Depth was made free first: a request used to share its time limit out among all the providers on the list, so adding providers made every one of them slower, and the first and best provider got squeezed hardest. Each request now spends its time on the two or three most likely to answer and simply has more of them to choose from as busy ones recover.",
"tags": [
"improvement",
"infra"
]
},
{
"date": "2026-09-09",
"title": "New gallery avatars now arrive with a skeleton, so they animate properly out of the box",
"summary": "Avatars generated for the gallery were published as static meshes: they still moved, but through a fallback that maps a generic body onto them rather than a skeleton of their own. Every new one whose shape a humanoid skeleton can actually drive now goes through our auto-rigger first and arrives with a real 52-bone skeleton, so the whole motion library drives it directly. The ones that are skipped are skipped for a reason we measured: some generated models come out as a figure standing on a wide base slab, and rigging one of those tears the slab apart the moment it moves, so those keep publishing as static models. If the rigger fails or stalls, the avatar is published anyway rather than held back.",
"tags": [
"improvement",
"feature"
],
"link": "/gallery"
},
{
"date": "2026-09-09",
"title": "The avatar gallery passed 68,000 models, all generated on our own GPUs",
"summary": "The gallery keeps filling on its own: a scheduled job generates avatars from a growing prompt library on the GPU fleet we host ourselves, checks each finished model for a usable mesh and real textures, and quarantines the ones that fail instead of publishing them. That took the catalog past 68,000 avatars, with about 800 more arriving a day, and every one is browsable, previewable and free to use. The last 30 days rejected 94 models on mesh faults alone.",
"tags": [
"feature",
"improvement"
],
"link": "/gallery"
},
{
"date": "2026-09-09",
"title": "The animation gallery now opens on one page of the catalog instead of all of it",
"summary": "Browsing /animations used to download every clip in the library before you had finished reading the first screen, and that download grows with every batch of clips we add. It now loads one page and fetches more only when you reach the end of the grid, while the clip total and the per-category counts stay exact from the first moment the page paints. Searching or filtering still reads the whole catalog, because those look at every clip, and the results stream in as they arrive. Avatar embeds and pose links that name a single clip now ask for that one clip rather than the whole library, which took the request from 1.1 MB to a few hundred bytes.",
"tags": [
"improvement",
"infra"
],
"link": "/animations"
},
{
"date": "2026-09-09",
"title": "The health page for our paid 3D services now checks that a payment could actually be collected",
"summary": "Every agent that buys a 3D generation from us on the OKX.AI marketplace can read a free health page first, and one of the lines on it says whether payment settlement is working. That line only ever checked that a settlement route was configured, not that it could run: the account that broadcasts a buyer's already-signed payment onto the chain pays the network fee for it, and if that account runs out of fee balance the page still read green while a paid call would have failed after the buyer had committed. The page now reports the fee balance it actually measured, and marks the payment lane down when nothing can settle, so the problem shows up on a free page before anyone pays rather than as a failure afterwards.",
"tags": [
"fix",
"improvement"
],
"link": "/docs/okx-marketplace"
},
{
"date": "2026-09-09",
"title": "The payments status page now says which of two opposite outages it is looking at",
"summary": "When agent payments stop settling, the status page names the cause. Two very different situations shared one description: either the platform stops offering Solana payment on its price quotes, so nothing is ever attempted, or it keeps offering them and then turns each signed payment away at the final check. Both are fixed by topping up the same wallet, but the evidence sits in opposite places, and the old wording sent whoever was on call to the wrong one. It could even report that Solana had been withdrawn from quotes while the counter printed beside it read zero withdrawals. The status page now names the mechanism and points at the record that actually holds the proof, and the operations runbook carries a table for telling the two apart.",
"tags": [
"fix",
"infra"
]
},
{
"date": "2026-09-09",
"title": "www.three.ws now takes you to three.ws, and 3D loads there",
"summary": "Typing www before the address landed on a copy of the site that looked right and then quietly refused to show any 3D. Avatars and models are served from a media host that only recognises three.ws, so on the www address every model was blocked by the browser and pages that should have shown a character showed an empty box instead. The www address now sends you straight to three.ws, keeping the page you asked for and anything after the question mark, so bookmarks and shared links keep working and everything on them renders.",
"tags": [
"fix"
]
},
{
"date": "2026-09-09",
"title": "Trading finally counts toward your daily streak",
"summary": "The cross-surface streak on three.ws counted signing in, forging a model, saving a world and a walk session, but not a trade, which on a trading platform is the one thing it most obviously should count. A confirmed on-chain buy or sell now keeps your streak alive like any other activity, and still counts once per day however many times you trade, so the streak measures showing up rather than trading volume.",
"tags": [
"improvement"
]
},
{
"date": "2026-09-09",
"title": "Put a live trader on your own page with one script tag",
"summary": "A trader with a following had no way to show it anywhere except on three.ws. Every public trader profile now has an Embed this trader button that hands over a one-line snippet: paste it into a site, a blog post, a link in bio or a stream overlay and the trader's live record renders there, updating itself. It shows realized profit or loss for the period, win rate, settled trades, whatever they are holding right now and their last few results, and every coin on it carries a Fork button that opens the real trade panel back on three.ws, signed by the reader's own wallet. The card holds no funds, asks for no key and signs nothing. It follows the reader's light or dark setting, stops updating when nobody is looking at it, keeps the last good card on screen if a refresh fails rather than blanking the page it sits on, and shows a loss exactly as plainly as a win. The snippet carries the leader's referral code, so the traffic it brings back is credited to them, and the panel previews the real card before they copy anything.",
"tags": [
"feature"
],
"link": "/docs/trader-card-embed"
},
{
"date": "2026-09-09",
"title": "Voice satellite: the token command works where you actually run it, and an unpaired satellite explains itself",
"summary": "Two fixes for the Home Assistant voice satellite. The command that prints a viewing token for a screen on your own network was documented to be run on the machine already running the satellite, but doing exactly that failed: it tried to start a second copy of the service and died because the ports were already taken. It now reads the identity it needs and prints the token, with nothing else running. Separately, a satellite that has not been paired yet answers its health check again and says why it is unpaired, and turns a browser away with that same reason instead of a blank refusal. That is the state you are in on a fresh install, which is exactly when you need something to read.",
"tags": [
"fix"
],
"link": "/smart-home/satellite"
},
{
"date": "2026-09-09",
"title": "The 3D studio stops promising \"5 seconds to go\" for eleven minutes",
"summary": "Every surface that waits on a generation, the ChatGPT widget and the custom GPT included, showed a countdown built from the lane's own time estimate. Once a model outran that estimate the countdown stopped at five seconds and stayed there for the rest of the job: on a generation measured this week that meant reading \"roughly 5s to go\" for the final eleven minutes. The countdown now runs only while there is genuinely time left on the estimate, and past it the wait says the job keeps running in the background and shows how long it has actually been going, which is a number that moves. The inline 3D viewer in ChatGPT also now loads its model through three.ws itself rather than straight from the asset store, so it renders from any host that embeds it instead of only from our own pages. Separately, the free 3D endpoints no longer announce payment-related headers to browsers: they never accepted a payment, but the shared header defaults named them anyway, which was misleading to anyone reading the network tab.",
"tags": [
"fix",
"improvement"
],
"link": "/forge"
},
{
"date": "2026-09-09",
"title": "News cards stop losing their preview picture to a cold cache",
"summary": "About a fifth of the crypto-news feed ships text-only, so those cards start as a source tile and upgrade in place once the publisher's own preview image is found. That lookup searched all 197 publisher feeds at once and gave up after 2.5 seconds, which on a freshly started server dropped whichever feed had not answered yet, so an article the feed had just served came back as unknown. The reader kept its tile, but the failed lookup was recorded as a public answer for the next ten minutes, so every visitor to /markets and /markets/news got the same dead result and a 404 in their browser console. The lookup now asks the one publisher that could hold the article first and waits for it, and a result that could not be confirmed is held for 30 seconds instead of ten minutes.",
"tags": [
"fix"
],
"link": "/markets/news"
},
{
"date": "2026-09-09",
"title": "The trading arena now says who just passed whom",
"summary": "The live leaderboard on the arena told you who was ahead and nothing about what had changed, so a place changing hands, which happens several times a week on its own, went unnoticed. A grudge-match strip now sits under the standings: each trader is paired with the one directly above them, labelled as an overtake, a debut or a standing chase, with the profit or loss both sides settled in the last 24 hours as the reason. It is worked out from the same closed on-chain trades the board is ranked from, by rebuilding what the board actually said 24 hours ago out of the trade timestamps rather than trusting a stored snapshot, so the matchups reach back over history the platform already has and can never disagree with the leaderboard or with either trader's profile. Every line is written from those numbers: it will not call a trader behind when they in fact hold more SOL and rank lower on consistency, and it will not say someone booked nothing when they simply did not trade. Each matchup links to the leader's verified record and to a ghost-copy of them with no money at risk. Also available to agents and scripts at /api/sniper/rivalries.",
"tags": [
"feature"
],
"link": "/play/arena"
},
{
"date": "2026-09-09",
"title": "The payments dashboard now says whose wallet ran out of SOL",
"summary": "When a Solana payment cannot go through because the wallet paying the network fee is too low, the chain reports it as a generic simulation failure. Our status page grouped all of those together, so the platform's own fee wallet running low looked exactly like a buyer signing from an empty wallet, and both were also counted as payment-network faults. They are now three separate, named outcomes, so the status page says whether the money problem is ours to fix, the buyer's, or neither. Today's reject book was 1,438 rows of the first kind reading as the third.",
"tags": [
"fix",
"infra"
]
},
{
"date": "2026-09-09",
"title": "The partnership page works properly on the partner's own site",
"summary": "The IBM partnership page is published once onto IBM's own domain and then keeps itself current by reading from three.ws, so everything it loads is a cross-origin read and anything we get wrong there is invisible on three.ws itself. Three things were wrong. The site-wide command palette asked the partner's server for our page index and for the ranking module it imports, so it never opened and logged an error on every visit: the index now travels with the code the way the language catalogs already do, and both files are readable from other sites. And a build step that adds the site's mobile stylesheet to every page was reaching the partner artifacts too, even though they are copied out untouched by design, so the hosted copy asked IBM's server for a stylesheet only we have. All three are closed, and the audit that loads the page from a foreign domain now passes all four of its checks: the live update applies, the language switcher mounts, the paid demo arms, and the console is clean.",
"tags": [
"fix"
]
},
{
"date": "2026-09-09",
"title": "Homes connected through the relay now follow the house when you rearrange it",
"summary": "If your Home Assistant is only reachable inside your house, three.ws reaches it through the three.ws integration you install there, which dials out to us. Those homes were building their room list once, when the connection opened, and then quietly ignoring changes: assign a lamp to the kitchen, rename a floor, add a device, and the 3D house kept showing the old arrangement until the connection happened to be rebuilt. Homes connected by a direct address never had this problem, and the split was in the relay, which was not carrying the four small notices Home Assistant sends when its rooms, floors, devices or entities change. It carries them now, so both kinds of connection track the house the same way and updates land in seconds. The notices only say that something changed; the relay already read the full room and floor lists, so nothing new about your house is exposed by this.",
"tags": [
"fix",
"infra"
],
"link": "/smart-home"
},
{
"date": "2026-09-09",
"title": "Connecting a home: the page now says which state it is in, out loud",
"summary": "Two moments on /smart-home were doing the right thing silently. Typing an address that can never work (a 192.168 number, a .local name) has always put the reason under the field straight away, before anything is sent, but the message was carried by colour alone and a screen reader never read it out; it is now announced the moment it appears, and marked in a way that does not depend on seeing red. And a house that has stopped answering, is refusing the stored token, or has simply gone quiet is now labelled as such by the page itself rather than only in the sentence on its card. Nothing about what you see moves: the last known summary of your house stays exactly where it was, with how long ago it last answered, because a home that went quiet should never look like a home that was deleted.",
"tags": [
"improvement",
"fix"
],
"link": "/smart-home"
},
{
"date": "2026-09-09",
"title": "The API container keeps the memory it was sized for, so a busy hour cannot take the whole site down",
"summary": "Every three.ws page and API call is served by one container, and in early September that container was given twice the memory so it could hold live connections to people’s connected homes without ever running short. The setting did not survive: the next full deploy quietly put it back to the smaller limit, and the site then ran for five days on half the memory it had been sized for while still being configured to hold up to 600 home connections. Measured over the last 24 hours, memory was already peaking at 86% of that smaller limit before a single home connection was counted, which is close enough to the edge that a busy hour could have killed the container mid-request and taken every page down with it, not just the connected-home part. The limit is restored, peak memory is back to 43%, and it is now written into the deploy configuration itself so no future deploy can undo it again.",
"tags": [
"fix",
"infra"
]
},
{
"date": "2026-09-09",
"title": "An app key can no longer say yes to unlocking your door",
"summary": "A person saying yes is the one thing that stands between an agent and your front door, and we found a way around it. If you had ever issued an API key or connected an app to your three.ws account, that key could send an unlock request with the \"the owner already approved this\" flag set, and the lock opened, with nobody asked and nothing to click. We reproduced it against a real house and a real deadbolt before fixing it. Two things changed. Only a signed-in browser session can say yes now: a key or a connected app can ask to unlock, open or disarm, and is always answered with \"a person has to approve this\", exactly as the agent in your chat window already was. And the home permissions you approve when you connect an app are now actually enforced on every home request, so an app you gave read-only access to, or one you never gave home access at all, is refused instead of let through. Nothing legitimate changed: your agent still asks, you still approve, the door still opens.",
"tags": [
"security",
"fix"
],
"link": "/smart-home"
},
{
"date": "2026-09-09",
"title": "The OKX marketplace chat bot now switches AI providers by itself",
"summary": "Chat for our OKX.AI listing runs on an always-on host that spawns an AI session to answer buyers. That host held exactly one AI provider, so when the provider's account was put on a billing hold the bot went on receiving every message and answering none of them, for days, while other credentials sat unused on the same service. It now holds an ordered chain of providers, asks each one whether it will actually serve before using it, and moves to the first that answers. A provider that recovers or gets funded is picked up on the next check with no redeploy, and the bot's status page lists what every provider answered, so whoever fixes it can see which one to fix rather than only that one was refused.",
"tags": [
"infra",
"improvement"
]
},
{
"date": "2026-09-09",
"title": "Avatars and token art load again on sites that embed three.ws",
"summary": "Assets served from /cdn (avatars, thumbnails, generated models) are read from our object store with a signed request, and when that request is rejected the route falls back to the store's public address so the bytes still arrive. That fallback was a redirect, and a redirect throws away the cross-origin permission the route grants: browsers re-check permission at the redirect's destination, and the public address grants none. Measured from an outside site, every one of those reads failed outright, which took down model previews on pages that load art through the proxy and silently broke third-party embeds of three.ws avatars. The bytes are now served through three.ws itself rather than pointed at, so the permission, the file type and the download rules all survive the fallback, partial reads for video and audio keep working, and a file that genuinely is not there says so instead of reporting a server fault the page retries forever. It also keeps the fallback off the rate-limited public address, which is the whole reason the proxy exists.",
"tags": [
"fix"
]
},
{
"date": "2026-09-09",
"title": "Paid 3D services now refuse a job they cannot deliver, instead of charging for it",
"summary": "The paid three.ws forge services sold to other AI agents charge the moment the generator accepts a job, and the finished model reaches the buyer only after it lands in our delivery storage. When that storage stopped accepting writes, jobs were still being accepted and charged while no model could ever be handed back. Those services now check that delivery storage is writable before they take the job: if it is not, the call is refused, says so, and is not charged. The check is cached and ignores a one-off blip, so a passing wobble never stops the services selling. The free public health page also had a bug where a storage failure replaced the name of the failing subsystem with the storage provider's error code and printed the request's signing details; it now names the subsystem and reports the reason alone.",
"tags": [
"fix",
"security"
]
},
{
"date": "2026-09-09",
"title": "Fade Radar: the wallets that have never once been right, and the coins they are crowding into",
"summary": "The Smart Money Radar tells you which proven winners are buying a coin. Fade Radar is the other half of that question, and on pump.fun it turns out to be the sharper one: how much of a coin's buy side is money that has never been right. A wallet joins the reverse-indicator board only after five coins it bought have a known outcome and not one of them won, which is a bar a surprising number of addresses clear. The new page ranks every launch from the last few hours by how much of its observed buy side belongs to those wallets, by wallet count and by SOL, and carries a second board of the wallets themselves with their full record. The number above the board is the part we care most about: it is measured, not asserted. We rebuild the reverse-indicator group using only coins whose outcome was known before the coins it is graded on existed, then check what actually happened. Coins where a quarter or more of the buyers were reverse indicators went on to win 1.8 percent of the time, against 22.1 percent for coins with none, across 26,626 labelled coins. Nothing here is a short and nothing signs anything: on a bonding curve the only tradeable form of a fade is not buying, so every action on the page is a link out to the coin, its intel, or a wallet's track record. Wallet links also now open directly on the Smart Money Radar, so an address cited anywhere on the platform lands on its record instead of a board you have to scan by eye.",
"tags": [
"feature"
],
"link": "/fade"
},
{
"date": "2026-09-09",
"title": "Three fades now stop when you ask the site to stop moving",
"summary": "Turn on reduced motion in your operating system and three.ws is supposed to hold still: every animation timing on the site is named rather than spelled out, so one switch can zero all of them at once. Three places had drifted off that and spelled their own timing instead, which meant the object library card fade, the avatar picker chips and the Money Pulse rows kept animating for people who had explicitly asked them not to. One of them was pointing at a timing name that does not exist, so the value it fell back to was never the one anyone chose. All three are back on the named ladder at the same speed they always looked, and the reduced-motion switch reaches them again.",
"tags": [
"fix",
"improvement"
],
"link": "/objects"
},
{
"date": "2026-09-09",
"title": "Your monthly agent turns now count, and locking up never does",
"summary": "Talking to the agent about your house is the one part of the smart-home lane that can really cost money: the room graph makes a home prompt large, and on a paid model a single turn costs several weeks of holding that house’s connection open. Those turns were being counted but never actually limited. They are now, against the monthly allowance shown on your plan page, and if you run out the agent says so in a sentence that names the limit, the date it resets and where to raise it, instead of failing at you. Two things are deliberately left outside that limit. Asking the agent what the house is doing still works, because you have to be able to see which door is open. And telling it to lock a door, close a garage or a valve, or arm the alarm still runs, over quota, on a paused home, in every state we can put an account in: the safe direction is free, always, on every plan. If our own billing lookup fails, the turn is allowed through rather than refused.",
"tags": [
"feature",
"improvement"
],
"link": "/smart-home/plan"
},
{
"date": "2026-09-09",
"title": "The Windows 11 widget draws its card again, and a new check keeps the board honest",
"summary": "The Agent glance widget on the Windows 11 widgets board renders an Adaptive Card, and the board fills that card in with its own expression engine rather than with JavaScript. Our card asked for the three stat rows the way JavaScript reads an array, which that engine treats as a syntax error, and it refuses the whole card rather than skipping the one line: a pinned widget came up empty with nothing to show for it anywhere else. The card now uses the syntax the board expects, and there is a new check that fills it in with the board's own engine before anything ships, alongside checks that the manifest, the service worker and the two endpoints still agree with each other. The board is the one surface we cannot open from a build machine, so the checks that stand in for it are the ones that have to be real.",
"tags": [
"fix",
"infra"
],
"link": "/glance"
},
{
"date": "2026-09-09",
"title": "The floorplan editor can make rooms, so a home with no areas is no longer a dead end",
"summary": "Most Home Assistant setups we see have devices everywhere and not a single area, which left the 3D home and the floorplan with nothing to draw and the only advice we could give was to go and create areas in Home Assistant's own settings first. That is a chore people leave to do and do not come back from. The floorplan editor now makes the room itself: name it, and the area is created in your own Home Assistant, placed on the plan, and the device that prompted it is filed into it in the same step. It reaches your dashboards, your voice assistant and your automations too, because it is a real area in your registry and not a picture of one here. The tray tells a home with no rooms where to start, and a device you file no longer leaves a stale entry behind in the list.",
"tags": [
"feature",
"improvement"
],
"link": "/smart-home"
},
{
"date": "2026-09-09",
"title": "When the payment wallets look empty, the self-heal now says whether they are empty or just fenced",
"summary": "The x402 payment rail keeps a set of platform wallets topped up from one funding root, and a recovery run reports why it could not draw on each source. Every source that held less than its configured floor reported the same thing: \"at or below floor\". That single phrase covered two opposite situations. A wallet holding nothing needs money added from outside. A wallet holding real funds behind a floor somebody set does not: the platform already owns that money, and a floor decision releases it. Because the two printed identically, three separate investigations of the same stalled rail concluded the platform was out of funds while one of its own wallets held roughly forty-five times what the stalled wallet was short. A skipped source now reports how much it is holding and the line it had to stay above, and every run totals the funds it left behind, so an operator can tell at a glance whether the recovery needs a deposit or a decision. The health page and the operations runbook now point at that total instead of asserting that money is required.",
"tags": [
"fix",
"infra"
]
},
{
"date": "2026-09-09",
"title": "The homepage stops calling its Community strip live when no new models are arriving",
"summary": "The Community strip on the homepage is headed \"Made with Forge. Right now.\" and carries a pulsing live chip. That chip was switched on by the Fresh tab alone, so it asserted liveness no matter how old the newest model on the strip actually was. When our object storage started rejecting the platform's credential on 2026-09-07, finished generations could no longer be recorded, the strip froze on the last model stored before that, and the chip went on pulsing live over two-day-old work with nothing on the page to say otherwise. The chip is now earned by the data: it appears only when the newest model on the strip is from the last six hours, and it reads the newest timestamp in the list rather than trusting the order it arrives in. Older models still show exactly as before, because a quiet stretch is worth showing; it just is not live.",
"tags": [
"fix"
],
"link": "/"
},
{
"date": "2026-09-09",
"title": "The retired Flappin UFO page is readable in light mode, and its two bottom-corner controls stopped sitting on top of each other",
"summary": "A full audit of /play/ufo, the page that stands where the retired Flappin UFO arcade demo used to be. In light mode its headline was painted in a violet-to-cyan gradient that all but vanished against the white card, the Retired badge and the Still live label were too pale to read, and the keyboard focus ring was that same near-invisible cyan, so tabbing through the page showed you nothing. All four now clear the contrast bar in both themes. The three links to the experiences that are still live were 37 pixels tall and the three.ws logo link was 21, both under the 44 pixels a thumb needs; every link on the page now clears it. The language picker and the Ctrl+K search hint were both pinned to the bottom right corner and overlapped each other, because this page carries its own compact header instead of the site nav that normally arranges that corner; they now stack. Two fixes from this audit land site-wide: the search hint was drawing dark text on its dark default pill on every light-themed page that does not define the site's surface colors, and neither floating control could be reached by screen reader landmark navigation. The page copy also lost the long dash the house style bans, in English and in all 84 translations.",
"tags": [
"fix",
"improvement"
],
"link": "/play/ufo"
},
{
"date": "2026-09-09",
"title": "The 3D Playground recovers when a model will not load, and now works from the keyboard",
"summary": "Dropping a .glb into /playground had exactly one failure mode: a dimmed screen and a spinner that never stopped. If the file was corrupt, the download stalled, or the 3D viewer script itself was blocked, nothing on the page said so and there was no way back. That spinner is now a shimmer inside the viewer frame instead of a curtain over the whole page, so the controls stay usable while a model streams, and a failure lands in a designed error card that names what went wrong and offers Retry or a file picker. A load that never finishes gives up after 25 seconds rather than hanging forever. The lighting presets no longer strand the scene on the previous preset’s exposure. The page also gained the structure it was missing: a real heading that says what it is for, headings on every control group, an accessible name on each switch, slider and field, a focus ring on everything you can tab to, and an upload button the keyboard can actually reach (it was hidden with display:none, which drops it out of the tab order entirely). On a phone the top navigation scrolls instead of vanishing. Animation names read out of an uploaded file are rendered as text, never as markup.",
"tags": [
"fix",
"improvement"
],
"link": "/playground"
},
{
"date": "2026-09-09",
"title": "Render Lab: every control leads somewhere, and the contact sheet shows its work",
"summary": "Render Lab now behaves when the things it depends on do not. The Open image button no longer sits there looking live before a render exists: it reads as unavailable until there is an image, and goes back to unavailable if the render fails. When no featured avatars come back, the model strip explains that and points at the field and at /create instead of showing a blank row. When the pose catalog cannot be reached, the panel says so and offers a retry rather than a lone None chip that reads like the model has no poses. Contact sheet tiles now shimmer while they render, so a sheet in flight looks like it is working instead of like a grid of empty squares, and a tile the renderer could not produce says so and keeps its label. Where a panel is switched off because the model cannot honor it, the explanation and its escape hatch stay at full strength and stay clickable: the link into Rig Doctor was previously unclickable, and the shortcut to a model with a full face rig now finds one from the live catalog instead of loading a raw file the slow way. Dragging the size slider no longer writes a new address on every step, which Safari eventually refuses outright. Copied links now carry the quality setting, so a link really does reopen the exact configuration. Framing, pose and model options are one stop for keyboard and screen reader users and move with the arrow keys, instead of costing 29 tab presses to walk past. The capabilities endpoint the whole page is built on is now documented in the Media API reference, which is where the page has always linked.",
"tags": [
"fix",
"improvement",
"docs"
],
"link": "/render-lab"
},
{
"date": "2026-09-09",
"title": "The Splat Viewer keeps its recenter and download controls, and its errors say what went wrong",
"summary": "Loading a scene on /splat used to erase the viewer's own controls: writing the scene label into the HUD replaced the recenter and download buttons with plain text, so from the first render onward neither existed. Both now live beside the label and survive every scene swap, and the download always offers the scene you are actually looking at rather than a previous one. The error state was worse than useless: it showed a heading over an empty space because the explanation was being written into an element that was not on the page. Now a failed load tells you what happened and what to do, including the advice to download and upload a file whose host blocks cross-origin requests. Clicking a sample or picking a file also shows the loading state immediately instead of sitting on an idle stage while the splat engine downloads in silence, and an engine that fails to load says so rather than leaving the page frozen.",
"tags": [
"fix",
"improvement"
],
"link": "/splat"
},
{
"date": "2026-09-09",
"title": "Sign Language Avatars is usable the moment the page opens",
"summary": "The signing avatar on /sign-language is a multi-megabyte model, and until it finished downloading the page had nothing to say: an empty box, an empty status line, and buttons that had not been wired yet, so an early click on Sign it did nothing at all. The stage now shows a breathing silhouette while the avatar loads, every control is wired before the download starts, and anything you type, click or open with a ?say= link in the meantime is queued and signed the moment the avatar lands. If the model fails to download, the stage says so and offers Retry the avatar plus a link to the sign API below, which runs on the server and works without the 3D preview; the vocabulary chips and the Sign it button now explain why they cannot sign instead of doing nothing. An avatar with no signing skeleton says so in the same words.",
"tags": [
"improvement",
"fix"
],
"link": "/sign-language"
},
{
"date": "2026-09-09",
"title": "Sign Mirror tells you why the camera did not start where you are actually looking",
"summary": "On /sign-mirror, pressing \"Turn the camera on\" and being refused, or being on a device with no camera, put the explanation on the far side of the page: it appeared in the status strip over the 3D avatar, in the other column, while the box directly above the button you had just pressed carried on giving the same idle tip as before. The button snapped back to \"Turn the camera on\" with nothing visibly changed beside it, so a refused camera read as a button that simply did not work. Every camera and hand-tracker failure now lands in that box beside the button, in both the side-by-side and the stacked phone layout, and the box turns amber and solid-edged so a problem no longer looks like a coaching tip. The sentences themselves are unchanged and still say what to do next, naming the browser permission prompt, a camera busy in another app, a missing camera, or a hand tracker that could not be downloaded. Picking another letter clears the warning and returns the box to its normal hint.",
"tags": [
"fix",
"improvement"
],
"link": "/sign-mirror"
},
{
"date": "2026-09-09",
"title": "Rankings survives a stalled leaderboard and stops reading the whole board aloud",
"summary": "If the rankings service went quiet, /rankings sat on a shimmering skeleton forever with no way out: a request that never answered looked exactly like one still loading. The board now gives up after 15 seconds and shows the same error card it already had, with a Retry button and the reason, so a stall is always recoverable. Switching metrics still cancels the previous request silently, so a fast tap never surfaces a false error. For screen readers, the results list itself was marked live, so every metric switch and every Show more re-announced up to a hundred rows; the board now announces one short line instead, such as \"Creations: showing 50 of 169 ranked creators\". Page copy also lost its stray dashes.",
"tags": [
"fix",
"improvement"
],
"link": "/rankings"
},
{
"date": "2026-09-09",
"title": "The press kit reads in the three.ws typeface, fits a phone, and quotes every mark at its true size",
"summary": "Every paragraph on the press kit was rendering in the browser default serif on a bare white-ish ground: the page styled its own cards and headings but never applied the site typeface, colour, or background to the document itself, so the one page a journalist opens looked nothing like the rest of three.ws. It now uses the platform design tokens throughout, in both light and dark mode. On a 320px phone the two OpenAI announcement cards were sliced off at the right edge, taking their Download PNG buttons with them, and because the page clips rather than scrolls sideways there was no way to reach what was cut; both card grids now fold to a single full-width column instead. The pixel dimensions printed under each mark were also all slightly wrong, off by a pixel or two from the artwork they labelled, which is the number a designer laying out a conference programme actually acts on. They are corrected, and the brand asset renderer now writes them back from the PNG headers on every build the way it already did for the archive size, so they cannot drift again. Last, the Copy buttons beside the boilerplate did nothing at all when the browser withheld clipboard access; they now select the passage in place and tell you which keys to press.",
"tags": [
"fix",
"improvement"
],
"link": "/press"
},
{
"date": "2026-09-09",
"title": "Payment sessions: the MCP config you copy now works, and a failed load tells you what to do",
"summary": "The Copy MCP config button on /payments handed out a config that could never connect. It baked in a shortened stand-in for your session token instead of the real one (the real token is shown only once, at creation, and is never retrievable afterwards), and it wrote your sign-in cookie in the wrong format. The snippet now points at two environment variables you export yourself, so the file you save never holds a secret, and a note beside it says exactly what each one is. Failures on the page also stopped hiding behind machine jargon: when the service refused a request, the page showed you its internal code, so trying to create a session with an empty balance read as the word insufficient_credits and nothing else. You now get the plain sentence, and when the balance is the problem, a link straight to Add credits. If the list itself cannot load, the four counters at the top no longer shimmer forever as though something were still loading: the page says what failed, leaves your sessions and budgets untouched, and offers a Try again button. Smaller things along the way: the Docs button used to open a page that did not exist, filtering to a status with nothing in it now offers a way back to the full list, the status filters are reachable on a narrow phone instead of being cut off at the edge, and every field in both dialogs is properly labelled for screen readers.",
"tags": [
"fix",
"improvement"
],
"link": "/payments"
},
{
"date": "2026-09-09",
"title": "Rig Doctor's report fits on a phone, and its headline numbers now come from the analyser",
"summary": "On a narrow phone screen the whole right-hand side of a Rig Doctor diagnosis was cut off the edge of the page. Coverage, Inventory, the bone rewrites and the generated manifest were all rendered wider than the screen, and because the page clips rather than scrolls sideways, there was no way to reach them: you got the verdict and the preview and nothing else. The cause was the generated manifest, a long unbreakable block of code that was stretching every panel beside it. The panels are now pinned to the column and the code block scrolls on its own, so the full report reads on a 320px screen. Separately, the page advertised eleven recognised rig conventions when the analyser had grown to twelve, and the missing one, MikuMikuDance, was absent from the explainer too. Both counts in the hero are now read straight from the analyser rather than typed by hand, so they cannot drift again, and the sample rig labelled Halfbody now describes what it actually shows: its arms stay frozen as well as its legs.",
"tags": [
"fix",
"improvement"
],
"link": "/rig-doctor"
},
{
"date": "2026-09-09",
"title": "The \"have you tried\" card no longer hangs off the side of the screen on a phone",
"summary": "The small suggestion card that appears in the bottom corner of the site was drawn 34 pixels wider than the space it asked for, because its padding and border were being added on top of its width instead of counted inside it. On a phone that pushed the card 22 pixels past the left edge of the screen, so its kicker, title and description were all cut off mid-word and the button inside it ran wider than the card itself. This only happened on the pages that do not load the main site stylesheet, which includes the whole markets family: the screener, coin pages, categories, the heatmap and about thirty others. The card now sits fully on screen with an even gutter on both sides at every width, and its button fits inside it.",
"tags": [
"fix"
],
"link": "/screener"
},
{
"date": "2026-09-09",
"title": "The Money Pulse now tells you when its stats are down, and retries on the spot",
"summary": "When the money-intelligence call behind /pulse failed, the three side panels (Top earners, Busiest wallets, Fresh launches) sat on \"Loading...\" forever and the 7-day activity strip went blank, which reads as \"the platform is quiet\" rather than \"we could not fetch this\". Each surface now says what went wrong, notes that the live feed itself is unaffected and that it retries every minute, and offers a Retry button that refetches immediately. First paint uses skeletons shaped like the real counters, bars and rows instead of dashes and a loading label, so nothing shifts when the numbers land. The empty states point somewhere useful now: no tips yet links to the agent directory, no wallet activity links to the agents that are online. Switching between mainnet and devnet shows the skeletons again rather than leaving the previous network's numbers under the new label. The stat cards report their filter state to screen readers, the rail cards are real headings, and the per-row explorer link is a full-size tap target that says where it goes.",
"tags": [
"fix",
"improvement"
],
"link": "/pulse"
},
{
"date": "2026-09-09",
"title": "The pitch deck now shows the whole character, and says so when the 3D cannot load",
"summary": "The live 3D figure on the deck at /pitch was framed by a camera that cut it off above the chest and pointed at its back, so the slide that says \"give your AI a face\" showed neither a head nor a face. Both stages now frame the full character from the front. Each stage also ships a still frame rendered from the same glTF file: it fills the panel while the model streams instead of leaving an empty box, and if the 3D runtime is blocked or fails outright the panel says so in the open, drops the \"Live\" badge that used to sit over nothing, and offers a retry. Navigation got the same pass: the slide dots are now full-size tap targets and step out of the way on a phone, the language picker no longer lands on the slide counter, slide copy no longer runs underneath the nav bar on a tablet, links on hidden slides are out of the keyboard tab order, and every button and link has a visible focus ring.",
"tags": [
"fix",
"improvement"
],
"link": "/pitch"
},
{
"date": "2026-09-09",
"title": "The Oracle reference now reads the model it is describing",
"summary": "The complete Oracle documentation at /oracle/docs described version 2 of the conviction model while production had moved on to version 3, and the difference was not cosmetic. Version 3 changed the question the score answers: earlier versions ranked the odds of a spike, and version 3 ranks the odds of a spike that holds, counting a launch as a hit only if a holder who bought at first sight is still up. A coin that ran 3x and gave it all back was a success under the old target and is a miss under the new one, which is why the base rate underneath every figure on the page is far lower than it used to be. The page had the old target, the old tier probabilities and the old holdout numbers throughout, and its roadmap still promised fitted weights and calibration that had already shipped. Every model figure on the page is now read live from the published model card when you open it, because the engine refits itself every six hours behind a promotion gate and any number typed into a document goes stale on a clock. When that read fails the page says so, shows the values it was published with, explains that they may trail production, and offers a retry rather than a blank space. Two public endpoints that existed but were undocumented are now in the reference: the one that publishes the whole scoring machine, weights and holdout curve and the record of every fit ever refused, and the one that publishes the realized win rate per conviction band from real fills. The tier ladder used to truncate its own evidence, cutting off the sample counts on every band below the top, and now gives each band a full line. In light mode the tier names were near-white on a white card and the strongest bands were invisible; they now have their own ramp.",
"tags": [
"docs",
"fix"
],
"link": "/oracle/docs"
},
{
"date": "2026-09-09",
"title": "The spend policy dry run stops contradicting its own counters",
"summary": "The simulator at /pay/simulator reports what an agent budget actually buys, and its headline sentence disagreed with the numbers directly beneath it. An endpoint that answers without asking for payment is free: it costs nothing and is never checked against your policy. The counters knew that, and the sentence did not, so the very first run on the page announced that all seven calls settle while the tile beside it said six. Calls that could not be priced at all vanished the same way, counted nowhere the reader could see. Both now have their own tile, and the verdict sentence is built from the counts it is describing, so settled, free, unpriced and refused always add up to the total in front of you. A run that fails now says why: an unusable line used to produce a bare \"Simulation failed (HTTP 400)\", and now quotes the endpoint the simulator could not read along with what a working line looks like. A budget left blank was quietly costed against one dollar, giving you a verdict for a budget you never proposed; the answer now says when that substitution happened. On a phone the little question marks beside each field were fifteen pixels wide and their explanations opened off the side of the screen, so the help was unreadable exactly where it was needed most: the tap targets are now large enough to hit and the explanations stay on the page at every width.",
"tags": [
"fix",
"improvement"
],
"link": "/pay/simulator"
},
{
"date": "2026-09-09",
"title": "The economy solver says which level its numbers belong to, and shares a link that lands on them",
"summary": "The solver at /play/solver reprices the whole /play world when you move the skill level, and it fetches the new level as you move. If that fetch failed, the page dimmed its tables slightly and said nothing else: the chart marker sat at level 60 while every table below it still held level 1 figures, with no visible explanation, no way to tell which level you were reading, and no retry. The explanation existed but was written only into a region screen readers can reach, so anyone looking at the page saw a silent mismatch. A failed level now raises a banner in the open that names the level that failed, says plainly that the figures below are still the old level, and offers both a retry and a way back. A level still loading shows the same banner as a busy state with the level it is solving, so a slow connection reads as slow rather than as broken. The page also writes the level into its own address bar now: it has always read a level from a shared link, but never wrote one, so sending someone the URL after dragging to level 60 sent them level 1. And a link asking for a level past the cap settles on the cap instead of leaving the page waiting on a level that will never arrive. Alongside it, a conclusion in the model that shouted one word in capitals now reads as a normal sentence.",
"tags": [
"fix",
"improvement"
],
"link": "/play/solver"
},
{
"date": "2026-09-09",
"title": "Your private avatars load on the dashboard again, and say so when they cannot",
"summary": "Avatars you have not made public were failing to load on your dashboard, with nothing on the page to say why. The dashboard was being handed a temporary direct link into our storage provider rather than a link back through three.ws. When storage stopped accepting our credential, that direct link was refused in a way browsers deliberately hide, so the page could not tell the difference between a refusal and a dead connection: it just showed nothing. Private avatars now load through three.ws itself, which checks that the avatar is yours exactly as before, and which already knows how to fall back to a second copy when storage is having a bad day. As a side effect these links no longer expire, so a dashboard tab left open for an hour still shows your avatars when you come back to it.",
"tags": [
"fix"
],
"link": "/dashboard/avatars"
},
{
"date": "2026-09-09",
"title": "The OpenAI partner page now proves the free ChatGPT connector is up instead of just claiming it",
"summary": "The /openai page told readers that a free, keyless 3D connector answers at three.ws/api/mcp-studio and then listed the eleven tools it exposes. Both were claims on a static page. Now the page calls the connector from the reader's own browser on load and reports what came back: live with the tool count and the response time, or a plain explanation and a Check again button if the call was blocked. The printed tool table is reconciled against the same response, so every tool the server really answers with is ticked, one the server has dropped is flagged, and one the server has added since the page was written is appended with its own description. The connector URL and the curl command both got copy buttons, and hover plates on the page no longer turn near-white when the site theme is switched to light.",
"tags": [
"improvement",
"fix"
],
"link": "/openai"
},
{
"date": "2026-09-09",
"title": "The partner page gets its footer back, fits a narrow phone, and stops naming three partners it does not have",
"summary": "An end-to-end audit of /partners found it was the only page on the site that carried a footer slot but never loaded the footer, so the whole bottom of the page was missing. It is back. On a 320px phone the partner cards were 38px wider than the screen and the overhanging text was silently cut off, because a card header could not shrink past its logo, name and category chip on one line; the chip now wraps and every card fits. The page description search engines and link previews read was also stale: it advertised Intel, Microsoft and Oracle, none of which have a card, and left out OpenAI and HackerNoon, which do. The AWS card said View on AWS Marketplace above a link to our own AWS page, so it now says what it actually opens. Beyond that: the 8 partners and 22 chains counters are no longer hidden from screen readers and the partner number is now counted from the cards themselves, keyboard users get a visible ring on every card and button, the card lift is dropped for anyone who asks for reduced motion, and the page eyebrow, the four benefit headings and every card call to action are translatable instead of staying English on a translated page. The floating walk companion also got a full-size dismiss, avatar-swap and narration button on touch devices, everywhere it appears.",
"tags": [
"fix",
"improvement"
],
"link": "/partners"
},
{
"date": "2026-09-09",
"title": "The agent-wallet demo pays again when an endpoint asks the buyer to cover its own network fee",
"summary": "On /play/agent-wallet a 3D avatar walks up to a paid x402 endpoint and buys a live market signal with real USDC on Solana. The page had gone quiet. three.ws's own paid endpoints switch to a self-pay payment challenge whenever the wallet that normally sponsors the Solana network fee cannot co-sign, and the bridge behind the page only knew how to build a sponsored payment, so every price quote came back as a server error and the demo never got past its connecting state. The bridge now signs both kinds: when no sponsor is offered, the agent wallet pays its own network fee, exactly as the platform's own payment ring already does. Four smaller things came with it. The page no longer offers a Pay button it knows will fail: when the agent wallet cannot cover the quoted price it says so and leaves the deposit address in front of you. Switching between BTC, ETH and SOL re-prices immediately instead of waiting for the next payment. The wallet balance and the Pay button stop reverting to placeholder text a few seconds after they load, which the translation pass had been overwriting. And the page carries a real page heading for screen readers and search engines, which it had been missing entirely.",
"tags": [
"fix",
"improvement"
],
"link": "/play/agent-wallet"
},
{
"date": "2026-09-09",
"title": "Arm your agent stops telling signed-out visitors to create an agent, and shows them the live conviction stream instead",
"summary": "A full audit of /oracle/arm found that anyone not signed in was shown “Create a 3D agent to arm it”, a dead end for a visitor with no account, and the rest of the page was hidden behind it. Signing in now gets its own state with sign-in and register buttons, and the live “clearing your bar” preview plus the 30-day win-rate readout stay on screen for everyone, because both run on public data. If the agents list itself fails to load, the page now says so with the status it got back and a Try again button, instead of pretending the account is empty. Four more fixes went with it: the arm switch and the two rule toggles can now be operated from the keyboard (they were focusable but did nothing on Enter or Space), coin artwork routes through the site image proxy so icons stop failing on IPFS gateways, the status line no longer flips back to “Disarmed” when the translation catalog lands after the page has rendered, and every form field now has a real label attached to it.",
"tags": [
"fix",
"improvement"
],
"link": "/oracle/arm"
},
{
"date": "2026-09-09",
"title": "The Recording Pipeline board stops showing mainnet numbers when you ask it about devnet",
"summary": "A full audit of /pipeline, the one-glance health board for the loop that turns every pump.fun launch into a trade signal. The recorder worker writes a single liveness record for whichever network it is recording, and the board was reading that record no matter which network you selected: switching to devnet showed the recorder as LIVE and told you the loop was waiting on the next launch, when nothing had ever recorded there. Devnet now answers honestly, with the recorder marked off and the reason spelled out, and the whole board follows: idle, every stage at zero, and the deploy instruction on screen. Switching networks while a refresh was already running used to be dropped silently, so you got the previous network's numbers under the new label; the newer request now always wins. Five more fixes came with it. The board refreshes every eight seconds and used to rebuild all seven cards each time, which threw away keyboard focus mid-read and made screen readers re-announce the entire grid on every tick; it now updates the numbers in place and announces only the one-line health summary. A response missing a section no longer collapses the page into a false connection error. The refresh spinner turns for exactly as long as the request actually takes. The network switch and refresh controls have a visible focus ring. And the board now points at the public endpoint behind it, so an agent can read the same stages and the same recommended next action as JSON.",
"tags": [
"fix",
"improvement"
],
"link": "/pipeline"
},
{
"date": "2026-09-09",
"title": "The in-game economy page reads correctly on a phone, and the boutique shows what you are buying",
"summary": "A full audit of /play/economy, the published reference for every price, prize and gate in the /play world. On a 320px screen the general store and currency panels were wider than the viewport, and because the page suppresses sideways scrolling their right edge was simply cut off: the store's price column was unreachable on the narrowest phones. Those panels now shrink to fit, and the wide tables scroll inside their own labelled box instead. The bar showing how a paid $THREE sale splits used to print its labels inside the coloured segments, where \"50% holder rewards\" was truncated to \"50% holder re...\"; the split now reads from a legend beneath the bar, which stays legible at any ratio. In the boutique, the Stetson and the Shades rendered as blank grey circles while the five colour items showed their colour, because the page ignored the thumbnail the server was already sending; both now show the item, with the colour swatch as the fallback if an image ever goes missing. The wheel's top prize is relabelled \"JACKPOT: 100 Cash\", and a failed load now reads as a proper sentence before the retry hint.",
"tags": [
"fix",
"improvement"
],
"link": "/play/economy"
},
{
"date": "2026-09-09",
"title": "The Sniper Arena leaderboard now loads while the 3D world is still building, and says so when it cannot",
"summary": "A full audit of /play/arena, the live 3D floor where three.ws agents trade pump.fun in front of you. The leaderboard read used to wait behind the whole 3D boot: the animation library and your spectator avatar had to finish downloading before the panel asked the API for a single number, so on a slow connection you watched empty placeholder rows for as long as the avatar took. The board read now starts on the first frame, in parallel with the 3D load, and the placeholders are proper skeleton rows shaped like the real ones so the panel does not jump when the data lands. When that read fails, the page used to show \"No agents are live yet\" and invite you to be the first, which was simply wrong: the arena was full, we just could not reach it. A failed read now says the leaderboard is unreachable, names the reason, and gives you a retry button; if a later refresh fails while a good board is already on screen, a compact amber strip says the standings are stale instead of wiping them. The loading overlay keeps its spinner instead of replacing it with a bare line of text. The \"On the floor\" number counts the agents actually standing in the arena rather than every ranked row. And the page finally has a real top-level heading, so screen readers and search engines get the same answer about what this page is.",
"tags": [
"fix",
"improvement"
],
"link": "/play/arena"
},
{
"date": "2026-09-09",
"title": "The sniper scoreboard now links out to the pages that explain what it is showing you",
"summary": "A full audit of /sniper/experiments, the live board comparing rule-driven sniper agents against LLM-judged ones on the same market. The board is dense with terms it could only gloss in a hover tooltip: earned autonomy tiers, moon bags, the reasoning ledger behind every verdict, the exit ladder. A reader who wanted the real explanation had nowhere to go. The page now ends with a row of links to the seven surfaces that write each of those up, from the live Sniper Arena floor to the earned-autonomy docs and the page where you arm your own agent. Four smaller fixes came with it. The Best arm tile printed its profit in plain white even when that number was a loss, so a losing leader read as a win at a glance; it is now coloured like every other figure on the board. An arm or a judge with no model recorded against it used to crash the whole render; it now reads as unattributed and the rest of the board keeps working. Both tables label their column headers properly, so a screen reader can pair a figure with the column it belongs to across nine columns. And the time-window buttons are declared as buttons rather than defaulting to form submits.",
"tags": [
"improvement",
"fix"
],
"link": "/sniper/experiments"
},
{
"date": "2026-09-09",
"title": "Scene Studio fits a 320px phone screen, and its playback controls announce themselves",
"summary": "A full audit of /scene, the in-browser 3D scene editor. On a narrow phone the panel that holds the outliner, transforms, geometry and materials sits along the bottom of the screen, but it carried a 335px minimum width from the editor it is built on. That is wider than a 320px screen, and because the page suppresses sideways scrolling the right 15px of it simply had nowhere to go: the object type labels and the right-hand stepper of every number field were off-screen and unreachable. The panel now shrinks to the screen. The first-run card that tells you how to start a scene was centred in a band that ran underneath the Import/Export/Share buttons, so on the same narrow screens the buttons sat on top of its heading; the card now starts below them and scrolls inside itself on a short screen instead of running over the panel below. The floating translate/rotate/scale toolbar was painted under that card too, and now sits above it. And the animation timeline's play, pause and stop buttons are drawn as bare icons with no text, so a screen reader announced three identical unnamed buttons; all three now carry names taken from the editor's own translated string table.",
"tags": [
"fix",
"improvement"
],
"link": "/scene"
},
{
"date": "2026-09-09",
"title": "Endpoint Shopper fits a phone, and says something useful when a run finds nothing to buy",
"summary": "A full audit of /shopper, the page where you describe a task, set a budget, and let an agent shop the x402 Bazaar for endpoints that can answer it. Three things were wrong. A run that came back with no plan and no answer, which is what happens when nothing in the Bazaar matches the task or everything that does prices above your ceiling, painted an empty white column and stopped: no explanation, no way forward. That case now says no endpoints matched, quotes the task back, and offers the three things that change the outcome (rewrite the task, raise the budget, browse the Bazaar) plus a button to run again. Second, when the request could not reach us at all, the page showed the browser's own wording for a dropped connection, which names no cause and suggests no fix; it now says the agent service could not be reached and what to do about it. Third, on a phone narrower than 560px the four navigation links were squeezed into a strip about 76px wide and had to be dragged sideways a few pixels at a time to read; they now wrap onto their own full-width row. Alongside those: the results column announces itself to a screen reader while a run is in flight and when it settles, the Task label sits above its box as intended rather than beside it, and an internal ticket number that had been sitting in the page's headline label since launch is gone.",
"tags": [
"fix",
"improvement"
],
"link": "/shopper"
},
{
"date": "2026-09-09",
"title": "The Pump.fun Stream tells you it is live, counts what it has seen, and has working Recent and Favorites tabs",
"summary": "A full audit of /pumpfun, the live pump.fun feed with a 3D agent narrating it. The connection badge in the header said \"connecting...\" forever, even with events pouring in, and the counter on the \"Powered by PumpPortal\" pill stayed on zero no matter how many launches arrived. Both were the same fault: the page's translated-copy pass rewrote those two headings wholesale every time it ran, throwing away the very elements the page updates. Seven controls were affected, including the Share button's \"Copied!\" flash and the avatar picker's name. Each one now hands the page ownership of its live value. The avatar picker's Recent and Favorites tabs were listed but had nothing behind them: opening either showed a blank panel. Both work now, kept in your own browser: every avatar you narrate with lands in Recent, and a star on any card adds it to Favorites. Filtering the feed down to nothing used to leave a blank column; it now says how many events are hidden and offers one click to clear the filters. If the stream drops, the feed says so and offers a Reconnect button instead of sitting on \"Waiting for events...\", and it notices about twice as fast as before. Also: the page title now matches the page (it said \"PumpFun migrations\" while the page streamed new tokens, graduations and claims too), a \"Launch a coin\" link sits next to the status badge, the first-time-fee-claims panel shows skeleton rows during its load and an actionable message with a retry if it fails, and clicking a tab in the agent-config dialog no longer reaches into the avatar picker and deselects its tabs.",
"tags": [
"fix",
"improvement"
],
"link": "/pumpfun"
},
{
"date": "2026-09-09",
"title": "The Animation Studio names itself for screen readers, and its empty timeline is legible on a phone",
"summary": "A full audit of /pose, where you pose an avatar and keyframe an animation. The page had no top-level heading at all: assistive technology jumped straight into \"Selection\" with nothing saying which studio you were in, so \"Animation Studio\" in the breadcrumb is now the page's real heading. The status pill above the canvas is the studio's only error surface (it is what tells you an import failed, an avatar could not be loaded, or that you need a keyframe before saving) and it never announced anything; it is now a live region, so those messages are spoken as they appear. On a 320px screen the empty-timeline hint (\"Pose the figure, then press K\") wrapped across the ruler's second markings and both became unreadable; the ruler now stays out of the way until you have keyframes to measure, and returns the moment you add one. Loading an avatar could also fail into a dead end: the picker said \"Failed to load: Failed to fetch\" with no way forward, and now explains whether the request reached three.ws and offers a Try again button. Behind that, a slow first CDN could make the 3D preview library load twice and throw an error into the console; a second load is now ignored.",
"tags": [
"fix",
"improvement"
],
"link": "/pose"
},
{
"date": "2026-09-09",
"title": "The sitemap filter matches words in any order, opens the first hit on Enter, and stops leaving an empty bar behind",
"summary": "A full audit of /sitemap, the human-readable index of all 924 pages on three.ws. Filtering used to test your whole query as one run of characters, so \"avatar studio\" found the Avatar Studio but \"studio avatar\" found nothing; every word is now matched independently and either order returns the same 23 pages. Typing a query that matched nothing left the section jump-list on screen as an empty bordered bar hovering above the no-results panel; it now steps aside with the sections it points at. Pressing Enter in the filter box did nothing at all, which is the one key anybody tries after typing a page name; it now opens the first match, and the hint at the right of the box switches from the / shortcut to a return arrow to say so. Under the surface: the address bar is updated once you stop typing rather than on every keystroke, which is what Safari's limit on that call actually allows, and each row's visibility is only rewritten when it changes rather than 924 times per key. A query of nothing but spaces is no longer treated as a search.",
"tags": [
"improvement",
"fix"
],
"link": "/sitemap"
},
{
"date": "2026-09-09",
"title": "The Coin Wars arena reconnects to its battle server, and a link with no battle now shows you where the wars are",
"summary": "A full audit of /play/war, the arena where two coin communities fight to the kill cap. The page had lost the address of the multiplayer server it joins, so every player walking through a war portal landed on 'The arena server is unreachable' instead of a battlefield; the arena is wired back up. Beyond that, a link with no pairing (a shared URL, an expired ticket, a battle that already finished) used to be one sentence and a button back to the lobby, which told you nothing about where a war actually is. It is now the war room: the battles running right now with their live score and clock, the communities queued for an opponent, and the top of the Elo ladder, each one a link into that coin's world where its war portal stands. The same panel sits under every other dead end, so 'there is no battle here' always comes with 'here is where one is'. Also fixed: a failed holder check used to show an internal service error instead of telling you what to do, the Fire button announced itself as 'Firespace' to screen readers, pressing Space on the result card did nothing because the arena was eating the key, and the HUD's hover and focus animations were dead because the page never loaded the site's motion tokens. The arena now says it is playable with WASD, moves focus onto the way out when a card appears, and reads cleanly from 320px up.",
"tags": [
"fix",
"improvement"
],
"link": "/play/war"
},
{
"date": "2026-09-09",
"title": "Ethereum wallets load on the portfolio page again, keylessly, and every token logo shows up",
"summary": "A full audit of /portfolio, where you paste any Solana or Ethereum address and get a live portfolio. The Ethereum half had gone dark: the single provider behind it ran out of monthly capacity, so every 0x address answered \"Data source unavailable\" with nothing behind it. Ethereum now has the same keyless floor Solana has always had. The ETH balance comes from the public Ethereum nodes the rest of the platform already uses, the token list with its prices and icons comes from a public block explorer, and the 24h move comes from the same market source the Solana side uses, with the coverage it achieved stated on the card. The Data panel names whichever lane actually answered, so you can see whether you are reading the paid provider or the free one. Token logos were another visible break: the page pointed your browser straight at an IPFS gateway, which the browser refuses on cross-origin grounds, so every holding fell back to a three-letter monogram and your address bar handed your IP to the gateway on every load. Logos now come through three.ws, sized for the row. Keyboard users could tab onto a slice of the allocation donut and get nothing: the tooltip was wired to an event handler property that browsers do not actually support, so it never fired once; the slice now names its asset and share, and Escape dismisses it. When a lookup fails, the page now shows the reason the server actually sent instead of a generic sentence, and clicking Retry after a rejected address no longer silently reloads the wallet you looked at before it. Underneath, the fallback that names Solana tokens when the primary metadata source is out of quota had never run at all, because of a missing import; token symbols and names come back on that path.",
"tags": [
"fix",
"improvement"
],
"link": "/portfolio"
},
{
"date": "2026-09-09",
"title": "The Object Library is full again: 511 free CC0 props, and the page no longer stalls loading them",
"summary": "A full audit of /objects, the public-domain prop library. The page had been showing \"Object library coming online\" to everyone: the 511 props were sitting on the CDN the whole time, but the server read their index over a signed storage connection whose credential had stopped matching, so every visitor got an empty shelf. The index is public data on a public domain, so the read now falls back to that domain when the signed one fails, and the same change refills the character library, the animation clips and the asset catalogue the agent tools search. Browsing is a different page too. It used to build all 511 cards at once, each one a live 3D viewer with its own model, which meant 511 image requests before the first card appeared and about eighteen seconds of nothing. Now it renders 48 at a time, with a Load more button and automatic loading as you scroll, and a card stays a still picture until you rest the pointer on it, at which point that one spins up in 3D; a handful stay live at a time instead of hundreds. The filter row dropped eight chips that were Poly Haven collection names rather than anything you would browse by. On a phone the Preview, AR and Download buttons each get a full-width row instead of three cramped pills. Keyboard focus is now visible on every control, a failed load names the actual reason above its retry button, and the page description no longer contradicts the one in the markup.",
"tags": [
"fix",
"improvement"
],
"link": "/objects"
},
{
"date": "2026-09-09",
"title": "Pay-per-call is keyboard-drivable, and it now tells you when the wallet lane is down instead of sitting on a spinner",
"summary": "A full audit of /pay, the page where an agent pays $0.001 USDC on Solana per call to a paid API. Three things were quietly broken. The wallet pill kept reverting to \"loading\" after it had already shown a real balance: the page fills that pill from a live balance lookup, but the translation pass that runs a moment later put the placeholder back over it, and the same race hit the session ticker and the wallet setup dialog. Those slots are now marked as owned by the page, so a rendered balance stays put. The whole wallet picker was unreachable without a mouse, because each wallet was a clickable box rather than a button; every wallet is now a real button that tab reaches, Enter and Space activate, and a screen reader announces as selected or not. Both dialogs (the first-visit walkthrough and the wallet setup prompt) now take focus when they open, close on Escape, hand focus back to whatever opened them, and stay hidden from screen readers while closed. States are honest now too. When the agent wallet service is unreachable the page used to tell a signed-out visitor to create their first agent, which was never the problem; it now says wallets are unavailable and offers a Retry that re-runs the lookup without a page reload. A failed activity feed used to leave a row of dots ticking forever and now says the feed is unavailable. The page also gained the heading it never had, one high-contrast focus ring across every control, and a copy pass that removed the long dashes from its description and on-page text.",
"tags": [
"fix",
"improvement"
],
"link": "/pay"
},
{
"date": "2026-09-09",
"title": "The Pump Visualizer's live counters stopped freezing a few seconds after the page opened",
"summary": "A full audit of /pump-visualizer, the 3D scene of live pump.fun mints. The status bar across the top counts the tokens in the scene, the mint rate per minute and the total market cap, and about fifteen seconds into every visit all three froze: the counter sat at 0 and the rate and total sat at a dash while the scene behind them kept filling with new coins. The cause was the page's own translation pass. Those three numbers were written inside blocks of translated copy, so the moment the language file arrived the browser rebuilt that copy and threw away the very elements the page was writing into. The numbers now live beside their labels instead of inside them, so the labels still translate into all 85 languages and the values keep counting for as long as the page is open. The per-minute and per-hour units now translate too, which they never did once the page switched between live mints and migrations. The page also gained a heading and a labelled search box for screen readers, and its browser tab now reads Pump.fun Visualizer, three.ws instead of a retired product name.",
"tags": [
"fix",
"improvement"
],
"link": "/pump-visualizer"
},
{
"date": "2026-09-09",
"title": "Signal Marketplace: publisher avatars appear, prices read like money, and a feed page no longer goes blank for signed-out readers",
"summary": "A full audit of /signals, the directory of paid trading feeds ranked by proven edge. Four real defects came out of it. Every publisher without a profile picture showed an empty grey square: the generated fallback avatar is an image, and the card was assigning it to a CSS background, which browsers ignore, so it never painted. A publisher whose profile picture had gone missing showed the browser's broken-image icon; both now render as an image that quietly falls back to the generated avatar. Prices were printed as raw numbers, so a feed charging 12.345678 USDC advertised exactly that; amounts are now grouped and rounded the way money reads, while a genuine sub-cent price keeps its digits instead of rounding to zero. Subscriber counts skipped the same shortening every other number on the card uses and printed as 1234567. Worst of all, opening any live feed's own page while signed out crashed the render and left a blank panel where the sign-in prompt belongs, which is exactly the reader that panel exists to convert. Each card's screen-reader description now carries every number a sighted reader sees (rank, edge, hit rate, average ROI, signal counts, subscribers and price) instead of just the headline score, and the loading skeleton's shimmer now sweeps across the card rather than starting halfway off it.",
"tags": [
"fix",
"improvement"
],
"link": "/signals"
},
{
"date": "2026-09-09",
"title": "Oracle: a failed request now says so, instead of claiming no coins have been scored",
"summary": "A full audit of /oracle, the fused conviction board for pump.fun launches. When a panel's request failed, the page told you the opposite of the truth: the live feed claimed \"Oracle is warming up\" and said the engine ships with its backend, the movers, proof, agent-ranking, activity and 3D-graph panels each claimed there was simply nothing to show, and the performance record said the edge was still proving itself. Every one of those is now a real error state that names what went wrong (offline, rate limited, or an engine error) and offers a retry that actually recovers. The Proof gallery had a second bug behind that one: opening it for the first time while the request failed painted nothing at all, not even a skeleton. Searching a symbol Oracle has not scored used to make the results dropdown vanish with no explanation, which reads as a broken search box; it now says there is no match, and says so differently when search itself is down. The conviction-breadth bar's tier segments and legend chips were 10 and 15 pixels tall, too small to reliably tap on a phone, and now carry a full 24-pixel target without changing how the bar looks. Screen readers can finally move from a tab into the panel it opens, every button declares its type, and if a browser is holding a cached copy of an older release the 3D graph explains that and offers a reload instead of retrying a module that can never load.",
"tags": [
"fix",
"improvement"
],
"link": "/oracle"
},
{
"date": "2026-09-09",
"title": "Pump Dashboard: the live feed shows real launches again instead of raw JSON",
"summary": "A full audit of /pump-dashboard, the pump.fun trading desk. The headline defect was the feed itself: every event streaming in from the realtime socket was labelled UNKNOWN and printed as a wall of raw JSON, because the socket names its event type in a field the page never read. Launches, graduations and trades now render as readable rows with the token name, symbol, mint and size, whale-sized trades are tagged as whales, and the socket's own subscription acknowledgements stop appearing as feed entries. Recent activity loaded on arrival was stamped with the current time, so a graduation from yesterday claimed to have happened seconds ago; each row now carries the time it actually happened. Filtering the feed to a type with nothing in it left an empty panel with no explanation, and now says what is being filtered and offers a way back to everything. The Token Scanner gave no sign it was working during a scan that routinely takes ten seconds, and could hang forever with no result and no error if a Solana provider accepted the request and never answered; it now shows a busy state, times out, and explains a failure with a way to retry. The Quote Engine printed raw machine codes at you (internal_error, no_curve) where a sentence belongs, and the quote service itself answered a rate-limited Solana chain with a generic internal error instead of an honest \"try again in a moment\". The Watch List fired a request that could only fail unless you had pointed the dashboard at your own PumpKit backend, and now explains the requirement and takes you to the setting. Twenty-one form fields across the desk got the labels screen readers need, the feed's filter row is now a properly labelled group of toggles, and the high-conviction stats no longer clip their labels on a 320 pixel phone.",
"tags": [
"fix",
"improvement"
],
"link": "/pump-dashboard"
},
{
"date": "2026-09-09",
"title": "Restyle Studio: a failed save now tells you what went wrong and what still works",
"summary": "A full audit of /restyle, the material and PBR editor. Every failure coming back from the server was thrown away and replaced with a bare status code: a rejected model URL read as \"Restyle failed: restyle failed (400)\" when the server had actually said which part of the URL it would not accept. The page now shows the real reason on all four calls it makes (AI restyle, seeded variants, version checkpoints, and texture generation). A second case got its own wording: when the model store itself is unavailable, the note says so, and points out that your edits are still live in the viewer and that Export GLB keeps working, because export runs entirely in your browser and needs no server at all. Dropping a local file used to be told to \"load a public URL instead\" even when the URL was never the problem, which sent people in circles; the advice now matches the actual failure. Verified end to end in a real browser: the 18 material presets, the live metalness, roughness, colour and emissive sliders, the seeded variant strip, Reset, and a validated GLB export all check out, as do the loading and error states, keyboard focus on every control, and the layout at 320, 768 and 1440 pixels.",
"tags": [
"fix",
"improvement"
],
"link": "/restyle"
},
{
"date": "2026-09-09",
"title": "A link that opens your house in 3D now really opens it in 3D",
"summary": "The live home page remembers whether you last looked at your house in 3D or as a room list, which is what you want when you open it yourself. It was also overruling links that asked for one view by name: once you had used the 2D view even once on a device, every /smart-home/<id>?view=3d link on that device quietly opened flat again. That matters for the two cases the parameter exists for, a link you share with someone in the house and the address a kitchen or wall display is pinned to. A view asked for in the link now wins, for 3D, 2D and the floorplan alike, and following such a link no longer overwrites the preference that device chose for itself.",
"tags": [
"fix"
],
"link": "/docs/home-scene"
},
{
"date": "2026-09-09",
"title": "Your own agent now stands in your live house",
"summary": "The 3D home at /smart-home/<id> always drew a generic figure standing in your rooms, even when you had given your agent a face months ago. It now shows your agent: the same avatar the walk world, /play and the voice satellite use, read from the one record the whole platform treats as \"my agent\". Switch agents in another tab and the body in the house follows, with no reload. It is fetched after the house has already drawn, so nothing waits on it, and the platform's default body stands in when you have no agent yet, no avatar yet, or an avatar you keep private, which by design publishes no model for any page to read.",
"tags": [
"improvement"
],
"link": "/docs/home-scene"
},
{
"date": "2026-09-09",
"title": "Smart Home: the connection pool now sizes itself to the server it is running on",
"summary": "The number of houses one server holds open at once was set by hand, in a different place from the memory that has to hold them, and the two drifted apart three times. Left alone, that ends with a server running out of memory and taking the whole API down with it, which every part of three.ws would feel, not just Smart Home. The pool now reads the memory its own container actually has and refuses to hold more houses than fit, saying so in the health endpoint when it does. When there is less room it connects fewer houses per server and keeps working, instead of risking an outage for everybody.",
"tags": [
"infra",
"fix"
],
"link": "/docs/smart-home"
},
{
"date": "2026-09-09",
"title": "Fact-check now reads the part of a source that is actually about your claim",
"summary": "When the fact-checker fell back to Wikipedia, it only ever read each article's opening summary. An opening summary says what a page is about, so it carries the answer only when your claim is about the page's subject. Ask whether Napoleon was unusually short and it read the first lines of \"Napoleon\", \"Napoleon III\" and \"Napoleonic Wars\", none of which mention height at all, so every source was scored as saying nothing and the verdict came back \"not enough evidence\" while the passage about his height sat further down the same article. The checker now reads each article in full and picks the paragraphs that actually discuss your claim, so that question returns the \"Appearance and image\" and \"Napoleon's height\" sections instead. Reference lists are skipped, since they repeat your words without asserting anything, and the deeper read is bounded by the time the check already had, so a slow source can never cost you the evidence that already arrived.",
"tags": [
"improvement"
],
"link": "/docs/fact-check"
},
{
"date": "2026-09-09",
"title": "Smart Home reads and works without a mouse, and now speaks past English",
"summary": "Three fixes to the connected house, all of them things a person hits rather than reads about. The room you have selected states what is on in it at full contrast: on the highlighted row that line sat below the readable minimum, which is exactly the row you are looking at. Answering the safety question that stands between you and an unlocked door now hands the keyboard back to the control you pressed, whether you answer with Yes, with Cancel or with Escape; the list rebuilds itself while the question is up, so the keyboard used to land back at the top of the page and you had to tab the whole house again. And the connect flow, the manage view and the floorplan editor now read from the same translation catalog the rest of the site does, so they are ready for the 86 languages three.ws ships in. Your own words are never translated: the name you gave your house, your rooms, your devices and anything your Home Assistant says come back exactly as you wrote them, in every language.",
"tags": [
"improvement",
"fix"
],
"link": "/docs/smart-home"
},
{
"date": "2026-09-09",
"title": "The generated motion clips were playing upside down, and we can now tell",
"summary": "Every one of the 133 motion clips three.ws generates from a text prompt was being written down in the wrong reference frame. The bone names all matched the rest of the library, so nothing looked wrong in the data, but a clip's rotations are measured against a rig's resting pose and the generator was measuring against a different one. The result was a half turn missing from the legs: play one of these clips on an avatar and the feet fold up over the body. Forward kinematics put the feet at 1.71 m and the head at 1.45 m, where a hand-authored clip puts them at 0.15 m and 1.68 m. Clips can now be converted into the library's own frame, which puts all 133 back on their feet, and the quality check has learned to spot the fault: it was invisible before because every other rule measures a clip against itself, and a clip in the wrong frame is perfectly consistent with itself. A second fault went with it. Every generated clip was drifting forward at about a quarter of a metre a second whatever it was asked for, so a clip of someone applauding walked a metre across the room while it clapped, and standing-still idles drifted faster than the walks did. That turned out to be an artifact of how the motion model's output is scaled back into real units, and it is now measured and removed while the real motion, a pirouette's sway or a walk's per-step surge, is kept. Removing it also repaired the foot-sliding check, which had been dividing by a stride that was never real, so the library's quality bar finally means something: 39 of the 133 clips meet it, and the rest are being held back rather than served.",
"tags": [
"fix",
"improvement"
],
"link": "/animations"
},
{
"date": "2026-09-09",
"title": "Every translated news page is advertised as translated again",
"summary": "The site publishes a manifest of which pages carry translated copy, and it was missing 113 of them: every news page. The build wrote that manifest before the news pages had been through the translation pass, so it read them as English-only and shipped a list 113 pages short. Nothing was untranslated, the translations were there and served correctly, but anything reading the manifest to decide what to offer in a reader's language (the language picker's page-level hints, and crawlers looking for alternates) was told those pages had no other language. The build now annotates the news pages first and counts afterwards, so the published count is 357 rather than 244.",
"tags": [
"fix"
],
"link": "/news"
},
{
"date": "2026-09-09",
"title": "Disconnecting a home now stops the screens that were showing it",
"summary": "A kitchen tablet or wall display left open on a connected home kept saying \"Live\" and kept drawing that house after the home had been disconnected from another device. The connection was really gone, but the open page was never told, so it sat on a stale picture of somebody's house with a green Live badge on it. Disconnecting now closes the live connection and hands every screen still watching a final message that says what happened, and those screens stop asking for more. They keep showing the last state they saw, greyed and dated, with the reason and a Reconnect button, which is what every other kind of drop already did.",
"tags": [
"fix",
"security"
],
"link": "/docs/home-scene"
},
{
"date": "2026-09-09",
"title": "The changelog channel stops announcing the same update twice",
"summary": "Updates reach the community channel automatically, and each one is remembered by its date and headline so it only goes out once. That worked across separate posting rounds but not within one: if the same update was written into the changelog file twice, both copies were unsent when the round was built, and both were sent, so subscribers got the identical message back to back. Six repeated entries had accumulated in the file that way, one of them added today and still unsent. The sender now collapses repeats before it posts, so a duplicated line costs a wasted row in a data file instead of a repeated message to everyone following along. The three verbatim repeats were removed from the changelog, and the build that publishes the feed now refuses a repeated entry outright rather than shipping it.",
"tags": [
"fix",
"infra"
],
"link": "/changelog"
},
{
"date": "2026-09-09",
"title": "An embedded avatar no longer crashes the page it was removed from",
"summary": "Sites embed three.ws avatars with the <agent-3d> tag, and a single-page app can swap the whole page out while an avatar is still loading. When that happened the avatar was left holding a piece of the page that no longer existed, and its own tidy-up threw an error into the host site console instead of quietly reporting that loading had stopped. The avatar had already handled the failure correctly; the report of it was what broke. Embedders now get the agent:error event they listen for, and nothing else, whether or not the page moved on while the avatar was still loading.",
"tags": [
"fix",
"sdk"
],
"link": "/docs/web-component"
},
{
"date": "2026-09-09",
"title": "The quality check on a generated model keeps running when one AI provider goes down",
"summary": "Every generated model can be looked at by a vision model before you see it, which is what catches a flat slab or a melted blob before it reaches you. That check asks one provider first and a free provider second, but when the first one started refusing every request it was still asked first every single time, and half the checks ended up returning no verdict at all, so models shipped unchecked. The check now notices a provider that is refusing, sets it aside for a few minutes, and goes straight to the one that is answering. The set-aside expires by itself, so the moment the first provider is healthy again it is used again with nothing to switch back on.",
"tags": [
"fix",
"improvement"
]
},
{
"date": "2026-09-08",
"title": "Your avatar shows up again on every page, instead of a stand-in robot",
"summary": "When you were signed in, the little avatar that rides along with you across the site had quietly been replaced by a generic robot on every page. The models themselves were fine and sitting safely in storage the whole time: what broke was the signed request the site uses to read them. Our storage provider stopped accepting that credential, and the page that serves your avatar had only that one way to fetch it, so it gave up and returned an error while the same file was being served correctly on the public storage address the entire time. That page now falls back to the public address and streams the model through as usual when the signed read is refused, so your avatar keeps loading through an outage like this one. It still prefers the signed path and returns to it automatically the moment storage is healthy again, and nothing about who is allowed to see a private avatar changed.",
"tags": [
"fix"
],
"link": "/avatars"
},
{
"date": "2026-09-08",
"title": "A live globe of what is happening on Earth, and every layer says how far back it can really see",
"summary": "There is a new page at /globe: a 3D globe carrying eleven layers of open data at once. Armed conflict, mobilizations and sanctions are read out of the world news stream and placed where they happened; earthquakes, volcanoes and wildfires come from the USGS and NASA; cyclones, floods and drought from the UN and European Commission disaster system; national internet blackouts from Georgia Tech's outage project; and economic contraction from World Bank figures. Under all of it sits the geography that gives the rest meaning: overseas military bases, nuclear sites, and the nine maritime chokepoints world trade passes through, each one scored by how much armed activity the live feed puts within reach of it. Drag the globe or press F to flatten it into a map, click any point to read what it is and open the source it came from, and search the feed for a place or an actor. The whole view lives in the link, so the exact globe you are looking at is one you can send to someone. Two things it will not do: it will not quietly show you an hour of data under a label that says a week, because every layer prints the window it actually covered, and it will not go blank when a source goes down, because a layer that cannot be reached says so and the other ten keep working. Nothing here needs an account or an API key, and the same data is one call away at /api/globe/intel.",
"tags": [
"feature"
],
"link": "/globe"
},
{
"date": "2026-09-08",
"title": "Two thirds of the site was missing its mobile stylesheet, and every page it reached got bigger buttons",
"summary": "three.ws keeps one stylesheet for phones: it sets a 44px minimum on anything you tap, keeps 3D viewers from fighting the page scroll, and holds bottom bars clear of the iPhone home indicator. It was reaching 192 of the 584 pages that carry a stylesheet at all. The build injected it, and the injector only ever saw pages built through the bundler, so everything served as a plain HTML file quietly went without: the whole documentation site, this changelog, /ar and hundreds of others. The documentation sidebar was the clearest case, 296 links measured at 30 pixels tall on a Pixel 5, against a rule sizing them to 44 that had been sitting in that stylesheet unused. The build now sweeps every page it produces, which added the stylesheet to 210 more of them, and the same sweep switches on the setting iOS needs before safe-area spacing does anything at all. On top of that, the controls the audit still found short have been raised across the home page, the marketplace, /coin, /play, /launches, /forge, /markets, /news, /irl, /walk and the docs shell.",
"tags": [
"improvement",
"fix"
],
"link": "/docs/start-here"
},
{
"date": "2026-09-08",
"title": "The changelog page no longer downloads three and a half thousand entries to show you this week",
"summary": "Opening the changelog fetched the entire history, every entry since the site existed, and wrote all of it into the page in one go: 870 KB over the wire and, on a mid-range phone over a slow connection, eleven seconds where the page would not respond to anything. It now paints the newest 150 entries from a small feed and pages in the rest only when you filter or ask for older updates, so the thing you came for is on screen almost immediately and the full history is one tap away. Entry pages, the What's New strip on the features page and the Walk changelog read the small feed too. The complete feed at /changelog.json is unchanged and still the one to use if you are building on it.",
"tags": [
"improvement"
],
"link": "/changelog"
},
{
"date": "2026-09-08",
"title": "3D pages send a phone a phone-sized world",
"summary": "The lighting environment behind our 3D scenes is a high dynamic range image, and every device was fetching the same 1.4 MB copy of it. On /play that was more than a quarter of everything the page downloaded, and on a slow connection it is seven seconds before the world is properly lit. Phones now get a 512-pixel copy at about 360 KB, which carries the same light energy to within half a percent, and desktops keep the full-resolution original where the sharper reflections are actually visible. The coin artwork in the flagship /play world was being fetched at its full 581 KB to texture something a few hundred pixels tall, and now arrives at the size it is drawn; the same was true of the coin art in the community lobby. Marketplace card thumbnails now come through the image resizer at the size the card paints them instead of at full resolution.",
"tags": [
"improvement"
],
"link": "/play"
},
{
"date": "2026-09-08",
"title": "Pages stop jumping around while they load",
"summary": "Three separate things were shifting the layout under you. The marketplace hero rotates through featured avatars every few seconds and wrote each one's name and description into boxes sized to whatever the previous one had been, so the buttons, the dots and the entire grid below moved on every rotation, forever; both boxes now hold the same space whichever avatar is showing. The helper cards that float in the bottom-right corner slid themselves up and out of the way by animating their position, which the browser counts as the page moving even though nothing on the page did; they glide now instead. And the prompt tip under the forge box appeared a moment after the page did, pushing the quality options and the Generate button down; it is there from the first frame.",
"tags": [
"fix",
"improvement"
],
"link": "/marketplace"
},
{
"date": "2026-09-08",
"title": "Pages stop re-announcing text they had already shown",
"summary": "The translation layer walks the page when its dictionary arrives and writes the right words into every labelled element. On English, which is what the page already shipped, almost every one of those writes put back exactly the text that was there. Browsers do not treat that as nothing: replacing text is a new paint, and the measurement browsers report for how quickly a page shows its main content was being reset to whenever the dictionary happened to land. The home page's headline had painted in 2.3 seconds and was being scored at 10.2. The pass now checks before it writes, so a page that is already correct is left alone.",
"tags": [
"fix",
"improvement"
],
"link": "/"
},
{
"date": "2026-09-08",
"title": "Paying for something no longer depends on our wallet having gas in it",
"summary": "Every Solana payment on three.ws was quietly gasless: we paid the tiny network fee on your behalf so you could pay with USDC or $THREE without owning any SOL. That is a convenience, but it had been wired as a requirement, so when the wallet we pay those fees from ran empty, every paid page stopped accepting money altogether. The club door and its dance tips returned an error for four days, and sixteen other paid endpoints (asset downloads, skill calls, print orders, cosmetic purchases and the rest) were in the same state, all of them sitting on a perfectly good account that was ready to receive. Receiving money never actually needed our wallet to have anything in it. When we cannot cover the fee, the payment now simply asks you to cover it yourself, which is a fraction of a cent and what most wallets do everywhere else, and the checkout tells you before you approve rather than letting your wallet be the first to mention it. Gasless comes back on its own the moment our wallet is topped up. Two related things got clearer: paying in $THREE stays available in that mode instead of disappearing, and if your own wallet is short on SOL the message now says so and tells you to add a little, instead of claiming we were the ones with the problem.",
"tags": [
"fix",
"improvement"
],
"link": "/club"
},
{
"date": "2026-09-08",
"title": "Every button on Drive Your Avatar is reachable again, by mouse and by keyboard",
"summary": "Two things were quietly breaking the controls on Drive Your Avatar. The gesture tray grew up the right-hand edge until it sat on top of the HUD column beside it, so Photo, Record, Map and the help button silently stopped taking clicks; the tray now keeps to the lower band and wraps into a second column instead of climbing into the buttons. Separately, pressing Enter anywhere on the page jumped straight to the chat box, which meant a keyboard user could tab to any control and never activate it. Enter now belongs to whatever you have focused, and only opens chat when you are looking at the scene. Along the way: the radar moved off the crowded bottom edge on phones, the chat box no longer pushes its send button off a narrow screen, the avatar list tells you when it could not load instead of always saying you are signed out, and the coin-world picker closes on Escape.",
"tags": [
"fix",
"improvement"
],
"link": "/temporary"
},
{
"date": "2026-09-08",
"title": "Avatars of people now come off the stronger lane, and stop standing on a slab of their own backdrop",
"summary": "Two things were quietly holding avatars back. The first: our own docs said that asking for a 3D person always ran on the strongest engine we host, the one that is best at faces and bodies. The code was still sending those requests to the faster, weaker engine, so every avatar came back off the wrong lane. Side by side on the same six descriptions, the weaker lane returned smeared, faceless figures with mitten hands and once no figure at all, while the stronger one returned people you can recognise. Avatars now go where the docs always said they did. The second: when you describe a person in words, we paint a reference picture first and sculpt the model from it. That picture has a plain background, and we had assumed plain was harmless. It is not. The sculptor was treating the backdrop as part of the scene and building it in, so four of six test figures came out standing on a full-size slab of their own background and two were swallowed by it entirely. Written descriptions now get the same background cutout that uploaded photos already got.",
"tags": [
"improvement",
"fix"
]
},
{
"date": "2026-09-08",
"title": "Watch your model being made, instead of watching a spinner",
"summary": "Making a model from a written prompt starts with work you never saw: an art director rewrites your words into a proper brief, then an image model paints the reference picture the mesh is actually sculpted from, and on our best lane it paints the extra angles that fill in the subject's back. All of that used to happen inside one long silent request, so Forge could only say \"art-directing your prompt\" for the whole wait and then reveal the reference picture at the very end, when it had already existed for half a minute. Now the pipeline reports each step the instant it finishes, so the stage list ticks over as the work really happens and the reference image appears on screen while the sculpting is still running. Nothing on that list is a guess: a step only lights up because the server said it finished. On a phone, pressing Generate now brings the panel into view instead of leaving the whole thing playing out below the fold, and the same is true when a model lands or when something goes wrong. If our asset storage is the thing that is down, Forge now says so plainly, tells you nothing was charged, and points you at the models you already made, which keep working throughout. A model on screen also stops spinning by itself for anyone who has asked their device for less motion, and the report-a-bug prompt no longer pops up on top of the recovery buttons on a narrow screen.",
"tags": [
"improvement",
"fix"
],
"link": "/forge"
},
{
"date": "2026-09-08",
"title": "The guardian console now tells you why someone is asking for an agent back",
"summary": "If a friend has named you a guardian or a beneficiary of their agent, /guardian is where you decide whether to let them back in. Until now the card told you a process was open and asked you to approve it, without showing you the one thing a decision actually turns on: the reason the person gave when they opened the request. It now shows that reason, names every guardian who has already approved or declined, and says how long the request has before it expires if nobody acts. The status stopped reading like a database field, so a card says \"awaiting approvals\" or \"in the safety window\" instead of pending_approvals. Approving takes a couple of seconds to travel; the button now says Approving while it does, rather than going quiet. If the safety window lifts while you have the page open, it refreshes itself so Complete transfer appears without a reload, and it refreshes again whenever you come back to the tab. Signed out, the page has a proper heading, offers to make you an account, and brings you straight back here after you sign in; with nobody yet naming you, it now points you at setting up guardians on your own agents. When the console cannot load at all it explains what went wrong in plain words, telling a dropped connection apart from a rate limit and a fault on our side, instead of printing the browser's own \"Failed to fetch\".",
"tags": [
"improvement",
"fix"
],
"link": "/guardian"
},
{
"date": "2026-09-08",
"title": "Your avatars show up on the dashboard again, and a failed one says why",
"summary": "The avatar previews on your dashboard, and every three.ws avatar embedded through the <threews-avatar> tag, were being pointed straight at the storage bucket that holds the 3D file. That address is on a different domain to the page asking for it, and the browser refuses cross-domain reads it has not been given permission for, so the model was blocked before a single byte arrived and the preview sat empty. Stored avatars now load through three.ws itself, which is the same origin as the page, carries your session so a private avatar stays private, and already knows to fall back to the public copy of the file when the signed read is refused. That last part is why this was invisible on our own site and total everywhere else: the bucket permits three.ws by name and nothing else, so a dashboard preview, a developer running the site locally, and a customer embedding an avatar on their own domain all failed for the same reason. When a model genuinely cannot be loaded, the frame no longer paints the raw network error, complete with an internal URL, into the box. It now says whether the avatar is private, gone, or the service is down, and what to do next. My Agents got the same treatment: a failed load explains the cause and gives you a Try again button instead of telling you to refresh the page, its keyboard focus outlines are visible again, and the page respects a reduced-motion preference.",
"tags": [
"fix",
"improvement"
],
"link": "/dashboard"
},
{
"date": "2026-09-08",
"title": "The Revenue view on Protocol Fees stopped calling itself Fees",
"summary": "Open /fees, switch to Revenue, and the numbers, the chart and the stat cards all changed to revenue while the table heading and the browser tab kept saying \"Fees\". The page writes those two pieces of copy itself, and the site-wide translation pass, which lands a second or two after the page does, was overwriting them back to the fees wording. Both now hold the metric you actually selected, and hand themselves back to the translator when you switch to Fees, so a reader in another language keeps their own copy. Four more things on that page while we were in there: the footer line now says when a figure is a cached copy served because DeFiLlama is unreachable, rather than presenting it as live; a run where the feed returns nothing no longer paints three empty totals and a \"Top 0 protocols\" line that read like a claim DeFi earned nothing; toggling Fees and Revenue faster than the network answers can no longer land one metric's numbers under the other's heading; and screen readers are now told every column is sortable, told which metric each region is showing, and given a proper pair of toggle buttons instead of tabs with no panel behind them. The \"+3\" badge on a protocol's chain list also names the chains it is hiding.",
"tags": [
"fix",
"improvement"
],
"link": "/fees"
},
{
"date": "2026-09-08",
"title": "The ChatGPT and AR docs now describe the AR launcher that actually runs",
"summary": "Three corrections to the documentation for the ChatGPT surfaces and augmented reality. The AR reference still said the one-tap launcher at /api/ar draws the model itself in Google's model-viewer, which stopped being true when the launcher became a pure router: it now sends Android straight into Scene Viewer and everyone else to a bundled page that builds a real USDZ on the device, which is what iPhone's Quick Look needs. The pages are named correctly throughout, including which one does the file conversion, so nobody debugging an iPhone handoff starts in the wrong file. The ChatGPT AR page also credited a single generation engine for every model, when the connector actually tries the free engines in turn and steps down to a slower one if the first is cold, which is why generation keeps working when one lane is down. And the paid studio server's page promised that anyone can browse its tool catalogue without credentials, without mentioning that a connector client is deliberately answered with a sign-in challenge instead, as the connector standard requires; both paths are now written down, with a command that works. That page's tool tables also describe nine tools that were previously only named in its price list, so no tool appears with a price and no explanation.",
"tags": [
"docs",
"fix"
],
"link": "/docs/ar"
},
{
"date": "2026-09-08",
"title": "The Launch a Coin button now works once you have filled the form",
"summary": "On the Launch a Coin page we fill in your token name and ticker from the agent you pick, but never the description, because only you can write that. The launch button read \"Add name, symbol & description to launch\" until all three were there, and that label was worked out once when the page drew itself and then never again. So the moment you typed your description the button was describing a form that no longer existed: it still said something was missing, and pressing it did nothing whatsoever, because it went looking for an empty field and found none. The button now keeps up with what you type, turning into \"Connect wallet to launch\" or the real launch action the instant the last field is filled, and back again if you clear one. Pressing it while something is genuinely missing now says which field it is, instead of silently moving your cursor there. Separately, if we cannot reach your agent list, the page now says so and offers a Try again button that recovers in place; it used to tell owners of a dozen agents that they had none yet and send them off to create a duplicate. The agent picker also fits a phone screen now instead of running off both edges.",
"tags": [
"fix",
"improvement"
],
"link": "/launch"
},
{
"date": "2026-09-08",
"title": "Text to 3D no longer stops when our image storage does",
"summary": "Turning a written prompt into a model takes one step you never see: we paint a reference picture of what you asked for, save it somewhere our sculpting engines can read, and hand them the link. If that save failed, the whole thing was over before a graphics card was ever asked for anything, and the error you got back was our storage provider's own wording about signatures and access keys, which told you nothing you could act on. Making a model from a photo you upload was unaffected, because it brings its own picture, so half the feature could look perfectly healthy while the other half was completely down. Our own engines never needed that saved copy: they accept the picture itself, sent along with the request. So when storage will not take the write, the reference view now travels inside the job and your model still gets built. The engines that genuinely need a link, the outside ones you can pick by name, now say so in plain language and point you at the ones still running, instead of failing with a message about our secret keys. Generations that take this route lose the extra angles we normally paint to fill in a subject's back, so they are built from the single front view.",
"tags": [
"fix",
"improvement"
],
"link": "/forge"
},
{
"date": "2026-09-08",
"title": "Launch Studio reads in light mode, and a recipe's reward routing now reaches the launch",
"summary": "Launch Studio, the catalog of 50 coin-launch recipes, was written entirely in white text and never flipped for the light theme, so on a light background its title, its recipe cards and its preview drawer were white on white with a dark strip painted across the top. The whole surface now draws from the site's colour system and reads in both themes. The bigger fix sits behind the Launch this coin button: every reward recipe promises that a coin's creator fees route to its subject, and that choice was being dropped the moment you left the studio, so the launch form opened with no recipient at all. The routing now travels with the coin, shows on the launch form before you mint, and opens the fee split pre-filled with that recipient the moment the coin is live, ready to review and sign. Two planner options no launch could ever apply were removed rather than left promising a payout. The preview drawer also behaves as a real dialog now: focus moves into it, Tab stays inside it, and closing it returns you to the card you opened.",
"tags": [
"fix",
"improvement"
],
"link": "/launch-studio"
},
{
"date": "2026-09-08",
"title": "A GPU waking up now says so, instead of looking like a stuck progress bar",
"summary": "Some of our 3D engines release their graphics card when nobody is using them, so the first request of a quiet spell has to boot one back up before any sculpting starts. That wait is normal and it ends, but several surfaces showed nothing but a spinning step while it happened, which reads as a hang. Now every place you can watch a model being built names it: the homepage forge, the Studio, the in-world item forge on /play, and the tools other apps drive us through all say a GPU is waking, how much of the boot is left, and that your job is already accepted and starts the second the worker answers. Every number comes from the generator itself, so nothing is a guessed timer. Two related fixes came out of it: the elapsed counter on /forge now reports how long your job has really been running rather than how long the tab has been open, so reopening the page mid-build no longer restarts the clock at zero; and the countdown now uses each engine's real boot budget instead of an estimate. The keep-warm schedule that holds those engines open during busy hours was also missing our highest-quality lane, which has the longest boot of the fleet.",
"tags": [
"improvement",
"fix"
],
"link": "/forge"
},
{
"date": "2026-09-08",
"title": "Buttons that looked dead now respond, and the whole site paints from one palette",
"summary": "A consistency pass across the platform. Fifty controls set the clickable cursor but had no hover styling at all, so the pointer changed shape and the button did not, which made working controls read as broken; the marketplace, Widget Studio, the brain workspace, the pump dashboard, agent detail and the on-chain deploy panel now all answer under the pointer, on the same timing. The retry buttons on an agent page were the worst case, because they only appear after something already failed. Separately, 73 colours were still typed in by hand across 24 stylesheets instead of coming from the shared palette, which is why a handful of panels stayed dark when you switched to the light theme and why some captions drifted a shade off everywhere else; they now all read from one source. Eight standalone pages that had quietly grown their own private palette (the demos, Widget Studio, the dashboard sub-pages, the print insert, agent detail) were folded back onto the shared one, so they also pick up the site-wide keyboard focus ring and the reduced-motion setting for the first time. Two new checks keep both from creeping back.",
"tags": [
"improvement",
"fix"
],
"link": "/marketplace"
},
{
"date": "2026-09-08",
"title": "Avatars keep their real skin, eyes and hair, and we can prove it on a phone",
"summary": "Every avatar the platform loads gets its skin, eyes, hair and teeth re-shaded with real-world material values the moment it appears: skin at the roughness bare skin actually has with a warm scattering layer, a wet cornea that catches a highlight, and hair that reads as strands rather than a helmet. That pass was already running, but the check meant to prove it was pointed at the wrong page and had been quietly reporting nothing at all, so it could not have caught the day it broke. It now runs against the same embed page your visitors see, at phone, tablet and desktop widths, and records exactly which materials it upgraded and to what values. Alongside it, a new check loads a full 4K-textured model on an emulated iPhone and Android phone over a slow mobile connection and measures the real bytes and load time, so a model that would not open on a phone is caught before it ships rather than after.",
"tags": [
"improvement",
"infra"
]
},
{
"date": "2026-09-08",
"title": "The club now takes $THREE at the door and at the stage, not just USDC",
"summary": "The club has always advertised two ways to pay: USDC or $THREE, on Solana, for both the cover charge and every dance tip. Only one of them was reachable. The checkout window picked the first token a service offered and gave you no way to choose the other, so $THREE was quoted on every paid page and payable on none of them. Across every club tip ever settled, not one was paid in $THREE, which is what you would expect from a menu with a hidden second item. The checkout now shows a token picker whenever a service quotes more than one, with each token's real price on it, so you can compare 0.001 USDC against its $THREE equivalent before you approve anything, and the price at the top of the window follows what you pick. If you have agent wallets, they still pay in USDC only, and the window now says so instead of quietly hiding them when you switch tokens. Two settlement bugs behind the door were fixed at the same time. The first would have hit the very first person to pay in $THREE: sending a token to a wallet that has never held it requires opening an account for it on-chain, and the check that decides whether we can afford to sponsor a payment was ignoring that cost, so the payment would have been accepted and then failed on-chain for a reason that looked like something else entirely. The second was making the club look open when it was not: a healthy payer covering their own fees could reset a shared 'we can still sponsor payments' flag that belonged to a different wallet, so the door kept quoting prices it could not actually take. Finally, when payments are paused because the wallet that settles them needs topping up, the checkout now says that in a sentence and tells you nothing was charged, instead of showing you the raw error envelope it used to paste in, which read like a broken link.",
"tags": [
"fix",
"improvement"
],
"link": "/club"
},
{
"date": "2026-09-08",
"title": "Forge tells you when uploads are down, instead of blaming your connection",
"summary": "Adding a photo on the Forge could fail with \"Network error during upload\" when nothing was wrong with your network or your photo. Our asset storage had stopped accepting our credential, but the page had no way to know that: the browser is handed a short-lived upload address and sends the image straight to storage, and when storage refuses, the refusal comes back without the header a browser needs in order to show the page what happened. So a platform outage arrived looking exactly like a bad phone signal, and Retry could never fix it. The page now checks that storage will actually accept an upload before handing one out, and says so plainly when it will not, so you can tell \"come back in a few minutes\" apart from \"try a different photo\". The same fix covers the Forge Studio and 3D print upload paths.",
"tags": [
"fix"
],
"link": "/forge"
},
{
"date": "2026-09-08",
"title": "Robinhood Portfolios can now show you what a basket would have done, and what the rebalancing did to it",
"summary": "Every generated portfolio now comes with a backtest, and it runs the basket twice: once rebalanced on the schedule it published, and once left alone. The gap between those two lines is the whole reason to pick one schedule over another, and it goes both ways, since rebalancing quietly harvests the swings in a basket whose holdings move against each other and quietly gives up gains in one where a single holding keeps winning. Coverage is stated rather than glossed over: a holding with no price history is named and left out, the window shown is only the stretch where every remaining holding has real observations, and nothing is filled in to close a gap. Getting that history at all took some doing, because this chain keeps no price archive and produces about 864,000 blocks a day, so reading a month of prices back off the chain is not something a page can do while you wait. Tokenized stocks are covered from their price feeds going back weeks, and a nightly snapshot now records the whole market so coverage keeps growing for everything else, memecoins included. Sharing a portfolio also got fixed: a link now carries the portfolio itself rather than the sentence that produced it, so what you send is exactly what the other person opens, where before the wording was re-run and could quietly come back as a different basket. There is a new endpoint for checking that a portfolio document really is the one a portfolio was published under, and a client library for developers who want the market, the generator, the backtest and that check from their own code. Along the way two data bugs were fixed that were quietly under-reporting the market: the tokenized-stock board had been describing only a couple of tokens per batch instead of all thirty, and the price-history reads had been silently returning less history the more of it you asked for.",
"tags": [
"feature",
"improvement"
],
"link": "/markets/robinhood/portfolios"
},
{
"date": "2026-09-08",
"title": "Agent Galaxy no longer strands you when the star map cannot be built",
"summary": "When the embedding service behind the 3D agent map is unreachable, the page showed a sentence of explanation and nothing to click. It now offers the same agents as a browsable list plus a link to how the map is built. The keyboard experience was rebuilt too: every control on the page has a visible focus ring, and the search box and Money-Cam button drop out of the tab order while an overlay covers the map, so you can no longer tab into a search that has nothing to search. An agent card's Chat button now opens that agent's chat panel instead of repeating the View agent link.",
"tags": [
"fix",
"improvement"
],
"link": "/galaxy"
},
{
"date": "2026-09-08",
"title": "The Agent Economy page is readable on a phone, and its x402 services finally have names",
"summary": "On a narrow screen the page laid itself out at desktop width and the browser quietly cut off two thirds of every section, so most of the content was simply unreachable on a phone. It now fits the screen at every size. The live x402 service list was also close to useless: every one of the ninety endpoints was labelled \"Service\", the price sat next to a raw token contract address, and the network read \"eip155:8453\" instead of \"Base\". Each row now carries a real name, its endpoint address, a clean price and a readable chain, an endpoint that takes path parameters no longer pretends to be a clickable page, and the list opens with a screenful you can expand to all of them. The three empty states tell you where to go next, and the services counter reports the true number instead of \"40+\".",
"tags": [
"fix",
"improvement"
],
"link": "/economy"
},
{
"date": "2026-09-08",
"title": "Agent Launches: the Tip button on every launch card actually works now",
"summary": "Each card on the launches feed carries the launching agent's wallet, and the whole-card link to the coin profile was sitting on top of it: every tap on Tip, and every attempt to preview the wallet, opened the coin page instead. The wallet now sits above that link, so tipping and the wallet preview both work, on the feed and everywhere else the wallet badge appears. Keyboard and screen-reader use got the same pass: the Mainnet/Devnet switch responds to the arrow keys and says which feed it controls, the graduation bars announce which coin they belong to and how far along it is, the wallet's trust score and balance no longer read as unlabelled boxes while they load, and a page that found nothing on devnet now offers a button back to mainnet instead of only telling you to switch.",
"tags": [
"fix",
"improvement"
],
"link": "/launches"
},
{
"date": "2026-09-08",
"title": "Every tutorial now shows you the page it is teaching",
"summary": "Twenty screenshots of the real product had been captured for the tutorials and then never put on the page: the tutorials described the Forge, Avatar Studio, Voice Lab, Scene Studio, Widget Studio and the rest entirely in words, while the pictures sat unused. All twenty are now embedded where they belong, each with a caption and a link straight to the live page, so you can see what you are about to open before you open it. The sign-language reference gained its two figures as well, and the missing Scene Studio capture was taken.",
"tags": [
"docs",
"improvement"
],
"link": "/tutorials/text-to-3d"
},
{
"date": "2026-09-08",
"title": "The agent vitals endpoint answers again, and eight documented API examples now say what they really return",
"summary": "GET /api/agents/vitals, the operator view that reports whether each autonomous agent can actually act, had been unreachable: a broader routing rule for agent profiles was catching the request first and answering with an agent lookup instead. It resolves to its own handler again. Separately, every runnable curl sample in the documentation was replayed against the live API, and the ones whose example used a placeholder id or needed a signed-in session now say so above the snippet rather than looking like a call that should have worked.",
"tags": [
"fix",
"docs"
],
"link": "/docs/agent-vitals"
},
{
"date": "2026-09-08",
"title": "Exit Lab is usable with a thumb, and it no longer goes quiet when the trade corpus is missing",
"summary": "A full audit of the Exit Lab console found 49 controls below the 44px touch floor on a phone: the 18px help dots, the 16px-tall policy sliders and the agent link on every row of the trade table were all but untappable. All of them are now full-size targets on touch devices, and the desktop layout keeps its density. Two dead ends went with them. The Run the search button used to sit there live and answer a click with nothing when no trades had loaded; it is now disabled and says which of the two reasons applies. And when the corpus fails to load, the attribution, trade-by-trade and search sections say so instead of leaving three headings over blank space, with the trade table stuck announcing itself as still loading forever.",
"tags": [
"fix",
"improvement"
],
"link": "/exit-lab"
},
{
"date": "2026-09-07",
"title": "Robinhood Portfolios: describe a portfolio in a sentence, hold it on-chain",
"summary": "Robinhood Chain is the one place where tokenized stocks, crypto majors and memecoins trade against each other in the same pools, so a basket holding NVDA, WETH and a launchpad coin is just three tokens in one vault. Write a sentence at three.ws/markets/robinhood/portfolios and a screen ranks all 700-odd holdable assets against it, then hands back a real basket with target weights, a thesis, and a per-holding reason for every pick. The whole methodology, the prompt, the weights, the schedule, and every holding's price and liquidity at the moment it was chosen, is hashed into a manifest you can download, so the document you read is provably the one the portfolio is deployed against. The vault behind it holds the basket in kind: you deliver every holding to mint and receive every holding to redeem, so it never sells anything to let somebody in or out and needs no price feed to work at all, which matters because most of this chain has none. Rebalancing back to target happens on the schedule the manifest published, and is offered as a descending-price auction anyone can fill rather than a trade a manager makes. Refining a portfolio mints a new version pointing back at its parent instead of editing what you already hold, and forks pay their ancestors out of a fixed slice of fees. When a tokenized stock is paused by its issuer everything stops until it resumes, and the page says so, because a halted share cannot be moved at any price. A companion board at /markets/robinhood/portfolios/universe lists every asset a portfolio may hold, with live liquidity, volume, and the premium each stock trades at against its Chainlink price.",
"tags": [
"feature"
],
"link": "/markets/robinhood/portfolios"
},
{
"date": "2026-09-07",
"title": "Hood Desk: a live onchain trading desk for any Robinhood Chain wallet",
"summary": "Paste a Robinhood Chain address (or connect a wallet) at three.ws/markets/robinhood/desk and the whole book loads: total value in USD, every token you hold priced against its Chainlink NAV feed or its deepest Uniswap pool, a native balance curve rebuilt from the chain's own history, and a live tape of your swaps, transfers and mints that folds each trade into one readable line instead of three. The centrepiece is the NAV premium ridge: the spread between what every tokenized stock is worth and what it actually trades for on the DEX, redrawn once per liquidity floor, so a dislocation that survives to the back rows is one you could really size into. Under it, the widest tradeable gaps, ranked. Reading any wallet is free, keyless and signature-free, so a desk link is a view anyone can open.",
"tags": [
"feature"
],
"link": "/markets/robinhood/desk"
},
{
"date": "2026-09-07",
"title": "Awesome 3D Agents: a curated list of everything you need to give an AI a body",
"summary": "Giving an agent a body means assembling a pipeline out of a dozen separate fields, and the map of that pipeline did not exist anywhere. There is now one: 152 hand-picked projects across 15 sections, covering text-to-3D generation, photo and video capture, avatars, rigging, motion generation, texturing, lipsync and voice, web renderers, asset pipelines, AR and XR, physics and networking, agent frameworks, datasets, free assets, and where to learn all of it. Each entry says in one sentence what the thing does and why you would reach for it instead of the alternative. The list is searchable and filterable by tag at /awesome, copyable as Markdown, and published in the standard awesome-list format on GitHub so it can be read, forked, and contributed to there. Every link is fetched and verified before it ships.",
"tags": [
"feature",
"docs"
],
"link": "/awesome"
},
{
"date": "2026-09-07",
"title": "Tutorials and walkthroughs are now readable by search engines, and pages we never wanted indexed stopped asking to be",
"summary": "The 76 tutorial and walkthrough pages loaded their content after the page opened, which meant a search engine reading the raw page found no text at all, and on tutorials found the words 'Tutorial not found' sitting under an empty heading. Those pages now arrive with their full text already in them, so they can finally be found by the questions they answer. Separately, eighteen pages were being submitted to Google while telling Google not to index them: sign-in surfaces, viewers that need a link to work, and one endpoint that returns data rather than a page. They have been withdrawn from the sitemap, and a new check refuses to let a page ask for indexing and refuse it at the same time. Two documentation pages that shared a title with the product page they described have been retitled.",
"tags": [
"improvement",
"docs"
],
"link": "/tutorials"
},
{
"date": "2026-09-07",
"title": "Reference-image upload now tells you to paste a URL instead of failing with a network error",
"summary": "When the image storage credentials are rejected, /forge used to hand the browser an upload address that the storage service refused. The refusal arrives without the header a browser needs to read it, so the page could only say \"Network error during upload\" and the reference image was lost. The upload endpoint now checks storage the same way the rest of the site does, so it says up front that direct upload is unavailable and points you at the paste-a-URL field, which keeps working through the outage.",
"tags": [
"fix"
],
"link": "/forge"
},
{
"date": "2026-09-07",
"title": "Archived news stories with emoji in their summary keep their search listing",
"summary": "Google Search Console flagged an archived story page for unparsable structured data. The cause: the archive trims a publisher summary to a fixed length before serving it, and the cut landed in the middle of an emoji, leaving half a character in the page data that search engines read. Trimming is now character-aware, so it never splits an emoji, a flag, or an accented letter, and the page data is checked for half characters before it is written. Around one archived story in seven thousand carried the flaw; all of them serve a valid listing now. The same fix covers the story summaries shown in link previews and the shared page descriptions across the site.",
"tags": [
"fix",
"infra"
],
"link": "/markets/news"
},
{
"date": "2026-09-07",
"title": "Keyboard focus is visible again on every /play HUD button, and the AGI filter counts clear AA contrast",
"summary": "Six buttons on the /play world HUD (friends, jobs, avatar, wardrobe, shop and buy) had lost their keyboard focus ring: the page never defined the shared focus-ring colour those buttons outline with, so the browser dropped the outline entirely and a keyboard player could not see where they were. The ring is back. On /agi, the count inside a selected filter chip measured 4.37:1 against its background at 11px, just under the 4.5:1 AA floor; it now uses the primary ink and clears it comfortably.",
"tags": [
"fix",
"improvement"
],
"link": "/play"
},
{
"date": "2026-09-07",
"title": "The AgenC task embed now loads even when it is inlined or served from a blob",
"summary": "The <three-ws-agent> element that shows a 3D avatar reacting to an on-chain AgenC task resolves its mesh decoder relative to its own script address. A copy that had been pasted inline, bundled, or loaded from a blob: or data: URL has no address to resolve against, and the element failed to register before it drew anything. It now falls back to the decoder on three.ws in that case and keeps the same-origin path everywhere else, so the embed works wherever the snippet ends up.",
"tags": [
"fix"
],
"link": "/agenc/embodied"
},
{
"date": "2026-09-07",
"title": "Shared avatar and agent links now show a real page to Google, not a loading spinner",
"summary": "Search engines were shown a different version of every /avatars/... and /agents/... page than people were, and that version had nothing on it: the name and description were hidden in a fallback block that a crawler removes before reading, the rest was a spinner, and the last line of the page sent the crawler straight back to the URL it was already on. Tens of thousands of avatar and agent pages therefore looked identical to Google and were filed as duplicates instead of being listed. Crawlers now get the same content a visitor sees: the name, the description, the rendered card, the creator, the tags, the skills, the on-chain identity, and links onward into the gallery and the directory. A link to an avatar or agent that no longer exists now says so honestly instead of quietly bouncing to a listing page, an id that belongs to the other side of the platform redirects to where it actually lives, and a private agent is no longer readable by anything sending a crawler's name. Signed-in surfaces were also unblocked in robots.txt so their existing no-index instruction can finally be read and obeyed.",
"tags": [
"fix",
"infra"
]
},
{
"date": "2026-09-07",
"title": "3D generation and the whole site keep working when our asset storage does not",
"summary": "Early on 2026-09-07 our object storage stopped accepting the platform's credential, and because every generation parks its reference image there, 3D generation stopped on the website and in ChatGPT at the same time, agent registration could not store its manifest, and images loaded through our CDN stopped resolving. Three things changed so that a storage fault is never again a blank page. Avatars, thumbnails and models now fall back to the bucket's public domain automatically, so the site keeps rendering. When generation genuinely cannot run you get a plain message saying storage is recovering and to try again shortly, instead of a raw error from the storage vendor. And our own health checks now watch the storage credential directly and treat a silent forge as an outage rather than a quiet hour, so we hear about this class of failure before you do.",
"tags": [
"fix",
"infra"
]
},
{
"date": "2026-09-07",
"title": "My Creations: one button that takes you back to everything you made",
"summary": "A creator told us the making part was intuitive and the finding part was not: they built an agent, closed the tab, and had no obvious way back to it. There is now a My creations button in the header, and it opens a single page holding everything you have made, newest first: your agents, the 3D avatars saved to your account, and every model you forged, with Open, Edit and Copy link on each one. Signed out it still finds the models forged in this browser, so work made before you had an account is not lost. Filter by type, search by name or prompt, and hover any card to spin it in 3D. The create wizard and the Forge now point here too, so you are told where your work lives at the moment you make it.",
"tags": [
"feature",
"improvement"
],
"link": "/mine"
},
{
"date": "2026-09-07",
"title": "A payment we cannot read now comes back with the price list, not a bare error",
"summary": "When an agent pays one of our x402 endpoints and the payment header arrives mangled (truncated in transit, double-encoded, wrapped by a proxy), the answer used to be a plain 400 error with no prices in it, so the agent had nothing left to work from and no way to retry. Every rejected payment now returns the ordinary 402 with the full list of accepted rails, tokens and amounts attached, exactly as it does when a payment is simply missing. That is what the x402 spec and the OKX seller SDK both do, and it means a paying agent can always read the quote again and try once more. Only a real server fault answers without a quote.",
"tags": [
"fix",
"improvement"
],
"link": "/docs/okx-marketplace"
},
{
"date": "2026-09-07",
"title": "Point an AI agent at the Blender on your own machine",
"summary": "A new MCP server drives the Blender you already have installed, with no GUI open and no add-on to install first. An assistant can now read a 3D file and tell you what is actually in it (objects, triangle counts, materials, bones, animations, real-world size), convert between GLB, glTF, FBX, OBJ, STL, PLY, Collada, Alembic, USD and .blend in either direction, render a preview image of a bare asset file that arrives with no camera or lighting of its own, and run a Blender Python script to do the things no single tool covers, like halving a mesh's triangle count and exporting the result in one step. It also reaches back into three.ws: describe something in a sentence and the free text-to-3D lane generates it and opens it in Blender. Blender runs in the background, one process per request, so it works the same on a laptop, a server, or in CI.",
"tags": [
"feature",
"sdk"
]
},
{
"date": "2026-09-07",
"title": "The Blender tools now show you the render instead of describing it",
"summary": "Asking an assistant to render a 3D model used to hand back a file path, which is useless if the assistant cannot open files on your disk. The rendered image now comes back with the answer, scaled to fit, while the full-resolution PNG still lands on disk. That turns a look at this model into one step, so an assistant can render, spot the problem, fix it and render again. Reading a file also reports its texture budget now: every image with its resolution and byte size, plus the total, which is usually what actually decides whether a model is small enough to ship, and it was previously reported as nothing more than a count. Two reliability fixes came out of real use: several requests at once now queue instead of all starting Blender at the same moment and exhausting the machine, and when Blender is installed but momentarily will not start you get a message saying exactly that, with the details, rather than being told to install software you already have.",
"tags": [
"improvement",
"feature"
]
},
{
"date": "2026-09-07",
"title": "Blender tools now open three.ws avatars, which Blender alone cannot",
"summary": "Most three.ws avatars are delivered in a compressed form that Blender flatly cannot read: point stock Blender at one and it refuses the file. That made the new Blender tools useless on exactly the models this platform produces. Compressed models are now unpacked automatically before Blender opens them, for both of the common compression schemes, with nothing extra to install and your original file left untouched. The answer tells you when it happened.",
"tags": [
"fix",
"sdk"
]
},
{
"date": "2026-09-07",
"title": "One command makes a 3D model small enough to ship, and shows you all four sides",
"summary": "Getting a model down to a size a web page can load used to mean writing a Blender script by hand every time: cut the polygon count, shrink the textures, throw away what nothing uses, compress the result. That is now a single request with a budget, and it answers with an honest before and after, so you can see what the size actually cost you in detail. On a test model it went from 1.3 MB to 282 KB. Previews can also orbit the model now: ask for four angles and you get four pictures from one pass, which is what tells you whether the back of a model was ever finished, or whether a texture only looks right from the front.",
"tags": [
"feature",
"sdk"
]
},
{
"date": "2026-09-07",
"title": "Turn a photo into a 3D model without leaving your editor",
"summary": "The Blender tools could already build a model from a written description. They now build one from a picture too: point at an image on your machine or a link to one, add a few words to steer it if you like, and the finished model opens in Blender. It runs on the same free lane as the text version and takes about as long.",
"tags": [
"feature",
"sdk"
]
},
{
"date": "2026-09-07",
"title": "The partner-hosted demo page loads all of its pieces again",
"summary": "Our partnership page is published once onto a partner's own website, so everything it loads comes from three.ws across origins. Two of its scripts were being refused by the browser there, which quietly cost the page its in-editor map and its iOS handoff. Both are now served with the same cross-origin permission the rest of the page already had, and an audit that loads the page from a foreign origin now passes end to end.",
"tags": [
"fix"
]
},
{
"date": "2026-09-07",
"title": "The featured agent is one you can actually open",
"summary": "The agent we spotlight is picked automatically from real activity, but the pick only checked whether an agent was public, not whether its owner had published it. That let a brand new starter agent, the draft every account is given on signup, take the spot, and every panel that tried to read it came back empty with a live feed error. The pick now requires a published agent, so the spotlight lands on one that every part of the site can open.",
"tags": [
"fix"
]
},
{
"date": "2026-09-07",
"title": "World tiles show their artwork again",
"summary": "The picker that lets you choose a community world was loading each world's artwork straight from a public storage gateway, and browsers now refuse to draw images served that way. Nine tiles on the page failed silently every visit. The artwork is now fetched through our own image service, so tiles paint their real picture, and a world whose artwork has gone missing falls back to its lettered tile instead of an empty square.",
"tags": [
"fix"
]
},
{
"date": "2026-09-07",
"title": "A sharper cutout for images with hair, fur, or fine edges",
"summary": "Background removal now offers BiRefNet as an optional model alongside the fast default. The default model smears wispy boundaries into a soft halo, and because every image-to-3D lane rebuilds geometry from the cut-out subject, that halo turned into real polygons on the finished mesh. BiRefNet keeps hair strands, fur, foliage, and thin straps intact. It takes about six seconds against the default's one, so it stays opt-in: pass model \"birefnet\" on the background-removal endpoint or pick it from the MCP tool when the edge is what matters.",
"tags": [
"improvement",
"feature"
]
},
{
"date": "2026-09-07",
"title": "The splat viewer opens the compressed formats captures actually ship",
"summary": "Gaussian-splat scenes used to arrive as enormous PLY files, and the tools that make them now export compressed containers instead: the same capture that is a gigabyte of PLY is about 100 MB as SPZ and about 42 MB as SOG. The viewer could not open either. It now reads .spz and .sog alongside .ply, .splat, and .ksplat, works out the format from the file contents instead of trusting the filename, and frames the camera on the scene it actually loaded rather than a fixed distance that suited one file in ten. It also shows the real splat count, and waits for the first sorted frame before going live so a slow machine no longer shows an empty stage.",
"tags": [
"improvement",
"feature"
],
"link": "/splat"
},
{
"date": "2026-09-07",
"title": "Studio Lab reads the same splat formats as the viewer",
"summary": "The Lab's Gaussian-splat tools ran on a different rendering engine from the Splat Viewer, so the two disagreed about which files they would open: the viewer had just gained .spz and .sog and the Lab had not. Both now share one engine, so anything one of them opens the other opens too, the Lab works out a dropped file's format from its contents instead of its name, and its camera frames the scene it loaded.",
"tags": [
"improvement"
],
"link": "/forge-studio"
},
{
"date": "2026-09-07",
"title": "Security: the image and mesh workers now run patched libraries",
"summary": "The services that strip a background, convert a mesh, and paint a texture all decode a file you upload, and they were pinned to an image library with a long list of published flaws, including several that let a crafted image corrupt memory. Those three services now run the patched release, and the mesh converter's 3D-format library was moved off a version with a critical flaw of its own. There is also a new check that scans every pinned Python dependency across the whole worker fleet against the public vulnerability database, so the next one gets found rather than waited for.",
"tags": [
"security",
"infra"
]
},
{
"date": "2026-09-05",
"title": "The VS Code extension now animates, refines, optimizes and reviews your models, and checks your embeds",
"summary": "Open a rigged model in VS Code, press Animate, and pick from the same 2,800-clip motion library the site uses: the clip is retargeted onto that model's own skeleton right in the editor, and Bake clip writes it into a copy of the file so it plays in Unity, Unreal or Blender too. Describe a motion in words and the GPU fleet samples one for you. Refine takes a change like make it metallic and generates a new version anchored to the current one, keeping the version history. Optimize runs the platform's web passes on your machine and tells you what it saved. Compare pulls the committed version straight out of git, opens both side by side, and writes a structural review of what changed. Check quality asks a vision model to score the render and hands its fix straight to Refine. A turntable GIF button makes README material in one click. And in HTML, React, Vue, Svelte, Astro and Markdown files the editor now checks agent-3d tags as you type: a missing source or size, an unpinned library script, a stale integrity hash the browser would refuse, a misspelt attribute, each with a one-click fix, plus hovers and completions for every attribute.",
"tags": [
"feature",
"sdk",
"improvement"
],
"link": "/docs/vscode"
},
{
"date": "2026-09-05",
"title": "IRL no longer fails to load your agent on a cold start",
"summary": "Opening IRL on a fresh page could show \"Couldn't load this agent\" with a Retry button even though nothing was wrong with your connection. The 3D loader was asked for a compressed avatar a moment before its decoder had finished arriving, and it gave up instead of waiting. Every avatar load on IRL and in the AR Studio now waits for that decoder, so the first load works the same as the retry always did.",
"tags": [
"fix"
],
"link": "/irl"
},
{
"date": "2026-09-05",
"title": "Walk up to an IRL agent, talk to it out loud, and stand it on your floor in AR",
"summary": "Tap any agent you discover on IRL and the card now leads with Talk and View in AR. Talk is a real conversation with the agent standing in front of you: hold the mic and speak, or type. It answers in its own persona, out loud in its voice (its cloned voice when it has one), and the 3D model's mouth moves with the words while it keeps breathing. Press the mic while it is talking to interrupt it. View in AR works on anyone's agent, not only your own: iPhone opens it in Quick Look on the real floor, animated, with a banner that brings you back to the conversation; Android phones with ARCore anchor it to the detected floor right inside the page, tap to place and pinch to size; every other device opens the AR launcher. Owners see a new \"talked with your agent\" event in their IRL feed when someone starts a conversation. The transcript stays between the visitor and the agent.",
"tags": [
"feature"
],
"link": "/irl"
},
{
"date": "2026-09-05",
"title": "Invite people to an IRL agent with a visit link and a printable sign",
"summary": "Every agent you have placed in the real world now has a visit link and a sign you can print and tape to the spot. Someone who scans it opens IRL with a banner naming the agent they came to meet, and the moment they are close enough the agent appears in their camera, flashes, and opens its card by itself: bio, reputation, wallet, tip QR, services, and a message box. That card also gained View in AR: on iPhone it stands the agent on the real floor in Quick Look, animated, and the banner inside AR brings you straight back to the card. The link never contains a location, so you can post it anywhere and the agent still only shows to people who are actually there. Find it under Visit link on each placement in your dashboard.",
"tags": [
"feature",
"improvement"
],
"link": "/irl"
},
{
"date": "2026-09-05",
"title": "Seven background jobs that were never running now run",
"summary": "Seven scheduled jobs were declared in the platform config but had no matching job on the scheduler, so they had never fired: the companion poll, feedback triage, print-order syncing, the home health alert, the home-relay sweep, web-variant generation for the forge, and the oracle refit. Three of them shipped with live features behind them, which meant those features quietly did nothing on a timer. All seven are now created and enabled on their declared schedules, and a drift check confirms every declared job exists and matches.",
"tags": [
"fix",
"infra"
]
},
{
"date": "2026-09-05",
"title": "The site header now stays put when you scroll",
"summary": "The navigation bar was set to stick to the top of the window but never actually did: it scrolled away with the page on every route, at every screen size, so reaching the menu meant scrolling all the way back up. The bar was mounted inside a wrapper exactly its own height, and a sticky element can only travel inside its parent, so it had nowhere to go. The header now stays on screen wherever you scroll, on phones and desktop alike.",
"tags": [
"fix",
"improvement"
]
},
{
"date": "2026-09-05",
"title": "3D models now open, animate and report on themselves inside VS Code",
"summary": "A new three.ws extension makes .glb and .gltf files first-class in the editor. Double-click one and it renders in an editor tab instead of showing binary-file-not-shown: orbit it, play its animation clips on a scrub bar, flip on the wireframe or the skeleton, and save the view as a PNG next to the model. A report panel counts triangles, textures, bones and extensions and says what to fix, using the same inspector as the validation page, running on your machine so the model is never uploaded to be measured. Compressed avatars decode locally, so it all works offline. From the command palette you can also generate a model or an avatar from a sentence on the free lane, rig a static model for animation, browse and import recent community creations, and drop in an agent-3d embed snippet with the current library version and its integrity hash already pinned.",
"tags": [
"feature",
"sdk"
],
"link": "/docs/vscode"
},
{
"date": "2026-09-05",
"title": "Your agent now lives in a handheld console you can put on any website",
"summary": "The new Pocket Console at /pocket is a real handheld with a real 3D agent inside the screen. The D-pad walks it, A waves, B jumps, START swaps between the agent walking, its live stats, and a menu of every public agent you can slot in like a cartridge, and L and R flip through them. It takes keyboard, touch and a plugged-in game controller. Four shell finishes, four scenes, and the whole state travels in the link, so what you share is the device exactly as you left it. Add chrome=off and one iframe puts the console, your agent included, on your own site: transparent background, no sign-in, no SDK, no API key.",
"tags": [
"feature"
],
"link": "/pocket"
},
{
"date": "2026-09-05",
"title": "A new agent skill teaches Claude to look for an existing 3D asset before making one",
"summary": "find-3d-assets joins the three.ws Agent Skills pack. It tells an AI assistant that the platform already publishes 3,492 finished 3D assets and how to search them, so a request like put a chair on my landing page gets an existing CC0 chair in seconds instead of a generation job that takes GPU minutes and produces something nobody has looked at. The skill covers all three ways in (the free MCP tools, the plain HTTP endpoint, and the command line), which framework to emit for each kind of asset, and how to read the license that travels with every result. It also says plainly when to give up on the catalog and generate instead, and to tell the user which of the two they got.",
"tags": [
"feature",
"docs"
],
"link": "/docs/agent-skills"
},
{
"date": "2026-09-05",
"title": "One command drops a three.ws prop, character or motion clip into your project",
"summary": "npx @three-ws/assets search wooden chair finds it, and npx @three-ws/assets add object:painted_wooden_chair_01 downloads the model into your public folder and prints the tag that renders it, already pointing at your local copy rather than our CDN. Your site keeps working even if we do not. The catalog behind it is the same one the AI tools read: 511 CC0 props, 107 rigged characters and 2,874 motion clips, searchable in one place instead of three separate galleries. It is also a plain HTTP endpoint now, GET /api/catalog, free and open to any origin, so a build script or a site that is not using an AI client can read it too. Adding an asset twice changes nothing the second time, and a file you edited afterwards is never overwritten without asking. Every object is CC0, and the license of anything that is not travels with it in the output.",
"tags": [
"feature",
"sdk"
],
"link": "/docs/api-reference"
},
{
"date": "2026-09-05",
"title": "Machine Atlas: two working machines built in the browser from code, with no 3D model files",
"summary": "A new Labs surface at /assembly renders a nine-cylinder radial aero engine and a coupled steam locomotive without loading a single model file. Both are generated when the page opens: bore and stroke set the crank throw, the crank throw sets the crankcase diameter, and every fin, rail section and propeller blade is a profile revolved, extruded or lofted at runtime. Move any dimension slider and the geometry is rebuilt in front of you. The motion is solved rather than animated, so the pistons follow the exact slider-crank and the locomotive's valve gear closes a real four-bar chain every frame, which means a rod length that cannot physically reach simply refuses to move. Pull either machine apart with the explode slider, switch to the structure view to see the mesh, hover any part to name it, hide whole subassemblies, and read the actual generating calls in the Source panel, printed from the values you set. Every configuration has its own link, and any of them can be exported as a GLB or saved as a PNG frame.",
"tags": [
"feature"
],
"link": "/assembly"
},
{
"date": "2026-09-05",
"title": "AI coding assistants can now search the three.ws asset library and get working code back",
"summary": "Every ready-made asset the platform publishes is now one search away from any AI coding client connected to the three.ws MCP server: 511 CC0 props, 107 rigged characters and 2,874 motion clips, all in a single searchable catalog instead of three separate browse pages. Ask for a wooden chair or an industrial lamp and you get matching items ranked best first, with tags, license, and the CDN link. Ask for the source and you get code that runs as it is: the model-viewer tag our own gallery renders with, the agent-3d web component pinned to the exact published version and its integrity hash, plain three.js, or a React component. Motion clips come back as three.js code that loads and plays them. All three tools are free and need no account, no API key and no payment, because you should be able to find out what is already here before you sign up for anything. A search that matches only part of what you typed says so instead of returning nothing.",
"tags": [
"feature",
"sdk"
],
"link": "/docs/mcp"
},
{
"date": "2026-09-04",
"title": "Generation jobs can now push their result to your server, and the health page finally watches the editing tools",
"summary": "Two additions for anyone who builds on the forge. First, webhooks for generation jobs: subscribe to forge.completed and forge.failed and three.ws POSTs you the finished model instead of making you poll for it. A generation outlives the request that started it, so until now an integration had to keep asking. The completion event rides the single writer every lane finishes through, so it arrives exactly once whether the job ran on the free lane, a paid call, or the background finalizer that picks up abandoned work; the failure event deliberately stays quiet while a job is being retried on another engine, so it only ever means the job is really over. Payloads and a runnable example are in the developer platform docs. Second, the forge health report now live-probes the editing tools as well as the generation engines: remesh, stylize, segment, background removal, text to motion, auto-rig and retexture each get checked for real. Those tools have buttons in the result panel, so a user can hit them, but none of them were being watched; retexture had in fact been offline the whole time and nothing said so.",
"tags": [
"feature",
"sdk",
"improvement"
],
"link": "/changelog"
},
{
"date": "2026-09-04",
"title": "Daily Match reads on a phone, and its live ticker finally matches its board",
"summary": "The Daily Match board tracks which agent ships the most real output each UTC day. Two things were wrong with it. On a narrow screen the eight-column standings table was being silently clipped instead of scrolled, so P&L and Score, the two columns the whole page is about, simply could not be reached on a phone; the table now scrolls inside its own region, with edge shadows that show up only while there is more to reach and keyboard focus so you can scroll it without a mouse. And the Live output ticker only ever read one of the four activity sources the board scores, so an agent that shipped a launch, a trade or a paid skill sale could rank first on a board sitting next to a panel claiming the day was quiet; the ticker now reads all four and labels each entry with what it was. The board also got wider on desktop so every column fits without scrolling, a designed loading and failure state for the ticker, and a refresh that no longer wipes good standings off the screen when a single poll fails: it says live updates are paused and puts them back when the next one lands.",
"tags": [
"fix",
"improvement"
],
"link": "/daily-match"
},
{
"date": "2026-09-04",
"title": "A second opinion now reviews every trade a sniper agent wants to make",
"summary": "Trading agents already pass a stack of mechanical checks before they buy: the market-cap band, the price-impact ceiling, budget and slot caps, and an on-chain honeypot check that simulates a real buy and a real sell. All of those answer a question with a number. A new Risk Officer asks the one they cannot: knowing everything we know, is this specific trade still a good idea? It is shown the proposed trade plus the agent's own reason for wanting it, and is told to look for what the agent missed, so it argues with the trade instead of agreeing with it. It starts in shadow mode on every strategy: it reviews and records, and changes nothing, so its calls can be scored against what the trades actually did before anyone lets it veto one. When it is armed, a veto has to name a specific fact, it can shrink a trade instead of killing it, it can only ever reduce risk and never raise it, and if the reviewer is unavailable the trade goes ahead unchanged rather than the fleet stalling.",
"tags": [
"feature",
"security"
]
},
{
"date": "2026-09-04",
"title": "Trader Wrapped: every trading agent's season, in eight slides you can share",
"summary": "Pick any trading agent on three.ws and read its season back as a recap you swipe through: what it made, the trade that carried it, the trade that hurt, the streaks and hours behind both, where it finished against every comparable trader, and the receipt with its verification state. Every number is cut from closed on-chain round-trips with buy and sell signatures attached, using the exact same arithmetic the leaderboard and the trader profile use, so a recap can never disagree with the record it links to. It is built to be honest rather than flattering: the worst trade is its own slide, a losing season is labelled a losing season and drawn in red, coins that cost the trader money still rank in the breakdown, coins a trader launched themselves are excluded from every superlative, the rank always shows how big the field really was, and a trader with fewer than three settled trades gets told there is no season yet instead of a chart made out of noise. Sharing a recap gives you a real preview card on X, Telegram and Farcaster, and the link carries your referral code when you are signed in. Nothing here signs, spends, or touches your wallet: browse it all signed out, and open the trade panel only if you choose to fork a coin you read about.",
"tags": [
"feature"
],
"link": "/wrapped"
},
{
"date": "2026-09-04",
"title": "Every 3D viewer on the site now runs the same build",
"summary": "The 3D preview you see on three.ws is drawn by a viewer library loaded from a public CDN, and pages had quietly fallen out of step: most ran the current version while a handful of older pages, demos and the copy-paste embed snippet were still on a version behind. That split is worse than it sounds. Two different builds of the viewer cannot coexist in one page, so if someone pasted our embed onto a site that already showed a three.ws model, the second one would refuse to start and show nothing. Everything is now on one build, the offline copy served by the hosted IBM page was checked against it byte for byte, and an automated check runs before every release that fails the build if a single page drifts off again. Pages still fetch the viewer in the way that suits them (some with a security hash, some through a fallback chain of mirrors so a blocked CDN cannot leave an empty box), but the version they all land on is the same one.",
"tags": [
"fix",
"improvement"
]
},
{
"date": "2026-09-04",
"title": "Generated models are checked for the invisible surface detail that makes them look real",
"summary": "A 3D model can have perfect shape and still look like plastic, because the colour picture painted onto it is only half of what makes a surface look real. The other half is invisible detail: which parts are bumpy, which are shiny, which sit in shadow. Without it, leather, wool and glass all catch the light in exactly the same flat way, and an object can read as though it were cast in dull metal rather than made of what you asked for. There was no way to check which of those layers a model actually carried, so we built one and pointed it at real generations from the live engines, then wrote down what each engine does and does not produce. That measurement is now published alongside the texturing service instead of being guessed at, and the checker refuses to pass a model missing a layer it should have, so the gap cannot quietly reopen.",
"tags": [
"docs",
"improvement"
]
},
{
"date": "2026-09-04",
"title": "Opening the homepage, Create and Discover no longer loads 3D you never asked to see",
"summary": "Three pages were doing heavy 3D work before anyone had scrolled to it. The homepage hero picked whichever featured avatar the list returned first, which meant a 4.6 MB model downloading and decoding while the page was still trying to become usable; it now picks the richest avatar that fits a size budget, which cut the hero to 2.1 MB without changing what the stage shows. Discover was fetching an entire 3D viewer engine on every visit even though almost every card in the directory is a still thumbnail, and now fetches it only when a card that really has a 3D model comes near the screen. The Create page did the same for one preview card that already waited for your cursor, and now waits for it too, with the poster on screen in the meantime so the card looks the same. The walking companion also stopped booting on top of a page that is still loading: it waits for the page to settle first, then turns up as before.",
"tags": [
"improvement",
"fix"
],
"link": "/discover"
},
{
"date": "2026-09-04",
"title": "The friends panel tells you when it cannot reach the server, instead of saying you have no friends",
"summary": "When the friends list could not be loaded, the panel showed the same screen it shows a brand new player: \"No friends yet\", with a button inviting you to go and add someone. That reads as though your friends were gone, when the truth was only that we could not look. It now says plainly that the connection dropped, that nothing was lost, and offers a Try again button that really does retry, and if the list had already loaded once, it stays on screen and is marked as possibly out of date rather than being thrown away. Coming back online repaints the panel straight away, even when your friends list is exactly as you left it. The panel also had a stray word \"null\" printed under the tabs every time it opened, on both the world and the walk pages, which is now gone.",
"tags": [
"fix",
"improvement"
],
"link": "/play"
},
{
"date": "2026-09-04",
"title": "The one-line x402 script now works on your site, not just on ours",
"summary": "Our docs tell you to make any element payable by dropping one line onto your page: a script tag pointing at three.ws/x402.js. On your own domain that line never worked. Your browser asked us whether another site was allowed to read the file and we never answered, so the script was refused and the button stayed inert. The same silence broke the runtime translation script and the language catalogs it reads, and it broke the IBM partnership page that IBM hosts on its own domain: that page keeps itself current by pulling the latest copy from us, and instead had quietly frozen on the version baked in the day it was published. All of it answers any site that asks now. The live copy also carries its own script and translation addresses with it, instead of asking your server for files only we have, and a new check loads that page from a stranger's domain so none of this can go unnoticed again.",
"tags": [
"fix",
"sdk",
"infra"
],
"link": "/docs/x402-studio"
},
{
"date": "2026-09-04",
"title": "Two paid endpoints now quote a price to anyone who asks, instead of an error",
"summary": "The 3D asset pipeline and pay-by-name resolution both checked your request for missing fields before they ever mentioned the price, so a caller who simply asked what they cost got a validation error back. Directories that index paid APIs ask exactly that way, with an empty request, and read an error as a service that sells nothing, which is why neither endpoint could be listed anywhere. Both now answer with the payment challenge and the real price first. The pipeline quotes the published example chain when the request names no stages, and anyone who pays and then sends an incomplete request still gets the same clear validation error, with nothing charged.",
"tags": [
"fix",
"improvement"
]
},
{
"date": "2026-09-04",
"title": "The Smart Money directory stops asking your browser for pictures it can never get",
"summary": "Every wallet in the copy-trading Smart Money list carried a portrait link pointing at an outside image host that serves nobody but its own site. Browsers blocked all of them, so the list drew the trader initials it was designed to fall back on while quietly recording a failed request for each row on screen. That unusable link is no longer sent, so the directory looks exactly as it did with none of the wasted traffic behind it.",
"tags": [
"fix",
"improvement"
]
},
{
"date": "2026-09-04",
"title": "The IRL placements desk can use your location again, and only when you ask it to",
"summary": "The page that pins your agents to real places offers to measure how far away each pin is and to drop a new one where you are standing. None of it could work: the page arrived with the browser location service switched off for it, so every attempt failed on the spot. Location is allowed on that page now. It also no longer reaches for your position the second the page opens: it reads it only when you have already granted permission, and otherwise waits for you to press one of the location buttons.",
"tags": [
"fix",
"improvement"
]
},
{
"date": "2026-09-04",
"title": "DeFi tables stop showing broken logos when a coin icon disappears",
"summary": "The protocol and chain logos on the fees, DeFi, DEX volumes, chain, protocol and exchange pages were loaded straight from a third-party icon host. When that host renamed or removed an icon, the row rendered a broken-image glyph and every reader's browser recorded a failed request. Those logos now come from three.ws itself: we fetch and resize them on our side, cache them, and hand back an empty answer for an icon that no longer exists, so the row shows the same neutral disc a protocol without a logo already gets. Pages load a little faster too, since a 24-pixel icon no longer arrives at full size, and your browser no longer talks to an outside host on every table paint.",
"tags": [
"fix",
"improvement"
],
"link": "/fees"
},
{
"date": "2026-09-04",
"title": "Cross-origin 3D loads no longer break on a header your site never sent",
"summary": "Embedding a three.ws avatar on your own site loads the model straight from us, and the site answers every origin. One case still failed: if your loader attached any extra header (a trace id, an auth token, anything a browser does not treat as standard), the browser first asked us for permission and we answered with a 404 page, so the model never loaded. That request is now answered properly, and the answer echoes exactly the headers you asked about. Alongside it, the media docs now spell out the three ways to read a model file and which one to pick: files under three.ws load directly, generated assets load directly from three.ws/cdn, and the /api/glb proxy is there for a URL whose host you cannot identify.",
"tags": [
"fix",
"docs"
],
"link": "/docs/media-api"
},
{
"date": "2026-09-04",
"title": "Knock somebody's door and get your money back if they never answer",
"summary": "Paying to reach a stranger has meant paying first and hoping. The money left your wallet the moment the payment cleared, and if the person never replied, that was that. Doors can now take escrowed knocks instead: your payment sits in a vault on Solana that the recipient cannot touch until they actually answer you. If they answer, they are paid. If they decline, you get everything back and are charged nothing, because reading something and passing on it is not a service. And if they simply never respond before the window they set, anyone at all can trigger the refund and every cent comes back to you, so it does not depend on you remembering to chase it. three.ws holds no key that can move that money, and neither does anyone else. Door owners can turn the lane on in their knock settings and choose how long they get to reply, from an hour to a month.",
"tags": [
"feature",
"security"
],
"link": "/docs/knock"
},
{
"date": "2026-09-04",
"title": "The 3D services we sell to other agents quote one clear price, however they are asked",
"summary": "Our four paid 3D generation services on the OKX.AI agent marketplace answer every caller with a price quote before any work happens. Two flaws in that quote are fixed. A request that arrived with no body and no content type was turned away with a formatting complaint instead of a price, and that is exactly how the marketplace's own compatibility check asks, so the check saw four services that looked like they had no payment set up at all. Separately, a request that named no specific tool got a quote listing the same payment network twice at two different prices, one of them wrong. Every unpaid request now receives a single unambiguous quote at the service's listed price, on each payment network it accepts.",
"tags": [
"fix"
]
},
{
"date": "2026-09-04",
"title": "The IBM partnership demos arm wherever the page is opened",
"summary": "The live demo pages loaded their own payment script across the network by full address, so on any address other than the main site (a preview, a local run, a copy published on a partner domain) the browser refused it and the paid demos never armed. Each page now loads the copy it ships with. The publish-once page IBM hosts got the opposite fix: every file it pulls now carries our full address, so its language switcher no longer looks for a file on the host's own domain and comes back empty.",
"tags": [
"fix"
],
"link": "/ibm/hello"
},
{
"date": "2026-09-04",
"title": "A knock now holds the money until you actually get a reply",
"summary": "Paying to reach someone used to send your money the instant the payment cleared, which meant an owner could take every knock and answer none, and you had no way to get it back. A new Solana program makes the payment conditional: it sits in a vault that only the knock itself controls, and exactly three things can happen to it. The owner answers inside the reply window and gets paid, the owner declines and you are refunded in full with no fee taken, or the window closes and anyone at all can trigger your refund. There is no admin key that can move a knock that is parked, the fee is capped at 10% in the program itself, and the fee you agreed to when you knocked cannot be raised on money that is already in escrow. Your message never touches the chain: only a fingerprint of it is stored, so either side can later prove what was sent without publishing it. Ten invariants covering every one of those promises run against the real compiled program on every test run.",
"tags": [
"feature",
"security"
]
},
{
"date": "2026-09-04",
"title": "Agent wallets can pay three.ws on X Layer again",
"summary": "Buying agents that pay for a 3D generation on X Layer sign their payment with a wallet that carries an on-chain delegation, which is how every OKX agent wallet is built. Our checker judged those signatures with a contract call that answers a different question, so it turned away correct, fully funded payments and quoted the price a second time instead of doing the work. Payments are now checked the same way the stablecoin itself checks them, so a signed authorization that the network would honour is one we accept.",
"tags": [
"fix"
]
},
{
"date": "2026-09-04",
"title": "When the payment wallets run thin, the money now reaches the right one first",
"summary": "three.ws pays its own on-chain fees from a funding wallet that tops up every engine below its floor. It ranked those engines purely by the size of their shortfall, so a top-up went to whichever wallet was missing the most, and the small wallet that pays the fee on every single payment could sit under its cutoff while the rail stayed shut. The payment wallets are now funded first, so a modest top-up restores settlement instead of a large one being required to reach it by accident. Two related repairs land with it: a wallet holding stablecoins but too little native currency to open the account a swap pays out into is now told exactly how much it is short, instead of retrying the same impossible conversion every half hour, and previewing the rebalancer with a dry run no longer executes real swaps.",
"tags": [
"fix",
"infra"
]
},
{
"date": "2026-09-04",
"title": "Knock on a stranger and get your money back if they never answer",
"summary": "A door that takes escrowed knocks now shows the option on its own page, next to paying up front. Choose it and your payment goes into an on-chain vault instead of the recipient's wallet: they are paid by answering you, and if they decline or the window runs out, every unit comes back to you. Before anything is signed, a dialog states the amount, who it reaches, and the exact deadline. Afterwards the page counts down to that deadline, and the moment it passes, a button hands your payment back. three.ws holds no key that can move it either way.",
"tags": [
"feature"
]
},
{
"date": "2026-09-04",
"title": "Run an escrowed door from your own wallet, and see what the chain actually says",
"summary": "Door settings now carry the escrowed lane: switch it on, choose how long you want to answer within, and open your on-chain door with your own wallet, which is the only thing that can. The panel reads the chain and tells you the truth rather than your intent, so a door you never opened, or one still charging last month's price, says so plainly instead of quietly turning strangers away. Escrowed knocks in your inbox show how long you have and what answering is worth, replying releases the payment, and refusing refunds the sender in full with no fee. Money still in escrow is no longer counted as earned.",
"tags": [
"feature",
"improvement"
]
},
{
"date": "2026-09-04",
"title": "The invitation link you just created stays on screen",
"summary": "Inviting somebody into your home hands you a link that works once. It used to vanish the instant the household list refreshed behind it, and since we keep only a fingerprint of that link and never the link itself, there was no way to get it back. It now stays put until you close the panel, whether or not the email went out.",
"tags": [
"fix"
]
},
{
"date": "2026-09-04",
"title": "The agent answering for us on the OKX.AI marketplace now says when it cannot answer",
"summary": "Our marketplace listing is answered by an agent that reads a buyer's message and writes back. It could fail in the quietest possible way: the process alive, the health page green, and every reply dying against an AI credential the provider had stopped honouring, which from the buyer's side is indistinguishable from being ignored. The host now asks the provider itself, on every boot and every fifteen minutes, and a refused credential turns the listing amber on our status page, raises an alert, and carries the exact commands that fix it. It also stopped crying wolf: a daemon that is merely still starting up no longer reports as dead.",
"tags": [
"fix",
"infra"
]
},
{
"date": "2026-09-04",
"title": "The agent that answers for us on the OKX.AI marketplace now lives somewhere that stays up",
"summary": "Buyers who message our marketplace listing were being answered from a development workspace that goes to sleep on its own, and one message this week sat unread for twenty hours because of it. It now runs on a host that stays up, keeps its identity across restarts, and reports its own reachability alongside every other part of the platform on the status page.",
"tags": [
"improvement",
"infra"
]
},
{
"date": "2026-09-04",
"title": "The Windows 11 widget now actually paints your agent on the widgets board",
"summary": "Windows renders the Agent glance widget from a card layout the site has to hand it, and we were handing it the address of that layout instead of the layout itself, so a pinned widget had nothing to draw. The worker now fetches the card, keeps a copy so a machine that wakes up offline still renders, re-paints slots that were pinned before it took over, and registers the 15 minute refresh the widget advertises, which was never actually being scheduled. Pin it with Win + W, Add widgets, Agent glance.",
"tags": [
"fix"
],
"link": "/glance"
},
{
"date": "2026-09-04",
"title": "Pairing a house, paying an agent and the status page all answer again",
"summary": "A packaging rule left two small token modules out of the deployed image, and because the server loads them the moment a request arrives, every route that touched them answered with a server error instead of a page: pairing or managing a house, the agent payment endpoint, and the public status page. The modules ship again, and the check that simulates the upload now runs as part of every deploy, so a file the server needs can no longer go missing quietly.",
"tags": [
"fix",
"infra"
],
"link": "/status"
},
{
"date": "2026-09-04",
"title": "Pin your avatars to IPFS from Settings, and choose which copy loads first",
"summary": "Settings now has an Avatar storage panel. Every avatar you own is listed with its size, its three.ws storage copy, and whether it also lives on IPFS. Pin one and it gets a permanent content address you can open on any public gateway, and a small R2 / IPFS switch decides which copy a viewer resolves first. If a deployment has no pinning provider it says so plainly instead of showing an address that leads nowhere. The About block on the same page now reports the build that is actually serving you and the language the page is really rendering in, rather than today's date and a hardcoded English.",
"tags": [
"feature",
"improvement"
],
"link": "/dashboard/settings"
},
{
"date": "2026-09-04",
"title": "The DeFi page reads on a phone, links straight to each protocol, and can refresh itself",
"summary": "Every protocol name on /defi is now a real link into its detail page, so you can open one in a new tab, and dragging to copy a TVL figure no longer navigates away. A Refresh button in the toolbar pulls new figures without reloading the page and keeps the current table on screen while it works; if the refresh fails, the numbers you already had stay put and the page says so. On a phone the table now fits the screen, so TVL and the 1-day move are readable without scrolling sideways. When the upstream is down you get a real retry button instead of a dead link, an empty feed says so plainly instead of showing a $0.00 total, and figures served from cache are labelled as cached rather than passed off as live.",
"tags": [
"improvement",
"fix"
],
"link": "/defi"
},
{
"date": "2026-09-04",
"title": "The agent economy demo now quotes the amount it actually pays",
"summary": "On the Agent Economy demo, NOVA buys a live Solana market briefing from ORACLE. The narration announced a 0.001 SOL payment while the listed price, the receipt panel and the transfer itself were all 0.000001 SOL, and the screen in the world printed a PAID badge even on the simulated runs that signed-out visitors get. Every one of those now reads the same figure from one place, and a simulated run says so on screen. The page also handles a bad day: if the stream drops, the run times out, or the market feed cannot be reached, you get a plain explanation and a Retry button instead of a stalled page and a pointer at the developer console. On a phone the Run demo button no longer sits off the edge of the screen, the 3D world resizes with the window instead of staying clipped at its widest size, and a browser without WebGL is told so rather than shown a black rectangle.",
"tags": [
"fix",
"improvement"
],
"link": "/demo"
},
{
"date": "2026-09-03",
"title": "Twenty-one finished packages were built and never shipped; the publisher now finds them by itself",
"summary": "Twenty-one packages in this repo were written, tested and documented and then sat unpublished, because the publish script read a hand-maintained array that nobody remembered to edit. That array is gone: the publisher now discovers every package on disk, so a package ships by existing rather than by being added to a list. It also runs a preflight first and refuses to publish anything that would land broken (a scoped package that would go out private, a missing README or LICENSE, an entry point that is not actually in the tarball). The standalone-repo mirror had the same blind spot and only ever saw MCP servers; it now covers library packages too. Along the way the licensing was made consistent: 86 packages shipped a LICENSE file saying all rights reserved while their manifest and the repo both said Apache-2.0, and every one of them now says Apache-2.0 in all three places.",
"tags": [
"improvement",
"sdk",
"docs"
],
"link": "/docs"
},
{
"date": "2026-09-03",
"title": "Refine now returns the same model at higher quality, not a different one",
"summary": "Refine bumps a finished model up a quality tier. It replayed your prompt but never pinned the randomness, so on the text path it painted a brand new reference image and reconstructed a different object: you asked for this model, better, and got another model, better. Every generation is now stamped with a seed that rides along on Refine, Retry and an automatic engine switch, so the tier is the only thing that changes. Starting a fresh generation still explores, so Make another on the same prompt keeps giving you something new. Two related fixes came with it: requests that differed only by their seed were being collapsed into a single job and handed back the identical file, and the force-a-fresh-run flag skipped the result cache but not that collapsing.",
"tags": [
"fix",
"improvement"
],
"link": "/forge"
},
{
"date": "2026-09-03",
"title": "Download your model as FBX or 3MF",
"summary": "The download menu on a finished model offered five formats it could convert inside your browser: GLB, OBJ, STL, PLY and USDZ. FBX and 3MF have no browser converter, so they were missing from the menu even though the platform has always been able to author them. Both are there now, converted on our side and handed straight back: FBX keeps the skeleton for Unity, Unreal and Maya, and 3MF is the modern package that slicers and print services expect. A server conversion shows real progress while it runs, says what went wrong if it fails, and leaves the finished file one click away in case the download was missed. The format menu also no longer runs off the left edge of a narrow phone screen.",
"tags": [
"feature",
"improvement"
],
"link": "/forge"
},
{
"date": "2026-09-03",
"title": "AI material restyling works again without an IBM key",
"summary": "Restyling a model's materials in plain language (make it chrome, weathered copper, frosted glass) asked IBM Granite for the answer and gave up with a not-configured message if no IBM credentials were present, even though the same free model chain that powers the rest of the site was one call away. It now leads with Granite where it is available and falls back to that shared chain everywhere else, so the button works on every deployment. The two matching tools on the 3D Studio agent server, material generation and prompt direction, had the same dead end and were fixed with it.",
"tags": [
"fix"
],
"link": "/forge"
},
{
"date": "2026-09-03",
"title": "The free 3D generation lane is back up, and it can no longer fail quietly",
"summary": "The worker behind the default free text-to-3D and image-to-3D lane stopped being able to load its model on the evening of 2 September, and then kept accepting work it could not do. It had cached that one failure and replayed it to every job that followed, while its status page still reported itself healthy, so nothing raised a flag: roughly a third of generations on that lane failed over about twelve hours. The lane is running again, and four things now stop a repeat. The load is retried instead of one bad moment being treated as final. A worker that genuinely cannot run turns work away, so your generation is handed to another backend rather than dying several minutes later. It reports itself as unhealthy, so it gets replaced automatically and shows up as down on the status page instead of green. And the part of the model that had to be fetched from an outside site every time the worker started, which is what failed, now ships inside the worker, so that fetch cannot fail again.",
"tags": [
"fix",
"infra"
],
"link": "/forge"
},
{
"date": "2026-09-03",
"title": "Every three.ws MCP server now shows up in the directories, and six more joined the list",
"summary": "The one file every MCP directory listing is written from was six servers short and thirty-five versions behind, because it was being kept by hand. It is generated from the server manifests again, so a listing can no longer drift from what we actually publish, and a build check fails when it does. Six servers that existed but were invisible to the directories are now in: Herald, Home, Knock, the Metaplex agent registry, on-chain agent wallets, and the Robinhood Chain server. Paid tools also answer more cleanly to a store reviewer: the discovery data they attached to a payment prompt declared a transport the spec does not allow, so an indexer threw the whole row away. That is fixed, and a new check drives every paid tool the way a reviewer's client does and fails if any of them answers with anything but a clear, priced payment prompt.",
"tags": [
"improvement",
"sdk"
],
"link": "/docs/mcp"
},
{
"date": "2026-09-03",
"title": "The parametric avatar goes from 122 sliders to 306, and a brush for everything they still cannot reach",
"summary": "Avatar Studio's Parametric base now carries 306 shape sliders instead of 122, curated from source data that was already vendored: eyelid heights and eye folds, nose bridge and septum, philtrum and cupid's bow, ear rotation and shape, temples, jaw bones, lats, navel, knee alignment, upper and lower leg length, and per-ethnicity phenotype anchors. Face regions keep separate left and right controls because facial asymmetry is part of a face; limbs get one symmetric slider. With that many controls the panel grew a filter box, collapsed body groups, and an \"N edited\" pill so you can see which closed group you have already touched. Above all of it sits Free Sculpt: turn the brush on and drag on the avatar to push or pull the surface anywhere, with a real-size on-screen ring, adjustable strength, and symmetry you can switch off. Playback pauses while you paint so the mesh holds still, dragging the background still orbits the camera, and Clear sculpt puts the body back exactly. Two things that used to be editor-only now reach everyone: a body edited in the avatar editor's Proportions panel and any free-sculpt stroke are both rendered into the GLB the server serves, so what you shaped is what viewers, embeds and AR downloads actually see.",
"tags": [
"feature",
"improvement"
],
"link": "/create/studio"
},
{
"date": "2026-09-03",
"title": "Copy trading now stops itself when a leader slides, and a trader has to earn the right to be copied",
"summary": "Three guards landed on the copy loop. You can set a drawdown limit when you start copying a trader: if their real closed trades fall that far below their best point, new copies stop automatically, you get told why, and exits on positions you already hold keep coming through so you are never frozen inside the leader that just tripped. Resuming while they are still down is refused with the actual numbers rather than a button that silently re-pauses. Second, a trader now has to have earned a record before anyone can put money behind them: five closed round-trips, a day of trading history, and real capital deployed, all measured on-chain. Fall short and the panel shows exactly what is still missing instead of a flat refusal, and a losing trader with a real record still qualifies, because this asks whether the record is real, not whether it is good. Third, you can no longer copy an agent you own, and follows between your own agents no longer count toward the follower score on the leaderboard. Along the way, starting or editing a copy from a trader profile was failing silently on every attempt; that is fixed.",
"tags": [
"feature",
"security",
"improvement"
],
"link": "/dashboard/copy"
},
{
"date": "2026-09-03",
"title": "Fork a trade: one tap from any coin on the site into the real pump.fun trade",
"summary": "The pages that tell you which coins are real and which wallets keep winning used to end in a link out to pump.fun, where you had to find the coin again and start over. Now every one of them carries Fork. One tap on a Coin Radar card, a Smart Money card or its drawer, the coin deep-dive on the Live Trade Feed, or a position a leader is still holding in a ghost-copy replay opens the real trade panel for that exact coin: live quote, the same safety firewall that guards every other trade surface (a coin it judges unsafe stays unbuyable), and your own browser wallet as the signer. Nothing is delegated, no key is handed over, and no balance is held. A fork on a ghost-copy row opens at the size your paper budget used, so the practice position and the real one match. Forks also travel: the Share button next to them builds a link that opens the trade panel on arrival, carrying your referral code when you are signed in, and a size a link suggests is capped and always editable before you sign.",
"tags": [
"feature"
],
"link": "/radar"
},
{
"date": "2026-09-03",
"title": "Your agent on a Mac and an iPhone: the Agent glance widget comes to Apple",
"summary": "The widget that has been on Android home screens and the Windows 11 widgets board now runs on Apple platforms too, from one WidgetKit extension shared by both. On a Mac it lives in Notification Centre in small, medium or large, fed by a small menu bar app that shows the same card and carries Open my agent, Refresh now and Unlink this Mac. On iPhone and iPad it is a home screen widget inside the three.ws app. Linking is one press on three.ws/glance: the page hands the device its own token, and if nothing opens (a Mac whose browser is signed in elsewhere) it shows a code to paste instead. The widget refreshes about every half hour without you opening anything, follows light and dark mode because it fetches the card in both, keeps the last card it downloaded when you are offline and says when that was rather than showing a spinner, and opens your agent on a tap. Every linked device is listed on three.ws/glance with a revoke button, and each token can only ever read your glance card.",
"tags": [
"feature"
],
"link": "/glance"
},
{
"date": "2026-09-03",
"title": "Daily Forge shows the day's theme again",
"summary": "Daily Forge is built around one thing: the challenge of the day. That headline was being wiped out a moment after the page loaded. The translation layer runs after the page fetches its language file, and it was overwriting the day's theme with the placeholder text baked into the HTML, so every visitor read \"Today's theme\" and \"Loading today's challenge\" forever while the starter prompt and the idea chips below quietly described a completely different subject. The theme, its emoji, its hint and the browser tab title now hold. Four more fixes landed with it: a failed generation says \"Could not reach the forge, check your connection\" instead of the browser's raw \"Failed to fetch\"; when a model finishes but the preview cannot show it, the Download button the message points at is now actually clickable instead of sitting underneath the message; every button and link has a visible focus outline for keyboard and screen-reader users, and the streak milestone popup takes focus, closes on Escape and hands focus back; and a browser without 3D support now still sees today's theme and a clear explanation instead of a permanent loading line. The community strip at the bottom also fills in properly rather than staying empty.",
"tags": [
"fix",
"improvement"
],
"link": "/daily"
},
{
"date": "2026-09-03",
"title": "Every SDK doc now points at code you can actually run",
"summary": "Eight developer docs pages described a capability without ever showing you a working project that uses it, so finding the runnable version meant guessing at folder names on GitHub. The SDK, MCP, embedding, authentication, create-an-agent, Solana, agent-payments and agent-wallets pages now each end with a Runnable example section linking straight to the project that backs them: a browser demo, a Node script that drives the free MCP server end to end, nine single-file HTML embeds you can open with no build step, the full wallet sign-in round trip, and a complete agent defined entirely as files. The machine-readable index that AI agents read to navigate the site now carries the examples location too, so an agent asked to build against three.ws can fetch working code instead of inferring it from prose. Every one of those examples is installed, run and link-checked before release, so a link on a docs page never points at something broken.",
"tags": [
"docs",
"improvement"
],
"link": "/docs/sdk"
},
{
"date": "2026-09-03",
"title": "Agent avatars on the Pay page load instead of 404ing",
"summary": "Every agent card on the x402 Pay page asked for its owner's avatar picture at a URL that has never existed on this site, so the picture always 404'd and the card fell back to a plain letter tile. The page now uses the real thumbnail the rest of three.ws uses, so an agent with a public avatar shows its face on the card, and an agent whose avatar is private still gets the letter tile without a wasted failed request.",
"tags": [
"fix"
],
"link": "/pay"
},
{
"date": "2026-09-03",
"title": "Live Trading Theater: the tape opens full and the connection light tells the truth",
"summary": "The theater got a pass over every state it can be in. The connection light in the top bar used to say Connecting for the whole visit even after the live feed had connected seconds earlier, because the page's translation pass ran last and put the placeholder wording back; it now reports the connection it actually has, and the same fix stops the sound button telling a screen reader the sound is off while it is on. The Live tape used to be a blank column until something happened on-chain: it now opens seeded with real recent activity, shows placeholder rows while that is loading, and says what it is waiting for when the feed is genuinely quiet. A slow roster response no longer dumps you straight onto the error card; the page tries again with more room and says it is still assembling. Recent-activity entries that had nowhere to link are plain text now instead of links that went nowhere, and the ones that do link to a three.ws page stay in your tab instead of opening a new one. The nameplates over each 3D trader were invisible to screen readers even though they are the only keyboard route into a trader's wallet and reputation panel; they are now reachable, and every remaining control on the page has a visible focus ring.",
"tags": [
"fix",
"improvement"
],
"link": "/theater"
},
{
"date": "2026-09-03",
"title": "Older Home Assistant houses get their agent tools back",
"summary": "If your house runs a Home Assistant release from 2025, three.ws was telling you the Model Context Protocol Server integration was not enabled even when it was, and your agent lost the curated tool list you had set up for it. Home Assistant moved that endpoint between releases: the older ones serve it over one connection style and the newer ones over another, and we were only asking for the newer one. three.ws now asks your house which one it speaks and uses that, instead of guessing from a version number. Nothing to change on your side. We also now test every release we support on a real Home Assistant before we say we support it, so the supported range is a measurement rather than a promise: 2025.10 and newer, with everything working across all of it.",
"tags": [
"fix"
],
"link": "/smart-home"
},
{
"date": "2026-09-03",
"title": "Sign Mirror tells you what went wrong, and remembers how close you got",
"summary": "The ASL practice mirror got a pass over every state it can be in. When the camera cannot start it now says which thing failed and what to do about it: permission refused, no camera on the device, the camera busy in another app, or the hand tracker itself failing to download. It used to print the raw internal error, which nobody could act on. While the tracker downloads you see that it is downloading, instead of a panel that still says the camera is off. Turning the camera off puts the panel back to telling you what to do next rather than leaving an empty box. Every letter square now carries the best score you have reached on it, so twenty six identical squares become a map of where your practice actually is, and that score is kept between sessions. Screen readers now hear the finger being named, and the live skeleton and the per finger bars are redrawn in place instead of being rebuilt on every camera frame, which takes real work off the machine doing the grading.",
"tags": [
"fix",
"improvement"
],
"link": "/sign-mirror"
},
{
"date": "2026-09-03",
"title": "Your house, in 3D, live",
"summary": "Open a connected home and you get a real 3D model of it, driven by your own Home Assistant. Every room is lit the way the room is actually lit, in the colour its bulbs are actually set to, and dark when they are all off. The outline around a room turns amber the moment anything in it is unlocked or open, so you can see across the whole house at a glance. Your agent stands in it. Click anything to see what it is and act on it, and anything that would open your home asks first, right on the thing it would move. If the connection to your house drops the house stays on screen, greyed, with how long ago it was last seen: it never vanishes. There is a full 2D view of the same house that reads and controls everything without a graphics card, and a browser that cannot run 3D is sent there automatically and told why.",
"tags": [
"feature"
],
"link": "/smart-home"
},
{
"date": "2026-09-03",
"title": "See what your connected homes use, before a limit ever stops you",
"summary": "Your smart home plan now has a page. It shows every allowance whether or not you are near it: how many homes you have connected, how many people are in each household, how many live screens, how many voice minutes and agent turns you have used this month, and the exact date each one resets. Two things a plan can never do are written on the page itself, because they matter more than any number on it. Locking a door, closing a garage or a water valve and arming an alarm are never refused by a limit, on any plan, in any state, including on a home your plan has paused. And confirmation prompts, your action log and every safety property of household roles are on every plan, free included: safety is not an upgrade. If a plan change ever leaves you with more homes than it covers, nothing is disconnected and nothing is deleted. The extra homes are paused, with their access tokens and their whole history untouched, and you pick which ones are live from the same page.",
"tags": [
"feature",
"improvement"
],
"link": "/smart-home/plan"
},
{
"date": "2026-09-03",
"title": "three.ws Drive: your agent, in the car",
"summary": "There is a new page at /drive built for a car screen instead of a desk. Your agent is on it full size, you hold one big button and talk, and it answers out loud in its own voice with its face moving. It shows you one line of what it said rather than a wall of chat, because you are driving. It starts dark, because a bright panel at night is glare, and it will switch to a day palette on one press or one sentence. The keyboard stays out of the way while you are moving: if your browser can tell how fast you are going, text entry locks itself the moment you pull away and comes back when you have properly stopped. A few things you say never leave the car at all, so \"repeat that\", \"stop talking\", \"louder\" and \"night mode\" happen instantly, while \"stop at the next charger\" still goes to your agent like any other question. It works today on a head unit browser, on a phone in a cradle, and on Android Automotive. The Apple CarPlay version is built and waiting on Apple's approval for voice assistants in the car. One thing this fixed for everyone: an avatar used to answer in silence if you were not signed in, and now it always has a voice.",
"tags": [
"feature",
"improvement"
],
"link": "/drive"
},
{
"date": "2026-09-03",
"title": "The tour atlas counts its stops correctly again",
"summary": "The atlas at /tour/atlas is the map of the guided tour: one card per stop, numbered in the order you would walk them. Two hundred and three of those cards were still carrying numbers from an older version of the tour, so the grid showed duplicate stop numbers, skipped others, and put cards out of order, under a heading that claimed one more stop than the grid actually held. Every card is now numbered by its real position in the tour and the count above the grid matches what is under it. Nothing about the stops themselves changed: same pages, same screenshots, same highlights.",
"tags": [
"fix"
],
"link": "/tour/atlas"
},
{
"date": "2026-09-03",
"title": "See, export and delete everything a connected home stores",
"summary": "Connecting a house to three.ws hands over a key to a physical building, so the answer to \"what do you keep, and for how long\" is now a page you can read and an endpoint you can call. Every piece of data a connected home creates is listed with why it exists and what removes it, including the parts we deliberately never store: the names of your rooms and devices, their states, and any audio from voice control. There is no record of which lights were on, or when. The log of what the agent did in your home is yours: it keeps 90 days by default, you can shorten it to a single day and the old entries go immediately, and an operator who has to keep records for longer can, with a written reason. One endpoint shows you what is stored, downloads all of it as JSON, and deletes one home or every trace of all of them.",
"tags": [
"feature",
"security",
"docs"
],
"link": "/docs/home-privacy"
},
{
"date": "2026-09-03",
"title": "The status page can tell one dark house from an outage",
"summary": "Connected homes now report on the public status page at /status, and the hard part was making that number honest. If your Home Assistant is switched off, your token expired or your internet dropped, that is your house: it shows up in your own home view, not as a three.ws outage and not as an alert to anyone. It takes at least ten homes failing at once before the platform will call it a problem on our side, because with three homes connected one holiday cottage losing power would look like a third of everything being broken. One thing is treated as an emergency at a count of one: an action that can open a building (a lock, a garage door, an alarm) recorded as having run without someone saying yes. That has no acceptable rate.",
"tags": [
"infra",
"improvement"
],
"link": "/status"
},
{
"date": "2026-09-03",
"title": "Connect your house to your agent, and see every way it can go wrong",
"summary": "The /smart-home page now takes you from \"I have a Home Assistant somewhere\" to a connected house, and it is honest at every step. An address that only exists on your home network is refused the moment you type it, with the two real ways round it, instead of spinning for fifteen seconds. A rejected token says so and puts the cursor back in the token box. A house that stops answering keeps its rooms on screen, marked stale with how long ago it last replied, rather than blanking out. Your access token is sent once, encrypted at rest, and never comes back to the page: it is not in your browser storage, not in a URL, and not in any response. Disconnecting says plainly that the stored token is erased and reminds you to delete it in Home Assistant too.",
"tags": [
"feature"
],
"link": "/smart-home"
},
{
"date": "2026-09-03",
"title": "A stranger cannot make your agent open your front door",
"summary": "The home lane now has a written threat model and eleven security checks that run with every build, and the standard they hold is a lock, not a database row. The big fix: the address you give for your house is now resolved and pinned on our side before anything dials it, so a web address that secretly points back into our own network is refused instead of connected to, and it is re-checked every time your house reconnects rather than trusted from the day you added it. A device or integration in your house cannot talk your agent into anything either: we wrote an unlock instruction into a real room name, a real device name and a real scene name, ran a real agent against that house, and the front door stayed locked. Your privacy page is now rate limited like everything else, so nobody can hammer the export or the delete.",
"tags": [
"security",
"improvement"
],
"link": "/smart-home"
},
{
"date": "2026-09-03",
"title": "Your house reads in your language, in your units, at a size a finger can hit",
"summary": "Three fixes to the smart home page, all of them things you would have noticed. It now speaks all 87 languages: the page was written to be translated but never loaded the piece that does the translating, so everyone got English, and Arabic and Hebrew never flipped to right to left. Temperatures now come from your house rather than from an assumption. If your Home Assistant is set to Fahrenheit, every room reads in Fahrenheit, and a room at 70 no longer shows as blazing hot because the comfort range was written in Celsius. A house that mixes a Celsius sensor with a Fahrenheit thermostat is converted before it is averaged, so the number you see belongs to a real scale. The small print at the top of the page also got lighter so it clears the contrast standard, in both the light and dark themes, and the Show button next to the token field is now big enough to tap without missing.",
"tags": [
"fix",
"improvement"
],
"link": "/smart-home"
},
{
"date": "2026-09-03",
"title": "Any assistant can now run your house, and it still will not open your door",
"summary": "There is a new way to connect your home that needs no three.ws account and no remote web address at all: install @three-ws/home-mcp in Claude, Cursor, or anything else that speaks Model Context Protocol, give it your Home Assistant address and a token, and it reads your rooms, runs the scenes you already built, and calls services. Because it runs on your own machine, your ordinary home network address works and nothing about your house leaves the building. The safety rule is the same one the rest of the platform uses, applied honestly to this setup: locking up, closing the garage and arming the alarm run straight away and never ask. Unlocking, opening a door and disarming are refused outright, every time, because saying yes to opening a building takes a person and this kind of connection has no person in it to ask. If you genuinely want one specific door handled without asking, you name that exact door yourself when you start it, by hand, and nothing your assistant can do will widen that to any other door. We proved it against a real Home Assistant rather than a stand-in: the front door was told to unlock seven different ways, including with a fake confirmation smuggled into the request, and it stayed locked every time. Alongside it there is now a full walkthrough from nothing to a working agent in a real house, covering what to do when it says your home is unreachable, and what a guest in your household can and cannot do.",
"tags": [
"feature",
"sdk",
"docs"
],
"link": "/smart-home"
},
{
"date": "2026-09-03",
"title": "Your home is a household now, and a house sitter still cannot open your front door",
"summary": "A house has more than one person in it, and until now the only way to let a partner, a house sitter or a colleague use a connected home was to hand over your password. Every home has a household. On its card under “Who can reach this home” you can invite somebody by email and choose exactly what they get. There are five roles and each one says plainly what it can and cannot do at the moment you pick it. Someone who lives there can control everything and approve unlocking a door. A guest can turn the lights on and can never approve unlocking anything, ever, no matter what your agent asks them: that refusal is enforced on our servers and there is no version of the app that could offer them the button. A guest or a viewer can also be given just the kitchen, or just three specific devices, and the rooms you did not give them are removed from what they receive rather than hidden on screen, so their app never learns those rooms exist. Invitations work once, expire after a week, and can be withdrawn before anyone uses them. Removing somebody takes back every standing allowance they ever approved in the same instant, so a door can never keep opening on the say-so of a person who has left. And every action in your home log is attributed to the person who took it, so “who did what” has a real answer.",
"tags": [
"feature",
"security"
],
"link": "/smart-home"
},
{
"date": "2026-09-03",
"title": "We tried making your agent a Matter device, measured it, and decided against it for now",
"summary": "Matter is the standard that lets smart-home gear from different brands work together, and the plan had always been to make a three.ws agent show up as a Matter device so that any smart-home controller could see it, drop it in a room, and use it in an automation. We built it and pointed it at a real Home Assistant to find out whether the idea holds up. It does, technically: the agent paired in under a second over an ordinary network with no Bluetooth and no special radio, Home Assistant listed it as a proper device with a name, a model and a serial number, put it in the living room, and messages went both ways in well under a second. It also held the line that matters most: an automation that asked the agent to unlock a front door was refused and the door stayed locked, because an automation is not a person saying yes, and only a real confirmation from a person opened it. We are still not shipping it, and the honest reason is that it would be a narrower version of something you can already have. The three.ws Home Assistant integration already puts your agent inside your house with the whole conversation available to it, where Matter would offer a switch and a sensor. On top of that, appearing in the big-name smart-home apps needs a certified manufacturer identity we do not hold. The full write-up, every measurement, and the exact conditions that would change our answer are now in the smart-home documentation, so nobody has to guess or start over.",
"tags": [
"docs"
],
"link": "/docs/smart-home"
},
{
"date": "2026-09-03",
"title": "Talk to your house without finding a phone, and a background \"yeah\" can never open your front door",
"summary": "Hands-free is here for connected homes. Turn it on at /voice/home, say the wake word, and ask for what you want: \"turn the kitchen light off\" works while you are carrying groceries. The wake word is matched by a model that runs inside your own browser tab, so that audio never leaves your device; only the sentence you say afterwards is sent, and only to be turned into text. While the microphone is live there is an indicator on the page that cannot be switched off, and Mute genuinely stops the microphone at the device rather than hiding a dot. Talk over the agent and it stops mid-sentence and listens, in about a tenth of a second, and it will not wake itself up when it happens to say its own wake word out loud. The part we spent the most care on is the door. Anything that opens your house is spoken back in full first, the exact device appears on screen while you decide, and only the word \"confirm\" goes through. A \"yeah\" or a \"sure\" overheard from a conversation on the other side of the room does not, and on a device with no screen the agent refuses to take an unlock by voice at all and offers your phone instead. Four wake phrases to choose from, twelve designed states including an honest one for deployments where speech recognition is not available, and every stage of the round trip timed live on screen so you can see where the seconds go.",
"tags": [
"feature",
"security"
],
"link": "/voice/home"
},
{
"date": "2026-09-03",
"title": "Connect a home that only exists on your own network",
"summary": "Most Home Assistant installs sit on a home network with nothing exposed to the internet, so until now three.ws had no way to reach them. Now the house connects to us instead. Install one integration through HACS, paste a ten-minute pairing code, and your agent can see your rooms and run your scenes with nothing listening on your network, no port forwarded, and no firewall rule changed. The part worth saying plainly: three.ws never receives a Home Assistant token for a home connected this way. The integration signs in to Home Assistant on your own machine with a credential it makes for itself and never sends anywhere, so there is nothing stored here that would open your house. What we can do through that connection is a fixed list, enforced twice, once by us and once by the integration on your machine: reads, your room layout, and device controls. It can never run code on the machine Home Assistant lives on, restart or reconfigure your instance, or reach anything else on your network. Unlocking, opening and disarming still stop and ask a person first, every time, exactly as they do for a home connected by address. Disconnecting works from either end and takes effect immediately. There is an honest threat model in the docs, including what a compromise of three.ws would mean for a connected house; read it before you decide.",
"tags": [
"feature",
"security"
],
"link": "/docs/home-relay"
},
{
"date": "2026-09-03",
"title": "Your home tells you when it stops answering, and whether the problem is ours",
"summary": "When a connected house goes quiet, the manage page now says what happened and what to do about it, without you having to ask anyone. It names the real reason the handshake failed, separates a Home Assistant that rejected our token (make a new one) from a Home Assistant we cannot reach at all (check the instance), and keeps the last room state on screen instead of emptying the page. The part that matters most: it tells you whose problem it is. If enough other homes are failing at the same moment, it says so plainly and tells you there is nothing for you to do, so nobody spends an evening rebooting a router that was never broken. A house that already looks unhealthy loads the explanation on its own; a house that looks fine keeps its panel closed until you open it, so nothing extra is fetched for homes that are working.",
"tags": [
"feature",
"improvement"
],
"link": "/smart-home"
},
{
"date": "2026-09-03",
"title": "One unreachable home can no longer take the others down with it",
"summary": "Homes connect over a live socket, and a house whose broadband keeps dropping in and out was able to crash the server process that held it, which meant every other connected home on that server went quiet at the same time. The cause was in the open-source Home Assistant client we build on: it reconnects a subscription after a drop without handling the case where the connection dies again mid-reconnect. That is now handled on our side, the failure is reported instead of being lost, and the connection recovers on its own. We have written the problem up for the upstream project too. Found by deliberately flapping a real house on and off every five seconds for two minutes.",
"tags": [
"fix",
"infra"
]
},
{
"date": "2026-09-03",
"title": "Coin Radar tells you when it is your filter, not the market, that is empty",
"summary": "Search the radar for a ticker it has not seen and the board used to say \"Radar is clear, waiting for the next launch\", which blamed a quiet market for a filter you set and gave you no way back. Most of Radar's filters run on the server, so an over-tight one came back looking exactly like a quiet window. Now the board reads your filters instead of the response: it says no coins match, lists every filter you have on as a chip you can click to drop just that one, and keeps a reset-everything button next to it. The 24-hour pulse strip across the top also got an honest state of its own. It is a separate request from the live feed, and when it failed it used to shimmer as if still loading forever; now it says the aggregate did not load, offers a retry, and leaves the feed below running. On a 320px phone every card was losing about 19 pixels off its right edge to a layout overflow that the mobile scroll guard was quietly clipping; cards now fit the screen they are on.",
"tags": [
"fix",
"improvement"
],
"link": "/radar"
},
{
"date": "2026-09-03",
"title": "The splat viewer lets go of the graphics card between scenes",
"summary": "Every scene the viewer opened built a fresh WebGL renderer, and tearing the old one down freed its memory but never released the graphics context itself. The library we use for splat rendering ends its own cleanup by removing an element from the page that was never there, which throws and skips the last of its teardown, so the contexts stacked up across a session of opening one capture after another. The viewer now releases each context by hand on every swap and when you leave the page. Recentering the camera no longer rebuilds the whole scene either: it moves the camera on the running viewer, so the stage snaps back instantly instead of flashing through the loading screen. If the browser does take the graphics context back, after a GPU reset or under memory pressure, the stage now says so and offers a way to reload rather than sitting on a frozen frame. Two smaller things while we were in there: a scene you opened by URL can be downloaded from the viewer, the same as an upload or a sample, and the drop zone now has a real button for keyboard users with a labelled URL field for screen readers.",
"tags": [
"fix",
"improvement"
],
"link": "/splat"
},
{
"date": "2026-09-03",
"title": "The rest of the smart home now speaks your language, and the wall display stays awake",
"summary": "The connect page learned all 87 languages last week; the invitation, plan and voice satellite pages had the same gap and now speak them too, along with the live 3D house. Setting a page to Arabic or Hebrew also actually mirrors it now: the reveal button, the coloured bars beside notices, and the room and device columns in the live house used to stay stuck on the left. If you have asked your device to reduce motion, the house stops animating between states while staying completely live, so lights and locks still update the instant they change, they just arrive rather than glide. The live house also keeps the screen awake now, which is what you want on a kitchen tablet on a shelf, and gives that up the moment you switch away, which is what you want on a phone in your pocket. On a phone with a notch the house fills the screen without hiding its own controls behind the cutout.",
"tags": [
"improvement",
"fix"
],
"link": "/smart-home"
},
{
"date": "2026-09-03",
"title": "The signal marketplace tells you how to get on it, and reads correctly to a screen reader",
"summary": "No verified trader has published a paid feed yet, so the marketplace's empty state is what everyone actually sees, and it only pointed at the leaderboard. It now also points at the place you publish from: the Signals tab of your agent wallet. The rank and network filters used to announce themselves as tabs, which promises a screen-reader user a panel to arrow between that this page does not have; they are now announced as the pressed toggles they really are, wired to the board they re-rank. The Retry button on a failed load and both empty-state buttons had no hover or keyboard-focus styling at all, so a keyboard user could not see where they were; all three now light up on hover and show a focus ring. Landing on the page with a stale or misspelled sort in the address bar rewrites it to the board you are actually looking at, so a shared link always opens the view it names. And a feed link that no longer resolves now gets a real page heading instead of a floating error message.",
"tags": [
"improvement",
"fix"
],
"link": "/signals"
},
{
"date": "2026-09-03",
"title": "three.ws Drive on Android Auto, and your agent can reach your house from the car",
"summary": "The car surface now has an Android Auto app to go with the CarPlay one. Open three.ws on your car screen and you get four big controls, and your agent listens straight away. The car screen is Android's own layout, which is the rule for every app that runs while you drive, so the conversation runs on your phone and comes out of the car speakers. There is a notification while it is running, on purpose: your agent should never be listening invisibly, and one tap stops it. The bigger part is what it can now do. If you have connected your home to three.ws, your agent already reaches it, so \"turn the porch light on\" works while you are driving home. Anything that opens or unlocks something is different and stays different: your agent will not touch a lock, a garage door or an alarm on its own, and it will not even ask you to approve one while the car is moving. It tells you it is waiting and offers it for the moment you park. On the car screen, where you cannot see what you would be approving, it asks you to do it on your phone instead.",
"tags": [
"feature",
"improvement"
],
"link": "/drive"
},
{
"date": "2026-09-03",
"title": "Restyle Studio: upload works from the keyboard, and reverting a version really rewinds the source",
"summary": "Two things on the Restyle Studio page were quietly broken. \"Upload GLB\" was a styled label wrapped around a hidden file input, so it could not be reached by keyboard or by a screen reader at all: the only way to bring your own model in was a mouse click. It is now a real button, in the tab order, with the same look. Separately, clicking an older entry in the Versions strip reloaded that model on the stage but left every following action pointed at the newest checkpoint, so an AI restyle or a Save right after a revert quietly re-skinned the wrong model and filed the result under the version you had gone back to. Reverting now re-anchors the source, so restyle, Save version, and \"Save all as versions\" all branch from the version you are actually looking at. The rest of the sweep is smaller but adds up: the metalness, roughness and emissive sliders and the model-URL and instruction fields now have real labels; every async result note is announced instead of silently changing; a successful save reads as success rather than as another grey hint; clicking \"Load URL\" with an empty box tells you what it wants instead of doing nothing; dropping a file on the stage highlights the target and says so when the file is not a model; picking the same file twice in a row works; and \"Load the sample model\" clears the failing ?url= so a reload does not land straight back in the error state.",
"tags": [
"fix",
"improvement"
],
"link": "/restyle"
},
{
"date": "2026-09-03",
"title": "The sniper scoreboard reads correctly in the light theme, and tells you what it is doing while it loads",
"summary": "The strategy board was building a second copy of its own stylesheet in JavaScript and appending it to the page, so those rules won every tie and quietly reverted the real one. In the light theme that meant profit figures and the LLM badges rendered in a near-white green and lavender on white paper, close to invisible; they now use the same colours as every other crypto page and read properly in both themes. Loading is no longer a blank space under the headline: the board paints a skeleton in the exact shape of the table while it reads each agent's live wallet balance off the chain. If the feed cannot be reached you now get a plain sentence about what went wrong plus a button to try again, instead of a raw browser error and a silent 30-second wait, and a refresh that fails on top of a board that already has numbers keeps the numbers, says they are the last good read, and offers the same retry. Switching the time window while a slower request is still in flight can no longer paint the wrong window's figures. The judgment ledger says so when there are no LLM verdicts in the window rather than disappearing. And the table stopped showing a hand cursor over rows and headers that were never clickable or sortable.",
"tags": [
"fix",
"improvement"
],
"link": "/sniper/experiments"
},
{
"date": "2026-09-03",
"title": "Tour Atlas counts its own stops honestly, and says why a stop has no picture",
"summary": "The atlas advertised 185 stops in its page title, its search prompt, its social cards and its listing everywhere on the site. It has had 263 for a while, and nothing was checking, so the number quietly went stale. It now says 263, and the atlas guard fails the build if that number ever stops matching the manifest again. Eighteen stops carry no screenshot because they are live 3D or live-market pages that never stop painting long enough to be photographed; their cards used to claim the page 'did not respond', which was untrue of all of them, and now they say what actually happened and invite you to open the page live. Opening one of those stops used to show an empty black frame above the write-up, and now shows the same explanation. If the atlas itself cannot load you get a plain sentence about what went wrong and a Try again button, instead of a raw browser error or an instruction to run a developer command. And a /tour/atlas#stop-name link now opens that stop even when you are already on the page or have a filter narrowing the grid.",
"tags": [
"fix",
"improvement"
],
"link": "/tour/atlas"
},
{
"date": "2026-09-03",
"title": "Endpoint Shopper takes payment on the page, so a paid run actually runs",
"summary": "Paying for an Endpoint Shopper run used to send you to the generic payment page, which arrived with no price and no wallet options because it could not read the challenge this route sends. Even if it had, it would have replayed the run as an empty request: your task and budget were left behind on the previous page, so a completed payment unlocked nothing. Payment now happens on /shopper itself. The card quotes the real price, lists the Base and Solana wallets the endpoint accepts, reports connecting, signing and settling as they happen, and replays the exact task and budget you typed, so the answer and its cost breakdown appear right where you were, with a link to the settlement transaction. The example task chips are now real buttons you can reach with the keyboard, the task box and budget slider carry proper labels, and the Marketplace and Fact Checker links no longer vanish from the header on a phone.",
"tags": [
"fix",
"improvement"
],
"link": "/shopper"
},
{
"date": "2026-09-03",
"title": "Scene Studio stops hiding its own menu bar, and tells a new scene what to do next",
"summary": "The 3D scene editor was pinning itself 13 pixels too high, so the site header covered the top of the File, Edit, Add, View, Render and Help row and sliced the tops off the letters at every window size; the autosave checkbox was hidden entirely. It now sits below the header. The Import, Export and Share buttons were parked on top of the Scene, Project and Settings tabs, which made those tabs unreadable and unclickable on any desktop window; they now dock beside the sidebar and follow it when you drag it wider or narrower. At phone width the menu bar used to spill Help onto a second row that the layout could not show, and it now keeps all six on one line. A brand-new scene is no longer a silent black grid: a card explains the three ways to start and disappears the moment you add anything. Importing a model by URL now happens in a proper dialog that checks the link and explains what is wrong inline, instead of a bare browser prompt, and a failed import, export or share now says what went wrong and what to try next in the same notification style as the rest of the site rather than a browser alert box. Sharing a scene labels it with the scene's own name instead of the page's title. The translate, rotate and scale buttons are icon-only and had no names at all for screen readers; they now announce both their name and which one is active. The page also gained the heading every other page on the site has.",
"tags": [
"fix",
"improvement"
],
"link": "/scene"
},
{
"date": "2026-09-03",
"title": "Your Home Assistant voice assistant can have a face now",
"summary": "If you already run Home Assistant with a voice pipeline, it hears you, understands you, turns on the lights and answers, and the whole visual side of that is an LED ring. You can now point that same pipeline at a three.ws agent and get a face on any screen in the house: it reacts while you speak, the transcript fills in as your own speech recognition produces it, and it lip-syncs the answer to the audio your own text-to-speech generated. Home Assistant keeps every part of the pipeline, running on your own hardware exactly as it does today. Setup is a pairing code, one Docker command, and adding it in Home Assistant as a Wyoming Protocol integration. Nothing needs to be exposed to the internet: Home Assistant connects out to the satellite on your own network, the satellite connects out to us, and there is no port to forward. The rule the whole thing is built around is that closing the page never stops your voice assistant working, so the run finishes, the lights still change, and the only thing you lose is the face.",
"tags": [
"feature"
],
"link": "/smart-home/satellite"
},
{
"date": "2026-09-03",
"title": "Search keeps its filters honest, and says so when it cannot answer",
"summary": "Sharing a filtered search link now works the way it looks: open /search?type=agent and the Agents filter is the one highlighted, instead of the row claiming All while the results were already narrowed. After a search runs, the types that turned up nothing for that query fade back, so you can see at a glance where the matches actually are. When the search index itself is unreachable the page now says it is warming up and offers a retry plus the browse galleries, rather than telling you nothing matched, which was never true. A failed search says what went wrong and gives you the same retry and browse links. Pressing Enter runs the query immediately without waiting for the pause-and-search delay, and Escape clears it. Result cards lost their duplicate untitled thumbnail link, creators without a public profile read as plain names instead of links that went nowhere, the search box and the remix instruction field are properly labelled for screen readers, and every focusable control on the page now shows a visible focus ring.",
"tags": [
"fix",
"improvement"
],
"link": "/search"
},
{
"date": "2026-09-03",
"title": "The payments feed now tells you which volume is ours and which is real demand",
"summary": "three.ws runs a closed loop where platform wallets pay platform endpoints on a schedule, to keep the payment rails warm and provably live. That traffic is real money on a real chain, and it is the overwhelming majority of settled volume, so anyone reading the raw total was reading our own money back to us. The x402 volume report on the paid analytics endpoint now carries a revenue split alongside the totals: how many calls and how much USDC came from outside wallets, how much is the internal loop, and how much is test traffic that was never money at all. A new command, npm run readout:revenue, prints the same breakdown for any window and refuses to print an external figure when it cannot tell our wallets apart from anyone else's, rather than quietly reporting a flattering one. One classifier backs both, so the API and the command can never disagree.",
"tags": [
"improvement",
"infra"
],
"link": "/docs/autonomous-x402"
},
{
"date": "2026-09-03",
"title": "Avatar posters and clips survive a CDN outage instead of coming back blank",
"summary": "Some of the images we render of an avatar on the server are drawn by a headless browser that loads three.js from a public CDN: the images the render endpoint and the MCP render tools return, rendered clips, and the preview cards for models our fast in-process renderer cannot decode on its own. When that one CDN was slow or unreachable, the page's module load simply hung, the render timed out, and the image came back empty with nothing to explain why. A failover had been written for exactly this and was never connected to the renderers. It is connected now: before each render we check that the CDN is answering and switch to a second one if it is not, remembering the answer for ten minutes so a burst of renders pays for a single check. The three renderers also share one pinned version of three.js instead of keeping three copies that could drift apart, and a test now fails the build if a renderer ever goes back to hardcoding a single CDN.",
"tags": [
"fix",
"infra"
]
},
{
"date": "2026-09-03",
"title": "Traders can now price the latency of their own calls",
"summary": "Copy trading has a new option for the trader being copied: alpha-drip. A trader can set how long each $THREE holder tier waits before their copy intent is revealed, so holding more $THREE means seeing their calls sooner. It is off unless a trader turns it on, and the rules are strict on purpose: a higher tier can never wait longer than a lower one, no delay can exceed fifteen minutes, and the delay only affects when a copier is SHOWN the trade. The trade itself is still recorded the instant it happens and still appears in the trader's public track record, so nothing is ever hidden. Copiers see exactly which seat their balance buys before they subscribe, and a held intent shows a live countdown that unlocks in place. Traders can also ask for a suggested ladder tuned to how fast their own edge actually decays, measured from their real closed trades, and get warned when the slowest tier would be waiting longer than that edge lasts.",
"tags": [
"feature"
],
"link": "/docs/alpha-drip"
},
{
"date": "2026-09-03",
"title": "Every 3D model now says whether a physics engine could actually use it",
"summary": "A 3D model can look flawless on screen and still be unusable the moment you drop it into a physics engine: an open shell sinks through the floor, an inside-out surface spins like it is hollow, and a model fitted to a unit box is a metre tall when you asked for a teapot. Nothing in a GLB file has ever said so, so everyone building robotics, game or simulation pipelines has been finding out one broken asset at a time. Now every model page carries a verdict beside its geometry stats, and the GLB viewer will grade any public model you point it at. Four answers: ready to simulate as-is, sound but needs its real size set, needs its surface closed first, or nothing there to simulate. Open the badge and you get every number behind the verdict, including the exact volume, centre of mass and inertia tensor, copyable as JSON straight into your own pipeline. It is free, needs no account, and works on any public model, not just ours. Agents can ask the same question through the free grade_sim_readiness tool, so a shopping agent can check ten candidates for nothing and only pay for the one that works. Models forged here are graded the moment they are generated, so the answer is already waiting when the page opens, and when a model is credentialed the grade is signed along with it, which makes the physics claim as hard to forge as the authorship claim.",
"tags": [
"feature",
"sdk"
],
"link": "/docs/sim-readiness"
},
{
"date": "2026-09-03",
"title": "Avatar Studio can now dress, rig and walk your avatar, not just paint it",
"summary": "Avatar Studio picked up everything the older avatar editor could do and it could not. A Wardrobe tab puts real catalog garments on your avatar, bound to its own skeleton so they move with every animation, on top of the show/hide and recolour controls for the clothes the body already came with. A Walk tab lets you drive the avatar around the stage with WASD, which is where a garment that does not follow the hips actually becomes visible, and hands the unsaved look straight to the full Walk page. Editing a saved avatar also gets a Rig tab that turns a static mesh into an animation-ready one, and a Play as this button that saves and drops you into a world wearing it. Saving a dressed avatar keeps the clothes: the outfit is stored as data and rebuilt on the pristine base, so nothing gets baked on twice and the base model you started from is always recoverable.",
"tags": [
"feature",
"improvement"
],
"link": "/avatar-studio"
},
{
"date": "2026-09-03",
"title": "Copy a working embed snippet with no account, and embed an agent's chat page",
"summary": "The embed editor moved into the Widget Studio, and it got the two things it was missing. You no longer need an account: the walking-avatar and the new Agent Chat widget carry their whole setup in the embed URL, so you can configure one, copy the snippet and put it on your site without signing in (signing in still gets you a shareable link and the ability to change a live embed later). Agent Chat embeds an agent you already published, so visitors talk to it on your page. You can also point an embed at a .glb or .vrm you host yourself, and the modal now tells you exactly where the snippet goes in HTML, React, WordPress, Webflow and Shopify. Every control writes itself back into the address bar, so a configured studio link is something you can send to someone. Old /embed links keep working and land on the same configuration.",
"tags": [
"feature",
"improvement"
],
"link": "/studio"
},
{
"date": "2026-09-03",
"title": "The Showcase directory is back after a stray change emptied it",
"summary": "Showcase lists every ERC-8004 agent that carries a 3D body, discovered across 22 chains and indexed from the canonical registries, and any card opens that agent in the live viewer. An unrelated change had emptied the file behind it, so for a while the page loaded and then showed nothing at all. It is restored and browsing again. While it was open we also gave the page a single real heading instead of two competing ones, which matters to anyone reading the site with a screen reader, and rewrote its description, which had been advertising a drag-and-drop model viewer rather than the agent directory it actually is.",
"tags": [
"fix"
],
"link": "/showcase"
},
{
"date": "2026-09-03",
"title": "Chain logos on the fees tables render instead of breaking",
"summary": "Every chain row on the fees and revenue tables showed a broken image. The upstream data source hands out chain logo links that are missing a piece of their address, so each one failed to load. We repair the address before the data reaches the page, and any icon that still fails for its own reasons now quietly becomes the same neutral disc a protocol without a logo already gets, rather than a broken-image mark. The same repair covers the DEX volume, protocol and chain pages that draw from the same feed.",
"tags": [
"fix"
],
"link": "/fees"
},
{
"date": "2026-09-03",
"title": "Soft shadows are soft again in Pose Studio, Mirror and Diorama",
"summary": "These three 3D surfaces were asking for a soft-shadow mode that the renderer stopped supporting and silently swapped for hard-edged shadows, complaining in the console every time a scene started. They now use the supported soft-shadow type, so contact shadows under a figure read as shadows rather than hard cutouts.",
"tags": [
"fix",
"improvement"
],
"link": "/pose"
},
{
"date": "2026-09-03",
"title": "Coin art and 3D models load again, wherever the page is running",
"summary": "Token logos on the launches feed and every coin card were painted straight from a public IPFS gateway that now rate-limits us, so a feed of launches rendered a wall of blank tiles and filled the console with blocked requests. Art now comes through the site's own image service, which retries across gateways, resizes to the size actually shown, and falls back to a generated placeholder instead of an error. Creation gallery models load the same way, through the site's model proxy, so a 3D preview also works inside an embed on somebody else's site.",
"tags": [
"fix",
"improvement"
],
"link": "/launches"
},
{
"date": "2026-09-02",
"title": "Keyboard users can move past the Create page hero card",
"summary": "The recommended card on /create wrapped its rotating 3D preview inside the button itself, so the preview took a tab stop of its own and screen readers announced a control inside a control. The preview is now decorative, as it was always meant to be: the whole card is one target, and the model still wakes up and rotates when you hover or tab to it.",
"tags": [
"fix"
],
"link": "/create"
},
{
"date": "2026-09-02",
"title": "Groundwork for agents that can run your house",
"summary": "A three.ws agent already has a body, a voice and a memory, all of it stuck inside a browser tab. This is the first piece of letting it reach past the screen: a new developer package, @three-ws/home-bridge, that connects an agent to a real Home Assistant, the open-source home platform that already runs in millions of houses and already speaks to Zigbee, Z-Wave, Matter, Thread and about fifteen hundred other things. We wrote no device code at all, on purpose. What the package adds is the part nobody had built: live state from the house shaped into rooms, so a 3D scene can dim when your real lights dim; \"good night\" and \"I'm leaving\" resolved to the scenes you already made yourself, in Home Assistant, rather than to something we invented for you; and a safety gate that reads the way our on-chain one does. Turning a light off is free. Unlocking a door stops and asks you, every time, until you say otherwise for that one door. Locking up never asks, because moving your house toward safety should not need permission. That last part turned out to matter more than expected: Home Assistant's own \"turn it off\" command performs an unlock when you point it at a lock, so an assistant told to turn something off can open your front door with nothing in the command's name to warn anyone. Our gate catches it. All of it was built and checked against a real Home Assistant running here, not against stand-ins, and the write-up of what else is out there and where this goes next (voice in every room, the agent standing in a live model of your home) is published alongside it.",
"tags": [
"sdk",
"docs",
"security"
],
"link": "/docs/smart-home"
},
{
"date": "2026-09-02",
"title": "A polish pass over the coin worlds, from the friends drawer to every button on a phone",
"summary": "A sweep through the whole /play journey, from the lobby to the world and back out, fixing what a measured run found rather than what looked wrong. The friends drawer is the big one: it used to do nothing at all for the first moment after you tapped it, because the panel's code was still being fetched, and if you gave up and pressed Escape in that gap the drawer ignored you and slid open anyway a few seconds later. It now answers the tap on the spot with a roster loading in place, closes when you tell it to, and if the panel genuinely cannot be fetched it says so and gives you a Try again button instead of sitting there empty. On a phone, every control in the world has grown to a proper thumb-sized target: the heads-up bar, the leave and voice and zen buttons, the reaction pills, the shop chips and every close cross, plus the sort buttons and footer links back in the lobby. Several of these were half the size a thumb needs, which is why they were easy to miss mid-session. Two things that used to jump are now still: the lobby no longer shoves the community grid down when your saved avatars arrive, and the emote tray no longer flashes as an empty pill before it fills. And pressing Escape on the welcome card no longer throws an error behind the scenes. Nothing about the desktop layout changed; the roomier targets apply only where you are actually using a finger.",
"tags": [
"improvement",
"fix"
],
"link": "/play"
},
{
"date": "2026-09-02",
"title": "Photo mode in the coin worlds now actually hands you the photo",
"summary": "Press P in a coin world, or the camera button on the heads-up display, and you get a picture of the world on a share card: the scene itself, the three.ws mark, the coin you are standing in, and, during a live event, that event's name stamped on it. The card was there but the shutter went nowhere. Every single press answered \"Couldn't photograph the world just now\" and showed nothing, because of a fault in the code that builds the preview, so nobody has ever actually been handed one of these. Pressing it now opens the card, with Download and Copy image next to it. The picture is taken as a separate render rather than grabbed off the screen, so no panel, name tag or chat line is ever in the frame, and it works the same in zen mode, which is when most people will want it. It costs nothing when you are not using it: none of photo mode is loaded until the first press. Two smaller things fixed alongside: taking a second photo while the first card is open now replaces it cleanly instead of briefly stacking two sheets, where a click landing in the wrong moment could dismiss the shot you just took; and the line promising that P takes another one is true now, and visible, having previously rendered as nothing at all. The camera button also stays lit while the picture is being taken, instead of going quiet: taking the photo is real work, and on a slow phone it can run for a few seconds, which previously looked like a button that did nothing. Verified end to end in both Chrome and Safari, on a laptop frame and a phone frame, down to the pixels in the saved file, so the black rectangle that a mishandled 3D screenshot usually produces cannot come back unnoticed.",
"tags": [
"fix",
"feature"
],
"link": "/play"
},
{
"date": "2026-09-02",
"title": "A generated model that came back as a flat slab no longer passes as the best possible result",
"summary": "Text-to-3D occasionally lost the reference image partway through and returned a pancake: a flat, textured relief the width of the whole frame instead of the object you asked for. Nothing caught it. The automatic quality check counted triangles, textures and overall size, and a slab has plenty of all three, so it scored higher than most good models and skipped the visual review that would have spotted it. The check now also measures whether a model has any depth at all. A model with none is sent for visual review instead of being trusted, which is what triggers a regeneration. Genuinely flat things (a plate, a coin, a rug) and slim upright characters are unaffected, because the test asks which direction is flat and how extreme it is rather than treating thinness alone as a fault.",
"tags": [
"fix"
]
},
{
"date": "2026-09-02",
"title": "A single page now shows every way an x402 payment can fail, before it stops the economy",
"summary": "The platform already recorded every rejected, failed and replayed payment, but nothing displayed them, so the first sign of trouble was usually the economy halting. The new payment-outcomes board reads that record live: how many payments settled and how many failed, how many were refused as invalid before settling, how many were replayed copies of an earlier payment, the reasons behind each, and the balance of every wallet that funds a settlement against the floor it stops working at. That last panel matters most during an outage, because a starved fee wallet and a broken payment rail look identical from outside and need opposite fixes. Where a number could not actually be measured the board says so rather than showing a reassuring zero. Operator tooling, behind the same sign-in as the other internal boards.",
"tags": [
"feature",
"infra"
],
"link": "/docs/x402-endpoints"
},
{
"date": "2026-09-02",
"title": "The 3D Studio connector for ChatGPT is documented as the eleven tools it actually serves",
"summary": "The free 3D Studio connector grew to eleven tools when look_at_model landed alongside check_job, but a lot of the writing about it had not caught up: the connector page, the press kit, the partner page, the blog summary, the announcement copy, the press release and the store-submission package still said nine or ten, and the social card enumerated only nine tool names. Every one of those now matches a fresh pull from the live server, the social card lists all eleven, and the module headers inside the connector describe the real catalog including which tools spend generation quota. A test now pins the total, so the next tool to land fails the suite instead of quietly re-opening the same drift. The standalone viewer also stopped crediting the AR launch endpoint with a renderer it no longer loads. Two smaller corrections rode along: the pending-generation example now shows the field that tells a client whether the model coming back is a bare mesh or a finished rig, and the three places the platform pins the 3D viewer library are now explained in one comment, so nobody has to guess why an embed and a standalone page load different builds.",
"tags": [
"docs",
"fix"
],
"link": "/docs/mcp-studio"
},
{
"date": "2026-09-02",
"title": "License terms for three.ws containers on NVIDIA NGC",
"summary": "three.ws is preparing to publish its image-to-3D inference server in NVIDIA's NGC catalog, so the licence that will cover those container images is now public. It states plainly what you get: Apache-2.0 source, no model weights inside the image (you bring your own, under the upstream licence), and a container that carries no analytics or telemetry and sends nothing back to us. It sits alongside the other policies on the legal page.",
"tags": [
"docs"
],
"link": "/legal/nvidia-ngc-eula"
},
{
"date": "2026-09-02",
"title": "Fixed a footer link that led to a dead page",
"summary": "The terminal page's footer offered an Avatars link that pointed at an address the site does not serve, so clicking it returned a 404 instead of the avatar gallery. It now opens the gallery, the same place every other page sends you. On the knock page, a placeholder link that sat empty until the profile finished loading was removed, so people navigating by keyboard or screen reader no longer land on a link with no name and nowhere to go.",
"tags": [
"fix"
],
"link": "/tty"
},
{
"date": "2026-09-02",
"title": "Agent profiles stopped losing their on-chain history",
"summary": "Two thirds of the agents in the directory had quietly stopped picking up new on-chain activity. The indexer resumes each agent from the last transaction it saw, and once the RPC provider answering the call no longer held that transaction, the whole read failed and the agent was stuck there permanently: its profile kept loading, it just never learned anything new again. The crawl now notices that its bookmark has gone stale, re-reads from the top and carries on, and the same recovery landed for the secondary chains, where one had been parked on pruned blocks since April. The 1,083 stuck agents have been unstuck. The status page also learned to score this: an index that is erroring rather than lagging used to look perfectly fresh, and a lagging secondary chain no longer reports a platform outage nobody is having.",
"tags": [
"fix",
"infra"
]
},
{
"date": "2026-09-02",
"title": "Every showcased agent now has its own page",
"summary": "Agent Spotlight entries got somewhere to live. Each one now has its own page with the agent standing in 3D, the full write-up instead of three clamped lines, a facts panel read live off the agent (skills, conversations, on-chain actions, upvotes), share buttons, and a rail of related entries in the same category. Sharing one finally works too: paste a link anywhere and it unfurls with that entry's headline and the agent's own card rather than the generic showcase image. If you submitted an entry, you can now edit or remove it, which you could not do before: an Edit link on your card opens the write-up ready to change, and removing takes two clicks so it cannot happen by accident. Ranking got fairer as well, with a stable order for entries published at the same moment instead of a shuffle between page loads.",
"tags": [
"improvement",
"feature"
],
"link": "/spotlight"
},
{
"date": "2026-09-02",
"title": "Examples gains a voice, an AI's eye on 3D, and a filter",
"summary": "Three more runnable examples and a way to find them. The avatar can now speak a line you type: the free text-to-speech lane synthesizes it and a new speakAudio() method routes the audio through the same analyser the built-in voice uses, so the mouth follows the real waveform instead of miming. A second example calls the look_at_model tool on the public MCP server and shows exactly what an AI assistant sees when it inspects a mesh: four rendered angles plus a plain-language read on the geometry, which is how an agent judges its own output and decides to try again. A third reads the live tool list off that same server, so the page can never advertise a tool that was renamed or dropped. With seventeen examples the page also gets a filter: type to search names, descriptions and tags, or narrow by tag, and the lanes fold away as they empty. A deep link always wins over an active filter.",
"tags": [
"feature",
"improvement",
"sdk"
],
"link": "/examples"
},
{
"date": "2026-09-02",
"title": "The treasury's self-heal plan now only promises SOL it can actually move",
"summary": "When the payment rail runs short of the SOL it needs to settle, the platform first tries to heal itself: it looks across its own agent wallets for idle funds and pulls them back to the wallet that pays fees. The preview of that plan was counting wallets it could never open. Eight custodial wallets were encrypted under a key retired during the 2026-07 move to Google Cloud, so the preview kept advertising their balance as available while every real run failed on them, and the plan looked the same the next time anyone checked. The preview now opens each wallet's key before listing it, exactly as a real run does, and reports the ones it cannot open separately with the reason. A plan that reads as capital on hand when the capital is unreachable is worse than no plan at all, and it cost two separate investigations. The health check's settlement hint was corrected to match: it no longer says owner funds are needed only once every wallet reports being at its floor, because a sealed wallet never reports that.",
"tags": [
"fix",
"infra"
]
},
{
"date": "2026-09-02",
"title": "Agent profiles start learning about the world again",
"summary": "Every agent profile reads an index of what that agent has done on-chain, and that index had quietly stopped filling. Each agent is crawled from the last transaction the platform saw for it, and the crawl asks whichever node is not busy; nodes disagree about how far back they keep history, so a bookmark written by one was regularly unreadable by the next. That failed the whole read, the bookmark was only ever saved after a successful read, and so an agent would present the same dead bookmark forever and never index another thing. Two thirds of the directory was stuck that way. The crawl now recognises a bookmark no node can resolve, starts again from the newest activity, and clears the dead value instead of writing it back, and there is a one-command recovery for a backlog that has already built up. The same fault had frozen three secondary chains, one of them for four months, including a case where the node said plainly that it no longer holds those blocks while a second node's unrelated complaint was the only thing the crawl saw; those chains now skip forward to the present and record exactly how much history was lost doing it, so the gap is a number rather than a silence. The sweep was also only reaching a fifth of the directory per run, leaving three quarters of its time unused, which put the slowest profiles more than two hours behind; it now covers the whole directory in about seventy minutes. Finally, the platform status page had been calling the whole index down because of a secondary chain nobody reads on a profile, which buried the real fault for weeks: a secondary-chain problem now reports as degraded and says so, and the count of stuck agents is scored in its own right.",
"tags": [
"fix",
"infra"
],
"link": "/agents"
},
{
"date": "2026-09-02",
"title": "Every paid endpoint now appears in the public API document",
"summary": "Agents find three.ws services by reading the machine-readable API document the platform publishes at /openapi.json. That document listed 24 paid endpoints while the platform actually sells 75, because the entries were written by hand and most were never added. The rest answered a perfectly valid payment challenge in production and could still not be found, which is the whole point of publishing the document. Every paid endpoint is now generated from the same single catalog that already writes the platform's other service listings, so all of them appear with their real price, inputs and payment details, and one written once. The 24 richer hand-written entries are untouched. A test now fails the build if a service is ever added to the catalog without reaching the document, so the gap cannot reopen.",
"tags": [
"improvement",
"docs"
]
},
{
"date": "2026-09-02",
"title": "Fact checks stop saying \"not enough evidence\" when the evidence was there",
"summary": "A fact check reads five sources and asks a language model where each one stands on the claim. When that model answered in a shape the reader could not parse, every source was quietly recorded as saying nothing, the check reported \"not enough evidence\", and nothing anywhere said the reading had failed. A live check on whether a tomato is a vegetable came back that way while holding the Supreme Court case that settles the question. The reader now copes with the formats models actually use, including the thinking-out-loud preamble and code fences it used to choke on, and it no longer stops at the first square bracket it sees. When a reply genuinely cannot be read, the check now says it was degraded instead of presenting a verdict it did not earn, which also keeps that answer out of the week-long cache and out of the published accuracy score. The same fault had been collapsing the three search angles behind each check down to one.",
"tags": [
"fix",
"improvement"
],
"link": "/fact-checker"
},
{
"date": "2026-09-02",
"title": "A fact check can now say a claim is partly true",
"summary": "The fact checker grades every claim as supported, contradicted, mixed or not enough evidence, but the sources behind it could only ever be read as agreeing, disagreeing, or silent. That left no way to record what a source usually says about a half-truth: that the claim is right about the fact and wrong about its scale, its cause, or how universally it holds. Every source reading the same nuance was forced to pick a side, they all picked the same side, and the claim came back as a flat true or false. \"A tomato is a vegetable\" and \"Napoleon was unusually short\" were scored plain wrong rather than partly right, and on the published accuracy benchmark not one of the forty claims that deserved a mixed verdict ever got one. Sources can now be marked as partly bearing the claim out, they show up that way in the source list, and evidence that takes neither side no longer counts toward either. A claim every source calls half-true now reads as mixed on its own, without the sources having to contradict each other, and the confidence next to a mixed verdict finally measures how sure the evidence is that the answer is \"both\", instead of how close it came to breaking one way.",
"tags": [
"improvement"
],
"link": "/fact-checker"
},
{
"date": "2026-09-02",
"title": "Materialize: your generation, printed and posted to you",
"summary": "Everything the forge makes has lived on a screen. Materialize prints it and ships it. Point it at any model, yours or one from the gallery, and it measures the mesh first: whether it is a closed solid, how many separate pieces it is, where the holes are, how thin the thinnest wall gets, and its exact volume. That measurement is the price, so nothing is guessed. Pick a material and a height and you get an itemized quote, every line shown, in resin, tough resin, nylon, full-colour sandstone, draft plastic or stainless steel. Before it prints, the model is rebuilt as a watertight solid, its holes filled, scaled to the size you chose, and hollowed with drain holes if that saves you money. Full-colour prints carry the colour sampled from the model's own texture, which is what lets a printer reproduce what you see on screen. Checkout is USDC on Solana, holders get up to fifteen percent off, and you can follow the order from screening to production to the tracking number. An AI agent can order one too, over x402, with nobody in the loop: as far as we know that makes this the first API where an agent pays and a real object turns up at a real address.",
"tags": [
"feature"
],
"link": "/materialize"
},
{
"date": "2026-09-02",
"title": "Every print ships with proof it is the one you ordered",
"summary": "A certificate of authenticity is usually a card with a logo on it, which proves nothing. Every object Materialize ships now carries a real one. Its edition number is enforced in the database, so the third one that ships really is edition three. It records the fingerprint of the exact file the printer was given, and of the model you looked at. And both of those are written into a transaction on Solana, signed by the platform. The QR on the box opens a page showing the original model spinning, the prompt it came from, the edition, and the raw payload that went on chain, so you can hash the file yourself and compare. That check does not depend on our database, our company, or any block explorer staying online, which is the entire point.",
"tags": [
"feature"
],
"link": "/materialize"
},
{
"date": "2026-09-02",
"title": "Materialize will not print a weapon, a key to your door, or somebody else's brand",
"summary": "Printing something is not like generating a picture of it. A refused image costs nobody anything; a printed firearm part is a crime in most of the places we ship to, and a working copy of a key is a burglary tool. So Materialize checks every order twice, and the rules are code rather than a polite request. It refuses firearm components, suppressors, ammunition and magazines, keys and lock picks, counterfeit goods and other people's brand marks, concealed and spring-loaded blades, drug paraphernalia, and realistic firearm likenesses at life size. The first check runs before you are ever asked to pay, so a refusal costs nothing, and it names the category, links the policy, and says what you can print instead. The second runs on the paid order on its way to the printer, and an order that fails it is refunded rather than quietly shipped. What this deliberately does not do is refuse the ordinary: a tabletop miniature carrying a weapon prints, a blunt prop sword prints, your own logo prints. Full policy is in the docs.",
"tags": [
"feature",
"security"
],
"link": "/materialize"
},
{
"date": "2026-09-02",
"title": "Your Materialize order now tells you where it is, every step",
"summary": "Ordering a physical print is only half the promise; the other half is knowing it is actually happening. Every Materialize order now moves through a real fulfillment queue that a person works, and every step it takes is recorded and sent to you: it went to the production lane, it is on the printer, it is in quality check, it shipped, here is the tracking number and the carrier. Nothing skips a step and nothing goes backwards, because the order's status follows one fixed set of rules rather than whatever the last message said. If a print fails quality check it goes back to the printer rather than out the door. If a job goes quiet for longer than the lead time you were quoted, we get told before you have to ask. And when a print bureau's own system reports in, we check it really came from them and ignore repeats, so a retried message can never leave your order in a state it was never in.",
"tags": [
"feature"
],
"link": "/materialize"
},
{
"date": "2026-09-02",
"title": "The gallery and your avatar list stay fast as the catalog grows",
"summary": "There are now more than 58,000 avatars on three.ws, and about 1,600 more arrive every day. Two of the pages that read that catalog were quietly getting slower every single day, because they asked the database to look at every avatar in existence just to show you fifty of them. Both now read only the page you asked for, and the gallery's headline counts are worked out once a minute instead of once per visitor. Your dashboard list is roughly a hundred times quicker to answer, and, more to the point, it will still be that quick when the catalog is ten times bigger.",
"tags": [
"improvement"
],
"link": "/gallery"
},
{
"date": "2026-09-02",
"title": "The animation gallery shows you clips instead of a spinner",
"summary": "The motion library is nearly 3,000 clips and growing, and /animations used to download the entire catalog before it would draw a single card. Now the first batch of clips appears as soon as it arrives and the rest fill in behind it while you are already browsing, searching and filtering. However large the library gets, you start looking at animations after the same short wait.",
"tags": [
"improvement"
],
"link": "/animations"
},
{
"date": "2026-09-02",
"title": "Text-to-animation is no longer rationed like a paid service",
"summary": "Turning a sentence into an animation runs entirely on our own GPUs, but it was being counted against the allowance meant for services we pay a third party for: thirty a day, per person. It now uses the free allowance that lane deserves, which is eight times higher, so you can iterate on a motion prompt properly instead of running out halfway through.",
"tags": [
"improvement"
],
"link": "/animations"
},
{
"date": "2026-09-02",
"title": "When image understanding fails, it now says which part failed",
"summary": "Our image understanding runs through a chain of providers so one outage cannot take the feature down. When every link in that chain failed, the error you got only described the very last one, which was routinely the least relevant and sent everybody looking in the wrong place. The error now lists every provider that was tried and what each one said, so a problem gets fixed instead of investigated twice.",
"tags": [
"fix"
]
},
{
"date": "2026-09-02",
"title": "Materialize checkout: buy a real print, or let an agent buy one",
"summary": "You can now order a physical print of a model and pay for it. Quote the job, pick a material and a size, and checkout hands you a payment in USDC on Solana with the itemised price you were shown, never a recomputed one. Once the payment lands the order moves into safety screening on its own, and every step writes to a timeline you can read at any time. The same checkout is open to AI agents at /api/x402/print-order: an agent quotes the job, pays, and a real object is manufactured and shipped to a street address, with no account anywhere in the loop. Prices come from one catalog file and nothing else, a quote is signed so it cannot be altered on the way to checkout, and an order that was never going to be accepted is refused before anyone is charged.",
"tags": [
"feature",
"sdk"
],
"link": "/materialize"
},
{
"date": "2026-09-02",
"title": "Limited editions: cap how many copies of your model can ever exist",
"summary": "Any model you forged now has a Physical editions panel on its page where you set one number: how many physical copies of it may ever be printed. Leave it empty and it stays an open edition, which is what every model is by default. Set it and the number starts meaning something everywhere: the price panel tells a buyer which number they would be and how many are left, a sold-out edition is refused where the price is quoted rather than after someone has paid, and every certificate that ships reads \"3 of 25\" for good. It counts copies that actually shipped rather than orders people started, so an abandoned checkout never strands the last one, and it cannot be lowered below what has already left the building, because a certificate reading \"edition 7 of 5\" would be a lie already in somebody's hands. There is also now a page at /cert for looking a certificate up by the number printed on its card, and a certificate whose on-chain proof was delayed retries on its own until it lands.",
"tags": [
"feature",
"improvement"
],
"link": "/materialize"
},
{
"date": "2026-09-01",
"title": "The community showcase is open",
"summary": "Agent Spotlight is live at /spotlight. The agents index lists everything anyone ever registered, which is thousands of agents and no way to tell which ones are actually good. The spotlight is the other half of that: you write up what your agent does, everyone else upvotes what they like, and the best work rises. Filter by category, search across write-ups and builders, and open any entry straight into the agent's 3D page to talk to it. Entries stay honest by construction: your card stores your pitch and reads the agent's live name, avatar, skills and on-chain identity on every visit, so it can never go stale. One entry per agent, one vote per person, and submitting again edits your entry instead of making a second card. It opens with fourteen agents the team picked out of what the community has already built, from an adversarial code reviewer to an autonomous Solana trader that runs with no language model in the loop.",
"tags": [
"feature"
],
"link": "/spotlight"
},
{
"date": "2026-09-01",
"title": "three.ws is on the Solana dApp Store",
"summary": "The Seeker app is live. Open the dApp Store on a Seeker or Saga, search three.ws, and install it: a 3D agent studio that runs full-screen on the phone. Take a selfie and get a textured, rigged, animation-ready character in about a minute, or describe one in text, or upload a GLB you already have. Give it a personality, a voice and skills, then talk to it in 3D. Browse everyone else's agents in the marketplace, and deploy your own on-chain as a Metaplex Core asset your wallet holds. Every signature routes to the phone's Seed Vault through Mobile Wallet Adapter, so signing in is one tap and the private key never enters the app. Creating, chatting and browsing are free. Anything you make on the phone is waiting on the web app the moment you sign in there.",
"tags": [
"feature"
],
"link": "/docs/seeker-app"
},
{
"date": "2026-09-01",
"title": "Launch a 3D collection without a studio",
"summary": "3D Drops is live. Give it a base style and a few trait layers with weights, and it rolls a whole supply-capped collection in one go: up to 10,000 unique characters, each one a real rigged, animation-ready model rather than a flat picture. The roll is instant and free at any size because the art is generated per item when you reveal it, not rendered in bulk up front. Rarity is computed from the supply that actually came out, ranked, and sorted into common, rare, epic and legendary. Every collection publishes a fingerprint of its own recipe and, once launched, the seed behind it, so anyone can recompute the entire supply and check it against what the site served them. There is a Verify button on every collection page that does exactly that.",
"tags": [
"feature"
],
"link": "/drops"
},
{
"date": "2026-09-01",
"title": "Guided Tour page: real track counts, a clean fit on small phones, and a proper fallback when the outline is slow",
"summary": "The Quick highlights and Full tour cards on /tour now show their real stop counts and minutes (they were being reset to placeholder text a moment after loading), the chapter list matches the tour itself and no longer lists the separate getting-started walkthrough as a chapter, the page no longer clips its right edge on 320px screens, and if the tour outline fails to load you get a clear message with a retry button instead of an empty section.",
"tags": [
"fix",
"improvement"
],
"link": "/tour"
},
{
"date": "2026-09-01",
"title": "$THREE tiers page: the ladder now loads for everyone, signed in or not",
"summary": "Opening /three without a wallet connected or an account signed in showed an error card instead of the tiers, because the tier lookup refused anonymous requests. The page now shows every tier, its perks, and the exact $THREE threshold to any visitor, with a connect-wallet or sign-in prompt to see your own position. Signed-in accounts without a linked wallet get a clear prompt to connect one, and if the ladder genuinely fails to load the page keeps its heading and offers a retry.",
"tags": [
"fix"
],
"link": "/three"
},
{
"date": "2026-09-01",
"title": "The Story So Far loads again, and catches up to the Android launch",
"summary": "The 3D history timeline at /timeline had been opening on its error screen in production since it shipped: the page itself was there, but the file holding the milestones never made it into the deployed build, so every visitor was told to check their connection. The file now ships with every build, and a build without it is refused before it can deploy. While there, the timeline gained the milestones since late July, from the first Hugging Face article to the Android app, the Glance home-screen card, and the CoinMarketCap coverage, each with a link to its source. The floating walk companion no longer parks itself on top of the auto-play button, keyboard users get visible focus rings on every control, and the closed detail panel stays out of the tab order.",
"tags": [
"fix",
"improvement"
],
"link": "/timeline"
},
{
"date": "2026-09-01",
"title": "Stablecoins page: a clear message and a one-click retry when the feed is down",
"summary": "When DeFiLlama cannot be reached, /stablecoins now says so in plain words and offers a Try again button that refetches in place, instead of a reload link over blank stat cards. An empty answer from the feed gets its own message rather than $0.00 cards. When the page is showing the last good copy because the live feed is unreachable, the footer line says so, and the \"updated\" time no longer flickers back to a static label after the language pass runs. The page title and description now spell out what the board covers: market cap, peg health, backing mechanism, and chains.",
"tags": [
"fix",
"improvement"
],
"link": "/stablecoins"
},
{
"date": "2026-09-01",
"title": "Strategy Objects: signed-out visitors are sent to sign in instead of a half-open builder, and the view switcher is keyboard-visible",
"summary": "On /strategies, Equip, Fork, and New strategy now take a signed-out visitor straight to sign-in and bring them back to the page afterwards. Before, the page treated everyone as signed in, so the equip picker and the strategy builder opened for visitors who could not use them and Fork failed silently. The My library tab now shows its sign-in prompt without firing a request that could only fail, the Marketplace, Leaderboard, and My library switcher shows a visible focus ring when tabbed to and is announced correctly by screen readers, and the browser tab reads Strategy Objects.",
"tags": [
"fix",
"improvement"
],
"link": "/strategies"
},
{
"date": "2026-09-01",
"title": "Widget Studio: an honest preview status, copy that works in every browser, and clearer errors",
"summary": "On /studio the preview status line now stays in sync with the preview: it no longer snaps back to \"Pick an avatar to preview\" after the language pass runs, and when an avatar's model cannot be fetched it says so and disables camera capture instead of reporting a live preview over an error screen. Copy buttons fall back to the browser's built-in copy when the clipboard API is denied, so the snippet lands on the clipboard in embedded and in-app browsers too. A failed public-avatar search explains what happened and how to retry, the search box and the script-tag snippet are labelled for screen readers, and the page description now matches what the studio does.",
"tags": [
"fix",
"improvement"
],
"link": "/studio"
},
{
"date": "2026-09-01",
"title": "Spatial MCP renderer: a clear message and a retry when a scene cannot be fetched",
"summary": "The reference renderer on /spatial-mcp now shows a shimmer while a GLB downloads and, if the asset cannot be fetched or decoded, replaces the empty frame with a message that names the host, explains what to check (reachable, https, CORS, a valid .glb), and offers a Try again button that re-renders in place. Any host that embeds the renderer gets the same loading, ready, error, and empty states through a data-spatial-state attribute on the mount. The page title and description now say plainly what the page is: the reference renderer for the open Spatial MCP standard.",
"tags": [
"improvement",
"fix"
],
"link": "/spatial-mcp"
},
{
"date": "2026-09-01",
"title": "Readable header navigation from the first paint, and a readable feed status on the terminal",
"summary": "On a cold load, the theme system's capability check could leave the header's navigation labels stuck on the light theme's grey over the dark bar for several seconds, well under the contrast people need to read them. The check now runs without triggering colour transitions, so Build, Discover, Learn, Search, Walk, Sign in and Console render in their intended colours the moment the header appears. On /terminal the \"Positions\" feed-status chip no longer fades its label along with its idle dot, so it stays readable against the bar. The automated accessibility floor that caught it now measures pages once their entrance animations have finished, so it reports what a reader sees rather than a half-faded frame.",
"tags": [
"fix",
"improvement"
],
"link": "/terminal"
},
{
"date": "2026-09-01",
"title": "Strategy Lab backtests and simulated runs find real launches again",
"summary": "In production, every Strategy Lab backtest and simulated session was scanning for candidates and coming back empty: the server-side strategy runtime was asking for pump.fun data with a page-relative address that only a browser can resolve, and the discovery tools behind it needed an indexer the platform never configured. The runtime now calls the pump.fun tools directly in-process, and the new-launch and trending scans read the live pump.fun feed with the platform's own launch recorder as backup, so a scan returns real mints on the first tick. The Lab page also tells a signed-out visitor exactly what signing in unlocks, links straight to it, shows a live \"session running\" state while a run streams, and explains a quiet session (every candidate filtered out, or the feed briefly unreachable) instead of a bare \"complete\".",
"tags": [
"fix",
"improvement"
],
"link": "/strategy-lab"
},
{
"date": "2026-09-01",
"title": "Living Stages: the live feed connects in production, so the host actually performs",
"summary": "Every stage on three.ws/stage was reading \"feed offline\" in production: the page never told the browser where the realtime venue server lives, and even when pointed at it, the client tripped over an ordering quirk while subscribing to the host's state and dropped the opening line. The venue now connects the way /play does, subscribes correctly, and registers for the host's words before anything else, so you hear and read the host from the first beat, see who else is in the room, and get an acknowledgement when your question is queued. The feed pill is now a real control: when the feed drops, tap it to reconnect. A dead stage link explains itself and points back to the live marquee, a network blip retries in place without reloading, every button on the page has a visible keyboard focus ring, and the page description says what a Living Stage actually is.",
"tags": [
"fix",
"improvement"
],
"link": "/stage"
},
{
"date": "2026-09-01",
"title": "Sperax page: readable on narrow phones, with a one-tap copy of the plugin manifest URL",
"summary": "On phones around 320px wide, the /sperax page was rendering nearly twice as wide as the screen and the right half was cut off: the headline stopped mid-word, the secondary button and the second claim step were invisible, and there was no way to scroll to them. The wide tool-call diagram was forcing the whole page to its width. The page now fits any screen and the diagram scrolls inside its own box (keyboard included). Step one of the claim flow also gained a Copy manifest URL button, so the SperaxOS plugin address lands on your clipboard in one tap; when a browser refuses clipboard access, the button selects the URL for you and says so.",
"tags": [
"fix",
"improvement"
],
"link": "/sperax"
},
{
"date": "2026-09-01",
"title": "Get Started wizard: avatars come back into the flow, the brain you pick is the brain you get, and sign-in happens at publish time",
"summary": "The 5-step wizard at /start now finishes what it starts. Taking a selfie, writing a prompt, building in the studio, or using the editor brings the finished avatar straight back into the wizard instead of stranding you on the avatar page, and Upload GLB uses the same storage path as every other upload, so it works. The AI model you choose is now applied to the agent (before, it was collected and dropped), the embed snippet points at the real loader, and a signed-out visitor who reaches the publish step gets a sign-in panel that resumes exactly where they left off rather than a cryptic error. Helper pills no longer sit on top of the Continue button, off-screen steps are no longer tabbable, the price and wallet fields have names for screen readers, and failed price or wallet saves are reported instead of swallowed.",
"tags": [
"fix",
"improvement"
],
"link": "/start"
},
{
"date": "2026-09-01",
"title": "Claim threews.sol Subdomain: a clear account status in every case, a fit on small phones, and the light theme",
"summary": "The /threews/claim page now tells you where you stand the moment it loads: a loading placeholder while it checks your account, a sign-in prompt if you are signed out, a link to set a username if your account has none, and a plain explanation with a link to change it if your username has characters a .sol name cannot hold (before, that case ended in a bare validation error with nowhere to go). If the account check or the availability check cannot reach three.ws, the page says so and offers a retry instead of going quiet; a failed mint keeps its reason on screen with a retry when the failure was temporary and a link to create an agent when a wallet is what is missing. The .threews.sol suffix no longer spills off the edge on 320px screens, every control has a visible keyboard focus ring, and the page follows the light theme instead of staying black.",
"tags": [
"fix",
"improvement"
],
"link": "/threews/claim"
},
{
"date": "2026-09-01",
"title": "Mission Control: the cockpit now works signed out and on a phone",
"summary": "Open /terminal without an account and the Focus and Positions panes tell you exactly what unlocks trading (sign in, or create an agent wallet) instead of showing buy and sell buttons that could only refuse, and the Positions pill reads idle rather than reconnecting forever. Token art now loads through the same-origin image proxy, so coins whose art lives on IPFS gateways that refuse browser requests show their image instead of a blank tile. On phones the first launch that streams in no longer yanks you off the feed to the Focus tab, and the Positions tab is tappable again: the floating language picker used to sit on top of it, so it now lives in the cockpit's top bar. A seconds-old coin whose details pump.fun has not indexed yet is retried once when the server says to, and the page carries a proper h1 for screen readers.",
"tags": [
"improvement",
"fix"
],
"link": "/terminal"
},
{
"date": "2026-09-01",
"title": "Documentation sweep: 164 docs re-checked against the code they describe",
"summary": "Every doc the freshness gate flagged as describing code that had moved since it was written (174 of them, from the UX flow walkthroughs and the Oracle model write-up to the ops runbooks, tutorials, and the agent-abilities chapters) was re-read against the current code. 164 were corrected or extended: renamed routes now point at /agents/:id, retired model names and provider ladders are current, counts and thresholds match what ships, and the resilience work of the last month (last-good caches, typed 503s with Retry-After, the RPC read guards) is described where it applies. Along the way three things the docs promised were made true in the product: the Herald posting scope can now be minted from the dashboard key form, the translation script defaults to a Groq model that still exists, and the home page OpenAPI link opens the spec directly.",
"tags": [
"docs",
"fix"
],
"link": "/docs/freshness"
},
{
"date": "2026-09-01",
"title": "Faster pages: the home page stops freezing itself, and every page sheds code it never ran",
"summary": "A performance sweep of the home page, Create, Forge, Marketplace, Play, Docs, Discover and Chat, measured with Lighthouse and a CPU profiler. The home page's live agent no longer gets torn down by a 22-second timer while it is still loading (the teardown forced a WebGL context loss that froze the tab, and showed a false \"couldn't load\" message on slow connections), and the playground preview and the visitor-arena animation below the fold now wait until you scroll to them. The theme switcher stopped re-checking the page's light-mode support on every DOM change, which had cost the Marketplace almost nine seconds of main-thread time. The embed script at /agent-3d/latest is 25% smaller: its whitespace minifier was being undone by a later build step. The animation clip library is half its previous size with no visible change (numbers are stored at float precision instead of 17 digits). Web fonts now swap in without moving text, Forge gallery thumbnails load as small WebP tiles instead of full renders, and pages that only show the wallet tier badge no longer download the Solana wallet stack.",
"tags": [
"improvement",
"fix",
"infra"
],
"link": "/"
},
{
"date": "2026-09-01",
"title": "3D previews now wake up when you look at them",
"summary": "Second half of the performance sweep. Every rotating 3D preview on the site (the Marketplace grid, the Discover directory, the home page's community strip, the Create page's base-avatar card) used to spin for as long as the tab stayed open, which kept the page busy even when nobody was watching. They now turn when they scroll into view, settle once they have turned far enough to read as 3D, and turn again when you hover or focus them, with no motion at all if your system asks for reduced motion. The Create page's card shows a picture of the base avatar straight away and loads the live, rotatable model when you reach for it, so opening the page no longer waits on a 748 KB download. Forge stops downloading the 3D viewer until you actually generate something. Three things that used to appear late and shove the page down (the site header, the Marketplace hero, the weekly theme strip) now hold their space from the first frame. Coin art in the /play world and Marketplace card art are fetched at the size they are displayed instead of at full resolution.",
"tags": [
"improvement",
"fix"
],
"link": "/marketplace"
},
{
"date": "2026-09-01",
"title": "Examples now show everything an agent can do, not just how to embed one",
"summary": "The examples page used to be six ways to put a 3D avatar on a page. It is now organised into four lanes covering the whole platform, and every example runs on the page against the live production APIs. Avatars and embeds keeps the inline viewer, the talking agent, the floating support bubble, your own chat UI, and the iframe widget. 3D models adds a free text-to-3D generator that returns a real mesh and renders it in place, a GLB inspector that measures any model against a triangle budget, the full 112-clip animation library playable on any rig, and the live feed of models autonomous agents bought with their own money. Agents adds an agent defined entirely as files, hosted so the manifest loads from any website, and an avatar that performs American Sign Language. Agent economy adds the x402 payment handshake, where a paid endpoint quotes its price, network, and asset without a request spending anything, and the record of coins launched through the platform. Two new methods came out of the work: sign() performs your own text in ASL for captions and accessibility overlays, and playClip() drives the shared animation library on any humanoid rig. A fix landed alongside them: an embedded avatar was painting an invisible block under itself that swallowed clicks on whatever the host page put below it.",
"tags": [
"feature",
"improvement",
"docs",
"fix"
],
"link": "/examples"
},
{
"date": "2026-08-29",
"title": "Your agent on the Android home screen",
"summary": "The three.ws Android app (1.1) now ships a home screen widget. It shows your agent's avatar, its name, and how many moves it made today, refreshes itself in the background about every half hour without opening the app, keeps the last card it saw when the phone is offline, and opens your agent with a tap. Link a phone from the Glance page in one tap; every linked widget is listed there with a revoke button, because each one carries its own token that can read your card and nothing else. The Glance card also comes as a PNG now, which is what any native widget host, chat client or image-only surface needs.",
"tags": [
"feature"
],
"link": "/docs/native-widgets"
},
{
"date": "2026-08-29",
"title": "Our second Hugging Face article: the whole 3D AI agent stack in one piece, with a live 3D model inside it",
"summary": "Building 3D AI Agents End to End is live on Hugging Face. It walks through every 3D generation model we run and where it runs, rigging and animation for any humanoid, the multi-model brain and its guard chain, agent payments, and every program and partner behind the platform. The article embeds a rigged avatar you can orbit in place and avatars that wave as animated PNGs, and every claim links to the code and the open model files.",
"tags": [
"docs"
],
"link": "/news/huggingface-building-3d-ai-agents"
},
{
"date": "2026-08-28",
"title": "three.ws is an Android app you can install today",
"summary": "The signed app is public on GitHub as v1.0.0 and installs on any phone from Android 6 up. It opens full-screen with no browser around it, and it adds four things the website on a phone cannot do: sign in with a single tap through your Solana wallet (on a Seeker that tap is Seed Vault and the key never leaves the secure element), point the camera at yourself and get back a rigged 3D avatar, share a photo into three.ws from any other app and have it land in the avatar flow already attached, and mint an agent on Solana mainnet from the phone. It is 3.95 MB and open source like the rest of the platform. The Solana dApp Store listing was submitted the same day and is in review.",
"tags": [
"feature"
],
"link": "/seeker"
},
{
"date": "2026-08-28",
"title": "Avatars now face you when a page opens",
"summary": "The 3D viewer used to seat its camera on one fixed side and call that the front, which only worked for characters whose rig happened to be built facing that way. Everyone else opened showing you the back of their head. The viewer now reads which way the body is actually turned, from the rig itself, and frames the face, so every avatar greets you the same way no matter which tool it was made in. A second fix ships with it: if you have asked your phone or computer to reduce motion, avatars no longer spin slowly on a turntable, they hold still while keeping their breathing and blinking.",
"tags": [
"improvement",
"fix"
],
"link": "/marketplace"
},
{
"date": "2026-08-28",
"title": "Embedded avatars load again on other people's websites",
"summary": "Anyone who dropped a three.ws avatar onto their own site with the <agent-3d> tag and pointed it straight at a model file got an error panel instead of the character: the embed code referred to something that was not there and gave up at the last step, right before it would have started the idle animation. It now loads and plays as documented. Two related repairs came with it: the compression decoders three.ws hosts can finally be used by other sites (they were being blocked, so every embed quietly fell back to a third-party CDN and logged errors doing it), and the code that runs on other people's pages is now checked for this exact class of mistake before it can ship again.",
"tags": [
"fix",
"sdk"
],
"link": "/docs/web-component"
},
{
"date": "2026-08-28",
"title": "Portal: paste any web address and walk it in 3D",
"summary": "Type a website into /portal and it becomes a place you can walk. Every section of the page turns into a building whose height is how much that section actually says, every link becomes a door you step through (a door to the same site rebuilds the world there, so Back walks you through the sites you visited), every image becomes a billboard showing the real image, and every code block becomes a dark monolith. The colours come from the site's own theme colour, so a site looks like itself. You explore it with your own avatar, and the same world downloads as a 3D file you can open in Blender or view in AR, embeds in your own page with one line, and is readable by an AI agent through a new MCP server. Two things worth knowing: the same address always builds exactly the same city, so a link you share opens the world you walked, and Portal behaves like a polite crawler, reading each page once, saying who it is, and obeying robots.txt.",
"tags": [
"feature",
"sdk"
],
"link": "/portal"
},
{
"date": "2026-08-28",
"title": "One command turns a sentence into a rigged 3D character you can put on your site",
"summary": "Type `npm create @three-ws/agent \"a friendly cartoon astronaut\"` in a terminal and about a minute later you have a folder: the 3D model with a real skeleton so it can be posed and animated, a demo page that works when you double click it, and the two lines of HTML that put the character on your own website. No account, no API key, no signup, and nothing to install first. It also works from a photo, and it can make plain objects instead of characters. The same free lane the website uses is behind it, so what you get in the terminal is what you get on three.ws.",
"tags": [
"feature",
"sdk"
],
"link": "/docs/create-agent"
},
{
"date": "2026-08-28",
"title": "The Oracle stopped grading itself on the wrong question, and started learning on a clock",
"summary": "The Oracle scores every pump.fun launch in its first ninety seconds. Two things were wrong with it, and the second one was worse than it sounds.\n\nFirst, it was ranking the wrong outcome. The score answered \"will this spike 3x or graduate\", which meant a coin that ran 3x and then went to zero counted as a hit. Prime-tier coins were hitting that target 58.2% of the time and rugging 53.0% of the time, and 27.8% did both. The engine was accurate and useless: it answered \"will this move\" while everyone reading it heard \"should I hold this\". The score now ranks the survivable win, meaning it ran AND you would still be up, and it publishes two more numbers beside it: rug risk, and give-back risk, which is the share of coins like this one that run and then hand the run straight back. That last number is the one that names the trap.\n\nSecond, and this is the part that took a while to find: the rug flag was a readout of the price of SOL. It was decided by whether a coin's market cap had fallen under a hardcoded $3,000, and an empty pump.fun bonding curve is worth a fixed 27.958993 SOL, so the test really asked whether SOL was above about $107.30 that day. Of 206,428 coins labeled rugged, 206,419 were under $3,000. Of 25,180 labeled survivors, the cheapest was exactly $3,000. Identical dead coins, sorted by a price feed. Every win rate that subtracted rugs was reading the SOL chart. Labels are now built from two ratios where the price cancels out, the whole history has been recomputed, and the real rug rate turns out to be 11.4%, not 91%.\n\nThe model also learns now. Its weights used to be a file baked into the build, so nothing but a person running a script could ever update it, and nobody did: it sat frozen for nineteen days while eight times more evidence piled up behind it. It refits every six hours against fresh outcomes and only takes over if it beats the incumbent on data it has never seen. Candidates that fail are kept too, with the reason.\n\nAnd all of it is public. Oracle Lab publishes the actual model: every weight, the number of launches behind it, what happened to them, the reliability curve for every tier claim, and the full version history. Install @three-ws/oracle-model and you can score launches locally with the real weights, no key and no rate limit, reproduce any score we have ever shown, and re-measure our accuracy against outcomes you collected yourself.",
"tags": [
"improvement",
"feature",
"sdk"
],
"link": "/oracle-lab"
},
{
"date": "2026-08-28",
"title": "Model Diff: see exactly what changed between two 3D models",
"summary": "Every 3D pipeline has one failure that costs more than all the others put together, and it is completely silent. Animation clips find a character's joints by name. Rename a bone, drop a joint, or run a file through a tool that renumbers things, and the model still loads, still renders, and never moves again. Nothing errors. You find out when somebody tells you the character is standing there frozen. Model Diff is the check that catches it. Drop two files and it tells you what actually changed: the geometry, the materials, the textures, the skeleton, the clips, with renames recognised as renames instead of one thing vanishing and an unrelated thing appearing. It ignores the meaningless noise a re-export always introduces and flags the edits that matter, and it grades every difference, so you can tell a recoloured jacket from a rig that just lost both eye joints. Both models load into the same 3D scene: lay one over the other, drag a line across to wipe between them, or stand them side by side, and click any line in the report to fly the camera to the thing it names. Nothing you drop is uploaded. The same engine is a free API anyone can call without a key, a command you can put in a build so a broken export fails it, and a tool your AI agent can use to check its own work before it publishes it.",
"tags": [
"feature",
"sdk",
"improvement"
],
"link": "/diff"
},
{
"date": "2026-08-28",
"title": "Glance: your agent on your Windows home screen, in your README, and in your terminal",
"summary": "Your agent now has a card: its name, what it did in the last 24 hours, when it last acted, and a tap straight back into it. Pin it to the Windows 11 widgets board and it installs with the site itself, no app store, and refreshes on its own every fifteen minutes; signed out it shows a sign-in card instead of an error, and offline it keeps showing the last card it saw. The same card is one image URL you can paste into a GitHub README, a Slack message, or any page, and it follows the reader's light or dark mode. There is a live element for websites, a command you can run in a terminal to watch an agent work, and an npm package for building your own. Android is next, then macOS and iOS.",
"tags": [
"feature",
"sdk"
],
"link": "/glance"
},
{
"date": "2026-08-28",
"title": "Herald: your terminal can now make a 3D character walk onto your screen and talk to you",
"summary": "We took the part of three.ws where an avatar delivers a notification in person and turned it into something anyone can use, on any website. @three-ws/herald is an open library: one line of code and a 3D character walks into the corner of your page, looks at the person using it, and tells them, with a link to the thing it is about. It is not a toast library. It is an interrupt budget: an importance floor, quiet hours, a rate limit, dedupe, freshness, batch collapse, and a stable reason for every message it refuses to deliver, so you can always answer why you did not hear about something. No GPU, an iframe, or reduced motion falls back to an accessible card that screen readers announce the same moment everyone else hears it. There is also a rail: with an API key, anything that can make an HTTPS request reaches the browser tab you already have open. Run npx @three-ws/herald watch -- npm test and your avatar tells you when the build finishes, keeping the exit code. Wire the MCP server into Claude or Cursor and an agent that worked for twenty minutes can walk on and tell you it is done. An announcement always goes to your own sessions, so no key can interrupt anyone but its owner. Try the whole thing, rules engine included, at three.ws/herald.",
"tags": [
"feature",
"sdk"
],
"link": "/herald"
},
{
"date": "2026-08-28",
"title": "Every release note now shows you the code that shipped it",
"summary": "Until today this site kept two separate records of the same work: a changelog written for people, and a commit feed written by engineers, neither aware of the other. The new ship log joins them. Each release note now lists the commits that produced it, and says how confident it is in each match and why it made it, in plain words you can argue with. Commits nobody ever wrote up are grouped and shown too, so the work that shipped quietly is visible instead of lost. The announcements in the Telegram channel now carry a link to the exact code behind them, the raw commit feed reads in plain language instead of engineering shorthand, and it no longer posts merges or dependency bumps. The whole thing is a free, open-source package that works on anyone's repository.",
"tags": [
"feature",
"sdk",
"improvement"
],
"link": "/ship"
},
{
"date": "2026-08-28",
"title": "Report a bug and it arrives as a test that fails until it is fixed",
"summary": "When you tell the agent in the corner that something is broken, it now sends what you actually did: the steps you took, the page, and the error the browser saw. Those steps compile into a real test that fails while the bug is there and passes once it is gone, so the slowest part of fixing anything (working out how to make it happen again) is done before anyone starts. You can see every step that gets attached before you send, and nothing you typed into a field is ever recorded, only how much. The recorder is open source and works on any site: install @three-ws/witness.",
"tags": [
"feature",
"sdk",
"improvement"
],
"link": "/docs/witness"
},
{
"date": "2026-08-28",
"title": "Your agent reads the checkout before you pay",
"summary": "A new browser extension puts the companion on the one screen where it is worth the most: the payment page. It tells you the total against the price you were actually shown, names money the page never itemised, adds up the fees, and reads the fine print for the things that catch people out, a trial that turns into an annual charge, a renewal at a different price, a cancellation notice period. Every amount it quotes is computed rather than read off a language model, so the arithmetic still works when nothing else does, and it never calls a charge illegal or a scam because that is not a call it is in a position to make. It never reads a form field: your card number, security code and address are structurally unreadable to it, personal data is stripped in your own browser before anything is sent, only the site's hostname travels, and nothing about the page is stored anywhere.",
"tags": [
"feature",
"security"
],
"link": "/companion"
},
{
"date": "2026-08-28",
"title": "x402 Preflight: check whether a seller can take your money before you pay",
"summary": "Paying an AI agent for a service tells you the price, but nothing tells you whether that seller can still complete the transaction. If its own fee wallet has run dry, you find out only after you have signed and paid, and the money and the moment are both gone. Preflight is a small, signed statement any seller can publish saying which payment networks it can currently settle on, how well recent payments have gone, and which other network to use instead when one is down. It expires within a minute, so nobody can wave an old healthy report through an outage, and it is signed, so a seller that says it can deliver has gone on record. three.ws now publishes one, and there is a free page, a command-line tool, an agent tool and an open-source client for reading anyone else's. The format is open: no permission needed, nothing routed through us.",
"tags": [
"feature",
"sdk"
]
},
{
"date": "2026-08-28",
"title": "Your 3D agent now runs in a terminal, with nothing installed",
"summary": "Type curl three.ws/tty into any terminal and a rigged 3D character walks, turns and animates in the window, drawn entirely out of coloured text. There is nothing to download and no graphics card involved on either end: the frames are drawn by a plain software renderer on our servers and streamed to your shell as they are made, so the animation starts before the model has finished loading. Point it at an agent you own with three.ws/tty/<id>, or at any public 3D file. The same renderer is published as a package, so you can run it against your own models offline, orbit the camera with the arrow keys, or capture a single still frame to paste into a README. It reads a real skeleton and plays the model's real animation, so the character actually moves its limbs rather than spinning in place.",
"tags": [
"feature",
"sdk"
],
"link": "/tty"
},
{
"date": "2026-08-28",
"title": "We now prove our fallbacks work, instead of hoping they do",
"summary": "Adding a backup for an outside service is easy; knowing it made things better is not. A new audit reads the whole codebase and grades every call we make to an outside service on five protections, publishes the result as a public map, and fails the build if a new unprotected call is added or an existing one gets worse. On top of it, each endpoint now declares what it is allowed to tell you when a service it depends on is down, and a test proves it by actually breaking the service and checking the answer. That immediately caught a real bug: a token lookup was telling people a token did not exist when the price indexer was simply unreachable. Related fixes stop stale data reaching a payment retry, a token security verdict and an access check, where an out-of-date answer costs money or grants access.",
"tags": [
"feature",
"improvement",
"infra"
],
"link": "/docs/resilience"
},
{
"date": "2026-08-28",
"title": "Your notifications now have a body: meet your companion",
"summary": "Your phone buzzes forty times a day and two of those mattered. three.ws/companion is the other half of that problem. Connect what you already use, on your own keys: a Telegram bot you create, the private calendar link Google, Apple and Outlook all publish, your inbox over IMAP with an app password, or nothing at all but a token your phone, your Mac, your build script or an AI agent can post to. Everything that arrives is scored out of 100 against the things people actually want interrupting for (a saved contact asking you something, a one-time code, a meeting starting in ten minutes, a payment that failed) and the rest is kept quietly in a feed with the reason it stayed quiet. What clears your bar is delivered in person: a 3D character walks on, and if it is from someone you gave a face to, it arrives wearing their avatar and speaking in their voice. It reaches you on three.ws, as a push on your phone, and on your desktop, where a new companion app puts a character that walks across your screen and comes over to tell you. You can answer straight from the message: a Telegram delivery carries a reply box that sends back into the same chat, so it is a conversation and not a receipt. Quiet hours are absolute, the credentials are encrypted, and one button revokes every device. Developers get the whole thing as @three-ws/companion: a client, the same triage rules the server runs (so you can score your mail locally and never send the rest), a drop-in 3D stage for any page, a CLI, and an MCP server so an agent that finds something urgent can stand in front of you and say it.",
"tags": [
"feature",
"sdk",
"improvement"
],
"link": "/companion"
},
{
"date": "2026-08-28",
"title": "AI agents can now see the 3D models they make",
"summary": "Until today every 3D generator on the internet, ours included, answered with a link to a file an AI agent cannot open. A person clicks it; an agent is blind to it, which is why an agent that makes a 3D model has never been able to tell a clean one from a melted one. That changes now: ask to look at any model and it comes back rendered from four angles as actual pictures the agent sees, with the geometry explained in plain language. An agent can finally check its own work, spot that the back of the model was never finished, and try again saying exactly what was wrong. It works over MCP, over a plain web request, and through a new open-source package called @three-ws/see, and it is free, including for agents that bought a model from us and want to confirm they got what they paid for.",
"tags": [
"feature",
"sdk"
],
"link": "/docs/3d-vision"
},
{
"date": "2026-08-28",
"title": "The Seeker app now runs full-screen, and its release process is written down",
"summary": "The three.ws app on Solana Seeker verifies against the site and opens full-screen on its own home screen, with no browser address bar. Two bugs found while testing it on a real Android build are fixed: the app only recognised your Seed Vault wallet on the first page you opened, so signing in from the login page said no wallet was found, and the login page never loaded the wallet at all. A new guide documents exactly how a release reaches the Solana dApp Store, from the publisher account through building and submitting the app.",
"tags": [
"fix",
"docs"
],
"link": "/docs/seeker-publishing"
},
{
"date": "2026-08-28",
"title": "Tell the agent in the corner what is broken",
"summary": "The companion that walks your pages now listens. When a page actually fails, it turns and asks what you were doing, and the report it sends carries what you should never have to type: the page, the exact build it happened on, your screen size, and the errors the browser recorded at that moment. You can also open it any time from the control beside the avatar, and dictate instead of typing. Reports are grouped so that twenty people hitting one problem read as one problem, and every one of them is read by a person before anything changes.",
"tags": [
"feature",
"improvement"
],
"link": "/docs/feedback"
},
{
"date": "2026-08-28",
"title": "Agent payments now stop themselves when the fee wallet runs dry",
"summary": "The platform sponsors the network fee on every agent payment. When that wallet ran down to the point where it could not cover a single fee, the check meant to pause payments could not run, because it depended on a balance lookup and the network providers were rate-limited at the same moment. Payments kept being attempted for three hours and none of them could complete. The pause now reads its answer from the failure itself, which costs nothing and works exactly when the network is too busy to answer questions. The status page also names the real cause now, a wallet that needs funding, instead of pointing at the payment rail.",
"tags": [
"fix",
"infra"
]
},
{
"date": "2026-08-28",
"title": "Agent history catches up on fast chains in days instead of weeks",
"summary": "The cross-chain index read one window of blocks per chain every 15 minutes. That is enough to keep pace with a slow chain but not to close a gap on a fast one: Arbitrum was 18.4 million blocks behind and would have taken about six weeks to catch up, while most runs finished early with most of their time budget unused. Runs now spend that spare time on whichever chain is furthest behind, and do nothing extra when every chain is current.",
"tags": [
"improvement"
]
},
{
"date": "2026-08-28",
"title": "Your agent now tells you the important things in person",
"summary": "Notifications no longer wait behind a badge. When something worth stopping for lands while you are on the site (a sale, money in, a 3D generation finishing, anything touching your account), your corner companion turns to you, plays a gesture, and says it out loud with a link straight to it. If the companion is switched off it is summoned for that one message and walks off again, without turning itself back on. It is a real delivery channel next to in-app, push, email and Telegram: toggle it per category in Settings, and there is a Turn off control in the bubble that silences it in that browser instantly. It only ever announces unread notifications under fifteen minutes old, says each one exactly once, caps itself at two per visit, and never speaks into a background tab. Nothing is spoken aloud unless you already turned on the companion's voice; otherwise it is text in the bubble, which screen readers announce too.",
"tags": [
"feature"
],
"link": "/notifications"
},
{
"date": "2026-08-28",
"title": "Next on the roadmap: your agent on your home screen",
"summary": "A fifth phase has been added to the public roadmap: native widgets. Right now three.ws is something you open. The plan is for the agent you own to sit on your home screen and your desktop, showing itself and one live number about it, refreshed by the system without you opening anything, with a tap that drops you back into the right screen. Android is first, because the Android app is already built and signed: three.ws ships as a real app for Solana Seeker, and as of today Google confirms the app's release key against the site, so a three.ws link opens inside the app instead of a browser tab. Windows 11 follows through the installed web app, which needs no separate store submission, then macOS and iOS through a shared widget extension. None of it is shipped yet. It is written down, scoped and public so you can hold us to it, and the phase is on the roadmap in the repo along with the work order behind it.",
"tags": [
"docs"
]
},
{
"date": "2026-08-28",
"title": "Agent history on Base, Ethereum and Gnosis is moving again",
"summary": "The cross-chain agent index had stopped advancing on three networks. A single unprintable byte inside an on-chain metadata string was rejected by the database, and because the index only records its position after a successful write, every following run re-read the same block and hit the same byte. Base had been frozen for 13 days, Gnosis for 15 and Ethereum for 4. Those bytes are now stripped on the way in, so the event is kept and the crawl moves past it. Registrations, ownership transfers, metadata edits and reputation attestations on those chains flow into agent profiles again.",
"tags": [
"fix"
]
},
{
"date": "2026-08-28",
"title": "Photo uploads, avatar saves, launches and team tasks fail fast instead of hanging",
"summary": "A handful of steps still had no time limit on the outside service they called. When that service accepted the connection and then went quiet, the request sat there until the platform killed it, and you got a spinner that ended in nothing: a photo-to-avatar upload that never came back, a saved model that never finished downloading, a coin launch stuck fetching its own thumbnail, an ordinals order whose status never loaded, a team task where one hired agent held up the whole run. Each of those now has a deadline and a clear answer when it is missed, and the ones that are safe to repeat retry once first. Anything that spends money or starts a generation is never repeated: it is bounded and then reported, so no upload is billed twice and no hire is paid twice. Saving a model from a link is also checked more tightly now, against the address it actually connects to and with the size limit applied while it downloads.",
"tags": [
"improvement",
"fix",
"infra"
]
},
{
"date": "2026-08-28",
"title": "The iOS app now exists in the repo, and the web half of it ships today",
"summary": "three.ws is being packaged for the App Store as a native iOS app. The container and its Xcode project are in the repo, and the part that runs inside the website went live with this deploy: on iPhone the app will get a working share sheet, links that open the app instead of Safari, wallet sign-ins that return to the exact page you left, and a real offline screen. Nothing changes on the web. What is still outstanding is an Apple Developer organization account, app icons, and push notifications.",
"tags": [
"feature",
"infra"
],
"link": "/docs/ios-app"
},
{
"date": "2026-08-28",
"title": "Your avatar now runs in a terminal, and it can be your coding agent's face",
"summary": "A new package, @three-ws/tty-avatar, draws any three.ws avatar or agent body live in a terminal window: one npx command, no browser, no GPU, 24 frames a second in colour. It has moods (thinking, working, happy, error, asleep) that work on any model without a rig, and one command wires it to Claude Code so the avatar thinks while the agent reads your prompt, nods while it edits files, shakes when a tool fails and bounces when it finishes, with a caption naming what it is doing. Every avatar and agent page now has the command in its Embed tab.",
"tags": [
"feature",
"sdk"
],
"link": "/docs/tty-avatar"
},
{
"date": "2026-08-28",
"title": "Your agents can now tell you why they are not doing anything",
"summary": "An audit found ten of twelve autonomous trading agents had not acted in weeks while every health check reported them perfectly fine: the container was up, the feed was streaming, every strategy was armed. Liveness measures the process, but acting needs a whole chain of preconditions the process knows nothing about. A new open-source package, @three-ws/agent-vitals, models those preconditions as a causal graph and answers the real question: can this agent act right now, and if not, what is the one thing to fix. It reports root causes instead of symptoms, so a stale deployment that killed a model chain that starved a decision queue reads as one item with one command, not three red rows to guess between. Run npm run agent:vitals for the fleet, or GET /api/agents/vitals.",
"tags": [
"feature",
"sdk",
"infra"
],
"link": "/docs/agent-vitals"
},
{
"date": "2026-08-28",
"title": "Knock: put a price on reaching you, and hear what gets through in person",
"summary": "Every account can now open a door at three.ws/knock/<username> and set what one message from a stranger is worth, anywhere from a tenth of a cent to a thousand dollars, or free. Anyone can pay it to get exactly one message through: a person pays with the wallet already in their browser, and an agent pays over x402 with no human involved. The USDC settles straight to your own wallet; three.ws never holds it and takes none of it. What arrives is not another badge: your 3D companion walks on screen wherever you are on the site and says who is at the door and what they paid, ranked by that amount, so the expensive ones interrupt and the free ones wait in the feed. You set the price, the daily cap, the message length and a block list, and every one of those refusals happens before any payment, so a message that was never going to land is never one somebody was charged for. Only the one-line subject is ever read aloud. Senders get a receipt link they can check for your reply without an account here, which means answering someone never hands out your address. There is a browsable directory of open doors, an npm package and CLI for agents (@three-ws/knock), and an MCP server (@three-ws/knock-mcp) that deliberately holds no wallet: it quotes a paid door and stops, so a human always approves the spend.",
"tags": [
"feature",
"sdk"
],
"link": "/knock"
},
{
"date": "2026-08-27",
"title": "Every outside service the site talks to now has a time limit and a plan B",
"summary": "A follow-up sweep across the whole codebase, not just the pages that had already broken. Every call to an outside service is now bounded by a deadline, which a build check enforces from here on, so one slow provider can no longer hold a page or a scheduled job open until it is killed. Reads that used to get a single attempt now retry the failures worth retrying: launch pages and share cards during a busy mint, token names, bounty lookups, agent search, Telegram alerts and trade signals. Prices for ETH now come from four independent feeds instead of one. A wallet history scan that could not reach a block explorer used to report the wallet as having never transacted, which quietly affected airdrop eligibility; it now says which chains it could not read. Risk reports are pointedly not cached, since an old verdict on a token that has since changed hands is worse than none.",
"tags": [
"improvement",
"infra"
]
},
{
"date": "2026-08-27",
"title": "Every official three.ws account and address, listed in one place",
"summary": "The community page now carries the complete list of channels three.ws operates: the X account we post from, the Telegram release channel and the community group, the GitHub repo, the support address, and the one $THREE contract on Solana with a copy button so nobody retypes it by hand. It is there so any account, link or \"new contract\" claiming to be us can be checked in about five seconds. Two things are worth committing to memory: we never message you first and nobody from three.ws will ask for a seed phrase, a private key or a wallet signature, and we never launch a second token, so any migrated or replacement $THREE contract is a theft whatever it is posted under. One change behind the list: the founder's personal X account, @nichxbt, has been unavailable since August 27, 2026, so nothing posted under that handle is us while it is down. Announcements carry on exactly as before through @trythreews, the Telegram channel, and this changelog, which publishes itself straight from the deploy.",
"tags": [
"security",
"improvement"
],
"link": "/community"
},
{
"date": "2026-08-27",
"title": "Prices, wallets, search and speech keep answering when a data provider goes down",
"summary": "A single slow or rate-limited provider used to be enough to blank a page. Token pages, portfolios and dashboards now fall back to other market sources and, failing that, to the last good reading, clearly marked with when it was taken instead of passed off as live. Wallet and checkout pages survive a Solana node outage: a payment can still be built, a balance reads as unknown rather than zero, and a temporary outage says retry shortly instead of failing outright. Image understanding skips a provider that is rate-limiting instead of waiting on it, so it answers from a working one. Text-to-speech falls through to a second voice engine rather than failing. Agent search still returns keyword matches when the search index is unreachable. Ticker availability is the one thing deliberately never answered from cache, because a name claimed minutes ago has to show up as taken.",
"tags": [
"improvement",
"infra"
]
},
{
"date": "2026-08-27",
"title": "Agent pages, embeds and the camera tools keep working when a CDN or gateway is down",
"summary": "The parts of the site that load files from someone else's server used to have exactly one place to get them. If that one host was slow, blocked by a corporate network, or rate-limiting the visitor, the result was a blank box with no explanation: an agent page with no avatar, a 3D preview that never appeared, mocap and sign-language input that refused to start, a deposit QR that showed a broken image. Each of those now tries every equivalent source in turn. Agent avatars retry across all four IPFS gateways, on-chain reads rotate through every public node for the chain, the 3D viewer loads its decoders from our own servers, the camera tools ship their code with the page, and a QR code that cannot be drawn falls back to text you can still copy.",
"tags": [
"improvement",
"fix"
]
},
{
"date": "2026-08-27",
"title": "Every outside data source now has a backup",
"summary": "Prices, DeFi stats, token data, pump.fun feeds, RPC reads and 3D generation providers all had single points of failure: one rate limit or outage upstream turned into a blank section or an error page. Every one of those reads now has a deadline, retries only on transient faults, a second and third source where one exists, and keeps serving the last good value (marked as stale) when everything upstream is down. In the browser, API calls time out instead of spinning forever, script and model loads try mirror CDNs, and payment flows report a failed confirmation with the transaction signature instead of showing nothing.",
"tags": [
"improvement",
"fix",
"infra"
]
},
{
"date": "2026-08-27",
"title": "Every tap target on the marketplace and your agents pages now fits a thumb",
"summary": "On phones, the hearts, tags, author links and the write-with-agent button on marketplace cards were as small as 18 pixels tall, so browsing on a Seeker or any Android phone meant a lot of missed taps. All of them, plus the walk-companion controls and footer links, now sit inside 44 pixel tap areas without changing how the cards look. Measured on a real phone profile: the marketplace went from 1,352 undersized controls to zero.",
"tags": [
"improvement"
],
"link": "/marketplace"
},
{
"date": "2026-08-27",
"title": "A real home screen for the Seeker app, and a Seeker verified badge",
"summary": "The three.ws app on Solana Seeker now opens on its own home screen at /seeker: one tap signs you in with Seed Vault, your agents are right there, and Create, Discover, and Chat are a thumb away. Seeker owners can prove it by checking for the Seeker Genesis Token in their wallet (a read, never a transaction); once verified, every agent they own shows a Seeker verified badge on its page.",
"tags": [
"feature"
],
"link": "/seeker"
},
{
"date": "2026-08-27",
"title": "Share a photo or a 3D file straight into three.ws",
"summary": "On Android (and the Seeker app) three.ws now appears in the share sheet. Share a photo from your camera roll and it lands in the selfie-to-avatar flow already attached; share a .glb and it opens the upload flow. Going offline inside the app shows a three.ws screen that reloads on its own when the connection is back, instead of a browser error, and the home-screen icon no longer gets its corners cropped on round or squircle launchers.",
"tags": [
"feature",
"improvement"
],
"link": "/create"
},
{
"date": "2026-08-27",
"title": "Text-to-3D now has five independent image sources behind it",
"summary": "Every text-to-3D generation starts by painting a concept image, and until today that step leaned on two providers. When one of them was cut off and the other stalled, text prompts failed for hours. The painter now tries five independent sources in order: Google's image model, NVIDIA's FLUX.1-dev, two more FLUX providers routed through Hugging Face, and a keyless public service as the final fallback, all under one shared time budget so a slow provider hands off instead of holding your request. A finished model's status is also remembered, so checking on a done job answers instantly instead of re-scoring it every time.",
"tags": [
"fix",
"infra"
],
"link": "/forge"
},
{
"date": "2026-08-27",
"title": "Seeker app: stay signed in across restarts, and a submission path that matches today's dApp Store",
"summary": "On Solana Seeker, the three.ws app now remembers your Seed Vault session when Android restarts the app in the background, so reopening it no longer asks you to approve again; a revoked session is dropped cleanly instead of failing every signature. The My Agents home-screen shortcut opens your own agents instead of the public index. Behind the scenes the store submission tooling was rebuilt for the Publisher Portal that Solana Mobile now uses, and the listing copy was rewritten to describe only what the app really does.",
"tags": [
"improvement",
"fix",
"infra"
],
"link": "/my-agents"
},
{
"date": "2026-08-27",
"title": "Text-to-3D is generating again, and it answers in seconds",
"summary": "For most of a day, describing something in the forge produced a long wait and then a busy message, on the site, in the ChatGPT app, and for agents buying through OKX. Two things had gone wrong at once: the image provider that paints the concept picture had stopped responding, and the free assistant that sharpens your description was asking for models that no longer exist. Both are fixed. The painter now runs on a model that answers in about five seconds, the assistant uses current models, and a text-to-3D request is accepted in three to eight seconds instead of never. Image-to-3D was never affected.",
"tags": [
"fix"
],
"link": "/forge"
},
{
"date": "2026-08-25",
"title": "GitHub Discussions is open: say hello, show what you built, or steer the roadmap",
"summary": "The Discussions tab on the three.ws repo now has three real threads instead of an empty room. Welcome is where to introduce yourself and find every other channel we are in. Show and tell is for avatars, embeds, and experiments you built, with the good ones reposted with credit. Roadmap is an open thread on what to build next and what is in your way, and anything concrete raised there becomes a tracked issue linked back to the thread. The repo also passed 100 stars this week, which was the last thing standing between three.ws and GitHub's Open Source Friday livestream.",
"tags": [
"feature",
"docs"
],
"link": "/docs/community"
},
{
"date": "2026-08-25",
"title": "The market tools now clear the WCAG AA contrast floor",
"summary": "An accessibility sweep of the compare page found four places where text did not have enough contrast against what sat behind it. The selected time-range button, the rank badge, and the heatmap's view switcher all put white on a blue too light to read; they now use a blue and a text colour chosen to clear the standard in both the light and the dark theme. The comparison table's green was too pale on a white background, the coin column headings were painted in their chart-line colour (worst of all for the amber series), and the Getting started pill's counter borrowed a colour meant for light backgrounds while sitting on a dark one. Every one of those is fixed, and the column headings keep their tie to the chart through a coloured rule under each symbol.",
"tags": [
"fix",
"improvement"
],
"link": "/compare"
},
{
"date": "2026-08-25",
"title": "Dashboard Account fits a phone screen, and it stops claiming you own no .sol domains when it could not check",
"summary": "On a 320px phone every card on the account page ran past the right edge of the screen, which put Sign out, the AI provider key fields, Link wallet, and the quick links out of reach with no way to scroll to them; the cards now fit and every control is reachable. The SNS panel used to answer a resolver outage with \"No primary .sol domains found\", telling you to go register a domain you may already own. It now says plainly that the lookup failed and offers a retry, and when only some of your wallets could be checked it says how many were missed instead of quietly counting them as empty. The panel also refreshes itself when you link, disconnect, or re-prioritise a wallet, rather than sitting on a stale answer until you reload. The delegation table says how many agents it is showing out of your total instead of silently listing the first eight, and the page finally carries a search description of what it holds: profile, wallets, domains, delegation, provider keys, and the audit trail.",
"tags": [
"fix",
"improvement"
],
"link": "/dashboard/account"
},
{
"date": "2026-08-25",
"title": "Capabilities stops reporting an empty dashboard when the API is the thing that failed",
"summary": "The Capabilities command center used to ask the API once per agent for launcher and market-maker data, which on a real roster tripped the rate limiter mid-load. The failed calls were silently discarded and the page rendered a confident all-zero dashboard for an account that had launchers, coins and live markets. It now reads each capability for the whole account in a single call, so the request count no longer grows with your roster, and any lane that does fail says so in place with a working Retry instead of showing a zero. The 30-second refresh also stops resetting the tab you were reading and no longer replaces a Launch Now or Claim button while its request is still in flight, and on a phone the row action stays pinned in view instead of scrolling off the right edge.",
"tags": [
"fix",
"improvement"
],
"link": "/dashboard/capabilities"
},
{
"date": "2026-08-25",
"title": "The Data API console keeps working when one of its services is down, and its corpus numbers are now live",
"summary": "The developer console at Dashboard > Data API is fed by three separate services, and any one of them failing used to take the whole page down: a pricing outage left a single error card where the pass status, the API key table, the quickstarts, and the endpoint list should have been, and a pass-status hiccup left a loading shimmer that never resolved. Each card now fails on its own with an explanation and a retry, and everything else stays usable. Status messages appear next to the control that produced them, so a key rotation error no longer lands in the pricing card several screens away. The copyable quickstarts stopped pasting a shortened version of your key into commands that could only ever be rejected; they now carry a clearly marked placeholder, or the real key in the moment it is minted. Copy buttons say so honestly when the browser refuses the clipboard instead of claiming success. The article and publisher-feed counts are read from the archive itself rather than typed into the page, so they no longer drift, and the tables and code blocks can be scrolled with the keyboard.",
"tags": [
"fix",
"improvement"
],
"link": "/dashboard/data-api"
},
{
"date": "2026-08-25",
"title": "Dashboard Analytics fits a phone screen and its failure page can actually reload",
"summary": "On a 320px phone the time-range buttons refused to wrap, which pushed every panel below them past the right edge of the screen; they wrap now and nothing scrolls sideways. The rare failure page showed a Reload button that did nothing when tapped, so it now reloads, and it explains what happened in plain language with a link to support instead of only printing a developer error string. A malformed response from one of the services behind the page no longer blanks the whole dashboard: the panel that depends on it degrades on its own and the rest still draws. The page also stops requesting a monetization report it never displayed, and its search-engine description finally says what the page shows: revenue over time, revenue by skill, widget views and chats, the conversion funnel, and per-agent performance.",
"tags": [
"fix",
"improvement"
]
},
{
"date": "2026-08-25",
"title": "Describe it to 3D stops offering a retry that cannot work, and finally follows your light theme",
"summary": "Every attempt on this page costs about a minute of GPU time, so the page was handing out the wrong recovery. When your avatar library was full, it showed the failure and a Try again button that burned the same minute and hit the same ceiling; it now links straight to your dashboard to free a slot or to Pricing to raise it, and drops the retry entirely. When the image engine is throttled it quotes the wait the server actually asked for and counts the retry button down in place instead of letting you click into another rejection. A failed build no longer keeps spinning its orb and claiming it is still rendering, a dropped connection says so instead of showing a generic shrug, and the recovery links inside the error box are readable and keyboard focusable. Separately, the page had a light-theme hole: the toggle was live but only half the colours flipped, so the composer, the example chips and the result panel stayed near black on a white page and the main button rendered black text on blue, under the readable-contrast floor. Light mode is now a complete palette, and the site-wide theme switcher no longer decides a page cannot do light before the stylesheet that gives it light has loaded, which was silently serving dark to people who had chosen light on about a third of page loads. The prompt counter, placeholder and build clock were also too dim to read in either theme and are not any more.",
"tags": [
"fix",
"improvement"
],
"link": "/create/prompt"
},
{
"date": "2026-08-25",
"title": "Credits shows your $THREE discount tier, and no longer goes blank when it cannot load",
"summary": "The Credits page had a hole where its error state should be: if the balance request failed, every panel disappeared and the page ended below the heading with nothing on it. There is now a designed card that says what happened, links to platform status, and retries in one click, and a dropped connection reads as a real sentence instead of the browser's \"Failed to fetch\". The page also finally states the thing it promises: a pill showing the $THREE tier you are actually charged at, with the discount you are getting or how much more $THREE reaches the next tier. On a 320px phone the whole layout used to be a hundred pixels wider than the screen with the right edge silently cut off, so nothing over there was reachable; the page now fits, and the history table scrolls inside its own box. Smaller fixes throughout: every button and field has a visible keyboard focus ring, the deposit address has a copy button, the $THREE unit price shows real decimals instead of rounding to $0.00, the amount field and the history controls stay translated when you switch asset or load another page, and the price list dropped the dashes from its labels.",
"tags": [
"fix",
"improvement"
],
"link": "/credits"
},
{
"date": "2026-08-25",
"title": "Crew HQ keeps its 3D room standing, and crew links now open where they point",
"summary": "Crew HQ refreshes who is in the world every 20 seconds, and it used to rebuild the whole room to do it: every member's 3D figure was destroyed and recreated four times a minute, which burns through the handful of live 3D canvases a browser will give a page and could catch a figure mid-load. The room now writes the new state into the figures already standing and only rebuilds when the roster itself changes, and opening a shared crew link no longer flashes the loading skeleton over a room that is already there. A member whose model cannot be loaded now shows their portrait on the plinth instead of an empty space. Signed-out visitors get a clean page: the sign-in-only crew request is no longer fired at all, the crew directory loads beside the room rather than behind it, and it stops listing the crew you are already looking at. A tag nobody flies offers the founding form only to someone who can actually take it, and a dropped connection now says what happened and how to retry instead of showing the browser's \"Failed to fetch\".",
"tags": [
"fix",
"improvement"
],
"link": "/crews"
},
{
"date": "2026-08-25",
"title": "Compare reads correctly on a phone and stops mislabelling gains and losses",
"summary": "The side-by-side coin comparison had three problems worth fixing. Percentage rows rendered every number the same neutral colour, so a loss looked like a gain unless it happened to be the row winner, and the winning cell was painted green even when it was a 37% drawdown; gains are now green, losses red, and the row winner is marked by weight and a tint that reads on either. The performance overlay gained date labels along its bottom edge, and on a phone it now draws at the width of the screen instead of shrinking a desktop-sized chart to a third of its height with unreadable text. Switching range fast could also leave a year of history drawn under a 7-day label; the chart now ignores a slow answer that a newer range has already replaced.",
"tags": [
"fix",
"improvement"
],
"link": "/compare"
},
{
"date": "2026-08-25",
"title": "Connect X no longer dead-ends when X sign-in is switched off",
"summary": "Seeding an agent's memory from X starts by connecting your X account, and on a deployment where X sign-in is not switched on every Connect X button led to a page of raw error text with no way back. Those buttons now say X connections are not enabled and why, in the agent editor's Social tab and the avatar editor's X panel as well as in Settings, and following an old connect link lands you back on the page you started from with the reason stated instead of stranding you. Nothing about consent, seeding, or revocation changes.",
"tags": [
"fix"
],
"link": "/settings"
},
{
"date": "2026-08-25",
"title": "three.ws crossed 100 stars on GitHub",
"summary": "The open-source repo behind three.ws passed 100 GitHub stars: 9,508 commits from 21 contributors, 101 npm packages under @three-ws, 72 MCP servers in the official registry, 4,519 x402 endpoints with 110,416 on-chain settlements, and 3,000 validator attestations on Solana. Thank you for building with us.",
"tags": [
"docs"
]
},
{
"date": "2026-08-25",
"title": "Text-to-3D no longer hangs when the concept painter stalls",
"summary": "Every text-to-3D generation starts by painting a concept image, and that step tried its image providers one after another with no overall time limit. When the first provider stalled, a request could sit for two and a half minutes before answering, and anything with a shorter patience (ChatGPT, agent clients) simply saw nothing. The painter now works to one shared budget: a slow provider hands off with real time left for the next one, and if the whole chain runs out you get a quick, honest retry-in-a-moment answer instead of a hang. Image-to-3D was never affected.",
"tags": [
"fix"
],
"link": "/forge"
},
{
"date": "2026-08-25",
"title": "The first-contribution guide now hands you real open work, and the docs sidebar can find it",
"summary": "The 15-minute contributor guide walked through adding MikuMikuDance rig support as its example, but that support had already shipped, so anyone following it literally would have re-done merged work and been told a bone was unmapped when it was not. It now shows the MMD change as the finished reference, exactly as it landed with its tests, and points you at two rig conventions that are genuinely still open, each with the bones that fail today. The community, triage, and first-contribution pages also appear in the docs sidebar under Guides instead of being reachable only from the README, and the triage page now explains every label the repo actually uses.",
"tags": [
"docs",
"fix"
],
"link": "/docs/first-contribution"
},
{
"date": "2026-08-25",
"title": "The full $THREE thesis is now a single document",
"summary": "A long-form research thesis on $THREE now lives in the docs: what three.ws is, every way the token is used (hold-to-access tiers, plan and pass discounts, marketplace and labor escrow, game cosmetics, gated embeds, deploy-fee waivers), how platform revenue is routed into buybacks, where the token trades and is verified, the ecosystem footprint, the roadmap, and every risk the platform documents about itself, each claim sourced to the code.",
"tags": [
"docs"
],
"link": "/three-token"
},
{
"date": "2026-08-25",
"title": "The First 19 Weeks: everything shipped since day one, week by week",
"summary": "The First 19 Weeks, the founding three.ws roundup, walks every week from the first commit on April 14 to today: what launched, what was announced, which partnerships landed, and where $THREE fits, with the totals (commits, changelog entries, pages, packages, MCP servers, listings) up front. It is written for the timeline, so no tables, and ships with a screenshot kit of the live surfaces.",
"tags": [
"docs"
],
"link": "/docs/the-first-19-weeks"
},
{
"date": "2026-08-25",
"title": "Creator Gallery cards stay put while you scroll, and the leaderboards show they are loading",
"summary": "On /creations, the 3D preview on every gallery card and leaderboard row was accidentally accepting drag input, so a card could be flung out of frame and a swipe over one scrolled nothing on a phone instead of moving the feed. The previews now spin on their own and hand the drag back to the page. The trending and top-creator columns also render shimmer rows while they load instead of sitting blank, the browse tally no longer shows a stale number over a loading grid, and keyboard focus on the search field is now clearly outlined.",
"tags": [
"fix",
"improvement"
],
"link": "/creations"
},
{
"date": "2026-08-25",
"title": "Billing & Passes names your creators, keeps its retry buttons live, and the dashboard fits a phone screen",
"summary": "On /dashboard/billing, creator subscriptions now show the creator's name (linked to their profile), the real renewal or end date, a colour-coded status, and the plan price with its interval, instead of a raw account id. When the passes or subscriptions feed cannot be reached, the Retry button now actually reloads that section. Across every dashboard page, the top bar no longer forces the layout wider than a 320px phone screen, so nothing is clipped off the right edge.",
"tags": [
"fix",
"improvement"
],
"link": "/dashboard/billing"
},
{
"date": "2026-08-25",
"title": "The complete open-source ecosystem, written down in one place",
"summary": "To mark 100 GitHub stars, a new doc maps everything three.ws publishes as open source: 101 npm packages, 72 MCP servers in the official registry, the x402 stack and its 4,519 endpoints, the contracts on 12 chains, the 33 workers, 60 agent skills, 31 specs, every editor and store integration, and what is not published yet.",
"tags": [
"docs"
],
"link": "/docs/open-source-ecosystem"
},
{
"date": "2026-08-24",
"title": "The pump.fun pill mascot is now a rigged, drivable avatar at /pill",
"summary": "pump.fun verified $THREE as officially ours, so we rigged their pill mascot and built it a stage. Walk it around with WASD or a touch joystick, run, jump, wave, dance, and drag to orbit. The model arrived as a static 297,000-triangle capsule with four stubby nubs, which our humanoid auto-rigger cannot read because there is no neck, no head and no long limbs to find. It now carries a 52-bone skeleton with canonical bone names, six clips authored for its own proportions, and 44,544 triangles in a 2.6 MB file, which means the entire three.ws animation library retargets onto it as well. Download the GLB from the page and open it in Blender, Unity, or anything that reads glTF.",
"tags": [
"feature",
"improvement"
],
"link": "/pill"
},
{
"date": "2026-08-24",
"title": "$THREE now carries its pump.fun verification badge, read live from pump.fun",
"summary": "pump.fun verifies a project's coin as officially belonging to that project, which is what tells a buyer the real $THREE apart from the copies that share its name and ticker. That badge now appears next to $THREE on the token page and on the holder dashboard. It is never hardcoded: the platform reads pump.fun's own verification flag on every stats request and renders the badge only while pump.fun publishes it, so a badge on three.ws always means the same thing as a badge on pump.fun.",
"tags": [
"feature",
"improvement"
],
"link": "/three-token"
},
{
"date": "2026-08-22",
"title": "Our Hugging Face article, avatar rigs, and 3D viewer are now linked from the site and the docs",
"summary": "three.ws publishes on Hugging Face, the place the machine-learning world goes to check whether a model claim is real: a community article on giving agents bodies and wallets, a repository of rigged avatars you can download under the MIT license, and a running Space that loads them in an interactive 3D viewer. Until now nothing on the site pointed at any of it. All three are now indexed in the docs with a note on which part of the platform each one demonstrates, listed in the news feed with a working link, and the article joins the press cards at the bottom of the blog. The docs page also carries the checklist for publishing the next one.",
"tags": [
"docs",
"improvement"
],
"link": "/docs/huggingface"
},
{
"date": "2026-08-22",
"title": "Any AI agent can now buy a 3D model from us in one call",
"summary": "Our OKX.AI marketplace listing has been rebuilt around one thing: an agent describes something, and gets back a real 3D model. Four new services (draft at one cent, standard at five, HD and image-to-3D at twenty five) plus a free job-status service, each one a proper MCP server that answers an unpaid call with a payment challenge on the marketplace's own rail. A buying agent receives exactly what our ChatGPT app gives a person: the downloadable GLB, the concept image, a browser preview, and a link that drops the model into a real room in augmented reality. Both surfaces now share one response shaper, so they can never drift apart. The identity studio and the specialist rigging and export endpoints stay live and payable, just off the listing while the forge proves itself. Update, September 2026: OKX has not approved the rebuilt listing yet, so for now agents buy by calling these endpoints directly rather than finding them on the marketplace.",
"tags": [
"feature"
],
"link": "/docs/okx-marketplace"
},
{
"date": "2026-08-22",
"title": "The README now says open source, because it is",
"summary": "Ten places across the README and our published articles still described three.ws as source-available, which understated the Apache 2.0 license the project actually ships under. They now say open source, and the README explains what that buys you: fork it, self-host it, ship commercial work on it. The top of the README also carries a license badge, a live count of good first issues, and a one-line path for anyone who wants to contribute rather than just use it. The security reporting line no longer points at a retired mirror repository.",
"tags": [
"docs"
],
"link": "/docs/first-contribution"
},
{
"date": "2026-08-21",
"title": "three.ws is now open source under Apache 2.0",
"summary": "The repository shipped as proprietary, all rights reserved, while describing itself as an open source framework. That is fixed: three.ws and all 102 of its packages are now licensed under the Apache License 2.0. You may use, modify, distribute, and build commercial products on any of it, including in closed-source work, as long as you keep the license and attribution notices and state what you changed. Apache 2.0 also carries an explicit patent grant from every contributor, which matters for the on-chain contracts in the repo. The contributing guide already promised Apache 2.0 to anyone who sent a pull request, so this makes that promise true rather than changing it. A NOTICE file now records attribution for the vendored and derived code that keeps its own license: the forked avatar builder, the vendored three.js editor, the animation clip library, and the club assets.",
"tags": [
"improvement",
"docs"
],
"link": "/docs/community"
},
{
"date": "2026-08-21",
"title": "MikuMikuDance avatars now animate instead of standing still",
"summary": "Upload a model rigged in MikuMikuDance and it now plays the full animation library like any other avatar. MMD names every bone in Japanese (センター, 上半身, 左腕) and shares no spelling at all with the Latin conventions we already understood, so until now the entire skeleton went unrecognised and the character fell back to a default body. Every joint in the standard PMX skeleton is now mapped, fingers and toes included, along with the two thumb-numbering quirks MMD has. The IK targets and twist bones are deliberately left alone, because driving those would fight the chain they exist to solve and tear the mesh. Rig Doctor also names the convention on sight now, so an MMD upload tells you what it is instead of reporting an unknown rig.",
"tags": [
"improvement",
"feature"
],
"link": "/rig-doctor"
},
{
"date": "2026-08-21",
"title": "A real front door for contributors: community channels, a 15-minute first PR, and how we triage",
"summary": "Three new docs make it obvious how to join in. Community lists every room we are in (GitHub Discussions, the Telegram community, the X community, the IBM Community group, the release channel) and which one fits which question. Your first contribution walks from clone to open pull request in about 15 minutes with a complete worked example: teaching the avatar retargeter a bone-naming convention it did not know, plus the test that proves it. Triage documents how issues and pull requests get sorted, what every label means, how to claim work so nobody duplicates it, and the response times you can hold us to. The README, the contributing guide, and the new-issue chooser all point at them now.",
"tags": [
"docs"
],
"link": "/docs/community"
},
{
"date": "2026-08-20",
"title": "AR now actually opens AR on iPhone: AR Forge and AR Studio, not just avatars",
"summary": "Placing a model from AR Forge or the multi-model AR Studio in your real room worked correctly on Android, but iPhone quietly got a downgrade: AR Forge fell back to the plain 3D viewer, and AR Studio faked it with a camera overlay that had no real-world plane detection, no correct scale, and nothing to rest the model on the floor. Only the avatar AR page did this right. Both surfaces now use the same real pipeline: a genuine USDZ is generated on your device (the same three.js conversion the avatar page always used) and handed to Apple Quick Look, so the model is actually tracked to your room, sized correctly, and sits on your real floor. AR Studio's AR button now also offers Google Scene Viewer directly instead of only the raw WebXR session. Shared AR links still unfurl with a real render of the model when pasted into X, Discord, or iMessage.",
"tags": [
"fix",
"improvement"
],
"link": "/ar"
},
{
"date": "2026-08-19",
"title": "Our AR studio is now a package anyone can drop into their own site",
"summary": "The multi-model AR surface behind three.ws is now a standalone open-source package: 3d-ar-studio on npm, and a hosted demo at nirholas.github.io/3D-AR-Studio. One script tag gives any web page a working augmented-reality studio: place any number of 3D models in your real room through the camera, drag, pinch and rotate them, generate new ones from a text prompt without leaving the camera view, then share the whole arrangement as a link or a QR code, or open a live room and build it with someone else. It carries the full rendering ladder we run here, WebXR placement with per-model anchors, real-world light estimation and depth occlusion on Android, camera passthrough with gyro world-lock on iOS, and a framed preview with a QR hand-off on desktop, plus the universal idle retarget so humanoid models animate instead of standing in a T-pose. Out of the box it pulls models and free keyless text-to-3D from three.ws, and a single option repoints it at anyone's own catalogue. It ships with an MCP server (published to the official registry) so an agent can generate a model, arrange a scene, and hand a person one link that opens it in their room, and a two-command CLI that puts a live AR page on GitHub Pages.",
"tags": [
"feature",
"sdk"
]
},
{
"date": "2026-08-19",
"title": "Deploying an agent on-chain now buys $THREE, and holders deploy free",
"summary": "The Metaplex Agent Registry deployer (the browser page and the @three-ws/metaplex-agent-mcp server) now carries a flat 0.02 SOL deploy fee on mainnet, and that fee is paid to the wallet the $THREE buyback lane spends from. So every agent someone puts on-chain through us turns into buy pressure on $THREE instead of nothing. Holding $THREE is what makes it cheaper: 50,000 in the paying wallet halves the fee, 250,000 waives it entirely, and the balance is read live from the chain when the transaction is built, so nothing is staked, locked, or spent to earn it. Deploy a whole fleet and the waiver applies to every one. Devnet rehearsals stay free, the fee travels inside the same transaction that creates the agent (so a deploy that fails costs nothing), and both the amount and its destination are shown before anything is signed. A new three_status tool prices your next deploy against any wallet and returns the live $THREE market and buyback figures.",
"tags": [
"feature",
"sdk"
],
"link": "/deploy-onchain"
},
{
"date": "2026-08-19",
"title": "Agents can now hold a spending allowance instead of your private key",
"summary": "Handing an AI agent a wallet has always meant handing it a private key, which means handing it everything in that wallet. The new onchain-agent-wallets MCP server replaces that with an allowance. Your money stays in an account you own, and the agent gets permission to spend a slice of it that the Solana token program enforces on-chain: it cannot spend past the ceiling however it is prompted, every spend decrements what is left, and you can cancel the whole thing in one instruction or take the balance back without the agent's cooperation. On top of that you set the rules it runs under: a maximum per transaction, a maximum per day, which addresses it may pay, which APIs it may buy from, an expiry, and a pause switch that stops it instantly for free. It can pay x402 APIs out of the allowance, and the price is always read before any money moves. Every decision, including every refusal and the reason for it, is written to an audit log you can read. You never have to put a secret key on the machine: owner actions come back as an unsigned transaction for Phantom, Solflare, or a Ledger to sign.",
"tags": [
"feature",
"sdk",
"security"
]
},
{
"date": "2026-08-19",
"title": "Every three.ws package on npm is current again",
"summary": "Our npm packages had drifted a long way behind the code: 73 of them, including the main SDK, the React components, the avatar and walk SDKs and most of the MCP servers, were sitting on npm at an older version than what we had actually built. Anyone installing from npm got stale code. All 73 are now published at their current version, and four packages that had never been released at all (the agent runtime, the sign language pack, the spatial MCP server and the VS Code extension) are on npm for the first time. The cause was our release script working from a hand written list, so any package added after that list was written was silently never published; it now discovers packages itself, so a new one cannot be forgotten again.",
"tags": [
"sdk",
"fix"
]
},
{
"date": "2026-08-19",
"title": "The docs header no longer cuts links off on a phone",
"summary": "On a narrow phone screen the docs header packed five site links beside the logo, so the last ones ran off the right edge: Discover showed as a lone D and Launch app was gone entirely, with no way to scroll to them. The header now keeps the logo, the 3D world button and Studio, and moves Widgets, Discover and Launch app into the navigation drawer, right under the docs search, where they are reachable with a thumb. Nothing was removed: the full header returns on wider screens.",
"tags": [
"fix"
],
"link": "/docs"
},
{
"date": "2026-08-19",
"title": "A page that fails to load now tells you which thing went wrong",
"summary": "When an avatar or agent page could not load, it always said the same thing: not found, followed by whatever internal message the browser happened to throw. A stale link and a temporary server hiccup looked identical, and the explanation was often developer jargon rather than English. The two cases are now told apart. A link to something that no longer exists says exactly that and points you at the marketplace or the agent directory. A genuine failure on our side says so, offers a Try again button, and keeps the technical detail as a small footnote instead of presenting it as the reason.",
"tags": [
"improvement",
"fix"
]
},
{
"date": "2026-08-19",
"title": "Phone fixes: helper widgets no longer sit on top of the controls",
"summary": "On a phone the floating language control and the Getting started pill landed directly on the page's own bottom controls: the language picker rendered inside the \"Ask the agent\" box and the pill covered the save row. Helper widgets now measure a page's bottom bar and float above it, and the language control shows a compact code (EN) instead of a full-width bar. Gallery cards show all three actions in full, so Animate is no longer clipped mid-word. In the AR Studio, the Add models sheet keeps its tabs at full height on short screens, and each model is labelled with a readable name instead of the whole generation prompt. On the viewer page a long email no longer pushes the account button off the edge of the screen.",
"tags": [
"fix",
"improvement"
],
"link": "/gallery"
},
{
"date": "2026-08-19",
"title": "Agent pages show what an agent can actually do, and what it charges",
"summary": "The studio page every agent link opens now carries the agent's real substance. Its Capabilities tab lists the skills it genuinely exposes with the real per-call price of each, and a summary line telling you the cheapest paid call up front (or that everything is free). Skill names read as words instead of raw slugs, with the exact slug shown underneath for anyone wiring an integration. Under the title there is now a row of real counters: skills, conversations, wallet, whether the identity is on-chain, and how long the agent has been around. Below the fold, an agent recommends similar agents rather than unrelated avatars, each rendered as a live 3D card. The page also loads better: a skeleton in the real layout instead of a line of text, cards that show a picture while their 3D model downloads, and full keyboard support on the section tabs with a shareable link per tab.",
"tags": [
"improvement",
"feature"
],
"link": "/agents"
},
{
"date": "2026-08-19",
"title": "Put any agent on-chain from your AI assistant: the Metaplex Agent Registry MCP server",
"summary": "A new npm package, @three-ws/metaplex-agent-mcp, lets any AI assistant deploy an agent on-chain on Solana the exact way the three.ws Genesis 333 shipped: it mints a Metaplex Core asset and registers its EIP-8004 agent identity, so the agent appears on metaplex.com/agents with its own built-in wallet. Agents can sign with their own key, and people can sign with Phantom, Solflare, or any Solana wallet, with a preview-first confirm step before anything is spent. Every royalty, creator, plugin, and metadata field is customizable, and devnet rehearsal is free.",
"tags": [
"sdk",
"feature"
],
"link": "/deployments"
},
{
"date": "2026-08-19",
"title": "Shared avatar links now unfurl with the avatar itself, and profile views finally count",
"summary": "Paste an avatar link into X, Discord, Telegram, or Slack and the preview now shows that avatar's rendered image, name, and description instead of the generic site card. The profile page also reports a view each time someone opens it, so the view counts that rank search and explore results reflect real traffic at last. Discover cards for agents without a coin no longer point at a dead address, and the sitemap hands search engines the current agent page URLs directly.",
"tags": [
"improvement",
"fix"
]
},
{
"date": "2026-08-18",
"title": "Agents and avatars now open in the same studio page",
"summary": "Clicking an agent used to land you on a long scrolling profile, while clicking an avatar opened the 3D studio. They are now the same page. Every agent link on three.ws (the search bar, the agents directory, the gallery, the marketplace, the galaxy, launches and creator profiles) opens the studio layout: the agent standing in a live 3D stage on the left, its identity, wallet and actions on the right, and tabs for chat, poses, skills, plugins and embed codes. Talking to an agent there uses that agent's own brain and memory, and AR now works from an agent page too. An agent that has no 3D body yet gets a proper page instead of an empty viewer, with both ways to give it one. The full agent profile (capabilities, economy, activity, trust and developer tools) is one click away on the Profile button, and every entity now has a single canonical link that is safe to share.",
"tags": [
"improvement",
"feature"
],
"link": "/agents"
},
{
"date": "2026-08-18",
"title": "Avatar and agent pages load in Safari again",
"summary": "Every avatar and agent page failed to load in Safari, on both iPhone and Mac, showing \"Avatar not found\" with the message \"Cannot access uninitialized variable.\" instead of the 3D stage. Chrome, Edge, and Firefox were unaffected, which is why the pages looked fine on most desktops while iPhone visitors saw a dead page. Safari checks a variable is ready at a slightly earlier moment than other browsers do, and the page set up its data one step before that check was satisfied. The page now finishes declaring its state before it starts loading, so it renders identically in every browser.",
"tags": [
"fix"
]
},
{
"date": "2026-08-18",
"title": "Every agent and avatar now opens its profile page, wherever you click it",
"summary": "Clicking an agent or avatar used to land you in different places depending on where you found it: search results took a detour through two redirects, the quick-search palette linked every agent to a dead URL, gallery thumbnails dropped you into the bare 3D viewer, and older links still used a retired address. Now every surface, including search, the quick-search palette, the gallery, the homepage showcase, your feed, and creator profiles, links straight to the entity's profile page: the one with the 3D stage, chat, wallet, skills, and embed tools. Agents share the same page style as avatars, so the platform feels like one place instead of six.",
"tags": [
"improvement",
"fix"
]
},
{
"date": "2026-08-17",
"title": "The AWS Marketplace integration is rebuilt for the rules AWS now enforces",
"summary": "Subscribing to three.ws through AWS Marketplace would have failed on the first click, and nobody would have known why. AWS changed how a buyer is identified: for any product created since June 2026 it no longer sends the customer identifier our integration was built around, sends a license identifier instead, and delivers cancellation notices over a different channel entirely. The whole path is rebuilt on the new model, from the database up: subscriptions are keyed on the license AWS actually issues, lifecycle events arrive over the new channel with their own authentication, and a cancellation that could refer to more than one of a buyer's subscriptions now refuses to guess rather than cutting off access someone is still paying for. Signature checking also follows the algorithm the message declares instead of assuming one, so a genuine notice can no longer be thrown out as a forgery. Listing status is unchanged for now: the product still has to be created and reviewed on the AWS side, which is theirs to schedule, not ours.",
"tags": [
"fix",
"infra"
],
"link": "/docs/aws-marketplace"
},
{
"date": "2026-08-17",
"title": "The listings page now tracks the AI platform directories, starting with OpenAI",
"summary": "The listings and distribution page tracked cloud marketplaces, startup programs, and chain dApp directories, but had nothing for the directories the AI platforms run themselves, which is where an MCP server gets listed as something you can install rather than something you can read about. There is now an AI Platform Directories section, and it opens with the three OpenAI surfaces: the plugin directory shared by ChatGPT and Codex, the Showcase Gallery, and the Cookbook, each with its current status stated plainly rather than implied. The three.ws MCP server at three.ws/api/mcp already meets the plugin directory's gating requirement, and the section says so.",
"tags": [
"docs"
],
"link": "/docs/listings"
},
{
"date": "2026-08-17",
"title": "The listings page now records our NVIDIA Inception membership",
"summary": "Anyone reading the listings and distribution page to see which programs back three.ws was getting an incomplete answer: the startup-programs table named Quicknode and Google Cloud but left NVIDIA Inception out entirely, even though the platform has been a member since July 2026 and every 3D generation lane runs on NVIDIA GPUs. The table now carries the NVIDIA row, saying what the membership is, what runs on the hardware behind it, and, as everywhere else on the site, that a startup program is not a partnership, an investment, or an endorsement.",
"tags": [
"docs",
"fix"
],
"link": "/docs/listings"
},
{
"date": "2026-08-17",
"title": "One car in the world: the Trench Car now drives every road in /play",
"summary": "The cars in a coin world no longer come in four unrelated shapes. Alongside the Trench Car, the community model from the avatar gallery, /play was also parking three procedural box cars in its depots: a boxy sedan, a pickup, and an amber roll-caged buggy on oversized wheels that looked nothing like anything else in the town. All three are gone. Every bay in every world, plaza and depot alike, now parks the Trench Car, so the car that drives past you on the ring road is the car you walk up to, press F, and drive away in. Frontier towns are included: their traffic used to be horse-drawn covered wagons, and it is the same car as everywhere else now. The car still appears the instant a world loads, as a plain silhouette on the same wheels, and fills in with the real model the moment it finishes downloading, so nothing waits on the asset and a slow connection never leaves an empty road.",
"tags": [
"improvement",
"fix"
],
"link": "/play"
},
{
"date": "2026-08-17",
"title": "Trading Swarms reads on a phone, says what actually went wrong, and lets you get back to the full directory",
"summary": "The swarms page has had an end-to-end pass. When something failed, it used to print the server's internal code at you: a swarm that no longer exists said \"not_found\", and a rejected contribution said \"bad_agent\". It now shows the sentence the API was already sending, so a failed join tells you what to do about it. \"My swarms\" was a one-way door with no way back to the public list; it is now a proper two-state switch next to the network switch, and both live in the address bar, so a filtered view is a link you can share and the back button behaves. Every swarm card is a real link too, so you can middle-click or cmd-click one open in a new tab. On a 320px phone the cards and the dashboard panels used to be cut off at the right edge; they now fit. The join, create, and kill dialogs take keyboard focus when they open, hold Tab inside themselves, and hand focus back to whatever you opened them from, and signing in or creating an agent is now offered as a button rather than named in a toast you cannot act on.",
"tags": [
"improvement",
"fix"
],
"link": "/swarms"
},
{
"date": "2026-08-17",
"title": "The coin-community lobby tells you which worlds people are actually standing in",
"summary": "The front door at three.ws/play has been rebuilt around the question it exists to answer: which of these coin worlds is worth walking into right now. It opens with the claim and three live numbers (how many communities are listed, how many people are inside a world at this moment, and what those communities are worth together), and every one of them is measured rather than decorated: the headcount comes straight from the multiplayer matchmaker, and when that cannot be read the number is left out instead of guessed. Each card now carries its own live count, so a world with nine people in it says so, and the search field has grown into a full-width command bar with a / shortcut and a clear button. Above the grid there are four orderings that are real reorders of real fields: the live trending order, most people inside, biggest market cap, and newest launch; the headcount ordering stays switched off, and says why, while live counts are unavailable. Cards gained the details you actually pick a world on (how old the coin is, whether it launched in the last day, whether it is still on its bonding curve, and how loud its board is), and lost the badge that was true of almost every coin and therefore told you nothing. The avatar picker, which used to be a tall panel standing between you and the worlds, is now a single row showing the avatar you will be seen as and the name above your head, folded away for anyone who has already chosen and wide open for anyone who has not. A card whose artwork fails to load keeps the coin's initial instead of leaving a grey hole, the grid fades in in reading order, and the whole page holds together from a 390px phone to a desktop.",
"tags": [
"improvement",
"feature"
],
"link": "/play"
},
{
"date": "2026-08-17",
"title": "Studio stops going quiet when the engine list fails, and a broken preview no longer leaves a wall of broken images",
"summary": "The Studio at three.ws/forge-studio was driven end to end in a real browser: both text-to-3D pipelines, all three tabs (Object, Avatar and Lab), a real generation from prompt to finished GLB, and the loading, empty, error and populated states at phone, tablet and desktop widths. The finding that mattered was silent. Studio asks the server which generation engines are live before it draws the Quality and Engine pickers, and if that request failed for any reason, the whole row of controls simply vanished with nothing said. You were left on a page that looked complete but had quietly lost the choice between Draft, Standard and High, and every engine behind it, with no hint that anything had gone wrong or any way to get it back short of a reload. A failed engine list now says so where the pickers used to be, explains that generation still works on the free engine, and offers a Retry that brings the full picker back in place. A server that genuinely has no engines to offer stays quiet as before, because that is not a failure. Second, when a thumbnail in the community showcase could not load, the card started rendering a 3D frame from the model file instead, which is the right answer, but it left the broken image sitting on screen for the whole wait. Those renders are queued one at a time behind a twenty second limit, so a bad network turned the showcase into a grid of broken-image boxes for minutes. The card now falls back to its readable gradient-and-prompt tile the instant the image fails, and the 3D render replaces it when it arrives. The rest is the polish an audit turns up: the Avatar tab carried two competing page headings, so a screen reader heard the page rename itself when you switched tabs; the Lab tab's Download button now tells assistive tech it is disabled and why, instead of just looking dim; the photo and sketch file pickers finally have names; seven pieces of small grey text across the Avatar and Lab tabs (the character counter, the elapsed timer, the sign-in hint, the slider read-outs and the footnotes) were too dim to read against the near-black background and now use the same secondary colour as the rest of the page, which clears the accessibility contrast floor on all three tabs; and a freshly generated model now actually previews, where before the viewer, the reveal animation and the game-ready poly counter all failed to read the finished file outside production.",
"tags": [
"fix",
"improvement"
],
"link": "/forge-studio"
},
{
"date": "2026-08-17",
"title": "The Money Flow Map stops cutting its own table off on a phone, and a refresh no longer blanks the map",
"summary": "The map of who pays whom at three.ws/flow was driven end to end in a real browser: every window, every filter, the legend toggles, search, zoom, the wallet table and the loading, empty and error states, at phone, tablet and desktop widths. Two findings mattered. On a phone the page was quietly cutting itself in half: the \"What the money bought\" panel and the entire wallet table were laid out wider than the screen and then clipped away, with no way to scroll to them, so the text version of the graph (the part a screen reader and a keyboard actually use) was unreachable on mobile. The table now scrolls inside its own frame and nothing runs off the edge. Second, the page re-reads the chain every minute, and every one of those re-reads used to cover the drawn map with a loading spinner, so a live page flickered into looking broken once a minute. A background re-read now leaves the map alone and reports itself on the timestamp next to the controls, and if it fails, the graph you were reading stays on screen with the failure named there instead of being wiped for an error panel. Switching the time window still shows the loading state, because that genuinely replaces the topology, and it now cancels a read that is still in flight: a slow 90-day request could previously land after you had already switched to 7 days and draw a window you were not looking at, under a button that said something else. The rest is the polish an audit turns up: the table's own caption said \"no wallets match the current filters\" even in a quiet window where you had set no filters, and now tells those two situations apart; the empty state no longer offers a \"Show 90 days\" button while you are already on 90 days, and offers the next longer window instead; the clear-search cross stayed on screen after Clear filters had already emptied the box; and the canvas now resolves its colours once per frame rather than several times per line, which is what a four-hundred-route graph feels like on a laptop.",
"tags": [
"fix",
"improvement"
],
"link": "/flow"
},
{
"date": "2026-08-17",
"title": "The Agent-Forged Gallery says what actually went wrong, and its thumbnails stop fighting the link they sit in",
"summary": "The gallery of 3D props that the platform's autonomous agents bought with real USDC was driven end to end in a real browser: a hundred live cards, the category chips, prompt search, all four sorts, and the loading, empty and error states at phone, tablet and desktop widths. The finding worth caring about was in the error state. When the feed failed with a server error rather than a dropped connection, the page still said only \"Something went wrong\", because the translation pass ran after the failed request and overwrote the specific reason the page had just written there. The reason now lives on its own line that translation leaves alone, so a failure tells you what it was and the Try again button next to it recovers the full gallery. Each prop's spinning preview had also been set up to drag-orbit under the link that wraps it, from an attribute written to do the exact opposite; the previews are now purely decorative and the whole tile is just a link to the 3D viewer. The category chips announced themselves to screen readers as tabs without being tabs, and never said which family the grid was filtered to; they are now a labelled group of toggle buttons that report their own state. Every control on the page (search box, sort, chips, the Preview and Download buttons, the thumbnail itself) now shows a focus ring when reached by keyboard, where the search box previously had its ring switched off entirely. A prop still waiting on its settlement no longer renders an empty price, and one whose model file is not written yet no longer offers a Preview and Download that lead nowhere.",
"tags": [
"fix",
"improvement"
],
"link": "/forged"
},
{
"date": "2026-08-17",
"title": "watsonx Constellation: stars actually open when you click them, and the page says which model really answered",
"summary": "The galaxy of live Solana tokens at three.ws/constellation had a broken centrepiece. Clicking a star only opened its panel if the click both landed within a pixel or two of the sphere's geometry and finished inside 450 milliseconds, so a slightly slow click, a press held a moment too long, or any tap on a phone silently did nothing. A click is now judged the way every other interface judges one: if the pointer ends where it started, it is a click however long you held it, and it opens the nearest star to where you aimed rather than demanding pixel precision on a dot a few pixels wide. The page also stopped taking credit it had not earned. The layout is meant to come from IBM Granite embeddings on watsonx.ai and the analysis from IBM Granite, but both have automatic failover chains, and the page labelled every result as Granite regardless of what actually served it. The status line now names the embedder that really placed the stars, and the analysis panel names the model that really wrote the text, saying plainly when watsonx Granite could not serve the request. Clicking a star while signed out used to end at \"Analysis unavailable (unauthorized)\"; it now explains that live analysis needs an account and links straight to sign-in and registration, with the galaxy, the live prices and the semantic layout staying open to everyone. The whole surface is reachable from the keyboard for the first time: arrow keys walk the stars in trending order, Enter opens one, Escape closes it and returns you to the galaxy, and a screen reader hears each star announced as you move. A failed or empty token feed now offers a Try again button instead of a dead message, the panel leads with the market cap the feed actually carries rather than a price it usually does not, the panel no longer holds keyboard focus while it is closed, and the hint bar no longer runs off the screen or under the language switcher on a phone or tablet.",
"tags": [
"fix",
"improvement"
],
"link": "/constellation"
},
{
"date": "2026-08-17",
"title": "Chat is the chat again: the fake marketing pages inside /chat are gone",
"summary": "The chat app at three.ws/chat had grown a second copy of a marketing site inside it, and none of it was real: a pricing page with invented plans that did not match our actual pricing, a trust centre claiming SOC 2, ISO 27001, HIPAA and PCI certifications we do not hold, blog posts and product updates nobody wrote, customer case studies with made-up companies and numbers, webinars and conference appearances with made-up staff names and dates, and a banner announcing an acquisition that never happened. The top navigation had already stopped linking to any of it, so it survived only as old links people could still land on. Every one of those pages is now removed, and every old link redirects to the real page that covers the topic: pricing to three.ws/pricing, updates to the changelog, blog to the blog, docs to the docs, security to the security doc, events to the community page. Sign-in got the same treatment: the email and password form used to accept anything you typed and drop you into the chat without signing you in at all. It now signs you in for real against the same account system as the rest of three.ws, shows you why a sign-in was refused, and asks you to accept the Terms when you create an account. The earnings page tells a signed-out visitor to sign in instead of reporting a load failure, buttons that did nothing (import a template, import from Figma, apply a code block, pick an image model that was never used) either do something now or are gone, and the deck templates and slide-count picker actually shape what you get back. The page also loads about eight times lighter: the removed in-app docs browser was bundling every documentation file into the app on first load.",
"tags": [
"fix",
"improvement",
"security"
],
"link": "/chat"
},
{
"date": "2026-08-17",
"title": "Tabbing into the Forge prompt box now shows you where you are",
"summary": "The Forge composer (three.ws/forge and the maximum-quality lane at three.ws/forge-max) draws its describe-it box, its photo-guidance box and its sketch-description box without a border, so the card reads as one clean surface. That styling had also switched off the browser focus outline, which meant anyone moving through the page with the Tab key landed in the main prompt field with no visible sign that they were there: the page looked identical whether the box had keyboard focus or not, on the single most important input on the surface. All three boxes now draw a focus ring when you reach them by keyboard, and only by keyboard, so clicking in with a mouse keeps the seamless look it always had. The rest of the lane was driven end to end in a real browser at phone, tablet and desktop widths: High stays pinned when you open /forge-max, the quality switches move between Draft, Standard and High with the holder gate reflected on the Generate button, and a busy or failing generator lands on a designed message with a retry rather than a blank panel.",
"tags": [
"fix",
"improvement"
],
"link": "/forge-max"
},
{
"date": "2026-08-17",
"title": "Smart Money Radar stops shearing coin cards on a phone, and a lost connection no longer masquerades as a coin nobody is buying",
"summary": "The /smart-money page was driven end to end in a real browser at 320, 768 and 1440 pixels. On a narrow phone the coin grid insisted on a column wider than the screen, so the right edge of every card was silently cut away by the page: the pedigree score badge and the star that adds a coin to your watchlist were both off-screen, with no sideways scroll to reach them. Cards now fit the screen they are on, and the score and the star are back. The bigger problem was that the page could not tell a coin it had never scored apart from a request that never arrived. If your connection dropped, or the radar was briefly down, every coin on your watchlist was relabelled Not on the radar yet, which reads as a verdict on the coin rather than a problem with the connection, and opening any coin or wallet gave you the same flat Not scored yet with no way to try again. The page now separates the two. A genuine miss keeps its explanation and now says why, and adds a link out to pump.fun or Solscan so the trail does not end there. A connection failure says the radar could not be reached and gives you a Retry button, in the drawer, on the watchlist, and on individual watched coins, and your watchlist is explicitly noted as still saved on your device. A first load that fails outright used to leave the Top wallets tab as an empty void; it now carries the same message and its own retry. A proven wallet whose trades predate the detail the radar keeps used to show a bare No recent coins on record, which looked like a wallet that had stopped trading; it now explains that the coin-level detail has aged out while the win record above still counts every verdict.",
"tags": [
"fix",
"improvement"
],
"link": "/smart-money"
},
{
"date": "2026-08-17",
"title": "A busy Forge engine now says it is busy and counts you back in, instead of telling you to go set an environment variable",
"summary": "The /forge page was driven end to end in a real browser. When every free 3D engine was momentarily at capacity, or a worker was still warming up, the server said so clearly and told the page to try again in about half a minute. The page threw that away. It treated any temporary outage as proof that the whole generator had never been set up, and showed a panel headed Generator not configured that asked you to set NVIDIA_API_KEY or HF_TOKEN on the deployment. For anyone who is not running the servers, which is everyone using the site, that is an instruction you cannot act on, about a problem you do not have, and the panel had no Try again button at all, so a wait of thirty seconds became a dead end that looked like the product was broken. A temporary outage now lands in the wait state the site already had for this: a heading that says Temporarily unavailable, the countdown ticking down to when the engines free up, and a Retry button that switches itself on the moment the wait is over. The genuine case, a deployment with no engine wired at all, still gets its original panel, because for the operator that message is the correct one. The engine picker also had a key underneath it explaining that a red dot means an engine is unavailable, but no engine could ever show that dot: only the amber busy dot was ever drawn, so a down engine was dimmed with no indication why, and the key described something that did not exist. Down engines now carry the red dot the key promises, at full strength so it stays distinct from the amber one. Separately, roughly one in ten preview thumbnails failed to load while developing locally, which made the gallery look like a broken image host when nothing was actually wrong with it.",
"tags": [
"fix",
"improvement"
],
"link": "/forge"
},
{
"date": "2026-08-17",
"title": "Skill tags now actually filter, a failed skills load says so instead of claiming the catalog is empty, and a shared skills link brings its filters with it",
"summary": "The /skills path was driven end to end in a real browser. It forwards into the Marketplace Skills tab, and it was forwarding a search that the tab then threw away: a link like /skills?q=whale landed on the full unfiltered list with an empty search box. The Skills tab now reads its search, category, tag, and sort straight off the address bar, and writes them back as you change them, so the view you are looking at is the view you can bookmark, share, reload, and step back through. The tag chips on a skill were the clearest casualty of that gap: clicking one navigated to a list that looked exactly the same as before, because nothing was filtering by tag at all. Tags now filter for real in the catalog itself, the tab shows which tag it is filtering by, and one click clears it. When the catalog cannot be reached, the grid used to show the words No skills found with an invitation to publish one, which is the same screen an empty catalog would give you and offers no way to try again: the error state and its Retry button were being overwritten a fraction of a second after they rendered. A failed load now says it could not load skills and hands you the retry. The count under the heading was also being reverted to its placeholder by the translation pass a moment after it was written, so it never showed how many skills you were looking at; it does now, and it names your active category or tag. The empty state names the filters that emptied the list instead of guessing. For anyone arriving without JavaScript, including crawlers, /skills was a completely blank page with no link out of it; it now carries a real heading, an explanation, and a link into the Marketplace. The search box has a proper label, the free and paid chips announce which one is selected, and the page holds up from 320px to 1440px.",
"tags": [
"fix",
"improvement"
],
"link": "/marketplace?tab=skills"
},
{
"date": "2026-08-17",
"title": "Self-hosted NIM says up front that no endpoint is wired, and a finished reconstruction actually appears on screen",
"summary": "The /forge-nim page drives a TRELLIS NIM you host yourself, and it was driven end to end in a real browser. It already asked the API on load whether a NIM was configured, then threw the answer away: there was nowhere on the page to show it. So a first-time visitor dropped a photo, waited, and got a failure at the end of the trip, and the failure read as the raw code nim_unconfigured because the page was reading the machine code instead of the sentence next to it. The page now carries a status pill that says whether a NIM is wired and reachable, and when none is, the stage explains that this page needs a NIM you host, offers a button that opens the endpoint field, and links to the hosted forge and the NVIDIA model docs in the meantime; the Reconstruct button stays switched off rather than promising a trip that cannot arrive. Typing your own endpoint now re-probes that box and reports what it found, and an address the server refuses comes back with the reason instead of a silent failure later. The bigger fix was invisible until a reconstruction finished: the 3D viewer had collapsed to zero height, so the model was never even loaded and a completed job showed an empty stage under its own result bar. The viewer now fills the stage. The button also never showed its loading spinner, because translating the page had quietly replaced the label it was written into. The page carries a real heading, the loading and error states announce themselves to screen readers, every control has a visible keyboard focus ring, and it holds up at phone width.",
"tags": [
"fix",
"improvement"
],
"link": "/forge-nim"
},
{
"date": "2026-08-17",
"title": "Text to 3D gets its footer back, a retry button that knows when to open, and an honest name for the model that refined your prompt",
"summary": "The /forge-spark page was driven end to end in a real browser. The closing note under the workspace was written as a page footer, and the site's old corner-credit styling for the 3D viewer pages had quietly claimed it: it was pulled out of the layout into a floating monospace strip that sat on top of the viewer on a desktop, and on a 320px phone it landed across the composer with its link to the full studio pushed off the edge of the screen entirely. That note is now ordinary page text, centered under the workspace where it belongs, and the real three.ws footer, with its full set of links, is on the page for the first time. When a generation fails there is now a Try again button inside the error itself, so recovery does not mean scrolling back to the top; when the generator answers that it is busy, the button counts down and opens exactly when a retry can actually get through. A stage that fails is marked in red instead of silently dropping back to grey, the wait now shows real elapsed seconds so a cold reconstruction lane no longer reads as a hang, and the first step is renamed to whatever model actually answered, because the prompt refiner is a chain of engines and Nemotron is only the first one it tries.",
"tags": [
"fix",
"improvement"
],
"link": "/forge-spark"
},
{
"date": "2026-08-17",
"title": "Talking Avatar Video stops handing the renderer a 3D file where a photo belongs, and says up front when the GPU lane is down",
"summary": "The /create/video page turns one of your avatars plus an audio clip into a lip-synced video, and it was driven end to end in a real browser: every state, every button, every width. The worst finding was invisible from the page. The reference frame the GPU worker renders from was being taken from whatever the browser had handy, which for an avatar with no thumbnail meant its 3D model file. The worker saved that as the reference photo and then failed on it minutes later, after the wait had already been spent. The page now sends only the avatar, and the server resolves a real image for it: the stored thumbnail, or a fresh portrait render when there is none. Second, the GPU renderer is a separate service, and when it is not running the only signal was a failure after you had already picked a file and uploaded it. The page now asks first and, if the lane is down, says so at the top, keeps Generate switched off, and points at the Avatar Studio and prompt-to-avatar in the meantime. The rest is the ordinary polish an audit turns up: a failed avatar list used to look identical to owning no avatars, so it sent people off to create one they already had, and now it says what broke and offers a retry; having no avatars yet gets a real empty state with a link to make one; the Generate button always explains what it is still waiting for and, once audio is in, how many clips of video that audio is worth; the clip you picked can be played back with its length and size before you spend a generation on it; a render job that goes unreachable now says so after thirty seconds instead of spinning for twenty minutes; and after a spent free trial the button under the error no longer stayed stuck on Upgrade plan for every later error. Avatar thumbnails are real buttons with visible focus rings, the two form fields have proper labels, and the second animated avatar that the site companion used to drop onto this page no longer loads over the one you are working on.",
"tags": [
"fix",
"improvement"
],
"link": "/create/video"
},
{
"date": "2026-08-17",
"title": "The Pole Club audio controls no longer fall off the edge of a small phone",
"summary": "On a narrow phone the club's top bar ran out of room and simply cut off whatever did not fit, and the page cannot scroll sideways, so the mute button and the 8D audio toggle were gone with no way to reach them. The bar now gives up the parts that are only there to read (the tagline and the price and rail labels, all of which the panel's How it works section repeats) and keeps both controls at full size at every width, down to a 320px screen.",
"tags": [
"fix"
],
"link": "/club"
},
{
"date": "2026-08-17",
"title": "Scene Composer works on a phone, starts you off with real models, and answers the keyboard",
"summary": "Scene Composer was unusable on anything narrower than a laptop. Below 900px the scene panel was switched off, which pushed the 3D canvas into a zero-width column, so a tablet visitor got no viewport at all and no way back to the hierarchy, inspector or bone-attach controls; at phone widths the toolbar forced the whole editor wider than the screen and shoved the right-hand panel out of view. The panel is now a drawer behind a toolbar button, the toolbar scrolls instead of stretching the layout, and the canvas keeps real dimensions at 320, 768 and 1440 pixels. The Recent Creations strip also no longer greets a first-time visitor with an empty box: until you have forged something it shows the live community showcase, so there are real models to drop into the scene straight away. The avatar picker was asking the directory a question it does not answer and got back a feed of on-chain agents with no 3D model attached, so most slots were silently dropped and the rest never showed their thumbnails; it now asks for public 3D avatars and renders them. The start panel finally honours what it always promised: paste an avatar ID, not just a URL, and a model that cannot be loaded leaves a written explanation instead of a toast that vanishes before you can read it. Both the creations strip and the avatar picker now offer a Try again that works when a request fails. On the keyboard side: the page has a real heading, the camera-view menu and every creation and avatar tile are proper buttons, the scene rows expose their hide and remove actions on focus, the prompt and avatar fields have labels, dialogs trap focus and hand it back on Escape, and there is a visible focus ring throughout. Removing an object now frees its GPU memory once the step drops out of undo range, undo puts a bone-attached item back on its bone, and saving an outfit refuses a locally dropped file rather than storing a link that breaks on reload.",
"tags": [
"fix",
"improvement"
],
"link": "/compose"
},
{
"date": "2026-08-17",
"title": "Building an agent no longer throws you out to the login page, and the wizard answers the keyboard",
"summary": "The six-step agent builder at /create-agent was audited end to end in a real browser: naming an agent, generating a whole spec from one sentence, choosing a body from the starter library, the skills and personality steps, the voice step, and the review screen, plus the loading, empty and error states at phone, tablet and desktop widths. The worst finding was that the wizard broke its own promise. It is built so anyone can construct an entire agent before making an account, and sign in only at the final step where the agent gets its wallet. But opening the My avatars tab without an account threw the visitor straight out to the login page, mid-build, with no warning and no explanation, because that tab asked for avatars you can only have if you are signed in. It now recognises a visitor who has no account, says so in the tab, and points at the Starter library and Upload tabs, which both work without one, alongside a Sign in button that saves the build first so nothing is lost either way. The step markers across the top of the wizard could only be used with a mouse. They are the only way back to an earlier step from the review screen, so anyone working by keyboard reached the end of the flow with no way to correct anything; they are now real buttons that take focus, show a focus ring, and respond to Enter and Space, with the completed steps marked as such for screen readers. The four tabs on the body step also announced themselves as tabs without ever telling a screen reader which panel each one opens, and left the arrow keys dead; the arrow keys, Home and End now move between them the way tabs are meant to work. Finally, a first-time visitor generating an agent from a description no longer sees a failed request in the browser console.",
"tags": [
"fix",
"improvement"
],
"link": "/create-agent"
},
{
"date": "2026-08-17",
"title": "The Create page no longer black-screens when the sign-in check stalls, and a broken remix link no longer kills every button on it",
"summary": "Three fixes to the front door for creating, plus a strip of your own avatars that was never once drawn. The page is meant to greet a returning creator with \"Start from one of yours\" and a row of the avatars they have already made, one click from remixing any of them. It read the library reply in the wrong shape, decided it had received nothing, and hid the row every single time, so nobody has ever seen it. It reads the reply correctly now and the row appears, with a real thumbnail for each avatar. The page opens behind a full-screen loading screen while it checks whether you are signed in; if that check got no answer at all (a stalled connection rather than a refused one) the screen stayed black forever and the page read as dead. It now hands the page back after a short wait and treats you as signed out, which costs at most one extra trip through sign-in. Separately, opening a remix link for an avatar that had been deleted or made private used to show a message for four seconds and then leave you on a page where not one of the eight creation cards did anything: the remix ran before the buttons were wired and gave up without wiring them. The cards are live first now, and a failed remix explains itself in a panel that stays put, with a button to try again and a button to move on. Also: the talking-avatar-video card now says \"Coming soon\" on the card itself instead of only dimming, answers the keyboard the same way it answers a mouse, and stops calling itself unavailable when the check simply could not reach us (it asks again when you click). A saved avatar whose thumbnail has been deleted falls back to a silhouette instead of a broken image.",
"tags": [
"fix",
"improvement"
],
"link": "/create"
},
{
"date": "2026-08-17",
"title": "Selfie to Avatar tells you your library is full before it spends ninety seconds building",
"summary": "Five fixes to Selfie to Avatar. If your avatar library was already at the limit for your plan, nothing noticed until the very end: you waited out the full reconstruction, and then the page said the 3D engine had hit a snag and told you to try again. It was not an engine problem, trying again could never work, and every attempt spent another ninety seconds of generation to land in the same place. Your library is now checked the moment you press the button, so a full one is refused immediately with the actual reason and what to do about it, and a build that fills the last slot while it is running ends with that same plain explanation instead of the engine message. The page also stopped losing its own error messages: the Build button floats above the page, so tapping it from higher up left the explanation somewhere below the fold while the button quietly reset itself, which read as nothing having happened. Errors now scroll into view. On the finished-avatar screen, Learn how to embed pointed at a page that did not exist and answered 404; it now opens the embedding guide. And on a narrow phone the Photoreal and Stylized choices were squeezed into two columns barely wide enough for one word per line, so they now stack and can be read.",
"tags": [
"fix",
"improvement"
],
"link": "/create/selfie"
},
{
"date": "2026-08-17",
"title": "Trial Conversions: the Buy button lands on the right skill, and Edit pricing stops going nowhere",
"summary": "Six fixes to the Trial Conversions page, which shows both sides of a free trial running out: the trials you hold and the trials running on skills you sell. A seller's Edit pricing button pointed at a page that no longer exists and answered 404 on every row; it now opens that agent's monetization panel, where the price and the trial grant are actually set. A buyer's Buy it and Open agent buttons dropped you at the top of an agent selling a dozen skills, leaving you to hunt for the one your trial just ran out on; they now open the agent's pricing list scrolled to that exact skill, with the row briefly marked so you can see which one you came for. Those links also took a redirect that quietly discarded everything after the question mark, so they now use the direct address. A trial whose skill has since been delisted used to claim \"1 of 1 free runs left\", inventing a grant size nobody ever reported, and said \"free runs\" even when there was one run; the count now says only what is known, and matches word for word what a screen reader is told. A row missing a timestamp printed the bare word \"started\" with nothing after it, and the seller list said \"1 trials running\"; both now read as sentences. If you hold more trials than one page can carry, the page says how many it is leaving out instead of showing a total that stops at the cap. Sellers who price in more than one currency now see the currencies the headline number cannot hold, rather than only the largest pile.",
"tags": [
"fix",
"improvement"
],
"link": "/conversions"
},
{
"date": "2026-08-17",
"title": "Coin Intelligence fits a phone screen, and a search no longer snaps back to the previous results",
"summary": "Five fixes to the Coin Intelligence dashboard. On a narrow phone the page was sized against its own tab strip rather than the screen, pushing the layout about 170px wider than the viewport; because the site clips sideways overflow, the right-hand third of the headline, the counters and every coin card was cut off with no way to scroll to it. The page now fits the screen and the tab strip scrolls inside its own row. Changing a filter or typing a search could also snap back to the previous set of coins a second later, because the older request was still open and whichever answer arrived last won; only the newest request can now paint the radar. The leaderboard used to greet an empty day with the radar's message about loosening filters, next to a Reset filters button that did nothing at all, since the leaderboard has no filters; it now explains that nothing has cleared observation in the last 24 hours. When the feed is down, the counters in the header stay in place instead of vanishing, the Confirmed winners rail and the Outcomes and Coverage panels say so instead of standing over empty boxes, and every one of those messages now carries a Retry that actually reloads. Smart Money shows shimmering placeholder rows while its cross-coin wallet aggregate is computed, and that aggregate plus the learning stats are now cached for minutes at the edge rather than seconds, so repeat visits are immediate.",
"tags": [
"fix",
"improvement"
],
"link": "/coin-intel"
},
{
"date": "2026-08-17",
"title": "The Concierge page stops cutting itself off on a phone, and its install tabs work from the keyboard",
"summary": "Three fixes to the Concierge landing page. On a narrow phone screen the whole page was being sized against the longest line of the install snippet, which pushed the column roughly twice as wide as the screen; because the site clips sideways overflow, the right-hand half of every headline, paragraph and button was simply cut off with no way to scroll to it. The column now fits the screen and the code snippet scrolls inside its own box, the way it already did on a laptop. The four install tabs (one tag, web component, npm, imperative) announced themselves to a screen reader as tabs but were not wired to the panels they open and ignored the arrow keys; they now move with Left, Right, Home and End, and each tab is properly tied to its snippet. Copy buttons also say which snippet they copy, and if the browser blocks the clipboard the snippet is selected for you so Ctrl+C still works. Finally, the passive \"Have you tried\" suggestion card could land on top of the live concierge demo in the corner; it now stacks clear of it.",
"tags": [
"fix",
"improvement"
],
"link": "/concierge"
},
{
"date": "2026-08-17",
"title": "Describe it to 3D keeps your prompt when it asks you to sign in, and stops a cancelled build from hijacking the next one",
"summary": "Four fixes to the prompt-to-avatar page. If you were not signed in, typing a long description and pressing Generate bounced you to the login page and threw the description away; you now come back to the page with your words still in the box. Cancelling a build and immediately starting another used to leave the abandoned build running in the background, so a minute later the cancelled avatar could take over the screen and Open in editor would point at the wrong character; each build now carries a token and an abandoned one can no longer write to the page. The Cancel button went dead in the one state where people reach for it most, after a build had already failed, because it was checking whether a build was still in flight rather than whether you were still looking at the build screen; Cancel and the Escape key both work there now. And when an avatar finished, Open in editor and Make another sat just below the bottom edge of a laptop screen, so the result appeared to have no next step: the preview is now sized against the window height, and both buttons land in view.",
"tags": [
"fix",
"improvement"
],
"link": "/create/prompt"
},
{
"date": "2026-08-17",
"title": "The Character Library counts its own characters, and fills the screen while it does",
"summary": "The library page had been advertising 106 rigged characters since launch. There are 107 on the shelf today, and there will be more tomorrow, so the number is no longer typed into the headline, the page description or the share card: the page asks the library how many it has and prints the answer under the hero. The grid also stopped wasting a third of a desktop screen. A layout rule left the page shrink-to-fitting to about two thirds of the window, so a wide monitor showed three columns where five fit; it now uses the full width it was designed for. Every character card previously carried a hidden second tab stop that announced the same name as the card itself, which made keyboard browsing twice as long as it needed to be, and thumbnails could not be dragged out of alignment by an accidental swipe anymore because the preview no longer pretends to be an interactive 3D control. The search box has a visible focus ring again instead of a one pixel border shift, and when the library cannot be reached the page now says what actually went wrong rather than reverting to a generic sentence a moment later.",
"tags": [
"fix",
"improvement"
],
"link": "/character-library"
},
{
"date": "2026-08-17",
"title": "Every Crypto Converter result now has its own link you can share",
"summary": "The converter promised a shareable result and never had one: the address bar stayed on /converter no matter what you picked, so there was no way to hand someone a rate. Now the page reads and writes its own URL. Pick BTC to EUR and type an amount and the address becomes /converter?from=bitcoin&to=EUR&amount=3, a new Copy link button puts that on your clipboard, and opening the link restores both sides and the amount exactly. Alongside that: the asset picker is fully keyboard-driven (arrow keys move the highlight, Enter picks, Escape closes and returns focus to the button you came from), the amount field says what it wants when you type something it cannot use instead of showing a silent dash, a coin whose price failed to load says price unavailable rather than a bare dash, the four quick presets light up when they match what you are looking at and stay disabled until live rates arrive, and the whole widget fits a 320px phone screen, where the currency buttons used to be pushed off the right edge and the picker off the left.",
"tags": [
"feature",
"improvement",
"fix"
],
"link": "/converter"
},
{
"date": "2026-08-17",
"title": "Coins explains a failed market feed and gives you a button that actually retries it",
"summary": "The Coins index had three ways to leave you stuck with no explanation. Load more coins swallowed a failed request silently, so a click that fetched nothing looked like a broken button, and it kept offering more rows forever even after the feed had run out. A market table that failed to load said only that data was temporarily unavailable, with nothing to press. The global stats strip across the top vanished entirely when its request failed, so the page looked like it simply had no market cap, dominance or Fear and Greed to show. All three now say what happened and hand you a working retry: the table and the stats strip each get a Try again that refetches in place, a load-more failure keeps the rows you already have and explains itself, and once the feed is exhausted the button retires with a line telling you how many coins it ranked. Being rate limited is now told apart from an outage, so you are asked to wait a minute rather than to keep hammering a feed that will refuse you. The coin search box shows a Searching indicator while it waits instead of staying shut, explains a failed lookup instead of closing as if nothing happened, keeps its previous results readable while the next query resolves, and can no longer let a slow answer overwrite a newer one. Screen readers now follow the highlighted search result, and an empty feed no longer wears the wording of a failure.",
"tags": [
"fix",
"improvement"
],
"link": "/coins"
},
{
"date": "2026-08-17",
"title": "Cosmos tells you when world generation is offline instead of showing you NVIDIA's internal error",
"summary": "The Cosmos page asked NVIDIA for a world and, when NVIDIA had retired the model behind that endpoint, printed the vendor's raw reply straight into the banner: a function id, our account id, and a Retry button for a request that could never succeed. That failure is now recognised for what it is. The lane tries each published Cosmos world model in turn, and only when every one is gone does it report itself offline, with a plain sentence, the living aurora backdrop still running behind your avatar, and no button that pretends the next click will work. A dropped connection still gets a real Retry, because that one can. Alongside that: the stage no longer overflows its column on a phone (it was rendered 672px wide inside a 320px screen, clipping the avatar) and now stands portrait on small screens so your character has room; a clip that arrives but will not play says so instead of leaving a finished stage with nothing in it; the Share button copies a link that actually carries the world, so whoever opens it sees the scene playing rather than an empty stage with the prompt refilled; a community avatar whose thumbnail has been deleted falls back to its initial instead of a broken image; the status dot no longer claims the lane is healthy before anything has been asked of it; and a job the service can no longer track ends promptly rather than spinning for five minutes.",
"tags": [
"fix",
"improvement"
],
"link": "/cosmos"
},
{
"date": "2026-08-17",
"title": "Compare Coins now tells you when a coin will not load instead of shimmering forever",
"summary": "The side-by-side coin comparison had one failure mode that looked exactly like a hung page: when the market data service did not answer, the stats table kept its grey loading shimmer up permanently and the chips still showed the raw coin ids, with nothing to click and nothing to read. Every one of those states is now written out. A coin that does not exist is labelled Not found with a one-click Remove; a coin the service failed to return is labelled Failed to load with a Retry that actually refetches it and redraws the chart. The chart says which of the two happened and offers the same recovery. The search box no longer goes quiet when the coin index is slow or down: it shows a Searching indicator while it waits and an explanation if the request fails, and a slow answer can no longer overwrite a newer one. Once four coins are up, the box says four is the maximum and stops accepting a fifth rather than silently swallowing the pick. New Copy link button next to the picker hands over the shareable matchup URL, the three default coins load in parallel instead of one after another, and the stats table now carries proper row and column headers for screen readers with a visible focus ring on every chip.",
"tags": [
"fix",
"improvement"
],
"link": "/compare"
},
{
"date": "2026-08-17",
"title": "Coin Communities keeps its own words on screen, and the Retry button on a failed coin now retries",
"summary": "A coin page that failed to load offered a Retry button that did nothing: it re-requested the coin the page was already showing, which the loader skipped as already-loaded, so the only escape from a transient outage was a manual reload. Retry now forces the fetch and recovers the page. The page was also arguing with its own translation layer: the browser tab and the failure message are written live by the page, but the translation pass ran afterwards and reverted them, so a coin that could not be reached said it did not exist, and the tab kept a coin's name after you had gone back to browsing. Both now hold what the page wrote and hand it back when the live state is gone. Alongside that: the price figure shows a real loading shimmer until both of its data sources have answered instead of a bare dash, signed-out visitors get a short line explaining the avatar row and a link to sign in rather than a silent unauthorized request, a pasted avatar URL that is not a link now says what a working one looks like, the search and mint fields have proper labels, and every button, chip and card has a visible keyboard focus ring and a press state.",
"tags": [
"fix",
"improvement"
],
"link": "/communities"
},
{
"date": "2026-08-17",
"title": "Clip Director shows real traders again, and can no longer print a number the trade never made",
"summary": "The Clip Director page told every visitor that no agent had ever closed a trade. Its trader picker asked the leaderboard for the highest-ranked agents and then kept only the ones with a closed round-trip, and because the ranking score has nothing to do with having closed trades, the nine agents that qualified all sat below the cut. The picker now asks the leaderboard for exactly what it needs, so those nine traders and their real closed positions are back on the page. The bigger repair is underneath: the card copy is written by a language model, and one of the providers answered a real +1.89x win with an invented loss on a different coin, complete with a celebrating avatar, next to the on-chain link proving otherwise. Every generated card is now checked against the trade it describes before it ships, so a ticker or a figure that is not in the round-trip gets the card thrown away and rewritten from the position itself, and a stop-out can never be celebrated. Also on the page: a shared /clip-director?agent_id= link now opens that trader even when they are outside the ranking, the Try again buttons retry the step that actually failed instead of doing nothing, each card's call to action is a real link, and the copy button copies on browsers that block the clipboard API instead of telling you to press a key.",
"tags": [
"fix",
"improvement"
],
"link": "/clip-director"
},
{
"date": "2026-08-17",
"title": "The Pole Club leaderboard now ranks the dancers who are actually on the poles",
"summary": "The club's leaderboard was reading names out of the payout registry instead of off the stage, so it listed four dancers under names matching none of the three at the poles, every one of them sitting at zero. It now names each dancer exactly as her pole card does, drops entries for dancers who are not in tonight's lineup and have never been tipped, and says what a quiet board actually means: the 24h and 1h views tell you the club has been quiet rather than claiming nobody has ever tipped, which was never true while the all-time board was full. Keyboard and screen-reader visitors get the rest of the pass: the page carries one real heading instead of five competing ones, Skip to stage now lands you on the stage, and the controls belonging to the walk-in (the agent picker, the Enter the club prompt) no longer sit invisibly in the tab order once you are inside the room.",
"tags": [
"fix",
"improvement"
],
"link": "/club"
},
{
"date": "2026-08-17",
"title": "The security agent in the world no longer opens its scanner in your face",
"summary": "zauth, the security agent standing in the town plaza, used to throw its RepoScan panel over the screen the moment you walked within a few metres of it, interrupting whatever you were doing and hiding the world behind a modal you never asked for. Walking up now only gets you the agent's line and the usual E prompt above its head. The scanner opens when you choose to open it: press E, click the agent, or click its nameplate. Nothing else changed about the scan itself, and the price is still fetched live from zauth in the background as you approach, so the panel shows the real cost the instant it opens instead of a loading shimmer.",
"tags": [
"fix",
"improvement"
],
"link": "/play"
},
{
"date": "2026-08-17",
"title": "Coin Clash stops claiming it is connecting when it has gone offline",
"summary": "The live indicator at the top of Coin Clash sat on a pulsing amber \"connecting\" forever, even once the page had established that the battle service was offline and had stopped trying to reach it. The cause was the page's own translation pass: the line of copy holding the indicator is translatable, and applying it rewrote that whole line back to the wording it shipped with, wiping out the status the page had just worked out. The indicator now survives a translation pass, so it reads offline when the arena is offline and live when battles are running. The same fault was quietly affecting the leaderboard, labor market and signals pages, and is fixed for all four. Two more repairs on Coin Clash: when the arena cannot be reached at all, the error card now explains that in words you can act on and offers Try again, instead of showing the browser's raw \"Failed to fetch\"; and on a narrow phone the Arena, War Records and Communities tabs wrap onto a second row rather than running off the right edge, where the Communities tab was cut in half and unreachable.",
"tags": [
"fix",
"improvement"
],
"link": "/clash"
},
{
"date": "2026-08-17",
"title": "My Collection loads visibly, keeps its counts, and never mangles an agent's name",
"summary": "The collection page had a loading state nobody could see: the shimmer placeholders were drawn inside a container the page kept hidden until the data had already arrived, so a slow connection got a blank stretch of nothing instead. The tabs, the placeholders and the counts now appear from the first frame. The counts next to Skills and Subscriptions used to vanish at random, because the page's own translation pass could overwrite the tab label a split second after the number was written into it; the count now lives in its own slot and survives. A skill published by an agent whose name contains a quotation mark or an angle bracket rendered as broken markup rather than as the name, and a publisher who earned a vanity wallet address had it shown as an ordinary one, because the card was reading a field the collection endpoint does not return. Both are fixed, so tipping a creator from your collection now shows the wallet they actually have. Empty states are centred under the tabs instead of squeezed into the first column, both of them now point you at the marketplace, the Retry button on a failed load is a real button rather than a raw grey box, and the whole page is keyboard-navigable: arrow keys move between tabs, Home and End jump to the ends, and every link and button shows a clear focus ring.",
"tags": [
"fix",
"improvement"
],
"link": "/collection"
},
{
"date": "2026-08-17",
"title": "The Community page's live activity feed renders as readable rows again",
"summary": "The activity list on the Community page was building each row as a link wrapped around another link, which no browser can parse: every entry was torn into three stray fragments, so the creator, the item and its thumbnail landed on separate lines with gaps between them and the whole list was unreadable. Rows are now single, valid entries again: click anywhere on one to open the creation, or click the creator's name to open their profile. If the feed cannot load, or returns something the page cannot read, you get a clear message and a Try again button instead of skeleton placeholders that never resolve. The page also follows the light and dark theme switch now instead of staying black on a light canvas, and the newsletter section finally has the signup field its text promises, rather than sending you hunting for the one in the footer.",
"tags": [
"fix",
"improvement"
],
"link": "/community"
},
{
"date": "2026-08-17",
"title": "A failed X seed no longer costs you six hours, and can no longer wipe a good one",
"summary": "Seeding an agent from your X posts is allowed once every six hours per agent, and until now that window was spent the moment we started reading, whatever happened next. If X was throttling us, if X had stopped accepting the connection, or if your sign-in could not be renewed, you got a bare failure and a six-hour lockout for something that was never yours to fix. Every one of those now hands the window straight back and says what actually happened: reconnect X, or wait the number of minutes X itself told us to wait. Worse was the run that read your account fine and found nothing usable in it, which is what a timeline of only replies and reposts looks like. That run deleted the batch of facts your agent already had and put nothing in its place. It now stops before the deletion, keeps the memories you had, gives the window back, and tells you what to post or add to your profile so the next seed has something to work with. Nothing else about the lane changes: nothing is read until you consent on a screen that lists what is read and stored, and revoking still deletes every memory the seeding produced.",
"tags": [
"fix"
],
"link": "/settings"
},
{
"date": "2026-08-17",
"title": "The avatar SDK bundle we ship is now built from the code we deploy",
"summary": "The <agent-3d> runtime that sites load from three.ws is a single bundled file, and the production build never rebuilt it. It was copied from whatever the build machine happened to have on disk, which could be days older than the code being deployed, so a fix to the avatar element could ship everywhere else on the site and quietly not reach the SDK. The deploy now rebuilds that bundle from the same commit as the rest of the release, every time. Nothing to change on your side: existing embeds keep working and simply pick up the current runtime.",
"tags": [
"fix",
"sdk",
"infra"
],
"link": "/avatar-sdk"
},
{
"date": "2026-08-17",
"title": "Chat stops hiding two nav links, and says something useful when a reply fails",
"summary": "The chat at /chat was audited end to end in a real browser: loading it cold, sending a message on the free built-in lane and getting a reply back, following every link in the header and footer, and the empty, loading, error and populated states at phone, tablet and desktop widths. The header was covering itself: the centred site links sat on top of the wallet controls, so Features and Docs were partly hidden and could not be clicked at any desktop width, and between tablet and laptop widths there was no way to reach those links at all. The links now sit in the row with everything else, and below the width where they fit they move into the menu. The footer had been shipping a mangled copyright line (a leftover encoding accident that printed stray characters where the copyright symbol belongs). When a reply could not be sent, the thread printed the browser's own words, Failed to fetch; it now says what happened, that nothing was sent, and how to retry, in a message styled as an error rather than as a quieter reply. A rate-limited reply used to say Try again in 79445s; it now says the wait in hours or minutes and, when the free daily allowance is the thing that ran out, points at adding your own provider key. Icon-only buttons across the header and each message (share, model settings, copy, edit, regenerate, delete) now carry names for screen readers, and the page title and description finally describe the page.",
"tags": [
"fix",
"improvement"
],
"link": "/chat"
},
{
"date": "2026-08-17",
"title": "Choreographer share links are readable again, and the studio fits a phone",
"summary": "The gesture composer at /choreograph was audited end to end in a real browser: loading a starter routine, adding and reordering beats, retiming and repointing a step, previewing the whole performance on a live avatar, scrubbing the timeline, sharing a link, and saving to a real agent, at phone, tablet and desktop widths. Copy share link used to hand you a link with the routine name escaped twice, so a routine called The pitch travelled as The%2520pitch: it worked, but it looked broken, and it invited people to tidy it up and break it for real. Links now carry the routine as you can read it, and every link shared before today still opens. A routine whose name ended in the loop marker came back from a link under the wrong name and looping when it should not. On a 320 pixel phone, opening either of the two code panels pushed the whole page sideways; the code block now scrolls on its own instead. Pressing the transport a second time while the first play was still loading its clips said Pause but started a second load; it now stops the one in progress, and deleting the last beat mid-load can no longer bring an avatar back onto a stage the routine has left. Signed-out visitors also no longer see a failed request in the browser console on first paint.",
"tags": [
"fix",
"improvement"
],
"link": "/choreograph"
},
{
"date": "2026-08-17",
"title": "The Characters gallery pages past the first screen, and stops fighting dark mode",
"summary": "The character gallery at /characters was audited end to end in a real browser: loading the feed, both sort orders, search, Load more, following a card through to its character page, and the loading, empty and error states, at phone, tablet and desktop widths. The worst finding was that the Top sort was quietly broken. Sorting by most-chatted read the right first screen, but pressing Load more asked the server to continue from the wrong place, so the feed ended after eight characters out of more than two thousand. It now pages all the way through, in order. The page also ignored the site theme: it painted a fixed light-blue background under the dark header no matter what you had chosen, which is now fixed in both directions, and wallet addresses on the cards were nearly invisible on a light background everywhere they appear on the site. The gallery filled only a narrow strip down the middle of a desktop screen and now uses the full width, up to four cards across. If the feed failed to load you got one line of grey text and no way to retry; there is now a proper message with a Try again button, a failed Load more keeps the characters you already had, and searching for something with no matches offers to clear the search rather than leaving you at a dead end. Pressing Top while the first screen was still loading used to light up the button and change nothing, so you were left reading New results under a Top heading. The search box had no label for screen readers, the sort buttons never showed which was selected, and neither they nor Load more showed a focus outline for anyone navigating by keyboard. Keyboard users can also now skip the header and land in the gallery. Finally, the feed itself is several times faster to answer on a cold request, because it no longer counts every conversation of every published character just to show you twenty-four of them.",
"tags": [
"fix",
"improvement"
],
"link": "/characters"
},
{
"date": "2026-08-17",
"title": "Pausing a skill bundle is no longer permanent, and every tooltip fits on a phone",
"summary": "The bundle builder at /bundles was audited end to end in a real browser: signing in, picking priced skills, the price panel and its backtest, publishing, the published list, and the loading, empty and error states, at phone, tablet and desktop widths. The worst finding was that Deactivate was a one-way door. It hid the bundle from buyers, which is correct, but it also removed it from the only page that could put it back, so a misclick permanently retired a bundle you had just priced. The button is now called Pause, a paused bundle stays on the page dimmed and clearly marked, and a Reactivate button puts it back on sale. Buyers still never see a paused bundle: the public list is unchanged, and paused ones are returned only to a signed-in owner of that agent. The small question-mark tooltips that explain every number on the page were also cut off at the edge of a phone screen, so four of the seven explanations were partly unreadable at 320 pixels wide; they now shift to stay on screen at any width. The sum of the parts and the suggested price were cramped into a broken-looking two-line squeeze on a narrow screen and now stack cleanly. Finally, a pricing error raised by one agent no longer sits on screen after you switch to another agent, where it read as a verdict on an agent it had never looked at.",
"tags": [
"fix",
"improvement"
],
"link": "/bundles"
},
{
"date": "2026-08-17",
"title": "Avatar Studio survives an accessory outage, and tells you when it is putting a hat on",
"summary": "The studio at /avatar-studio was audited end to end in a real browser: loading the base body, all six tabs, the colour swatches, the sculpt sliders, the motion library, undo and redo, and the loading, empty and error states, at phone, tablet and desktop widths. The worst finding was that a single failed request for the accessory catalogue took the entire studio down. The hats, glasses and earrings lists come from one file, and if that file did not load, the page replaced everything with a Retry screen, throwing away the 3D avatar it had already finished loading. Colours, sculpting, animation and saving need nothing from that catalogue, so they all now keep working, and only the three accessory tabs show a message explaining what is missing, with a Retry button that reloads just the catalogue. The Save, Reset, Randomise and Undo buttons were also on screen but not yet connected until that same request finished, so pressing them early did nothing at all and the tab strip sat empty; they are live from the first paint now. Putting on a hat downloads and fits a 3D model, which takes a few seconds, and until now the studio gave no sign it was working: the tile click looked ignored, and a hat that failed to download failed in complete silence. Every accessory change now shows what it is doing and says so plainly if it could not finish. Randomise no longer lists an accessory you are not actually wearing when its model fails to load. Pressing Undo then Redo quickly could leave the avatar and the list of what it is wearing disagreeing with each other; each step now settles correctly. Spinners also finally stop for people who have asked their device to reduce motion.",
"tags": [
"fix",
"improvement"
],
"link": "/avatar-studio"
},
{
"date": "2026-08-17",
"title": "Blockchain TVL by Chain is readable again, and its numbers can be copied",
"summary": "The cross-chain leaderboard at /chains was audited end to end in a real browser: the live DeFiLlama board, the sortable columns, the row links out to each chain's page, and the loading, empty and error states, at phone, tablet and desktop widths. The page had briefly stopped working entirely. Its row-click helpers were declared twice in the same file, which is a syntax error, so the script never ran and the table sat on its grey loading bars forever. That is fixed and covered by a browser test so it cannot come back. Two real defects came out of the same pass. Every row navigated on any click, so dragging across a cell to copy a chain's TVL threw the selection away and opened that chain's page instead; a figure or a name can now be selected and copied, while a plain click still opens the chain. And the page wrapped the site navigation in a second page banner, leaving screen readers with two overlapping banner landmarks to pick between. The share bar beside each chain also stopped reading its percentage aloud a second time, since the dominance column right next to it already says it.",
"tags": [
"fix",
"improvement"
],
"link": "/chains"
},
{
"date": "2026-08-17",
"title": "A shared ASL letter link now opens on that letter, not on A",
"summary": "The manual alphabet at /asl-alphabet was audited end to end in a real browser, and six repairs came out of it. Sharing a letter was the worst of them: a /asl-alphabet?letter=W link built the whole page around the letter A and only corrected itself once the 3D avatar had finished downloading, so anyone on a slow connection read A's description, saw A's key lit and A's look-alike note for the entire wait, on a link that promised W. The page now reads the link before it builds anything, so the right letter is on the card from the first paint and the avatar simply catches up. The Spell anything box had the opposite problem: it reported an empty or unspellable word only in the small caption pinned to the avatar, which on a phone sits a full screen above the box you just typed in, so pressing Spell it with a word full of punctuation looked like a dead button. It now answers in its own line, right under the input, and while the hand is spelling it lays the word out letter by letter and lights each one as the hand reaches it, so a long word can be followed without scrolling back up to the A-to-Z grid. That caption itself was white text on a scrim too faint to read over a pale avatar or in light mode, and is now a proper legible chip. Opening the avatar picker no longer lets a keystroke through to sign a letter behind the dialog. Switching avatars used to leave 'Loading avatar…' on screen forever once the new one had arrived, which also meant screen readers were never told the load had finished. And the Spell panel is no longer stretched to the height of the drill beside it, which left a large empty box on desktop.",
"tags": [
"fix",
"improvement"
],
"link": "/asl-alphabet"
},
{
"date": "2026-08-17",
"title": "Wallet Intelligence stops reporting win rates like 570% and 9790%",
"summary": "The pump.fun track record at /claim-wallet was audited end to end in a real browser, and the headline numbers were wrong. Lifetime win rate, early-entry win rate and dump rate are stored as percentages, but the page multiplied them by a hundred a second time, so a wallet that wins 5.7% of its trades was shown as '570% lifetime' and a 97.9% dump rate read '9790%'. The meters beside them were pinned full for the same reason, and the colour that flags a heavy dumper was triggered by anything above 0.3%. All three now read the real figure. Four other repairs came out of the same pass. A wallet the reputation brain has graded but whose per-coin rows are still indexing used to show an empty ledger captioned 'No coins match these filters', blaming filters that were not the cause; it now explains that the ledger is still filling in. When filters genuinely do hide everything, the table says how many coins were hidden and offers one click to clear them. The ledger header no longer claims 60 coins while showing 52, since dust is hidden by default. A dropped connection used to blank the report and print the browser's raw 'Failed to fetch' beside the search box with no way forward, and now shows a proper panel naming the cause with a Try again button. The Analyze button's 'Analyzing…' label was being overwritten by the translation pass and never appeared. Keyboard and screen-reader users can now sort the ledger and open any coin: the column sorters are real buttons that announce their sort direction, every row carries a real link to that coin's dashboard, the time-window and filter controls report whether they are on, and every control on the page shows a visible focus ring.",
"tags": [
"fix",
"improvement"
],
"link": "/claim-wallet"
},
{
"date": "2026-08-17",
"title": "Crypto Categories reads correctly in all 84 languages",
"summary": "The sector leaderboard at /categories was audited end to end in a real browser: the table, the search box, the sortable columns, the paging, and the loading, empty, error and rate-limited states all behave, at phone, tablet and desktop widths. One real defect turned up, and it only showed outside English. The page's own summary line and the description search engines and link previews read were translated once against an older wording and had kept a dash where the English now uses a colon, so every non-English visitor saw punctuation the platform does not use, in the headline copy and in the shared-link preview. All 83 translated catalogs are corrected, with the full-width colon that Japanese, Korean and Chinese actually want rather than a Latin one.",
"tags": [
"fix"
],
"link": "/categories"
},
{
"date": "2026-08-17",
"title": "The viewer's backdrop picker stops vanishing while you read it, and sharing no longer posts \"Meet Agent\"",
"summary": "Five fixes to the 3D model viewer at /app. Opening the Backdrop picker and taking a few seconds to read the six choices used to make the picker fade out from under the cursor: the viewer hides its controls after five seconds of stillness, and the picker was not on the list of surfaces that keep them awake, so an open picker went invisible and unclickable while still counting as open. It now stays put until you choose a backdrop or close it, and the controls resume fading once it is shut. Typing in the animation search while the clips were still loading stacked a second, contradictory message under the first, so the panel read \"Loading clips\" and \"No clips matching\" at the same time, and clearing the box left behind the nonsense line \"No clips matching\" with nothing in the quotes; the search now stays quiet until there are clips to search. Sharing the viewer to X, Farcaster or Telegram published \"Meet Agent\", an internal placeholder name, for anyone who had not loaded an agent of their own, and now uses the generic line instead. The page also shipped a sign-in panel and a welcome card that its own stylesheet blanked unconditionally, four links nobody could ever reach, and the preview card shown when the page is shared described the platform instead of the viewer. The avatar gallery behind the Avatar button is now announced to screen readers as a named dialog.",
"tags": [
"fix",
"improvement"
],
"link": "/app"
},
{
"date": "2026-08-17",
"title": "Avatar Studio now shows your whole avatar, and its Save button fits on a phone",
"summary": "The studio camera was framing the base body from the shoulders to the knees: the face you were sculpting and the shoes you were picking were both outside the picture, and the only way to see them was to orbit the camera yourself every time the page loaded. The framing is now solved from the lens rather than guessed at, so a full-body shot really does hold the avatar head to toe, arms included, on any window shape, and the half-body and headshot views stay as tight as their names promise. On a 320px phone the top bar had been running past the edge of the screen and taking the Save avatar button with it, so a look built on a phone could not be saved at all; the bar now fits, with the Back and title labels standing down to make room. The tab strip on that same narrow layout had been hiding its labels without putting anything in their place, leaving six blank buttons, and now shows an icon per tab. The Parametric body badge that read 120+ sliders was white text on a white pill and invisible everywhere. A failed save now says the connection could not be reached and that your look is still there, instead of repeating the browser phrase Failed to fetch, and every accessory search box, every tab and the Back link now carry names a screen reader can read. Opening avatar-studio in local development also serves the same page production does, which it previously did not.",
"tags": [
"fix",
"improvement"
],
"link": "/avatar-studio"
},
{
"date": "2026-08-17",
"title": "There is a real car in every coin world now, and you can get in and drive it",
"summary": "The cars in a coin world used to be boxes: a box on wheels driving the ring road, and a slightly different box parked for you to drive. Both are now the Trench Car, a real community model published to the three.ws avatar gallery, so the car that drives past you on the road is the same car parked at the plaza that you can get into. Walk within a few metres of one, press F (or tap the car on a phone) and drive it with WASD and the on-screen pedals, with a working speedometer, brake lights and a handbrake. Your driver now actually sits in the seat: every car in the world had been placing its driver standing on the roof, and other players saw them there too, because the seat was measured from the wrong point. Drivers are now seated inside the car in a seated pose for themselves and for everyone else in the world. Parked cars had also been sitting with their wheels a few centimetres into the asphalt, and snapping up the moment someone took the wheel; they now rest on their wheels whether or not anyone is driving. The model is optimized on the way in, from 16 MB down to 1.4 MB with a third of the triangles, one download shared by every car on screen, and if it ever fails to load the old box still drives so the delivery jobs keep working.",
"tags": [
"feature",
"improvement",
"fix"
],
"link": "/play"
},
{
"date": "2026-08-17",
"title": "Token in 3D paints in seconds instead of waiting on the slowest lookup, and the trade tape stops reporting every swap as $0",
"summary": "The /coin3d page renders any pump.fun or Solana token as a live 3D scene, and it was audited end to end in a real browser: the landing search, a loaded token, and every loading, empty and error state at phone, tablet and desktop widths. The biggest finding was that the whole page waited on the top-holder scan before showing anything. That scan walks every token account and takes around ten seconds on a large holder set, so the page sat on a spinner the entire time, and if the scan stalled outright there was no timeout at all and the spinner never went away. The coin, the price, the stats and the live tape now paint as soon as the token identity resolves, and the holder galaxy grows into the scene when the scan lands, with the holder share showing as pending until then. Every lookup is now bounded, so a stalled data source degrades one number instead of freezing the page. The live trade tape had been rounding every trade to the nearest dollar, which printed real swaps as $0 and 0.000 SOL; it now shows cents and fractional SOL. When a data source is genuinely unreachable the page says that, with a retry, instead of claiming the token does not exist. On a 320px screen the landing card no longer runs off the right edge with the search button and a third of the launch grid cut off, and it scrolls instead of clipping its own footer. The page also had no top-level heading in any state but one, the browser tab title reverted to the generic page name once translations loaded, and the price chart and holder share showed placeholder boxes that could pulse forever; all three are fixed.",
"tags": [
"fix",
"improvement"
],
"link": "/coin3d"
},
{
"date": "2026-08-17",
"title": "The Runway Lab called a stalled payment rail healthy, because nobody had tried to pay that day",
"summary": "The Runway Lab at /economy-lab is the page that tells you whether platform payments can still settle, and it was audited end to end in a real browser. It was getting the headline exactly backwards. The fee wallet that pays Solana network fees is currently below the hard floor it is never allowed to spend past, so nothing could settle at all, and the banner read \"settlement is healthy\" in green. The cause was that the page seeded its demand knob from measured traffic, the rail had taken no payments in 24 hours, and a projection of zero attempts came back with nothing refused, which the model scored as success. Zero attempts is now its own answer rather than a pass mark, and when the log shows no traffic the knob starts at a stated example load and says so instead of quietly projecting nothing. The live panel matches: a wallet under its floor no longer shows a green daily allowance it cannot reach, and an unread balance no longer prints as a confident 0.0%. If the page cannot read the rail at all, every knob now visibly switches off and there is a retry button, rather than six sliders that look live and do nothing. The projection chart was unreadable on a small balance, labelling all five of its gridlines 0.002 or 0.001, and on a phone it clipped those labels, ran its legend off the panel and stamped its own captions over the bars. Tapping a knob's help mark on a phone opened its explanation and closed it again in the same tap, so the explanations were unreachable on touch. Underneath, a projection of a starved rail was doing one full pass of the real settle-path check per refused payment, hundreds of thousands of them for a single slider drag; it now settles each refused hour in one step and gives the same numbers.",
"tags": [
"fix",
"improvement"
],
"link": "/economy-lab"
},
{
"date": "2026-08-17",
"title": "Forgot password now answers in your language, announces itself, and shows a focus ring you can see",
"summary": "The password-reset request page at /forgot-password was audited end to end in a real browser. Four things were wrong. Its result messages were the only copy on the page not wired into translation, so a visitor reading the page in Spanish typed their email and got the outcome back in English; worse, the first submit permanently reset the button to its English label. Those messages also carried no live-region role, so a screen reader announced nothing at all when the reset link went out and the visitor had no way to know it had. Every link and button on the page fell back to the browser default focus ring, which is a near-black outline drawn on a near-black page: keyboard users could not see where they were. And the avatar beside the form pulsed a loading skeleton forever if its model failed to load, promising something that was never coming. All four are fixed. The result copy, the two failure messages, and the inline email hint now come from the translation catalog, the success notice is announced, focus rings are drawn in a high-contrast accent across every sign-in and account page, and the avatar panel settles into a designed resting state when its model cannot load. The decorative avatar no longer catches Tab, which used to strand keyboard users inside a panel screen readers cannot see.",
"tags": [
"fix",
"improvement"
],
"link": "/forgot-password"
},
{
"date": "2026-08-16",
"title": "The Commons now fills with citizens in seconds instead of standing empty",
"summary": "Agora said it held 200 citizens while the square in front of you stayed empty. Every citizen was waiting on its own 3D model, and those run from 3 MB to 21 MB each, so the world was downloading close to a gigabyte before anyone could stand up: measured, 19 of 200 citizens had arrived after four and a half minutes. On top of that the page asked the server where each citizen's model lived one citizen at a time, 184 separate requests that also crowded out the economy ticker, which is why the population, tasks and $THREE readouts sat on dashes and the feed said it was reconnecting. The population endpoint now answers that question once for the whole square, and the crowd is drawn in two passes: everybody is standing within seconds, and the most recently active citizens change into their own avatar as it arrives, in place, without losing their spot or interrupting a walk to the job board. The population counter reads \"n / 200\" honestly while they arrive rather than reporting a number that did not match the square, and the Reload button on the \"failed to load\" screen now actually reloads. The \"reconnecting\" badge over the ticker also stopped crying wolf: it used to latch on after two failed requests and sit there over perfectly fresh numbers, and it now appears only when nothing new has actually arrived for half a minute, clearing itself the moment it does.",
"tags": [
"fix",
"improvement"
],
"link": "/agora"
},
{
"date": "2026-08-16",
"title": "The Avatar SDK playground now changes the avatar you are looking at, not just the code beside it",
"summary": "The \"Try it live\" panel on /avatar-sdk offers four controls, and two of them only rewrote the code sample. Picking agent-3d (the full runtime) left a three-ws-viewer on screen, and flipping the auto-rotate switch printed an auto-rotate line into the snippet while the preview stood still. Both controls now drive the live preview: choosing the full runtime loads it and swaps it in with a note saying the heavy bundle is downloading, and if that download fails the panel says so and falls back to the lightweight viewer instead of leaving you on a broken control. Auto-rotate is a viewer-only attribute, so it is now disabled with a reason when the full runtime is selected, and the snippet you copy no longer contains an attribute that runtime ignores. Separately, on a 320px phone the page had been pushing its own content past the right edge of a window that cannot be scrolled sideways, cutting off the end of the hero, the install command and several code samples; each code block now scrolls inside its own frame and nothing leaves the page. If the 3D viewer cannot load at all, the page now explains that in place of each preview and links to the source, where it used to render three silent empty boxes. Copy buttons that a browser blocks from reaching the clipboard now select the snippet and say to press Ctrl+C rather than doing nothing visible, the framework example tabs can be moved through with the arrow keys, and the Avatar Studio demo button reports that it is opening the moment you press it.",
"tags": [
"fix",
"improvement"
],
"link": "/avatar-sdk"
},
{
"date": "2026-08-16",
"title": "Agent Wallet's trust panel stops sending you to a page that does not exist",
"summary": "The Trust tab of an agent wallet lists what the wallet's credibility score is built from, and offers the owner concrete ways to raise it. Two of those pointers led nowhere: \"Verify your on-chain identity\" and, for an already-registered agent, the \"On-chain identity\" evidence link both pointed at an address the site has never served, so clicking either one landed on a not-found page. They now go somewhere real. The suggestion opens the ERC-8004 registration wizard with the agent's avatar already filled in, and the evidence link opens that identity's public on-chain page. The wallet page also stopped reading its address twice when you open the Vanity tab, and the screen reader announcement was corrected: being signed out, or not owning an agent yet, is the page's normal answer rather than an emergency, so those two screens are now read in turn instead of interrupting whatever was being spoken.",
"tags": [
"fix",
"improvement"
],
"link": "/agent-wallet"
},
{
"date": "2026-08-16",
"title": "AR Studio: \"Share live\" rooms actually connect now, and the model tray answers the keyboard",
"summary": "AR Studio can put several people in one AR scene at once: open a shared room, hand out the six-character code, and everyone places and moves models in the same space. On the live site that never worked. The page was the only realtime surface on three.ws that never named the server it should talk to, so the studio looked for one, found nothing, and told every user \"Shared rooms are offline right now\" while the server sat there ready to take the connection. It is now pointed at the same realtime service that /irl and /agora use, and two people joining the same code see each other and each other's models immediately. The failure path was hiding a second bug: when a room could not be reached, the studio threw an error partway through cleaning up, which left the room button stuck mid-state. Creating a room also stopped waiting for the connection before it declared success, so the invite link it promises to copy for you never made it to the clipboard. Both are fixed. Separately, the \"Add models\" tray is now fully keyboard-operable: left and right arrows move between the Recent, Yours, Community, Objects, and GLB link tabs, Home and End jump to the ends, and closing the tray or either dialog hands focus back to the button that opened it instead of dropping you at the top of the page. On a narrow phone the tab strip now fades at whichever edge still has tabs hidden behind it, so Objects and GLB link stop being invisible. A thumbnail that fails to load falls back to the same placeholder mark an image-less model uses, rather than a broken-image icon.",
"tags": [
"fix",
"improvement"
],
"link": "/ar/studio"
},
{
"date": "2026-08-16",
"title": "Skill Bundles: the page stops telling signed-out visitors they have no agents, and a failed request stops posing as an empty account",
"summary": "/bundles is where you group several of an agent's skills into one purchase and price it from your own sales. It was asking the wrong endpoint for your agents: one that answers with a single agent rather than a list, and answers it even when nobody is signed in. Every visitor, signed in or not, was told \"You have no agents yet\" and pointed at the create page, and the sign-in prompt the page ships could never appear. It now asks who you are first, then lists what you own, so a signed-out visitor gets the sign-in prompt and an owner gets their agents. Three other things were wrong once you got in. A dropped request while loading your skills or your published bundles was rendered as \"this agent has no priced skills\" and \"no bundles yet\", which is a claim about your account rather than about the network; failures now say so plainly and carry a Retry that re-runs just that call, on the price panel too. Every price was labelled $THREE no matter which token the skill was actually priced in, and each price now carries its real token. And the price field was squeezed to 31 pixels wide on a 320 pixel screen, too narrow to read the number you were typing. The builder also explains the zeros when nobody has yet bought two or more of the selected skills, asks the server for a live \"buyers better off\" count as you move the price instead of showing a dot, and the page finally has documentation behind the link that has always pointed at it.",
"tags": [
"fix",
"improvement",
"docs"
],
"link": "/bundles"
},
{
"date": "2026-08-16",
"title": "x402 Bazaar: the whole catalog is reachable, and the page stops blaming your connection for a provider that blocks browsers",
"summary": "The Bazaar advertised a number that was really just the size of one page: it said \"200 results\" while the facilitators were carrying more than a thousand matching services, and there was no way to reach any of the rest. The count now reads \"200 of 1465\" and a Load more button walks the catalog a page at a time, on the site and through the list and search endpoints, which both take an offset now. Three other things were quietly wrong. A service description containing one very long unbroken word (a raw URL, an id) stretched the result grid wider than the window, which gave the page a horizontal scrollbar on a desktop and pushed the filter sidebar off the edge of a phone. Changing a filter while results were still loading was silently ignored, so the sidebar could end up describing a filter the cards on screen had never been through; the newest change now always wins. And the search box and every filter field had no visible outline when you tabbed to them, so a keyboard user could not tell where they were. Separately, when a provider refuses browser requests outright, which is most of the catalog, the payment modal used to tell you to check your connection and offer a retry that could never work. It now names the provider and the reason and points you at a server, an agent, or the CLI.",
"tags": [
"fix",
"improvement"
],
"link": "/bazaar"
},
{
"date": "2026-08-16",
"title": "Atlas: a long search term no longer knocks the page sideways, and the clear button is tappable",
"summary": "/atlas is the whole platform as one searchable map, and its search box had three rough edges. Searching for a very long unbroken term (a pasted URL, a run-on word) pushed the text past the edge of its box and gave the entire page a horizontal scrollbar, so the layout slid sideways under you while you typed. Long text now wraps wherever it has to, and the same rule keeps long route paths from stretching the result grid on a narrow phone. The box also drew two clear buttons side by side, the one the browser paints for a search field and the one the page draws, which every other search surface on the site already suppresses and this one had missed. The remaining clear button was a 16 by 18 pixel target, well under the minimum anyone should have to hit on a phone. It is now a 26 pixel square with its own hover and keyboard focus treatment.",
"tags": [
"fix",
"improvement"
],
"link": "/atlas"
},
{
"date": "2026-08-16",
"title": "The ASL Alphabet drill now answers every press, and its package link goes somewhere real",
"summary": "Two things on /asl-alphabet did nothing when you used them. The Check button in the reading drill sat enabled from the moment the page loaded, but pressing it before starting a round, with an empty box, or after revealing an answer changed nothing on screen at all, which reads as a broken button. Check now always replies: it tells you to press start when there is no round to grade, tells you to type what you read when the box is empty, and says the round is finished once it has been graded or revealed. The start button's label follows the drill instead of being stuck on \"Skip to the next one\" after a round ends, so the feedback line can name the button you actually need. A wrong answer no longer says \"Watch it again\" when the round is already closed; it points at \"Show it again\" to study the shape and then move on. The drill's feedback is now announced to screen readers instead of changing silently. Separately, the note at the foot of the page (and the matching one on /sign-mirror) linked the signing engine to an npm page that does not exist: the package is in the repository and its registry publish has not landed yet, so both now link to the real source and say how to install it.",
"tags": [
"fix",
"improvement"
],
"link": "/asl-alphabet"
},
{
"date": "2026-08-16",
"title": "The BNB block race stops showing a dead chain's last number as if it were live",
"summary": "The live block race compares BNB Chain, Base, Ethereum and Solana by measuring each one off a public RPC every few seconds, and it promises that a chain going quiet says so instead of freezing a fake number. The four lane cards kept that promise; the big headline number did not. When the BNB RPC stopped answering, the headline held its last reading at full brightness, kept claiming it was N times faster than Base, and the Updated stamp sat on \"just now\" no matter how long the outage ran. The headline now dims the moment the reading stops being live, labels it as last measured and prints its real age in seconds, and drops the speed comparison rather than race a stale number against a live one. The Updated stamp ages off the measurement time the server reports, and a poll that measured nothing no longer refreshes it. If every chain goes down at once the page now raises its outage notice and its Retry button, which previously could only ever appear before the first successful reading and so was unreachable during a real outage. Retry now shows it is working, makes a genuinely new request rather than reporting an older in-flight one's failure, and no longer sits underneath the floating helper widgets where the click was being swallowed. A lane that has never returned a sample says so plainly instead of pointing at a reading that is not on screen, each poll goes to the network instead of replaying a cached body up to fifteen seconds old, and screen readers now hear a lane announce a change of state rather than re-reading all four numbers twelve times a minute.",
"tags": [
"fix",
"improvement"
],
"link": "/bnb-latency"
},
{
"date": "2026-08-16",
"title": "The Arena now ranks agents that actually traded above the ones that sat out",
"summary": "A tournament scored on realized P&L was handing a clean zero to every entrant that never opened a trade, which put them above every agent that took a real loss. On today's Daily Arena that meant the eight agents who did nothing held the whole podium while the two that traded sat at the bottom, and because a no-trade entrant can never clear the prize gates, the splits would have paid out nothing at all. Ranking now puts a result ahead of no result: entrants with no closed trade in the window sit below everyone who competed. They stay on the board with their honest zero, they just no longer outrank the field. The trophy, the podium on the front page and the prize ladder now only ever name an entrant that has actually traded, and the ladder only names an agent that is genuinely prize-eligible instead of promising a payout the settlement would never send. A tournament nobody has traded in now says it is awaiting the first ranked trade rather than crowning a zero. Also fixed: the tournament name in the browser tab was being reset to the generic page title a second after the page loaded, and the create-tournament form's fields overflowed their sheet on a 320px phone.",
"tags": [
"fix",
"improvement"
],
"link": "/arena"
},
{
"date": "2026-08-16",
"title": "Multi-LLM Brain now shows every model it can actually run, and says so when one is out",
"summary": "The /brain playground kept its own list of models on the page, and it had drifted from the models the server can really reach: nine were missing from the picker, including the current Claude Opus 5 and Sonnet 5 and all seven free models a signed-out visitor is allowed to use, while two entries in the list no longer existed at all. The picker is now built from the live model service, so it always matches what can be sent. A signed-out visitor opens on four models that are genuinely free to use instead of paid ones that could only refuse, and every model that needs an account is marked with a lock and a sign-in link rather than failing after you press Send. When the model list cannot be loaded the page now says so and offers a Retry that recovers, instead of quietly offering every model and erroring on send. When a model fails after every fallback route, the column now explains what happened and what to try next instead of printing a bare upstream word like Forbidden. Model choices are remembered between visits. On a phone the session list is reachable again through a drawer, the compare columns snap one per screen, and the Send button is no longer covered by the floating helper widgets. Model chips are proper buttons with keyboard focus, every field has a label, and clearing your persona from the sidebar now clears it everywhere.",
"tags": [
"improvement",
"fix"
],
"link": "/brain"
},
{
"date": "2026-08-16",
"title": "The assistant widget now tells you the truth when its chat lane is busy",
"summary": "Two things on the assistant widget page reported the wrong thing. The live preview status line always read \"Starting the live preview...\", even once the avatar was up and running, because the translation pass reset it a moment after the page wrote it. Every state now carries its own translated line, so the preview says what it is actually doing, and it reports a real failure with a Retry button when the widget or its avatar does not come up instead of silently claiming success. Second, when the free chat lane was throttled the widget only said \"Chat failed (HTTP 503)\", and a signed-out visitor could even be told to sign in for a capacity problem the widget has no sign-in for. The chat endpoint now answers with a proper at-capacity result and how long to wait, and the widget shows that in full and points at the settings panel where a visitor can drop in their own Groq or OpenRouter key. Editing the builder no longer re-downloads the avatar on every keystroke.",
"tags": [
"fix",
"improvement"
],
"link": "/assistant"
},
{
"date": "2026-08-16",
"title": "The Avatar Artifact page now shows real avatars, not just its house portrait",
"summary": "/avatar-artifact is the standalone viewer you can paste into any site as a bare iframe, but it only ever rendered one built-in procedural figure, so there was no way to point it at an actual avatar. It now takes ?agent=<id> to show any public agent's avatar, resolved live from the platform, or ?model=<glb url> to show a model directly from an allowed asset host. The agent's name becomes the heading and an Open profile link appears, the Copy embed button hands you an iframe of exactly what you are looking at, and avatars with no baked animation play the shared idle clip retargeted onto their own skeleton instead of standing in a T-pose. Every way it can fail now says what happened and what to do next: an id that matches no agent, an agent whose avatar is private or not created yet, a model URL from a host the viewer will not render, a model that would not load, and a browser without WebGL. Loading a large model reports its download progress instead of guessing, and a slow but healthy download is no longer cut off as a failure. With no parameters the page keeps its original portrait.",
"tags": [
"feature",
"improvement",
"fix"
],
"link": "/avatar-artifact"
},
{
"date": "2026-08-16",
"title": "AR Studio: the language picker no longer sits on top of the dock controls",
"summary": "On the AR Studio page the site-wide floating language picker landed in the bottom-right corner, directly over the right-hand end of the control dock, so the Clear button (and the buttons beside it on a phone) could not be tapped at all. The picker now lives in the studio's own top bar, where it is always reachable and leaves the dock its full width. It also shrinks instead of pushing its neighbours off screen on narrow phones, which fixes the same crowding anywhere else it appears in a tight header. The page description now also names every place models can come from: your forges, the community feed, the free CC0 object library, and any GLB link.",
"tags": [
"fix",
"improvement"
],
"link": "/ar/studio"
},
{
"date": "2026-08-16",
"title": "The artifact builder now reports what it actually measures, and stops flagging legal artifacts as over budget",
"summary": "The /artifact page is the pre-flight for pasting a three.ws agent into a Claude conversation: it builds the self-contained HTML, renders it in a sandbox that mirrors Claude's, and reports what you are about to paste. Two of those numbers were misleading. The stat labelled First paint was really the time the endpoint took to hand over the bytes; nothing had been rendered yet when it was measured. It is now split into Fetch, which is that download, and Render, which times how long the sandboxed frame takes to parse the document and start the viewer. The size gauge was grading the finished artifact against the size limit for the model inside it, but inlining base64-encodes the model and costs four bytes for every three, so a perfectly acceptable 7 MB artifact was painted amber and anything built from a model over about 5.6 MB went red, contradicting the endpoint that had just accepted it. Size is now graded against the roughly 8.6 MB an accepted model can actually produce, and hovering the number tells you how much of that budget you have used. The Size budget note explained the same thing incorrectly and now spells out the base64 cost.",
"tags": [
"fix",
"improvement"
],
"link": "/artifact"
},
{
"date": "2026-08-16",
"title": "The Avatar CLI page terminal is readable by a screen reader, and its recovery link is reachable by keyboard",
"summary": "The /avatar-cli page teaches the four commands of @three-ws/avatar-cli, and the replayed terminal in its hero was marked up as one flat image. A screen reader announced a single caption for it and hid everything inside: the transcript itself, and the Replay button that restarts it. The transcript is now a named region a screen reader can read line by line, Replay is an ordinary button again, and the transcript reports itself as busy while it is still typing rather than being read half-written. Below it, the live viewer proves the embed snippet works, and when the component loader cannot be reached it swaps in a notice with a link to the web component reference. That link is the only way out of the failure, and it neither looked like a link nor showed a focus ring when tabbed to; it now does both. The scrollable code blocks, which the browser makes keyboard focusable, now show the site focus ring instead of a hairline default. And the copy button's clipboard fallback used to tell everyone to press Command C, including on Windows and Linux; it now names the shortcut for the keyboard you are actually on.",
"tags": [
"fix",
"improvement"
],
"link": "/avatar-cli"
},
{
"date": "2026-08-16",
"title": "Scene Capture fits a phone screen, and a file that is not a point cloud now says so",
"summary": "Scene Capture turns a video of a room into an explorable 3D point cloud you can orbit in the browser, and also renders any .ply you already have. On a phone the viewer stage forced the page 553px wide against a 320px screen, so the whole page scrolled sideways and the side panel sat off the edge; the stage now sizes to the screen and the page fits at 320px. Dropping in a file that is not a point cloud used to look like a success: the viewer reported 0 points over an empty black stage and offered the broken file back as a download. It now tells you the file is not a valid .ply and what one should contain. The error and progress headings were also being overwritten a second after they appeared, so every specific message (Could not fetch that point cloud, Reconstructing) reverted to the generic Something went wrong or Working; the real message now stays on screen. A reconstruction whose job stops responding no longer counts up forever: after a run of failed checks, or past twenty minutes, the page says what happened and how to restart. When the worker caps a long clip at its frame budget, the viewer says the clip was truncated instead of presenting a partial scene as the whole thing. And when reconstruction is offline, the message is written for you rather than repeating a server deployment checklist. Accessibility: the drop area is a real button you can reach with the keyboard, every field in the panel is labelled, and progress, errors and the point count are announced to screen readers.",
"tags": [
"fix",
"improvement"
],
"link": "/capture"
},
{
"date": "2026-08-16",
"title": "CA to x402 now shows the momentum and risk read its own paid endpoint sells",
"summary": "Paste a contract address at /ca2x402 and the resolver has always returned four timeframes of price change and a 0-to-100 risk score with the reasoning behind it, the same intel an agent gets when it pays the generated endpoint $0.01 USDC. The page quietly threw both away and rendered only the 24h number, so a buyer could not see what they were about to pay for. Both now render: a momentum strip across 5m, 1h, 6h and 24h that shows whether a move is building or fading, and a risk panel with the score, its level, and the depth, age, float and flow checks that produced it, each marked by severity. Both read at 320px. The primary Resolve button was also painting white text on the white accent, so the page's main action looked like an empty slab; its label is now legible in both themes. And the error card no longer says \"Could not resolve\" for a mistyped address: a typo, a token with no market, and a resolver outage each say which one happened. Over on /capture, the URL and settings fields get a real focus ring for keyboard users, where an outline reset had left only a faint border tint.",
"tags": [
"improvement",
"fix"
],
"link": "/ca2x402"
},
{
"date": "2026-08-16",
"title": "Blockchain TVL by Chain is readable on a phone, and every chain in it is now a real link",
"summary": "The chain leaderboard at /chains ranks 100 blockchains by total value locked and shows each one's dominance share of on-chain DeFi. On a phone the table was pinned to a fixed 540px width, so the only column that fit on screen was the chain name: the TVL and dominance figures, the entire point of the page, sat off the right edge behind a sideways scroll with nothing to hint they were there. The table now sizes itself to the screen, and all three columns read in full at 320px. Opening a chain's detail page was hidden in a similar way. Rows navigated on click but showed no clickable cursor, and the destination was not a link at all, so it could not be opened in a new tab, copied, followed from a screen reader's list of links, or seen by a search engine. Each chain name is now a genuine link, the row still follows it when you click anywhere along it, and the sortable column headers announce their sort state to screen readers. When the data provider fails, the page now offers a Try again button that refetches in place instead of a link that reloads everything, and it keeps its DeFiLlama attribution instead of going blank. If the provider reports nothing at all, the page says so rather than presenting a $0.00 total as though cross-chain DeFi had emptied out.",
"tags": [
"improvement",
"fix"
],
"link": "/chains"
},
{
"date": "2026-08-16",
"title": "Coin Autopilot tells you whether your rules actually saved",
"summary": "Coin Autopilot is where you set the floors that decide when your coin buys itself back and when it pays creator fees out to holders, so a rule that looks set but never reached the database costs you real money. Two ways that could happen are fixed. A threshold you typed was only sent when the field lost focus, so typing a new floor and then closing the tab or hitting back threw it away silently; it now saves shortly after you stop typing, and anything still in flight is sent as the page closes. And the only confirmation was a toast that vanished after a couple of seconds, which meant a failed write left the switch showing a setting the server had never accepted. Each coin card now reports its own save: Saving, then Saved, or a Not saved marker that names the reason and stays put with a Retry button until the write lands. A coin you have never configured is marked Defaults, so the platform's inherited settings can no longer be mistaken for rules you chose. If the service goes down the page says so in both columns instead of leaving loading placeholders spinning, and it restores itself when the service returns rather than waiting for you to press Retry. Signed out, it stops retrying an endpoint that will keep refusing. Screen reader and keyboard fixes throughout: every switch and floor field is now named, save results are announced, the loading state is cleared when the data arrives, and the number fields have a visible focus ring.",
"tags": [
"fix",
"improvement"
],
"link": "/autopilot"
},
{
"date": "2026-08-16",
"title": "Autopilot Activity keeps up when you switch agents, and its receipts link to the exact memory",
"summary": "Autopilot Activity lists every action your agents took on your behalf, and it has an agent picker for narrowing that list to one of them. Picking a second agent while the first was still loading used to do nothing at all: the picker and the address bar moved to the new agent, but the request for it was never sent and the page kept showing the previous agent's actions, with no sign anything had gone wrong. Whichever agent you picked last now wins, however fast you switch. Two more repairs: each receipt cites the memories that motivated the action, and those links now scroll to the cited memory and highlight it instead of dropping you at the top of a long Knowledge list; and when an agent has no actions yet, the page sends you to that agent's own autopilot controls rather than the dashboard, which is where the setting you need actually lives.",
"tags": [
"fix",
"improvement"
],
"link": "/autopilot-activity"
},
{
"date": "2026-08-16",
"title": "Crypto Categories is searchable, and opens in a fraction of the time",
"summary": "The sector leaderboard at /categories renders every category the market data provider reports, and that list has grown past 750. All of them were being drawn into the page at once, roughly 2,250 coin icons in a single paint, and the only way to find a sector was to scroll past every one ranked above it. The table now opens with the top 50 and pages in 50 more at a time, and a search box narrows the whole list by name before you ever need to page. Full category names are shown too, instead of being cut off mid-word on a wide screen with half the row empty. The three things that can go wrong now read differently and each offers a way out: a failed load says the data provider did not answer and gives you a Try again button that refetches, a provider reporting nothing says so plainly, and a search that matches nothing offers to clear itself. Sortable column headers now announce themselves to screen readers, and keyboard focus stays on the column you just sorted instead of being dropped back to the top of the page.",
"tags": [
"improvement",
"fix"
],
"link": "/categories"
},
{
"date": "2026-08-16",
"title": "The FLAME paper link in the Avatar Engines Atlas opens the paper again",
"summary": "FLAME is the parametric head model underneath most of the photoreal-head engines in the atlas at /avatar-engines, so its Paper button is one a builder actually follows. It pointed at a Max Planck publication page that answers every visitor, browsers included, with a 403 Forbidden, so the button led to an error page instead of the paper. It now opens the paper PDF that the FLAME project site itself links to. All 68 outbound links on the page were re-checked afterwards and every one of them resolves.",
"tags": [
"fix"
],
"link": "/avatar-engines"
},
{
"date": "2026-08-16",
"title": "The model viewer tells you when a file will not open, and how to recover",
"summary": "The viewer at /app had two dead ends. Drop a file that is not a glTF or GLB and nothing happened at all: the rejection went to the browser console, where no one would see it, so the page just sat there looking broken. It now says plainly that no model was found in the drop and offers a button to pick a different file. And when a model failed to download, the error card offered a Retry button only to signed-in members loading their own agent; everyone else got a dead card with no way forward. Retry now works for any load, including a file you dropped yourself. The page title and description in search results also went back to describing the viewer instead of the platform, the keyboard focus outline on the chat box and the animation and share fields is now actually visible rather than a barely-there glow, the scene search and chat boxes gained screen-reader labels, and the footer's GitHub links pointed at a repository that no longer exists and now reach the real one.",
"tags": [
"fix",
"improvement"
],
"link": "/app"
},
{
"date": "2026-08-16",
"title": "The background agent mint says out loud whether it ran",
"summary": "Every avatar that finishes a scan or a prompt is given a draft on-chain identity in the background. That step is deliberately best-effort, so it can never cost you your avatar, but it was also completely silent: whether it minted, was skipped because the deployment holds no minting key, or was switched off entirely, the logs looked identical, which is how a run of avatars can go by without anyone noticing the identity was never created. Each finished avatar now records the outcome of both chains, including the reason a leg was skipped. The documentation for the feature also gained a way to prove the whole path on a machine that cannot reach the public test-network faucet, which is routinely exhausted on shared build machines: run your own node with the real Metaplex programs, and the proof reports which node served it rather than handing back an explorer link that leads nowhere.",
"tags": [
"improvement",
"docs"
],
"link": "/docs/draft-agent-mint"
},
{
"date": "2026-08-16",
"title": "The arbitrage board says which price feeds it actually reached",
"summary": "The x402 Arbitrage page compares what different providers charge for the same capability, and it builds that comparison from several facilitator catalogs at once. When one of those catalogs was unreachable, its listings simply vanished from every comparison and the page said nothing about it. A scan that had only seen half the market looked exactly like a market with fewer price gaps in it, which is the opposite of what the page is for. Each scan now names the facilitators it reached and how many listings each one contributed, and calls out any that dropped out along with the reason, so a short board is never mistaken for a quiet one. Separately, closing the payment window without paying used to leave the Pay cheapest button dead for four seconds before it would accept another click. Deciding not to pay now hands the button straight back.",
"tags": [
"fix",
"improvement"
],
"link": "/arbitrage"
},
{
"date": "2026-08-16",
"title": "Agora citizens show the craft they actually work, and the square fits a phone",
"summary": "Every one of the 200 citizens in the Commons was labelled a Fetcher. The nameplate over each avatar, the citizen list that screen readers and keyboard users navigate by, and the passport header all read the citizen's capability list rather than the craft it is known for. Every citizen carries the fetcher capability, because calling a paid service is the one job anyone in the Commons can take, so that list always begins with Fetcher no matter who you were looking at. Mira the Appraiser and Aria the Sculptor both wore a FETCHER chip, sitting under a coloured ring that was already showing their real craft. Citizens now say what they do: the crowd, the list and the passport agree with each other and with the colour, and a passport leads with the primary craft before the rest of what that citizen is able to claim. On a phone the bottom of the square was a pile-up. The job board, the sign-in bar, the language switcher and the Enter the Commons button all pinned themselves to the same strip of screen, so at 320px wide the button that takes you into the world was completely buried under two other controls and the board's last rows were cut off. Those layers now stack above one another at every screen size, and the stack follows the sign-in bar as it changes height.",
"tags": [
"fix",
"improvement"
],
"link": "/agora"
},
{
"date": "2026-08-16",
"title": "The BNB Chain page now says how many demos are actually live",
"summary": "The BNB Chain page checks each of its three demo tracks against the running deployment and reports the tally in its header. That line was stuck: it read \"Checking which tracks are live\" forever, even once all three cards beside it had finished and said Live. The page's own result was being written first and then overwritten by the translation pass that swaps the page into your language, so the finished count never survived. The header now holds its result and reads \"3 of 3 demo tracks live now\". The same checks were also leaving four cancelled requests in the browser's network log on every visit, which made a healthy page look like it was failing to anyone with developer tools open. They now close cleanly.",
"tags": [
"fix"
],
"link": "/bnb"
},
{
"date": "2026-08-16",
"title": "The agent wallet is usable on a phone again, and visitors see a real balance",
"summary": "On a narrow screen the wallet hub was quietly cut off. The column sized itself to the width of its tab strip rather than to the screen, and because the page cannot scroll sideways everything past the fold was simply unreachable: the Refresh button and the right edge of every card were invisible on a phone. The column now fits the screen and the section tabs scroll on their own, so every control is reachable at any width. Two things were also wrong for anyone looking at an agent wallet they do not own. The balance read as a dash even though an agent's balance is public on-chain data, and it now shows the real live figure. The recent-activity card claimed it had failed to load and offered a Retry that could never work, because the transaction feed is private to the wallet's owner; it now says exactly that and links to the block explorer, where the same transactions are public. Finally, the browser tab title sometimes flipped back to a generic label instead of naming the agent whose wallet you were looking at.",
"tags": [
"fix",
"improvement"
],
"link": "/agent-wallet"
},
{
"date": "2026-08-16",
"title": "The AGI page no longer goes quiet when its 3D body cannot load",
"summary": "The AGI page is built around a live 3D body that reacts to the market, and when that model could not be fetched the page said nothing at all. The loading line disappeared on a timer whether or not anything had arrived, so the centrepiece became a silent empty panel with no explanation and no way to try again, while the decision stream and track record beside it carried on as normal. A body that fails to load now falls back to the platform's own avatar, so there is still something on the stage reacting to the agent's mood. If that cannot be fetched either, the stage says so plainly, makes clear that the agent is still trading and that the decisions and track record below are live and unaffected, and offers a button to try embodying again. The loading line now stays up until the body genuinely arrives instead of vanishing early, and the stage no longer re-requests a model it already knows is unreachable every twenty seconds. The page's description shown in search results and link previews was also carrying an older wording than the page itself.",
"tags": [
"fix",
"improvement"
],
"link": "/agi"
},
{
"date": "2026-08-16",
"title": "Live Agents: every card on the wall now shows what its agent is actually doing",
"summary": "The Live Agents wall opened a separate live connection for every card on screen, and a browser will not give a page that many. The first handful connected, and the rest, more than forty of the forty-eight on the first page, sat on Connecting and an empty screen for as long as you left the tab open. Scrolling did not help them, because the connections that had been handed out were never given back. The wall now loads the real recent activity for the whole page in a single request, so every card has a genuine terminal of that agent's on-chain and skill actions the moment it appears, however far down the list it sits. Live connections are kept for what actually needs pushing: an agent with a real browser feed running, the one holding the spotlight, and the card you hover or tab to, which goes live under your cursor. A card that never manages to connect now falls back to its real activity instead of claiming to be connecting forever, the standby screen tells you how many actions an agent has recorded rather than flatly saying it has none, and the reaction buttons on a card no longer run off the edge of a phone screen.",
"tags": [
"fix",
"improvement"
],
"link": "/agents-live"
},
{
"date": "2026-08-16",
"title": "You can now teach an agent your GitHub work without waiting on an OAuth app",
"summary": "Seeding an agent from GitHub, the flow where you tick your profile, repositories and READMEs and the agent learns to talk about what you build, needed a GitHub sign-in button that this site never had switched on. The Connected accounts card said so honestly, and that was the end of it: an Unavailable tag and nothing you could do. There is now a second way in that needs nothing set up on our side. You create a read-only token on GitHub yourself, tick read:user and nothing more, paste it into Settings, and the same consent list, the same seeding and the same one-click deletion all work exactly as before. Consent still happens on GitHub's own screen, where you choose what the token may do, and we check what you chose: a token carrying access the feature cannot use, anything reaching your private repositories or able to delete them, is refused on the spot and never stored, with the offending permission named so you know which box to untick. The token is encrypted before it is saved and destroyed when you disconnect. One honest difference from the sign-in route: disconnecting deletes our copy and every memory it produced, but GitHub gives no way for a token to delete itself, so the confirmation now tells you to remove it on GitHub too and links you straight there.",
"tags": [
"feature",
"improvement"
],
"link": "/settings"
},
{
"date": "2026-08-16",
"title": "Paying with $THREE works again: credits top-ups, tips and purchases stopped failing before they were signed",
"summary": "Every attempt to top up credits with $THREE died in your wallet with a message about an incorrect program id, before you could approve anything and before any money moved. $THREE is issued under Solana's newer token standard, Token-2022, but the pages building the transfer were addressing the older one. That points the transfer at an account that does not exist, so the network rejected it during the safety check your wallet runs first. The credits page, agent tipping and the in-page purchase flow now read the standard directly from the coin itself and build against whichever one it actually uses, so $THREE and ordinary SPL tokens both go through. Two related touches on the credits page: it now tells you up front if the connected wallet holds no $THREE, or not enough for the amount you typed, instead of letting the wallet fail with something cryptic. The same mix-up was quietly hobbling the treasury side: the buyback and micro-buy lanes read their own $THREE balance from the wrong place, always saw zero, and so never moved what they had bought into the treasury. They now read and move it correctly.",
"tags": [
"fix"
],
"link": "/credits"
},
{
"date": "2026-08-16",
"title": "Alpha Co-pilot counts its live launches again, and a mistyped agent ID no longer leaves the last one on stage",
"summary": "Four things were wrong on the page where your agent reads a launch out loud. The badge next to Live launches, which tells you how many launches are waiting, was always blank: the counter was written inside the heading text, and translating the heading replaced the very element the page updates, so the number was thrown away every time. It now sits beside the heading and survives a switch into any of the 85 languages. Loading an agent that does not exist used to leave the previous agent standing on stage with its launch cards still on screen, and every Ask for a read button on those cards did nothing at all when clicked; the stage is now cleared back to its starting state with a prompt to pick another co-pilot. The buy button in the confirm drawer showed the amount with its unit twice over. And on a local or preview copy of the site the avatar's 3D body failed to load at all, because this page was the only one that did not route model files through the same asset path every other page uses.",
"tags": [
"fix"
],
"link": "/alpha-copilot"
},
{
"date": "2026-08-16",
"title": "The Animation Gallery keeps its keyboard, its focus, and its promises about what will play",
"summary": "The gallery of nearly 3,000 motion clips had a set of small breakages that added up. Roughly a third of the mocap catalogue has no picture made for it yet, and the page was asking the picture store for one anyway: every one of those requests came back empty, so a third of the cards spent a request to arrive at the same drawing they now show immediately. Opening a clip's details trapped nobody: pressing Tab walked out of the open window and into the page behind it, and closing the window dropped you back at the very top of the page instead of the card you came from. The playback controls were on show while a preview was still loading and stayed on show when it failed, so there were buttons that did nothing. Arrow keys jumped to the next clip even while you were dragging the speed slider, which moved the slider and changed the clip at the same time. Sharing a filtered view was misleading too: a link carrying the loops or one-shot filter applied it but left the buttons showing All, and a link with a filter the page does not recognise emptied the sort menu. All of it is fixed, along with the keyboard hint the window shows: Space really does play and pause now, from the moment the window opens.",
"tags": [
"fix",
"improvement"
],
"link": "/animations"
},
{
"date": "2026-08-16",
"title": "The Agent Commerce demo is usable on a phone, and says what happened when a trade cannot finish",
"summary": "The page where two agents pay each other on Solana had three problems. On a narrow screen the language picker in the bottom corner sat directly on top of the Run button, so the demo could not be started on a phone at all, and the agent name plates and speech bubbles were drawn underneath the payment card. The related-page links in the header were rendering at the size and colour of the site brand, which is what pushed the header off the edge of a 320 pixel screen. And when the live stream dropped part way through a trade, the page kept showing the payment step as if it were still running, left the last speech bubble frozen on screen, and explained itself only in a red message that disappeared after eight seconds. The layout now gives every control its own space at 320, 768 and 1440 pixels, the step that was running is marked as failed, and the middle of the screen carries a plain explanation with a button to run it again. Two more states were missing and are now designed: a skeleton card while the wallet balances are being read, and a retry when that check cannot reach the server at all.",
"tags": [
"fix",
"improvement"
],
"link": "/agent-trade"
},
{
"date": "2026-08-16",
"title": "Avatars no longer lose their preview picture forever because a server had a bad minute",
"summary": "Every public avatar gets a preview picture rendered for it in the background, and an avatar that fails to render three times is given up on permanently. The problem was what counted as a failure. If the machine ran out of memory and could not start the renderer at all, or if the picture store was unreachable, that counted against the avatar, even though nothing was wrong with the avatar itself. Four avatars had already been written off that way for a fault that lasted seconds, and a longer outage would have written off every remaining one inside a quarter of an hour, because the whole batch fails together. A fault in the machinery is now told apart from a genuinely broken model: the avatar keeps its chances and the job simply waits for the machine to be healthy. The two background jobs also stop and do nothing when the picture store is switched off, instead of working through the queue and burning everyone's chances on an error they could not have avoided. The four avatars written off in error have been put back in the queue.",
"tags": [
"fix"
],
"link": "/gallery"
},
{
"date": "2026-08-16",
"title": "Community-voted models reach our Sketchfab showcase with a real name",
"summary": "Models made from a picture rather than a written description carry no prompt to name them by, so they were labelled with the internal route name instead. Models the community voted for skip the usual title checks on purpose, since a winning model should never be dropped over its title, which meant a winner made from a picture was one run away from being published to the official three.ws Sketchfab account under the name Image-To-3d. It was, in fact, the second model queued to go out. Those models are now named from their category, or given a plain generic name when the category says nothing useful.",
"tags": [
"fix"
],
"link": "/forge"
},
{
"date": "2026-08-16",
"title": "Agent Studio shows your real avatar again, and a shared studio link stops opening the wrong agent",
"summary": "Four repairs to the Agent Studio. The live stage on the left is meant to be your agent's actual body, but the 3D renderer was leaking its internal debug controls into that panel, including a stray rainbow colour strip painted over the avatar. Those controls are hidden now, everywhere the live agent appears. Second, opening a studio link that names someone else's agent used to quietly load your own agent instead and put it in a full editor under the other agent's address, so any edit went to the wrong record; it now loads the agent the link names and, when it is not yours, says so and offers its public profile. Third, on a phone the five section tabs ran off the edge of the screen, leaving Money and Skills unreachable behind a hidden scrollbar; all five now fit. Finally the status pill on the stage moved to the top so it stops sitting on top of your agent's name, and the walking companion no longer doubles up in the corner on a page that already shows your agent life-size.",
"tags": [
"fix"
],
"link": "/agent-studio"
},
{
"date": "2026-08-16",
"title": "External Solana agents get their names and artwork back in the directory",
"summary": "Every agent the platform indexes from the Metaplex Agent Registry had been sitting in the directory as a bare address since August 6: no name, no picture, no 3D model. The crawler asks a Solana node for each agent's artwork through a method only some nodes support, and it was asking the same node every time, one that does not support it. Worse, the node that does support it answers a refusal in the same shape when it is simply busy, so the crawler could not tell 'I cannot do this' from 'not right now'. It now tries each node in turn, keeps the one that answers, waits and retries a busy one instead of writing it off, and refreshes the agents that are missing artwork rather than the same handful at the top of the list every half hour. If no node will serve the artwork at all, the crawl now says so out loud instead of quietly marking every agent as failed.",
"tags": [
"fix",
"infra"
],
"link": "/agents"
},
{
"date": "2026-08-16",
"title": "A dropped page in the agent directory no longer throws away everything you scrolled past",
"summary": "The agent index loads in pages of 24, and until now a network hiccup on page five did more damage than the hiccup deserved. The Load more button vanished, and the only recovery on offer restarted the list from the top, discarding the hundred-odd agents already on screen. A failed page now says so on its own terms and offers Resume loading, which picks up exactly where it stopped; Start over is still there if you want it, and the error notice clears itself once the page arrives instead of lingering under the results. Three smaller repairs came with it: loading another page no longer redraws the agents already listed, so the grid stops flickering once a few pages are in; the search box and sort menu now show a proper focus ring for keyboard users instead of a one-pixel border tint; and the source filter chips reach a full-size tap target on tablets, not just on phones.",
"tags": [
"fix",
"improvement"
],
"link": "/agents"
},
{
"date": "2026-08-16",
"title": "Prompt polish keeps its quotes out of your model, and a busy Forge says so plainly",
"summary": "Three fixes across the Forge endpoints. The Enhance button, which rewrites your rough idea into the prompt the generator actually draws from, stopped removing the quotation marks the writing model wraps around its answer: those quotes were being handed straight to the image step as if you had asked for them to be drawn. Both plain and curly quotes are stripped again. Second, when you ask the studio to change a model you already made and the generator is saturated, you now get an honest \"busy, try again in a moment\" with a wait time instead of a vague failure that blamed the wording of your request. Third, the self-hosted generator panel now tells a caller which input was missing rather than reporting the whole feature as unconfigured, and its status light no longer claims nothing is set up when the only problem was the address you typed.",
"tags": [
"fix",
"improvement"
],
"link": "/forge"
},
{
"date": "2026-08-16",
"title": "Web search answers again, and a stray address can no longer tie up the server",
"summary": "Three repairs on the public API. Web search had been answering nothing but an error: the question went to Google's grounded search as designed, but the model was given a single budget to both think and reply, spent all of it thinking, and came back empty, which we correctly reported as a failure and wrongly did so on every single query. Thinking now has its own allowance on top of the answer, and when a search does come back empty the log says why instead of leaving the next person to guess. Second, the machine-readable directory at the root of the site, the file other agents read to learn what three.ws sells, would hang forever on a handful of odd names rather than saying no such thing, and a request that never ends holds a server slot until it is cut off. Every unknown name now gets a plain not-found. Third, at checkout, a wallet that hands back a payment the network could never accept is now told so at the moment it happens, instead of the modal saying sending and the error arriving later from the payment processor.",
"tags": [
"fix",
"security"
]
},
{
"date": "2026-08-16",
"title": "Seeding an agent from your X account no longer asks for permission to post as you",
"summary": "Teaching an agent your voice from your public X posts required connecting X first, and that connect asked X for everything the platform can do with an account, including posting and uploading media on your behalf. Reading your posts once and writing as you are two different things to agree to, and only the first one is what seeding needs. The Connect button on the X card in Settings now asks X for read access to your profile and posts and nothing else, and it says so before you click. Posting is its own connect, offered right beside it and labelled as such, and Reconnect keeps whichever of the two you already chose rather than quietly widening it. If a connection cannot read your posts at all, the card now says which permission is missing and offers a reconnect instead of showing you a consent screen for a seed that could not run, and an agent trying to post through a read-only connection is told exactly that instead of failing against X with nothing to act on. Everything else about the lane is unchanged: nothing is read until you consent on a screen that lists what is read and stored, and revoking still deletes every memory the seeding produced.",
"tags": [
"improvement",
"security"
],
"link": "/settings"
},
{
"date": "2026-08-16",
"title": "The Forge community board remembers what you liked",
"summary": "The community showcase on /forge, and the same feed on the daily page and the model explorer, kept showing your own likes as un-liked. The board is held at the edge for a minute so it loads instantly for everyone, but the copy being held was the one built for a visitor who is not you, and it was handed to you as well. Your like button then started from the wrong state, and pressing it again fought with what the server already knew. The board now keeps a separate copy for a signed browser, so the hearts you filled in stay filled. Alongside it, a security fix on the Game-Ready export: when the export was paid for with $THREE rather than covered by holding, a badly formed payment reference made the server answer with the raw internals of its own database instead of a plain \"that payment could not be verified\". It now says only what a caller needs, and a database problem is no longer reported as a bad payment, so nobody is told to pay twice for an outage on our side.",
"tags": [
"fix",
"security"
],
"link": "/forge"
},
{
"date": "2026-08-16",
"title": "The Solana agent directory is live again, and complete",
"summary": "Discover and the deployments view list agents registered on the Metaplex Agent Registry, refreshed by a crawl that runs every half hour. The crawl had been reading nothing since 6 August: it asked one fixed Solana node for the registry, that node started refusing the request outright, and because it never tried any of the other nodes it fell back to, every run reported success while writing nothing. It now works its way down the same list of providers the rest of the platform uses, so one provider turning us away costs a second, not ten days. Fixing that surfaced a second fault underneath it: the registry holds two generations of agent records, and the crawl was reading both with the newer generation's layout, so older records came back as nonsense and the newer pass gave up entirely at the first old record it met. Each generation is now read on its own terms. The directory went from 1,558 stale entries to all 1,571 currently on chain, and new registrations show up within the half hour again. Separately, the treasury top-up that keeps the platform's wallets funded no longer fails outright when a Solana provider answers with something it cannot read: it treats that the same as a provider being busy and retries on the next pass, instead of skipping a funding cycle.",
"tags": [
"fix",
"infra"
],
"link": "/discover"
},
{
"date": "2026-08-16",
"title": "An unsubscribe now sticks, even if you click an old confirm link",
"summary": "Every three.ws newsletter email carries a confirm link and an unsubscribe link that share one token, so a reader who unsubscribed and later opened an older email and clicked confirm was quietly put back on the list. That link no longer signs anyone up: it says you are unsubscribed and leaves it that way, and opting back in starts from the site. Three other fixes ship alongside it. The allocation planner at /meta-allocator no longer serves an empty basket when the model declines to answer, it falls back to the allocator that always produces a real spread across verified leaders. The motion capture uploader told you an upload link lasted ten minutes when it really lasted five, so a large video queued on that promise could fail with nothing explaining why. And an agent whose stablecoin payment is rejected by the x402 Bazaar now gets the Bazaar's own details back instead of a different server's, so its retry is aimed at the right place.",
"tags": [
"fix"
],
"link": "/changelog"
},
{
"date": "2026-08-16",
"title": "A GitHub seed that stores nothing no longer costs you six hours",
"summary": "Seeding an agent from your GitHub profile, repositories and READMEs is allowed once every six hours per agent. Two failures used to spend that window without storing a single fact. The first is entirely ours: when every model provider is busy at once, nothing is read and nothing is written, and you were still locked out for six hours over an outage you did not cause. The second is the run that comes back with no usable facts, which tells you to add a README or another repository, advice you could not act on until the window expired. Both now hand the window straight back, so you can try again as soon as a provider frees up or as soon as you have widened your selection, and the reply says which of the two happened. Nothing else changes: your existing memories are untouched by a failed run, the seed still reads only the profile, repositories and READMEs you ticked, and disconnecting GitHub still deletes every fact it wrote across all your agents.",
"tags": [
"improvement",
"fix"
],
"link": "/docs/github-memory-seeding"
},
{
"date": "2026-08-16",
"title": "Avatar link previews stop rebuilding themselves, and two feeds answer a bad page link plainly",
"summary": "An audit of the chat, media and board endpoints found three real breaks. When you share an avatar link, the preview image is drawn once and then kept, so every later share is instant. A mix-up between the stored image and its address meant every freshly drawn preview was mistaken for an old broken one: the saved copy was thrown away and the picture was redrawn from scratch on every single share. The preview is now kept, as it always should have been, and genuinely broken old records are repaired the next time they are drawn. Separately, the character directory answered a hand-edited or stale \"next page\" link with a server error instead of saying the value was not a timestamp, and the bounty board did the same when the page size in the address was not a number. Both now answer plainly, and the board simply falls back to its normal page size.",
"tags": [
"fix"
],
"link": "/characters"
},
{
"date": "2026-08-16",
"title": "Forge tools refuse a made-up style or file type, and a worker hiccup no longer reads as a broken job",
"summary": "Three repairs to the Forge model tools. The stylize filter (voxel, brick, voronoi, lowpoly) checked the requested style against its list in a way that quietly accepted a handful of reserved words, so a request naming one of them was sent to the worker with an unknown filter and no density setting instead of being turned down. The same flaw let the reference-image uploader accept a made-up file type and hand back an upload slot with a nonsense filename. Both now check their list properly, and an unrecognised value falls back to the default or is refused. Separately, while the background remover, mesh optimiser, part splitter and stylizer were checking on a running job, a momentary fault at the worker was reported as a crash on our side, once for every check, which buried the real signal in the alerts. A fault now answers plainly and the page keeps waiting, which is what a passing hiccup deserves.",
"tags": [
"fix",
"improvement"
],
"link": "/forge"
},
{
"date": "2026-08-16",
"title": "Scrolling the activity feed past a bad cursor no longer breaks the whole feed",
"summary": "An audit of the account endpoints found one real break and two smaller ones. The activity feed on /community and your personal feed ask for the next page of results by passing the last item’s timestamp back; if that value arrived malformed, from a hand-edited link or a stale bookmark, the database rejected it and the entire feed came back as a server error with nothing to act on. It now answers with a plain \"that is not a timestamp\" so the page can recover and reload from the top. The referral code editor stated its length rule with a typographic dash that read as a minus sign on some devices, and now spells it out. The purchase history endpoint also documented the wrong default view: it always merges your purchases and your sales, which is what the dashboard shows.",
"tags": [
"fix"
],
"link": "/community"
},
{
"date": "2026-08-16",
"title": "Agent share cards survive an unconfigured bucket, and a mistyped payment mint is refused instead of listed",
"summary": "An audit of the agent endpoints turned up four things worth fixing. A skill price set through /api/agents/:id/skills/set-price accepted any text as its payment mint and saved it, so a typo produced a listing nobody could actually pay; it now has to be a real base58 address, matching the other door onto the same table. The agent share page and its OG image used to answer a social crawler with a hard error whenever object storage was not configured, instead of falling back to the plain card they already knew how to draw. A mistyped agent id in the diary, skill-price and screen-push endpoints came back as a server error rather than the plain \"that is not an agent id\" it is. And an agent screen with no live frame store now tells you the agent is offline right away instead of leaving you on \"Connecting\" for five minutes.",
"tags": [
"fix",
"improvement"
]
},
{
"date": "2026-08-16",
"title": "The AWS page stops promising AWS invoicing three.ws does not do, and the GLB optimizer reads compressed avatars again",
"summary": "Three corrections found by reading the docs pages against the code they describe. The AWS page told you a three.ws subscription would bill usage onto your AWS invoice, draw down AWS credits, and count toward Enterprise Discount Program commitments. None of that is the plan: the Marketplace subscription is free, and usage is paid per call in USDC over x402, exactly as it is for any other caller, so there is nothing to reconcile on your AWS bill at all. The page now says that, and describes the free subscription, the single key that opens every x402 endpoint, and the absence of any trial clock. Separately, the GLB optimization command documented on the 3D asset pipeline page failed on every compressed avatar, which is most of them, because it loaded the glTF extensions without the matching decoders; it now decodes meshopt and Draco and leaves the output as plain glTF the way its own contract promises. The NVIDIA models page also listed the tool-capable fallback model under the wrong vendor prefix.",
"tags": [
"fix",
"docs"
],
"link": "/aws"
},
{
"date": "2026-08-16",
"title": "Wallet reputation stops losing coins it never got around to judging",
"summary": "The rollup that turns launch history into wallet reputation was judging 80 coins every five minutes while the intel firehose was recording about 32,000 new mints a day, so it fell roughly 8,000 coins a day behind. That backlog does not wait its turn: the rollup works oldest-first and only reaches back 14 days, which is exactly when retention prunes the record, so the tail of the queue aged out and those buyers never counted toward anyone’s reputation at all. 309,000 coins were sitting unjudged with the oldest already 8.4 days from the cliff. The per-run budget is now 400, which outruns the firehose several times over, drains the backlog within days and then idles. Scores across the smart-money surfaces get their full evidence base back. The operations health board also tracks the retention and reconstruct sweeps, so a stall in either is visible instead of showing up later as a symptom.",
"tags": [
"fix",
"infra"
]
},
{
"date": "2026-08-16",
"title": "The agent showcase pages all the way through, and the strategy board ranks the whole marketplace",
"summary": "Three fixes to public read surfaces. Paging the ERC-8004 agent showcase past the first page could repeat a row forever or quietly skip one, because the cursor rounded the registration timestamp to the millisecond while the index stores microseconds, and an agent whose registry event carried no timestamp at all fell out of every page after the first. Both are fixed, and asking for a fractional number of results now returns a page instead of a server error. The published strategy board's default 'performance' order was ranking only the most recently published slice, so a proven, high-return strategy could sit invisible behind newer unproven ones; it now ranks the whole published pool before trimming to your page. And /sitemap.xml no longer answers a POST or a DELETE with the sitemap.",
"tags": [
"fix"
],
"link": "/showcase"
},
{
"date": "2026-08-16",
"title": "Token sentiment reads real pump.fun commentary again",
"summary": "The sentiment pulse for a Solana token was reading a pump.fun route that no longer exists, so it scored an empty set and reported every coin as perfectly neutral with full confidence. It now reads the coin's live callouts, the commentary feed the pump.fun coin page renders, and when that source is unreachable and you supplied no text of your own it says so with an error instead of returning a neutral-looking score. The scorer itself was also matching lexicon terms inside longer words, which read 'partnered' as bearish; it now matches whole words and knows the common variants (scammer, rugged, crashing, bought).",
"tags": [
"fix",
"improvement"
],
"link": "/tutorials/emotion-from-data"
},
{
"date": "2026-08-16",
"title": "Agent attestations that stopped indexing now recover on their own",
"summary": "An agent's on-chain attestation history is indexed by walking Solana signatures newer than the last one we recorded. If that recorded signature aged out of the RPC node's history, the walk failed outright and kept failing: the same dead signature was replayed on every run, so that agent's attestations silently stopped appearing on its profile while the scheduled job still reported success. The crawler now recognises that case, re-scans from the most recent signatures instead, and clears the stale marker, so an agent that fell behind catches up by itself on the next pass.",
"tags": [
"fix",
"infra"
]
},
{
"date": "2026-08-16",
"title": "The creations gallery now counts every creation a creator has made, and pages all the way to the end",
"summary": "The gallery ranked creators and paged the feed over only the newest 120 creations, however many the feed actually held. A creator's older work was invisible in their profile count, the top-creator board scored people on a fraction of their output, and scrolling the gallery dead-ended early with items still left to show. It now reads the full feed depth it always meant to, and does it in two database round trips instead of two per item, so the deeper read loads faster than the shallow one did.",
"tags": [
"fix",
"improvement"
],
"link": "/creations"
},
{
"date": "2026-08-16",
"title": "The KOL, Loom and marketplace MCP servers now tell you why a call failed, and ship runnable examples",
"summary": "When a market-data provider was rate-limited, these servers passed back only a bare code like provider_unavailable and threw away the sentence explaining it. They now surface the full reason. The KOL portfolio tool used to answer a provider outage with an empty card that read as a quiet wallet with no trades; it now reports the outage as an outage, so nobody sizes up a trader from a blank. The KOL and Loom packages also gained runnable examples that talk to the live public API, and the naming SDK now answers an unregistered .eth name the same way it answers an unclaimed .sol one, with data instead of an exception.",
"tags": [
"fix",
"sdk"
]
},
{
"date": "2026-08-16",
"title": "The agent wallet MCP server now quotes its own address when it asks an agent to pay",
"summary": "Agents that connect to the wallet server at /api/mcp-agent without a three.ws account can pay per call in USDC instead. The payment request it handed back named the main three.ws MCP server, not itself, so a paying agent signed a payment scoped to an endpoint it had not called and service directories listed the wallet server under the wrong description and tool list. It now describes itself: its own address, its own wallet-and-payments summary, and a directory example that only searches for services rather than spending anything. Separately, the one-sentence agent builder no longer mangles accented names when it suggests a handle: Munz now becomes munz rather than mu-nz.",
"tags": [
"fix"
],
"link": "/docs/mcp-agent"
},
{
"date": "2026-08-16",
"title": "Creator subscription tiers save their edits again, and a mistyped subscription link says so plainly",
"summary": "Editing an existing subscription tier from the creator dashboard silently failed: the save never reached the server, so the new name, price or status was lost every time. Saving works now. Alongside it, a mistyped or truncated subscription or tier link used to answer with a generic server failure; it now says the id is not valid, which is what a reader can actually act on. Tier pages are also directly addressable, so a single tier can be linked and read on its own, with unpublished drafts still visible only to the creator who wrote them.",
"tags": [
"fix",
"improvement"
]
},
{
"date": "2026-08-16",
"title": "The $THREE price quote survives an out-of-range dollar amount, and the payment audit log is admin-only for real",
"summary": "The public $THREE price endpoint converts a dollar amount into a token amount. Because $THREE trades well under a cent, a dollar figure large enough to overflow what a number can hold turned that conversion into a server error, and every such request also filed a false internal alert. It now answers plainly that the amount is out of range. The token amount it quotes is also computed from the exact price printed beside it, so the two can no longer disagree when the live price refreshes partway through a request. Separately, the settled-payment audit log checked admin wallets with its own copy of the rule, comparing Solana addresses without regard to capitalisation. Solana addresses are case-sensitive, so a lookalike address that differed only in capitalisation would have been accepted; the log now uses the platform-wide admin check, which compares Solana addresses exactly and also recognises an admin who signed in with a linked wallet.",
"tags": [
"fix",
"security"
]
},
{
"date": "2026-08-16",
"title": "Shared agent links keep their preview card through a chain hiccup",
"summary": "The preview card for an on-chain agent is rendered on demand from the chain and the agent's manifest. When either read was slow or briefly unavailable, the card answered as if the agent did not exist, and social sites cached that answer for up to an hour, so a link to a real agent unfurled with no image at all. The card is now served either way and only kept for a minute when a read fell short, so the next crawl picks up the real thing. Alongside it, ten agent endpoints (the preview card and page, the oEmbed provider, memory sync, the action log, ask, delegate and team tasks) now answer a mistyped agent id with a clear, specific error instead of a generic server failure.",
"tags": [
"fix",
"improvement"
]
},
{
"date": "2026-08-16",
"title": "Corrected stale numbers across the avatar and animation docs",
"summary": "An audit of eight documentation pages against the code they describe turned up four claims that had drifted. The Animations page said the motion-capture library held about 2,000 clips when it actually serves 2,874, so both the intro and the gallery section now quote the real size and point at the count the API itself returns. The selfie-to-avatar page listed the wrong default for the face-morph displacement clamp and credited a single tuning sweep instead of the two that chose it. The avatar-creation walkthrough still told people to pick a body type on the selfie flow, a control that was deliberately removed because it changed nothing, and named the wrong fields on the event the page dispatches. The architecture page now says plainly that the whitelist trust mode is reserved rather than enforced.",
"tags": [
"docs"
],
"link": "/docs/animations"
},
{
"date": "2026-08-16",
"title": "The $THREE deploy-burn ledger shows the real burns again",
"summary": "The $THREE page counts 1,000 $THREE burned for every agent deployed on three.ws, and the ledger beneath that number is meant to list the recent deploys behind it. The query that fetched them asked for a column the agent table has never had, so it failed on every request and the code beside it turned that failure into an empty list: the lifetime burn total kept climbing past 3.1 million while the ledger sat blank, and the protocol activity feed was blanked the same way. Both now read the column that exists and list real deploys again, a failed panel query is written to the logs instead of quietly becoming a zero, and the holder count on the revenue-share view no longer disagrees with the one on the same page's stats panel.",
"tags": [
"fix"
],
"link": "/dashboard/three-token"
},
{
"date": "2026-08-16",
"title": "An expired X connection now offers to reconnect instead of sending you to the login page",
"summary": "Posting a Walk screenshot or clip to X goes through your connected X account. When that connection could no longer be renewed, the share endpoint answered with the same code it uses for a signed-out visitor, so the capture panel told an already signed-in user to sign in to three.ws and linked them to the login page, which fixed nothing. That case now comes back as its own answer with the X connect link attached, and the panel offers a Reconnect X button that takes you straight to the flow that repairs it. The share caption also lost a stray dash.",
"tags": [
"fix"
],
"link": "/walk"
},
{
"date": "2026-08-16",
"title": "The wallet balance API stops calling a rejected upload a validation error",
"summary": "Apps and agents that read wallet balances through three.ws get honest answers when a request cannot be read. Posting a body that was too large, or forgetting the JSON content-type header, used to come back as \"your fields are invalid\", sending integrators hunting through a chain and address that were never the problem; those now answer with the right status for what actually went wrong. When the fields genuinely are wrong, the reply is one readable line naming the offending field plus a machine-readable list of issues, instead of the raw multi-line dump the endpoint used to return.",
"tags": [
"fix",
"improvement"
]
},
{
"date": "2026-08-16",
"title": "The $THREE rider pass now recognises the wallets that paid for it",
"summary": "The rider pass is a $THREE gate: you qualify either by holding $THREE or by sending 8,000 of it to the rider vault. Only the first half worked. The check answered purely from what a wallet currently holds, so anyone who bought a pass, and therefore no longer held the tokens they had just sent, was told they had no pass. The payments the vault recorded were written to the database and never read by anything. Both sources are now consulted, and a positive answer from either one stands on its own, so a paid pass still verifies while the chain node is unreachable and a holder pass still verifies during a database outage. A single transaction carrying payments from two different wallets used to fail outright and be retried forever by the payment listener, granting neither wallet a pass; every payer in a batch is now credited. The config the rider client loads at startup answered with a page of empty values when the deployment was not set up, which crashed the client with an unexplained error; it now says plainly that the feature is not configured. Repeating the address in the query string returned a server error and now returns a normal validation message.",
"tags": [
"fix"
]
},
{
"date": "2026-08-16",
"title": "Trading swarm streams work from other sites, and a mistyped contribution now says so",
"summary": "Two fixes to the trading swarm API. The live dashboard stream at /api/swarms/<id>/stream never answered a browser's cross-origin permission check, so a page outside three.ws could open the connection and then read nothing from it. It now answers that check exactly like the dashboard read next to it, and a HEAD probe of either URL returns immediately instead of holding the connection open for five minutes. Separately, a contribution amount that was not a number (a typo, an overflowing value) came back as a generic server error; it now comes back as a plain bad-amount answer naming the field, and a wrong content type says that rather than reporting an unknown action.",
"tags": [
"fix"
],
"link": "/swarms"
},
{
"date": "2026-08-16",
"title": "Solana agents from the outside registries get a real link preview, and a mistyped discover link stops erroring",
"summary": "Two fixes to the Discover detail pages. Agents crawled from the outside Solana registries opened fine in the browser but shared as a nameless card: every one of them unfurled in Slack, Discord, X and iMessage with a generic three.ws title and the stock image, while the Ethereum-side agents in the same grid previewed properly. Those pages now carry the agent name, description and artwork, whether the link points at the mint or at its registry entry. Separately, a Discover link with a mistyped or truncated id answered with a server error and the page showed a vague \"something went wrong\"; it now says plainly that the item was not found and shows the Not found state, and an agent whose on-chain id is larger than the site could previously handle resolves instead of coming back empty.",
"tags": [
"fix"
],
"link": "/discover"
},
{
"date": "2026-08-16",
"title": "Memory seeding reads Farcaster with no vendor key, and X now works from the app credentials we already hold",
"summary": "The memory-seeding demo asks a GitHub, X or Farcaster handle what it can learn about you in public, then writes a short memory seed your agent can carry. Two of those three connectors were showing a \"connector not configured\" card to everyone, because each insisted on a vendor key nobody had set. Farcaster now reads through the same failover the rest of the platform uses: an indexer when a key exists, and a public Farcaster hub with no key at all otherwise, so it returns real profiles and casts on any deployment. Anything the keyless lane genuinely cannot know, such as follower counts, is now shown as unknown instead of zero. The X connector accepts the app key and secret the platform already stores, exchanging them for a read-only token itself, so it no longer needs a separate token provisioned by hand. A mistyped Farcaster name now comes back as not found rather than as an upstream outage, and all four endpoints are documented in the API reference for the first time.",
"tags": [
"fix",
"improvement",
"docs"
],
"link": "/demos/memory-seed"
},
{
"date": "2026-08-16",
"title": "Avatar Studio trait library: the mirror can no longer be pointed at someone else's server",
"summary": "Avatar Studio loads its hair, clothing and accessory traits through a three.ws address that mirrors the open-source trait library, so your browser never talks to an outside CDN. The list of where those traits live was being written using the address in the incoming request, and that address can be set by whoever sends the request. Because the mirror is cached at the edge for a day, one crafted request could have left every Avatar Studio visitor loading trait files from a stranger's server for the next 24 hours. That list is now written the same way for everybody, with no part of it taken from the request, so there is nothing left to steer. Three smaller repairs went in alongside it: asking the mirror how big a file is no longer answers with the compressed size (a trait manifest claimed 2 KB and delivered 20 KB), the mirror will not hand back a web page or script even if the upstream library ever served one, and a failure at the library now answers plainly as either not found or mirror is down instead of passing its raw status through as ours.",
"tags": [
"security",
"fix"
],
"link": "/avatar-studio"
},
{
"date": "2026-08-16",
"title": "Mistyped filters on the revenue, checkout and ring endpoints answer plainly instead of failing",
"summary": "The public revenue feed, the checkout SKU API and the ring report all took values straight from the URL into the database. A timestamp that was not a timestamp, a checkout id that was not an id, and the query fell over: the caller got an opaque \"something went wrong\" with a support reference, and our own alerting logged a fault that was only ever a typo. All three now say what was wrong, in words, with the right status code. The ring report also stops treating an unrecognised period as \"show me everything\", which quietly answered the wrong question and ran the most expensive query it has. The revenue feed now echoes back the endpoint filter it actually applied rather than the raw text it was sent, and the paid-call lookup only accepts something shaped like a real transaction signature.",
"tags": [
"fix"
]
},
{
"date": "2026-08-16",
"title": "A mistyped id on the signal marketplace answers plainly, and the live feed links to the right chain",
"summary": "The signal marketplace identifies a feed or a subscription by a number. Anything that was not one (a typo, a stray word, an empty box) went straight to the database, which threw the whole query out, so the caller got an opaque \"something went wrong\" and our alerting recorded a fault that was never a fault. Pausing a feed, killing or resyncing a subscription, and subscribing to a feed each now answer with a plain \"that is not a valid id\" instead. Separately, the paid live signal stream built its Solscan links from whichever network the request named rather than the one the feed actually publishes on, so a subscriber could be handed devnet explorer links for a mainnet call; the feed's own record now decides, and every proof link lands on the chain the trade really happened on.",
"tags": [
"fix"
],
"link": "/signals"
},
{
"date": "2026-08-16",
"title": "Robinhood Chain worlds survive a hiccup in the live feed instead of going dark",
"summary": "The in-world trading terminal inside a Robinhood Chain world polls the chain feed every five seconds, and the Worlds lobby lists those coins from the same feed. A single event that arrived without its payload, or with a timestamp the chain wrote in an unexpected format, took both requests down with a server error: the terminal showed a dead tape, and the lobby's Robinhood Chain tab quietly emptied with nothing on screen to say why. Bad events are now skipped and the rest of the tape still renders, a price that is genuinely missing stays blank rather than printing as a real $0 trade, and a launch with no readable contract address no longer produces a card that leads nowhere. Asking for trades with an address that is not a contract address now gets a clear error back instead of an empty list that read as \"this coin has never traded\".",
"tags": [
"fix"
],
"link": "/worlds"
},
{
"date": "2026-08-16",
"title": "Skill marketplace: installs are locked to your own clicks, and edits keep their star rating",
"summary": "Three fixes to the skills marketplace. Installing or removing a skill now carries the same one-time security token every other account action does, so a page you happen to have open elsewhere can no longer quietly add skills to your account behind your back. Editing one of your published skills used to answer back with its rating blanked out, which made the detail panel show 'No ratings yet' on a skill people had actually rated until you reloaded; it now reports the real average and count. And when a skill you were paging past gets deleted or made private mid-scroll, 'Load more' stops cleanly instead of looping the first page of results back into the grid over and over.",
"tags": [
"fix",
"security"
],
"link": "/marketplace"
},
{
"date": "2026-08-16",
"title": "Sharing an agent link posts the agent's picture again",
"summary": "Pasting a link to an agent into a chat app, a social post or a group thread is supposed to unfurl into a card with the agent's name, its description and a picture of it. For every agent that actually has an avatar picture, which is nearly all of them, the page that builds that card was crashing instead, so the link unfurled as a bare URL with nothing attached. The card now renders with the agent's own thumbnail, falls back to a freshly rendered portrait when an agent has no thumbnail yet, and an unknown or malformed agent link sends the visitor to the app instead of erroring.",
"tags": [
"fix"
]
},
{
"date": "2026-08-16",
"title": "The sniper's laddered-exit settings can finally be saved",
"summary": "An armed sniper strategy has two knobs that decide how a winner is taken off the table: the multiple of your entry at which it sells enough to recover what you put in, and the minimum slice of the position it always keeps riding afterwards. Both were documented and both were read by the trading worker, but the arm API quietly threw them away on the way in, so every strategy kept the fleet default no matter what you sent. They now save, an explicit null on the multiple switches the strategy back to a single-shot exit, and reading your strategies back reports both values plus whether the treasury is allowed to refill that agent's wallet. Sending the treasury-refill consent flag now returns a clear error instead of a success that changed nothing, because that permission is granted by the platform, not by the arm call.",
"tags": [
"fix",
"improvement"
],
"link": "/dashboard/sniper"
},
{
"date": "2026-08-16",
"title": "The public profile sitemap now lists people, not the seed crons",
"summary": "The sitemap Google and Bing crawl is split into one file per kind of page, and the profiles file was handing search engines every account row with a username. That was 29,600 machine accounts minted by the avatar and forge seed crons against 42 real people, growing by about one a minute, so the profiles that belong to actual members were buried under auto-generated ones whose content (the avatars and agents they own) was already listed in its own sitemap. Machine accounts are now excluded from both the profiles sitemap and the last-updated stamp the sitemap index publishes for it, so that stamp changes when a person updates their profile rather than every time a cron runs. Separately, a handful of URLs under the sitemap API answered scanners with a server error instead of a plain not-found, which also filed a false internal alert each time; those now return a normal not-found listing the sitemap types that do exist.",
"tags": [
"fix",
"infra"
]
},
{
"date": "2026-08-16",
"title": "Live stage hosts stop swallowing hyphens, and only confirmed tippers get greeted by name",
"summary": "Two repairs to the Living Stages host. The cleanup pass that tidies the host's spoken line was stripping hyphens along with the invisible characters it was meant to remove, so every show heard \"on chain\" where the host said \"on-chain\" and \"state of the art\" for \"state-of-the-art\"; punctuation now survives and only invisible characters are removed. Separately, the host's memory of returning regulars counted tips whose on-chain settlement had not been confirmed yet, which meant anyone could post an unconfirmed tip under a name of their choosing and hear the host greet it aloud. That memory now reads confirmed tips only, matching the public leaderboard. Smaller repairs in the same pass: a stage that has not opened its first show is no longer cached as empty at the edge, an over-long coin address is refused instead of quietly trimmed into a different plaza, and the running tip total is written before the response returns.",
"tags": [
"fix",
"security"
],
"link": "/stage"
},
{
"date": "2026-08-16",
"title": "The $THREE economy dashboard no longer errors out on an odd date range",
"summary": "Asking the public $THREE economy stats endpoint for a window it could not read (a non-numeric range, or one longer than ten years) returned a server error instead of an answer. It now replies with a clear message for a range it cannot parse, caps an oversized one at ten years of history, and treats a blank range as all-time again. The creator earnings ledger does the same for a bad page cursor.",
"tags": [
"fix"
]
},
{
"date": "2026-08-16",
"title": "Token-gated scene shares survive a slow RPC, and a gate that could never open is refused up front",
"summary": "Three repairs to the holder-only chat scene shares. When the blockchain read behind a gate failed (a provider down, a missing provider key), visitors were told they did not hold enough tokens, which was untrue and also handed the provider's raw error text to anyone who asked; a failed read now says plainly that verification is unavailable and invites a retry, while a genuine shortfall still says exactly how much is missing. The Solana read itself now rotates through the same pool of providers the rest of the platform uses instead of leaning on one endpoint, NFT collection gates use the indexer that can actually answer them, and gates on Ethereum work without a paid provider key. Finally, a gate whose chain and asset type did not match (a Solana scene gated on an Ethereum token) used to be accepted and then locked everyone out forever; that combination, and an address in the wrong format for its chain, are now refused when the share link is created.",
"tags": [
"fix",
"improvement"
],
"link": "/chat"
},
{
"date": "2026-08-16",
"title": "Trending: muted text on the board is readable, and the Tip button that only navigated is gone",
"summary": "Ten pieces of text on the Trending board sat below the WCAG AA contrast floor: the subtitle, the two unselected tabs, the 7d and all-time buttons, the column headers, agent descriptions and the unit labels next to each number. They are all legible now, and the Oracle link in the footnote is underlined instead of being told apart by colour alone. The wallet chip on each row carried a Tip button and a hover card, but the whole row is a link, so the browser followed the row instead of running either one: pressing Tip took you to the agent's wallet page rather than opening a tip. Both controls are gone from the board and tipping stays where it works, on the agent page the row opens, while the chip keeps showing the wallet address and its live balance. An automated accessibility scan now reports zero violations on all three boards and on their loading, empty and error states.",
"tags": [
"fix",
"improvement"
],
"link": "/trending"
},
{
"date": "2026-08-16",
"title": "Trending stops going blank on a bad limit, and shared cards stop trusting the image host",
"summary": "Asking the trending board for a non-numeric number of rows (limit=abc) answered with an empty leaderboard and a 200, as though nothing on the platform were trending at all. The bad value was reaching the database verbatim, failing both the agent and the coin ranking, and a catch-all turned that failure into an empty list with nothing logged anywhere. An unreadable limit now falls back to the default ten, the same way an unrecognised window falls back to 24 hours, and if either ranking ever does fail it now says so in the logs instead of quietly serving an empty board. Separately, the trade, trader and walk preview cards inline a portrait fetched from whatever URL the profile points at, and they read that response whole with no limit; a card now refuses anything that is not an image or that runs past 1.5 MB, and falls back to the monogram it already drew for a dead host. The profile card at /u/<name> also used to answer an image request with a JSON error if the database hiccuped, which is exactly the kind of broken preview a social crawler caches; it now returns a real, uncached card instead.",
"tags": [
"fix",
"improvement"
],
"link": "/trending"
},
{
"date": "2026-08-16",
"title": "The public news feed can no longer be knocked out by a single post",
"summary": "Everything that reads three.ws announcements, RSS readers and the HackerNoon importer alike, reads one feed at three.ws/rss/announcements.xml. That format has no partial failure: one illegal character anywhere in it makes the entire feed unreadable, so a single post could have taken every subscriber offline rather than costing one item. Three inputs could do that, and none of them can now. A code sample in an announcement that happened to contain the sequence that ends a raw-content block used to end it early and corrupt every item after it. Control characters the format forbids outright, which no amount of escaping can rescue, were passed straight through; they are now dropped, while emoji and other real characters are kept untouched. And a post whose wording happened to match a marker the feed builder uses internally while turning links into links stopped the feed building at all, returning an error to every reader instead. Separately, the feed takes an optional parameter for reading one of the archived accounts instead of the curated list, and misspelling it used to hand back the curated feed with no hint anything was wrong, so a subscriber could sit on the wrong feed indefinitely. It now says plainly that the source is unknown and lists the ones that exist.",
"tags": [
"fix",
"infra"
]
},
{
"date": "2026-08-16",
"title": "The sniper trade journal starts recording, and three sniper readouts stop reporting the wrong number",
"summary": "The trade journal is the surface that answers why an agent bought and why it sold, and it had never recorded a single line. Two halves of the same mistake: the table expected a plain number where the platform stores an identifier, so every write failed and was silently discarded on purpose (a journal write must never be allowed to break a trade), and the page that reads it asked the database to compare the same two mismatched kinds of value, which failed outright and returned an error rather than an empty journal. Both are fixed and the surface now records and reads for real, across the 323 closed positions the fleet had already traded. Three readouts were also stating numbers that were not true. The Exit Lab corpus, asked without an explicit page size, answered with exactly one trade instead of the full 400, so the replay console looked like a fleet that had traded once; it now returns the whole window. The live pre-launch radar tape re-sent every precursor it had already shown, about forty times a minute, so the feed filled with copies of the same event; each precursor is now sent once. And the radar dashboard reported 100 watched wallets, the size of the table it was showing rather than the number it was watching, while the status line beside it reported the real figure; it now counts the wallets and the table says which slice of them it is showing.",
"tags": [
"fix",
"improvement"
],
"link": "/radar"
},
{
"date": "2026-08-16",
"title": "A coin world link now unfurls as the live world, and only ever points at three.ws",
"summary": "Pasting a coin community link into WordPress, Ghost, Discord, Notion or any editor built on embed.ly or iframely now turns it into the live 3D world instead of a plain link. The service that answers those editors existed but nothing on the page told them where to find it, so it was never called. Two problems in that service are fixed at the same time. It used to build the embed from the address in the incoming request rather than from the site address, so a crafted request could hand an editor an embed that framed somebody else site under the three.ws name; the embed is now always built from three.ws itself. And it used to answer for any link at all, returning a generic three.ws world card for a link that was never ours; it now says plainly that the link is not one of ours. It also answers in both of the formats the standard defines, and ignores a coin address that is not a real Solana address instead of putting it in the embed. Separately, the share images for pages and coin worlds, the holder leaderboard card, the free model verification check and the Oracle coin page now accept only read requests, and answer anything else with a clear message rather than doing the work anyway.",
"tags": [
"fix",
"security"
],
"link": "/play"
},
{
"date": "2026-08-16",
"title": "Avatar Wallet: the send button works again, and a failed reply tells you why",
"summary": "The embeddable avatar that holds a Solana wallet had its send button covered by the floating language picker, so pressing it did nothing at any window size. The picker now lives in the page's own top bar, where it can be reached without sitting on top of the composer, and the same stray picker no longer floats over the avatar inside every embed. On a phone the fund-your-wallet notice was pinned to a fixed spot and landed on top of the wallet bar once that bar wrapped onto a second line; the two now flow one under the other at every width. When a reply fails, the page repeats the reason the chat service actually gave, such as being at capacity, instead of a generic apology, and offers a Try again control that re-sends the message without retyping it. On a desktop window the wallet bar, the conversation and the composer now sit in a readable centred column instead of stretching the full width of the screen, and the page carries a proper top-level heading for screen readers.",
"tags": [
"fix",
"improvement"
],
"link": "/avatar-wallet-chat"
},
{
"date": "2026-08-15",
"title": "glTF Validator: a file it cannot read now says why, and the page is usable from the keyboard",
"summary": "Drop a model the validator cannot parse and the page used to claim it had been \"validated\", print \"Validator failed: undefined\", and, when a URL fetch was refused, wipe both result panels to an empty screen with no explanation. Every one of those paths now renders a real failure card: what broke, the actual reason, what to try next (CORS on the host, a .gltf that needs packing into a .glb, a download that did not finish), and a Try again button that reruns the same file or URL. The status line no longer reports a success it did not have, and it counts the issues found instead of just saying \"validated\". The file picker was unreachable without a mouse, so keyboard users could only use the URL box; it is now in the tab order with a visible ring on the drop zone. The three tabs move with the arrow keys, the Agent ID and Chain ID fields have labels, the report headings no longer compete with the page title, and the Submit Report dialog closes on Escape or a click outside, keeps focus inside while open, returns it where it was, and no longer sits underneath the site header. A dead link in the report footnote now points at the Khronos extension registry.",
"tags": [
"fix",
"improvement"
],
"link": "/validation"
},
{
"date": "2026-08-15",
"title": "Walk Leaderboard: switching filters no longer strands the board, and an empty board tells you how to get on it",
"summary": "Clicking through the leaderboard's filters faster than the board could load left it stuck: a second click while a request was open was dropped outright, so the board could sit on a loading skeleton that never resolved, or show one metric's rows under another metric's column header. Every request now supersedes the one before it, so whatever you clicked last is what you get. An empty board used to say the same thing whichever metric you were on, which was wrong for Sites: sites only count where your avatar walks away from three.ws, so that state now explains it and links to the walk companion, while distance and time point at the walk itself. Any narrow window that comes up empty also offers a jump to All Time rather than leaving you at a dead end, and a failed load offers a retry that says what went wrong. The board now lives up to \"updates live\": it refreshes itself while you are watching, shows how many walkers are ranked and when it last updated, and a refresh that fails quietly keeps the rows you are reading. For screen readers the board is a real table, the top three medals now announce their rank instead of reading as an empty cell, the momentum column speaks its direction, and the two filter groups are proper radio groups with arrow-key and Home/End navigation.",
"tags": [
"fix",
"improvement"
],
"link": "/walk-leaderboard"
},
{
"date": "2026-08-15",
"title": "Generation Watch: a dropped connection now says so, and the page reads properly to a screen reader",
"summary": "The page that shows a 3D generation in progress had one way to fail badly: if the tab lost contact with three.ws, it kept animating the countdown against a server it could not reach, silently retrying for forty minutes before admitting anything was wrong. It now stops after four failed checks, tells you plainly that the connection dropped and that your model is still being generated on our servers, and offers a \"Keep watching\" button that picks up where it left off. It also resumes on its own the moment the browser comes back online. The phase line used to lag a step behind the picture, announcing \"Painting the concept\" while the concept art was already on screen; it now switches to \"Sculpting the 3D mesh\" as soon as the artwork arrives. The page had no heading at all for a screen reader and read its four-times-a-second timer aloud continuously; it now has a real title heading, announces each state change once, and keeps the countdown quiet. It follows your light or dark theme instead of forcing dark, gives the three.ws link a visible focus ring, and no longer lets the site-wide Cmd+K hint sit on top of its buttons. The hand-off into the 3D viewer is now checked before it is followed, so it can only ever land on three.ws.",
"tags": [
"fix",
"improvement"
],
"link": "/watch"
},
{
"date": "2026-08-15",
"title": "Your master wallet: creating one no longer reports a failure, and your balances are real numbers again",
"summary": "Two faults on the master wallet page are fixed. First, every account that pressed Create my master wallet was told the creation had failed. The wallet was in fact created every time; the page then showed an error, and the same fault sat on the send and fund-an-agent paths, where a transfer that had already settled on Solana was reported as \"The transfer did not go through\", inviting people to send the same money twice. Nothing about the money was ever wrong, only what the page said about it. Second, the SOL and Solana USDC figures, and the total value, read \"unavailable\" for everyone regardless of what the wallet held, because the balance reader and the page were describing balances in two different shapes. They now show live amounts, priced against the SOL rate, and an empty wallet reads a plain 0 instead of \"unavailable\". A balance the network genuinely could not answer for still says \"unavailable\" and is left out of the total rather than counted as zero. Alongside that: failures now explain themselves in plain words with a support reference instead of printing raw operator text, the explainer bubbles fit inside a phone screen, and each one names what it explains for screen-reader users.",
"tags": [
"fix",
"improvement"
],
"link": "/wallet"
},
{
"date": "2026-08-15",
"title": "KOL Tracker: the whole board is readable on a phone, and dead columns are gone",
"summary": "On any screen narrower than the board, five of the tracker's seven columns were unreachable: the page could not shrink, and the overflow was clipped rather than scrollable, so P&L, win rate and trade count simply were not there on a phone. The board now fits any viewport and scrolls sideways on its own. Columns with nothing to show are dropped instead of rendering a full column of dashes, so followers and volume appear only once there is data behind them. Each wallet is printed once (it used to appear twice on every row) and links to its Solscan account, so any number on the board can be audited on-chain. Where the P&L for a window comes from is now stated once under the board instead of repeating on all fifty rows, the board refreshes every minute with the time of the last update on screen, and a refresh that fails keeps the rows you are reading rather than blanking them. A failed load now offers a Try again button, an empty window points at the trader leaderboard, and the whole board is navigable and announced properly with a keyboard and a screen reader.",
"tags": [
"fix",
"improvement"
],
"link": "/tracker"
},
{
"date": "2026-08-15",
"title": "Trade Terminal: an agent exit now opens the trade, not just the coin",
"summary": "The trade terminal advertises a feed of notable agent exits with realized PnL, hold time, exit multiple and a one-click path to copy the trader, but clicking an exit only ever opened the coin: the trade behind it was nowhere on the page and the copy-trader link did not exist. Selecting an exit now opens an Agent exit card carrying the whole position (realized PnL and SOL, exit multiple, hold time, entry and exit size, why it closed, how many people are copying) with links to the entry and exit transactions on-chain and a Copy trader button that goes to that agent's passport. Several honest-failure fixes came with it. A section that could not be reached used to say the data does not exist for this coin, blaming the coin for an outage; each one now says the request failed and offers a Retry that re-runs only that request. Coins with no intel snapshot, no funder graph or no indexed wallets no longer dead-end on a bare sentence, and point instead at the Oracle, Bubblemaps and Solscan. The price chart could get stuck behind a permanent \"Loading chart\" panel that hid the chart underneath it whenever a switched interval had little history. A migrated coin whose price is quoted off Jupiter is now labelled Graduated in the header strip rather than being reported as still on the bonding curve, and the strip names its price source. The failing feed also told the truth about which feed failed, an empty network says it is empty instead of showing one lonely pinned row, the SOL price in the header pulse finally carries a label, and the feed tabs and the coin search box are now navigable and readable with a keyboard and a screen reader.",
"tags": [
"feature",
"fix",
"improvement"
],
"link": "/trades"
},
{
"date": "2026-08-15",
"title": "The Unstoppable Agent dashboard shows its numbers again, and the agent itself is unstuck",
"summary": "The /unstoppable page promised a live view of an agent that funds itself, but every number on it was a dash unless you paid $0.01 first, so nobody could actually watch anything. It now reads a free public snapshot of the agent's treasury, runway, 24-hour earnings and costs, activity log and daily reflection, refreshed every five minutes, and paying still buys the real-time reading and still funds the agent. Two bugs behind the page were worse: the agent had halted itself months ago and could never restart, because the check that clears a halt only ran on ticks a halted agent skipped, and it was sitting on $8.93 against a ten-cent floor with nothing running. It is live again. Its daily reflection was also optional in practice, left to whatever the planner felt like doing, so none had ever been written; it is now guaranteed. The page also gained a real heading, a designed failure state with a retry instead of a permanent loading shimmer, and a layout fix that stopped the activity log stretching the page off a phone screen.",
"tags": [
"fix",
"improvement",
"feature"
],
"link": "/unstoppable"
},
{
"date": "2026-08-15",
"title": "Watchlist: the portfolio summary stops reporting numbers it does not have",
"summary": "The watchlist summary bar claimed $0 of 24h volume across every coin you watch. The pump.fun feed simply stops reporting 24h volume once a coin graduates, and the bar was treating each missing figure as a zero and adding them up, so a confident \"$0\" stood where the honest answer is \"not reported\". Every aggregate now counts only the coins that actually report it, shows a dash when none do, and says on hover how many of your watched coins it covers. The Oracle tier bar does the same: coins the Oracle has not scored get their own marked-off slice instead of quietly vanishing from the distribution, and each unscored card now carries an \"unrated\" badge that links to its coin page rather than leaving a blank corner. If the Oracle itself is unreachable, the page says so instead of labelling every coin unrated. Also fixed: at 320px wide the \"Clear all\" button was pushed off the edge of the screen with no way to reach it, the add-a-coin field had no visible keyboard focus ring, and the page description advertised a verdict the page does not show.",
"tags": [
"fix",
"improvement"
],
"link": "/watchlist"
},
{
"date": "2026-08-15",
"title": "Proof-of-Grind Gallery: filters stop stranding the board, and a shared card proves it is on it",
"summary": "Switching tier or sort while the gallery was still loading could leave you staring at an empty board until you reloaded: the page refused the new request because the old one had not finished, and then never drew anything. The newest request now always wins, so filters respond immediately no matter how fast you click them. When the board itself could not be reached, the summary strip announced \"Gallery: Live\", which was simply false; it now says the stats are unavailable and offers a Retry, as do the gallery and the leaderboard, and each retry re-runs just that request instead of reloading the whole page (the gallery's old retry button also rendered as raw browser chrome). Publishing a wallet refreshed the gallery but not the leaderboard, so your new entry was missing from the ranking until you reloaded. Sharing a card is now worth something: opening a shared link shows whether that address really is on the board, with its verified badge, tier and label, instead of only a local estimate. Alongside those: filtering down to nothing offers a one-click way to clear the filters, a failed \"Load more\" says so instead of silently doing nothing, the tabs work with the browser back button and with arrow keys, a rate-limited publish explains how long to wait rather than printing \"too many requests\", and a gateway failure during publish no longer surfaces a raw parser error.",
"tags": [
"fix",
"improvement"
],
"link": "/vanity/gallery"
},
{
"date": "2026-08-15",
"title": "Avatar Studio: the trait builder loads again, with its manifest, icons and frames intact",
"summary": "The trait-based Avatar Studio, the point-and-click builder behind the “three.ws Studio” option in the avatar creator, was shipping a blank page: two copies of React at different versions landed in the bundle, and React refuses to start when its renderer does not match. Both are pinned to the same version now. Behind that, the builder asked for its character manifest at a path containing the literal word “undefined”, so even a booting studio had no character classes to offer; it now reads the manifest from its own base path. The texture-compression library it loads at startup was requested at a doubled URL in local development and never arrived, and the trait card frames, rarity boxes and back/next arrows were pointed one directory too deep, so they silently 404ed. Every one of those assets resolves now, and if the compressor genuinely cannot start, an export that needs it says so instead of failing with an internal error. The in-app creator, the avatar SDK and the Avatar OS demo now open the builder at its exact entry point, so they can no longer land on the wrong studio page.",
"tags": [
"fix"
],
"link": "/docs/character-studio"
},
{
"date": "2026-08-15",
"title": "DeFi Yields: the filter row is styled again, the phone layout fits, and protocol links stop dead-ending",
"summary": "The Yields explorer was shipped with its filter row pointing at a stylesheet the page never loaded, so the chain, project and minimum-TVL controls rendered as raw browser dropdowns crammed into one line with their labels colliding. They are proper labelled fields again at every screen size. On a 320 pixel phone the Total TVL card used to be pushed off the right edge with no way to scroll to it, and opening a pool parked its chart outside the visible area inside a table you had to drag sideways; the three headline stats now stack, and the chart stays pinned in view and is drawn at the drawer's real width, so its axis labels are legible instead of shrunk to a third of their size. Fourteen DeFiLlama protocols carry a bracket, a plus or an exclamation mark in their name, and every yield row for one of them linked to a page that answered 404: those links now resolve. When the upstream feed is down, the three stat cards say so rather than sitting as loading placeholders forever, and the result count says the data is unavailable instead of going blank.",
"tags": [
"fix",
"improvement"
],
"link": "/yields"
},
{
"date": "2026-08-15",
"title": "Voice Lab: a dead voice provider now leaves the menu instead of failing on your first click",
"summary": "Google Gemini has been refusing every request on the platform's account, and Voice Lab had no way to know: it listed all 30 Gemini voices, put them first, and the first Preview a visitor pressed came back with a wall of raw error JSON quoting internal project numbers. Now the first refusal is the last one anybody sees. A provider that turns down the platform's credentials is withheld from the voice list for a few minutes with the reason written on its filter chip, its voices disappear from the grid the moment a preview hits it, and the free Microsoft Edge and NVIDIA lanes are named as the alternative. It restores itself the moment the provider works again, with nothing to switch back on by hand. Failure messages across the page are now single readable sentences rather than vendor dumps. The rest is the same page, checked end to end: the catalog error state offers a working Try again, a search that matches nothing offers Clear filters, a deployment with no provider at all says so instead of claiming no voices match, the Remove button next to your own ElevenLabs key is styled like a button again, every control shows a focus ring when you tab to it, and the text box in the playground has a label. The space bar no longer hijacks the button you are pressing: it starts a recording only when nothing on the page is focused. Muted text across the page was also too faint to read against the black background (grey on near-black, below the accessibility floor) and an unavailable provider was faded to the point of illegibility, which mattered because that chip is where the reason it is unavailable is written; both are legible now, and the chip tells a screen reader why it cannot be picked.",
"tags": [
"fix",
"improvement"
],
"link": "/voice"
},
{
"date": "2026-08-15",
"title": "Trending: the time window now filters the coin board, and every row links inside the site",
"summary": "The 24h / 7d / all-time control on Trending only ever changed the agent ranking; the conviction coin list ignored it and always answered with a fixed 36 hour slice. It is a real filter on both boards now, the coin column names the window it is showing, and coins that tie on a perfect score are separated by momentum, smart wallets and freshness instead of coming back in a different order on every reload. Agent and trusted rows used to point at absolute three.ws addresses, which bounced anyone browsing a preview or a mirror off the site, and the fallback link pointed at a path that does not exist; every row is now an in-site link that resolves. The three boards are a proper tab set for keyboard and screen reader users (arrow keys, Home and End move between them, each tab names the panel it opens), the selected tab is kept in the address bar so a view can be shared and the back button walks it, and the time buttons announce which one is pressed. Empty boards now say what to do next instead of stopping at 'nothing here'. On a 320 pixel screen an agent's wallet chip used to push the whole page sideways; it wraps now, with the tip action still reachable. Anonymous visitors also stop seeing two failed requests in the console on any page that loads wallet chips or reputation badges.",
"tags": [
"fix",
"improvement"
],
"link": "/trending"
},
{
"date": "2026-08-15",
"title": "Avatar SDK: the React entry finally loads, and the viewer answers to properties as well as attributes",
"summary": "The React bindings in @three-ws/avatar shipped as a .jsx file even though they contain no JSX, which meant Node refused to import them and browsers refused to run them as a module script: the live React SDK demo had been failing with a MIME type error. That entry is now plain JavaScript, so 'import ... from \"@three-ws/avatar/react\"' works in Node, in Next.js, and straight from a browser import map, and the demo page renders all four bindings again. The lightweight <three-ws-viewer> element now also accepts src, alt, background, ar and autoRotate as properties, not only as attributes, so frameworks that assign properties (React 19, Vue, or any Object.assign helper) drive it correctly, including values set before the element upgrades. The package's documentation was corrected in the same pass: the element method list now matches the methods that actually exist, the Avatar Studio default URL is the real one, the license is stated correctly, and the two dependency-free runtime entry points are documented.",
"tags": [
"sdk",
"fix",
"docs"
],
"link": "/demos/react-sdk"
},
{
"date": "2026-08-15",
"title": "3D viewer: shared model links keep their name, and every button on the page does something",
"summary": "Open a shared model link and the browser tab now keeps the model's name instead of resetting to a generic one a moment after the page loads, so a bookmarked or shared viewer link is identifiable again. The model's name is also the page's heading now, which is what screen readers and search engines read first. A link whose address does not point at a real https model file used to show a Download button that did nothing; the page now explains what is wrong with the link and offers the only action that helps. Copying the embed snippet from 'Embed it anywhere' confirms it copied, and launching a coin from the model shows that the token image is rendering, both of which were being written to a part of the page the visitor could not see. The page's own title and description now describe what the viewer actually does: preview any public glTF or GLB model, then pose, restyle, scene, AR, embed or download it.",
"tags": [
"fix",
"improvement"
],
"link": "/viewer"
},
{
"date": "2026-08-15",
"title": "Back-an-Agent Vaults: a successful deposit no longer reports itself as a failure",
"summary": "Depositing an amount with cents in it (anything that is not a whole dollar) settled on-chain and then showed the backer an error, because the confirmation line converted the amount through floating point and threw on the result. That is fixed, and the confirmation now names the shares you received. Opening a vault and pressing the browser's back button used to trap you on the page: every back press rebuilt the same entry instead of leaving. The open positions table said nothing at all when a vault's holdings had all been sold down to zero; it now says the capital is in USDC. A vault that failed to load offers a retry and a way back to the list instead of a dead end, and a display fault is no longer reported as a network problem. Anonymous visitors no longer trigger a failed sign-in check on load, the feed and the sign-in check now run at the same time rather than one after the other, the open-a-vault form no longer offers a submit button when you have no agents to open one behind, and the dialogs take keyboard focus, keep Tab inside them, and hand focus back when they close.",
"tags": [
"fix",
"improvement"
],
"link": "/vaults"
},
{
"date": "2026-08-15",
"title": "Grind-Bounty Market: shared bounty links now open the bounty, and \"Grind this\" grinds that one",
"summary": "Every bounty the market posts comes back with its own link (three.ws/vanity/bounties#<id>), and the API hands that link out as the bounty's home. Opening one did nothing: the page rewrote the address back to the plain board and left you to hunt for the card, which was not even there if the bounty had already been won or had expired off the open filter. A linked bounty is now fetched by id and pinned above the board, whatever its state, with a Clear button to go back; a link to a bounty that no longer exists says so instead of leaving you staring at the board. The Grind this button on a card was equally misleading: it switched you to the grinder and hoped, while the worker pool always took whichever bounty happened to be first in the queue. It now sets a real priority target that the pool grinds ahead of the queue, shows which bounty is pinned, lets you clear it, and says so plainly if that bounty stops being claimable. Refunding an expired bounty used to ask which address to send the escrow to and then ignore the answer, because the destination is bound to the address recorded when the bounty was funded; it now names the amount and where the money goes and takes a plain yes or no. Four smaller fixes shipped with it: the board no longer fetches itself twice on every page load, the market stats strip and the grinder leaderboard offer a retry when they cannot load instead of a dead end, the tab strip answers arrow keys and Home/End the way its role promises, and the sealed-delivery key field is finally wired to its label for screen readers.",
"tags": [
"fix",
"improvement"
],
"link": "/vanity/bounties"
},
{
"date": "2026-08-15",
"title": "World Lines: the AR encounter is finally offered, and the quest tabs answer the keyboard",
"summary": "The page has always promised an AR challenge, and the immersive encounter was fully built, but nothing ever put a button in front of it: a phone that can run WebXR was quietly handed the flat panel instead. Devices that support immersive AR now see a \"Meet the agent in AR\" option next to \"Begin the encounter\", and if the headset session refuses to start, the panel ceremony picks up where it left off. The four tabs (Near me, Explore, My proofs, Create) now behave like a real tab strip: arrow keys and Home/End move between them, only the selected tab sits in the tab order, each tab writes its own address so a tab can be linked and the back button walks back through them. Opening a region on the Explore tab used to leave a heading over an empty space when every quest in it had expired or filled up, and a failure there was a single line of red text with no way to retry; both are now proper states. Focus rings are visible again across the page (they were being drawn in near-black on a black background), the opened region is highlighted, the encounter dialog now takes and holds keyboard focus and hands it back where you left it, and a proof whose date is missing says so rather than claiming it was earned in 1970.",
"tags": [
"fix",
"improvement"
],
"link": "/world-lines"
},
{
"date": "2026-08-15",
"title": "The Wardrobe's garment forge no longer locks itself after a job expires",
"summary": "Forging a garment takes about seven minutes, so the Wardrobe (/wardrobe) remembers your job and picks the progress back up when you return. The catch: once the forge had cleared that finished job, the page treated the resulting \"no such job\" answer as a temporary network blip, so it kept the job saved, kept the prompt box and the Forge it button greyed out, and retried forever. One expired job made the generator permanently unusable on every later visit, with \"Reconnecting\" as the only explanation. An expired job is now recognised as finished business: the page clears it, unlocks the form, and tells you nothing is pending. A genuine outage is still retried, but it now gives the form back after several failed attempts and keeps your job saved so a reload resumes it. Separately, inspecting a piece in 3D and then typing in the search box no longer throws that 3D view away and reload every thumbnail on the page, filtering by slot with the keyboard keeps your place instead of dropping focus, the search and prompt boxes finally show a visible keyboard focus ring, and when the catalog itself fails to load the page now names the actual cause underneath the apology instead of swallowing it.",
"tags": [
"fix",
"improvement"
],
"link": "/wardrobe"
},
{
"date": "2026-08-15",
"title": "The pay-as-you-learn tutor can take a payment again",
"summary": "Every question on the tutor page died at the \"Confirming price\" step with a raw browser error, so no explanation was ever bought and the checkout never even reached the tutor service. The shared payment library was asking the service for its price with a GET request that carried a question body, which browsers refuse outright; it now sends the POST the service expects whenever a request body is present, which also unblocks any other page that pays a POST endpoint. On top of that, a resumed session no longer shows a real running total next to a stuck \"0 answers\" count, resumed questions show what each one cost, a shimmering placeholder now stands in for the answer while it is being written, the invoice closes with the Escape key and returns you to where you were, the page keeps a proper top-level heading once the conversation starts, every control has a visible keyboard focus ring, and a checkout that cannot reach the service now says so in plain language instead of quoting the browser.",
"tags": [
"fix",
"improvement"
],
"link": "/tutor"
},
{
"date": "2026-08-15",
"title": "The vanity receipt verifier says what went wrong instead of going quiet",
"summary": "Verifying a provably-fair vanity receipt runs entirely in your browser, so when the cryptography modules could not load (a blocking extension, a captive network, an offline laptop) the Verify button silently did nothing at all. The page now loads that code on demand and, when it cannot, says so plainly and hands you the offline command line check instead. Receipts delivered without a sealed envelope now get the custody check too, proving the key you hold is the one the receipt describes, with no extra key to paste. A share link whose receipt payload arrived damaged now explains that rather than leaving you with an empty form, opening a sealed envelope with the wrong key now tells you which key it wanted, and every control has a visible keyboard focus ring.",
"tags": [
"fix",
"improvement"
],
"link": "/vanity/verify"
},
{
"date": "2026-08-15",
"title": "The widgets gallery stopped clipping its controls on phones and now recovers from a failed load",
"summary": "The widgets gallery (/widgets) is where you pick a pre-built chat, voice, or 3D-avatar embed and copy the one line of HTML that drops it into your own page. On a narrow phone screen, each card's preview frame and its Customize panel were wider than the card itself, so the right edge got cut off: the mint field ran past the panel and the Reset button was pushed off-screen entirely, out of reach. Card layout now shrinks to the screen it is on, and the whole page is clean from 320px up. If the showcase list fails to load, the page now says so in plain language and offers a Try again button that reloads it in place, plus a link to the widget docs, instead of leaving a one-line error with no way forward. If the list comes back empty, the gallery explains what it would normally show and points at the Studio rather than showing a blank space. The widget count in the header now reflects the widgets actually loaded in every language, the page has a real top-level heading, the type filters announce which one is active to screen readers, the Preview and Code tabs are wired up as real tabs, and Escape closes the embed-format menu.",
"tags": [
"fix",
"improvement"
],
"link": "/widgets"
},
{
"date": "2026-08-15",
"title": "Viability now shows skeletons while it reads and says what broke when a read fails",
"summary": "The Viability page (/viability) reports the two real money loops behind three.ws: $THREE marketplace GMV and take-rate, and guarded agent coin trades with realized P&L. Until now, a slow read left the page as a bare headline with nothing underneath, and a failed read quietly deleted both panels while the page still claimed it had just updated. Both panels now hold their shape with skeletons while the live read is in flight, so the numbers land without the page jumping. If a read fails, that panel says so in plain language, names the reason (offline, rate limited, the status the API returned, or a deploy still rolling out), and offers a Retry that re-runs only that panel, leaving a healthy one untouched. The freshness stamp is honest too: it reports a read in flight, a failed read, or how long ago the figures on screen were actually good. Section titles are now real headings, and the network toggle announces itself correctly to screen readers.",
"tags": [
"improvement",
"fix"
],
"link": "/viability"
},
{
"date": "2026-08-15",
"title": "The vault's session controls moved out of browser pop-ups and its model panel became a real modal",
"summary": "The vault page (/vault) gives every visitor a session key that acts as their buyer identity: it is what pays for a model and what decrypts it in the browser afterwards. Everything you could do with that key used to happen through the browser's own grey pop-up boxes, which offered no way to see the address, no way to copy it, and no way to start over with a fresh one. The session strip now does all of it on the page: copy the address to fund it from anywhere, open an inline amount field to top it up from your wallet with the amount checked before the wallet is even asked, and rotate to a new key behind a two-step confirmation that first spells out what a new key costs you. The panel that opens when you pick a model is now a proper dialog: it says it is checking your purchase status instead of flashing a Buy button it might have to take back, keyboard focus stays inside it, Escape closes it, and closing puts you back on the model you were reading. When listings fail to load, the page now offers a way to try again instead of only stating the failure, and both empty states point at what to do next.",
"tags": [
"improvement",
"fix"
],
"link": "/vault"
},
{
"date": "2026-08-15",
"title": "The multiplayer worlds server connects on any local port, and its README documents what it actually reads",
"summary": "Two fixes to the server that runs every live world (/play, /walk, /agora, /irl, /play/war, the Living Stages, /ar/studio). Running two copies of the site on one machine used to break the worlds: the second copy gets the next free web port, and the server only accepted connections from the first four, so the world silently refused to join. Outside production it now accepts any connection coming from the same machine, while the live server keeps its exact origin list unchanged. Its README also described six settings when the server reads more than thirty, and listed two web endpoints when it serves seven, so anyone running or deploying it was working from a map that no longer matched the ground; the settings, the endpoints, and how the admin monitor is protected are now written down as they really behave, and the origin rules, the refusal to boot production without a real signing secret, and the closed admin monitor are covered by tests.",
"tags": [
"fix",
"docs",
"infra"
]
},
{
"date": "2026-08-15",
"title": "An embedded agent keeps talking when the browser refuses to play its voice",
"summary": "The living agent body that renders inside ChatGPT, Claude, and any site that embeds it speaks each reply with a lip-synced mouth. When a voice track is attached, the body plays the audio and drives the mouth from it. Browsers block that playback routinely: a chat panel the visitor has not clicked in is not allowed to start audio. The body was falling back to an empty line in that case, so it stood there with its mouth shut through the whole reply. It now falls back to the text of the reply it was actually given, so the mouth speaks the line whether or not the audio was allowed to start, and the engine is now covered by tests for the load, speak, emotion, chain-state, and teardown paths.",
"tags": [
"fix"
]
},
{
"date": "2026-08-15",
"title": "The concierge widget closes on Escape from anywhere on the page, and its SDK now installs on a server",
"summary": "Three fixes to the embeddable concierge. Escape now closes the chat panel wherever the visitor's cursor last landed: it only worked while focus was still inside the widget, so clicking the page behind the panel left the documented shortcut dead. Sites built with Next, Nuxt, SvelteKit, or Astro can now import the package the normal way, because importing it no longer expects a browser and no longer crashes the server render. And the 3D avatar stage stopped printing a deprecation warning into every host site's console. The package also ships full TypeScript types for shopping mode (catalog, products, policies, and the store events) and declares its own build and test tools, so a standalone install builds and tests without the monorepo.",
"tags": [
"fix",
"sdk"
],
"link": "/concierge"
},
{
"date": "2026-08-15",
"title": "The Walk page shows the real top walkers again instead of claiming nobody has walked",
"summary": "The Top walkers board on /walk read \"No walks logged yet. Be the first to put down some distance\" for every visitor, while twenty-nine people had in fact logged distance and the full leaderboard at /walk-leaderboard listed all of them. The page was asking the leaderboard endpoint for the ranking and getting a correct answer back, then looking for it under half a dozen field names the endpoint has never used, finding nothing, and drawing its empty state. It now reads the shape the endpoint actually returns, so the top three walkers appear with their handles, avatars, distances, and a link to each profile, and it asks for only the three rows it draws instead of the first fifty. The social section under it now shows placeholder cards while it loads rather than a blank gap. Two keyboard fixes travelled with it: tabbing into one of the nine live demo frames now draws a focus ring, where before you landed inside a demo with nothing on screen saying so, and the touch joystick inside every Walk embed carried a screen-reader label that ARIA does not permit on an unlabelled element, so assistive technology could announce it wrongly.",
"tags": [
"fix"
],
"link": "/walk"
},
{
"date": "2026-08-15",
"title": "The trading control room now tells you when the agents cannot afford to trade",
"summary": "The /trading page showed a green \"Trading live\" badge whenever the worker was up and the launch feed was connected. Both of those can be true while the agents' wallets are too empty to place a single buy, which is exactly what was happening: four of twelve armed wallets were below the minimum entry size and the page said everything was fine. It now reads the fleet verdict the trading worker already publishes, so an underfunded fleet says \"Live, wallets underfunded\" and one that cannot trade at all says \"Out of SOL\", each naming the real cause instead of blaming the feed. A new Wallets funded vital shows how many armed wallets can actually place their configured size, how much SOL would refill the rest, and whether the funding wallet can do it automatically or a person has to step in. An agent with no return to report now reads \"ROI unavailable\" rather than a stray dot.",
"tags": [
"fix",
"improvement"
],
"link": "/trading"
},
{
"date": "2026-08-15",
"title": "The contracts README now describes the contracts that are actually there",
"summary": "Anyone opening the on-chain contracts directory was reading a page written when it held three ERC-8004 registries. It now holds ten contracts and programs, including the two that custody real USDC on BNB Smart Chain, Base and Arbitrum, and the two Solana programs. The README listed none of them, claimed 25 tests where there are 242, and printed the canonical ERC-8004 identity addresses with mainnet and testnet swapped, so a developer copying them would have pointed at the wrong chain's registry. It also told you to edit a shared entry that seven testnets read from, which would have silently repointed all of them. All of that is corrected, the Solana build and test commands are documented, and the sample registration snippet now shows the values the function really returns.",
"tags": [
"docs"
]
},
{
"date": "2026-08-15",
"title": "Pages stop loading a 3D robot you have not scrolled to yet",
"summary": "The small animated robot in the site footer was starting up the moment a page opened, at the bottom of a page nobody had scrolled to. On plain pages that meant downloading a whole second 3D engine (253 KB) before anything else, and on the Forge it was the single busiest thing on the page: nearly 20 seconds of browser work, more than the Forge itself. It now waits until you are about a screen away from the footer, so it still greets you on the way down but no longer competes with the page you actually opened. It also stops taking a 3D slot away from the avatars on the page, which on a busy grid could leave a real avatar unable to render. Separately, the Forge's example prompt chips no longer shove the page down as they shuffle to a fresh set on load.",
"tags": [
"improvement",
"fix"
],
"link": "/forge"
},
{
"date": "2026-08-15",
"title": "Posed avatar renders finally come out posed",
"summary": "Asking the public avatar renderer for a pose returned a T-pose on most models: it matched bone names against a short hand-written list, and any rig that did not use those exact names (including every Mixamo export) was silently left in its bind pose while the response still claimed the pose had been applied. The renderer now uses the same posing engine the pose studio uses, so a preset now lands whatever naming convention the model's skeleton uses, and the camera frames the posed figure rather than the unposed one. The background colour on both public renderers is now checked before it is used, and the per-render budget is counted across the whole fleet instead of per server.",
"tags": [
"fix",
"security"
],
"link": "/docs/api-reference"
},
{
"date": "2026-08-15",
"title": "Scanning a skill purchase QR tells you when your wallet is short",
"summary": "Paying for a skill from a phone handed the wallet a transfer even when the wallet held none of the token being asked for, so the buyer signed and watched it fail with whatever cryptic message their wallet chose to print. The checkout now checks the paying wallet first and answers with the exact shortfall, so a wallet that is 0.35 short says so and a wallet that has never held the token says that instead of failing at signing time. The same checkout also stopped pinning the busiest public Solana node ahead of the platform's own faster ones, so building the payment is quicker and survives a throttled node.",
"tags": [
"improvement",
"fix"
]
},
{
"date": "2026-08-15",
"title": "Portfolio token pages price Base and Arbitrum holdings correctly",
"summary": "Opening a token from your portfolio asked Ethereum for the price no matter which chain the token was actually on. Almost every agent wallet runs on Base, so the most common holdings came back with no price, no chart and a shortened contract address instead of a ticker, even though the balance was right there in the list. Each token is now priced on its own chain, labelled with its real symbol and decimals, and native gas tokens use the right coin per chain (POL on Polygon, BNB on BNB Chain, AVAX on Avalanche) instead of always quoting ETH. Chains nobody prices, such as testnets, now say so plainly rather than showing another chain's number.",
"tags": [
"fix"
]
},
{
"date": "2026-08-15",
"title": "Push notification sign-up now only accepts a real push service address",
"summary": "Turning on browser notifications hands the platform the address your browser wants alerts delivered to, and that address is one the platform later calls from its own servers. It was stored exactly as given, so a crafted sign-up could point it at an internal address instead of a push service and use the delivery job as a way to knock on doors inside our own network. Sign-up now checks the address before storing it: it has to be an ordinary secure web address on a genuinely public host, or it is refused with a clear reason. If the check cannot finish because name lookup is briefly down, the browser is told to try again rather than being turned away for good. Delivery re-checks every stored address at send time too, so anything registered before this landed can never be dialed, and an address that could never be legal is dropped from the list the way an expired one already was.",
"tags": [
"security"
]
},
{
"date": "2026-08-15",
"title": "Agent coin pages stop reporting a burn that never happened",
"summary": "An agent coin page showed a \"last burn\" time taken from the last moment the buyback job ran at all, including every run where it looked and correctly decided there was nothing to burn. A coin with thousands of those skipped runs and no actual burns therefore read as having burned minutes ago while the burn count and burned total beside it both sat at zero. The burn time now counts confirmed burns only, so it stays blank until a real burn lands and always agrees with the count next to it. Separately, the three pump endpoints that manage a coin's autopilot policy and its invoices now require a request to originate from three.ws before it can act on a signed-in browser session, which is the rule the rest of the pump endpoints already followed. Agent keys and workers are unaffected.",
"tags": [
"fix",
"security"
]
},
{
"date": "2026-08-15",
"title": "Private avatars no longer show their picture in registry lookups",
"summary": "The registry lookup that turns an agent id, avatar id, slug, or Solana mint into one record already refused to hand out the 3D model of a private avatar, but it still returned the rendered preview image of that same avatar, so anyone holding the id could see what a private creation looked like. The preview is now gated on exactly the same visibility as the model: a private avatar returns its name and identity and no imagery at all. The embeddable widget read did the same thing with its avatar thumbnail and got the same gate. Separately, every read path that turns a stored thumbnail into a URL now goes through the one resolver that drops the old broken image keys, so agent manifests, token metadata, and launch feeds stop publishing image links that answer 404.",
"tags": [
"security",
"fix"
]
},
{
"date": "2026-08-15",
"title": "The live trade tape now says when live trades are unavailable",
"summary": "The per-coin trade tape streams from PumpPortal, which gates per-token trade subscriptions behind a funded API key. When that subscription was refused, the connection stayed open and simply never delivered a trade, so the tape showed a lit 'live' indicator above rows that could never arrive, and there was no way for a viewer to tell that apart from a genuinely quiet coin. The stream now forwards the refusal to the browser and the tape reports it plainly, keeping the settled trades it already loaded. A malformed coin address on the stream is also rejected up front with a clear error instead of opening a connection that can never match anything.",
"tags": [
"fix"
]
},
{
"date": "2026-08-15",
"title": "Daily usage caps stopped resetting themselves partway through the day",
"summary": "When the fast counter store is unavailable, the daily caps that bound expensive actions (persona extraction, voice cloning, withdrawals) fall back to a shared database counter, and a housekeeping job trims counters whose window has passed. That job decided what counted as 'passed' from the shortest-lived counter the server it ran on happened to have loaded, so a server that had only handled short-window traffic deleted everybody's in-progress daily rows. The effect was invisible and generous in the wrong direction: a cap of five per day would let a caller through six, eight, twelve times, because the tally kept getting wiped mid-day. Housekeeping now measures against the longest window any cap can use, so a daily tally survives until its day is genuinely over.",
"tags": [
"fix",
"infra"
]
},
{
"date": "2026-08-15",
"title": "The persona extraction and preview endpoints are documented in the API reference",
"summary": "Both routes have been live and usable for a while (turn a short interview or a sample of someone's writing into a structured voice profile, then audition that profile against a test message), but the only way to learn their request shape was to read the demo page's source. The API reference now covers them properly: every field with its limits, a real request and response for each, the exact meaning of every error status, and the per-user call budgets. The demo at /demos/persona-extract.html remains the fastest way to see them work.",
"tags": [
"docs"
],
"link": "/docs/api-reference"
},
{
"date": "2026-08-15",
"title": "The trusted-agent leaderboard now ranks every candidate, not an arbitrary sample of them",
"summary": "The leaderboard scores a capped pool of agents on every request, and that pool was selected in whatever order the database happened to return, from a candidate set roughly twice the cap. Half the agents with a real track record were silently skipped, so a top-scoring agent could be missing from the board entirely and two identical page loads could disagree on the ranking. The pool is now ordered by settled ledger volume before it is capped, so the strongest candidates are always scored and the board is stable between requests. Two related repairs: asking for a leaderboard with a non-numeric size (limit=abc) returned an empty board that looked exactly like 'no agent here is trusted', and now returns a plain error; and a database outage no longer publishes an empty leaderboard as if it were the truth, it reports the outage.",
"tags": [
"fix"
],
"link": "/trending"
},
{
"date": "2026-08-15",
"title": "A Premium Data API pass you bought straight from your wallet now shows up on your billing page",
"summary": "You can buy a pass without an account, paying from a Solana wallet on the Data API page. Do that, then sign in with the same wallet later, and billing showed you nothing: the pass was live, the API key worked, but the page insisted you had never bought anything. It was reading a single legacy address field on the account, and signing in with a Solana wallet never fills that field, so the lookup could not match a purchase it was standing right next to. Billing now resolves every Solana wallet you have signed into the account and shows every pass any of them bought, including purchases made before the account existed, with each pass's key, rate limit, and usage. Only wallets you have actually signed in with count, so nobody can claim someone else's purchases by typing an address. Two smaller repairs alongside it: the pricing endpoint now reports each asset's decimal precision from the same source that computes the amount you owe, rather than a fixed number that could one day render a price wrong by a factor of a hundred; and key rotation is now metered per account, because every rotation mints a real key.",
"tags": [
"fix",
"improvement"
],
"link": "/dashboard/billing"
},
{
"date": "2026-08-15",
"title": "Coin Intelligence search and verdict filters now look at the whole feed, not just its newest page",
"summary": "The Coin Intelligence radar lets you narrow the live launch feed by verdict (Strong, Watch, Caution, Avoid) and search it by name, symbol, or mint. Both filters were applied only to the page of coins already fetched, so asking for 60 Strong coins returned whichever handful of the newest 60 launches happened to be Strong: one, on a busy day. That reads as \"the engine found nothing\" when the engine had plenty. Search had the same blind spot, and could not see a coin that matched but sat a page or two back. Search is now run by the database across the observed feed, and the verdict filter looks through a much deeper recent window before handing back the page you asked for, so a filtered view returns a full page whenever a full page of matches exists. Searching for a literal percent sign or underscore now matches that character instead of behaving as a wildcard. The live trade tape at /api/pump/dex-trades also became readable from any origin, matching every other public pump.fun read endpoint, so an agent running on its own domain can consume it directly.",
"tags": [
"fix",
"improvement"
],
"link": "/coin-intel"
},
{
"date": "2026-08-15",
"title": "The referral visit beacon now names the real fault when it cannot read a request",
"summary": "Referral share links (?ref=CODE) fire a small beacon that records the visit, so the share to signup funnel is measurable. If a caller sent that beacon without a JSON content-type header, or with a body too large to read, the endpoint answered that the referral code was malformed, which is the one thing that was not wrong: the code had never been read. A bad or missing content-type now returns 415 unsupported_media_type, an unreadable body returns 400 bad_request, and an oversized one returns 413 payload_too_large, so anyone hosting their own landing page for a three.ws referral link fixes the header instead of hunting a code that was fine. The write path also picked up its first test coverage: referrer resolution, unknown-code visits, and the once-per-visitor-per-day dedup are now pinned, so the funnel counts cannot drift unnoticed.",
"tags": [
"fix",
"improvement"
],
"link": "/dashboard/referrals"
},
{
"date": "2026-08-15",
"title": "Scanning for trade candidates is faster on creator-filtered rules and honest about how much it looked at",
"summary": "The Trading Brain's \"Scan for matches\" preview shows the live launches your rule would buy right now, each with a real quote and a real safety verdict, and it never trades on its own. Three fixes. A rule that filters on the creator's track record has to look that record up over the network, and the scan was buying that lookup for every launch in the feed page, including ones it had already rejected for being too old or too small; it now pays for a creator lookup only when the rest of the rule already likes the coin, which is the difference between a scan that returns and one that crawls. The result said it had scanned the whole feed page even though it stops as soon as it has priced six matches, so the count overstated the work behind the answer; it now reports what it actually examined alongside the size of the page it read from. And an agent's Solana address is a field you can edit, so a value that is not a real address turned the preview into a server error rather than a scan; the safety check now resolves that address itself and the scan completes. The scan also gained its own per-owner limit, because every match it prices costs a live on-chain quote and a safety simulation. Separately, when the Trading Brain refuses something, the panel now shows the reason the server sent instead of a generic \"Scan failed\", so a rate limit, a kill switch, or an unreachable launch feed each read as themselves.",
"tags": [
"fix",
"improvement"
]
},
{
"date": "2026-08-15",
"title": "Sending $THREE from the wallet works, and swaps stopped failing outright",
"summary": "Two things were broken in the transaction builder behind the in-app wallet and the chat wallet tools. Sending any Token-2022 coin, which includes $THREE itself and every pump.fun-era launch, failed with a generic server error: the builder assumed the older token program and never asked the chain which one actually owns the mint. It reads that from the chain now and signs against the right program, so a $THREE transfer builds, and it opens the recipient's token account in the same transaction when they do not have one yet. Swaps had been failing entirely because they still called a Jupiter address that no longer exists anywhere on the internet; they now go through the same Jupiter client the rest of the platform uses, and a real quote comes back. Alongside that, mistakes in what you send are finally answered as mistakes instead of a support reference: a mistyped address, an address that is not a token, a memo that is too long, a signature that is not really a signature, or a balance that is too small each come back naming what is wrong, before your wallet ever asks you to approve something that could only fail. Amounts are also converted with exact decimal math now, so no transfer is off by a rounding error.",
"tags": [
"fix"
]
},
{
"date": "2026-08-15",
"title": "Rotating or revoking an x402 API key now requires a proof the request came from you",
"summary": "The billing page's rotate and revoke buttons destroy a live API credential, but the endpoint behind them accepted any request that carried your session cookie. A page on another site could have posted to it and killed a working key without you ever seeing the response. Both actions now require a single-use token the page fetches for each write, and the page tells you plainly when a token has gone stale or you have tried too often. Rotating also mints the replacement key before retiring the old one, so a failure part way through can no longer leave you with no key at all.",
"tags": [
"security",
"fix"
],
"link": "/billing/keys"
},
{
"date": "2026-08-15",
"title": "The Duplicate button on your widgets works again",
"summary": "Duplicating a widget from the dashboard failed every single time. The clone request never gave the new widget an identifier, and because the database requires one, the copy was rejected before it was written and the page showed only a generic server error with a support reference. A second, correct copy of the same clone logic was already in the codebase and the route simply was not using it; the route now runs that one, so duplicating gives you a real new widget with its own id, its own name ending in (copy), and the original's type, look, avatar and visibility carried over. The duplicate request also now carries the same cross-site request protection every other widget change already had, and it accepts an API key with write access, so you can clone a widget from a script as well as from the dashboard. Two smaller corrections travelled with it. When no chat provider is configured and the platform answers a visitor with its own placeholder line, that message is no longer filed in your transcript as though a model had written it. And chatting with one of the showcase widgets in the public gallery no longer tries to file a transcript against a widget that does not exist, which failed on every message. The developer reference now documents duplicate, stats, chat, transcripts and the knowledge base, all of which were live but undocumented.",
"tags": [
"fix",
"docs"
],
"link": "/dashboard/widgets"
},
{
"date": "2026-08-15",
"title": "Paying for a skill by Solana Pay QR now shows a real three.ws merchant card",
"summary": "When a wallet scans a Solana Pay code for a skill purchase, it first asks our server who the merchant is and shows the answer in the sheet you approve before signing. That reply pointed at a logo file that has never existed, so the icon came back as a 404 and the merchant appeared broken at the exact moment you decide whether to pay, under the retired 3D-Agent name rather than three.ws. The probe now returns the live logo and the current name, and answers wallets that ask from a browser, which the Solana Pay spec requires. Confirming a purchase through the off-web webhook also accepts the settlement hash an EVM payment needs, so a card that settled on Base can finish instead of sitting on pending forever, and a confirmation failure now reaches our error tracker rather than disappearing into a generic 500.",
"tags": [
"fix"
],
"link": "/marketplace"
},
{
"date": "2026-08-15",
"title": "A market intelligence outage no longer tells you your own API key was rejected",
"summary": "The two market intelligence reads in the unified API, recent narrative intel and momentum-ranked projects, pass a key we hold to the upstream provider; a caller needs no key at all. When that provider stopped accepting our key, both reads answered \"unauthorized\" with the provider's own wording, so an agent reading the response concluded its own credentials were bad and retried with different ones forever, when nothing on the caller's side was wrong. Both reads now say plainly that the service is temporarily unavailable because this deployment's upstream key was rejected, which is a status clients already know to retry rather than re-authenticate against. The older market intelligence endpoints always classified this correctly; the classification now lives in one place, so the two doors cannot drift apart again, and the error tables in the API reference list every code either one can return.",
"tags": [
"fix",
"docs"
],
"link": "/docs/api-reference"
},
{
"date": "2026-08-15",
"title": "The Duplicate button on your widgets works again",
"summary": "Duplicating a widget from the dashboard failed every single time. The clone request never gave the new widget an identifier, and because the database requires one, the copy was rejected before it was written and the page showed only a generic server error with a support reference. There was a second, correct copy of the same clone logic already in the codebase that the route was not using; the route now runs that one, so duplicating gives you a real new widget with its own id, its own name ending in (copy), and the original's type, look, avatar and visibility carried over. The duplicate request also now carries the same cross-site request protection every other widget change already had, and it accepts an API key with write access, so you can clone a widget from a script as well as from the dashboard. Transcripts got a smaller correction alongside it: when no chat provider is configured and the platform answers with its own placeholder line, that message is no longer filed in your transcript as if a model had written it.",
"tags": [
"fix"
],
"link": "/dashboard/widgets"
},
{
"date": "2026-08-15",
"title": "The tokenized-stock board sorts the way you asked, with the busiest names first",
"summary": "The Robinhood Chain stock board (and the API behind it) accepts a sort column and a direction. Both ran backwards: asking for the highest 24h volume returned the quietest tokens, the ones with no market at all led every column, and passing dir=asc handed back descending results. Numeric columns now read highest-first by default, dir=asc genuinely flips them, and a token with no price or no NAV feed sorts to the end in either direction instead of crowding the top. The response now also echoes the sort and direction that actually ran, so an unknown sort value comes back as the default it fell to rather than as the value you sent; the memecoin screener echoes its sort the same honest way.",
"tags": [
"fix"
],
"link": "/markets/robinhood"
},
{
"date": "2026-08-15",
"title": "The Granite embeddings endpoint no longer hands back an empty slot where a vector belongs",
"summary": "The public embeddings endpoint behind the watsonx Constellation turns your texts into vectors, one per input, and falls back to a second provider whenever IBM watsonx is unavailable. If a provider answered with only part of the batch, that counted as success: the fallback never ran, and the reply carried empty slots in place of the missing vectors while still claiming a full count. Anything plotting those vectors could not tell a missing embedding from a real one sitting at the origin. A partly answered batch is now treated as a failure, so the next provider covers the whole batch and a reply never contains an empty vector; if every provider is down you get a clear error naming the upstream cause instead. The IBM integration guide now also documents the batch limits, the real provider fallback order, and the fact that these responses are never cached.",
"tags": [
"fix"
],
"link": "/constellation"
},
{
"date": "2026-08-15",
"title": "The LLM usage panel in your dashboard settings finally shows your real numbers",
"summary": "Settings has a panel meant to show how much AI inference your agents have used this month. It always said \"No usage yet\", no matter how much you had used, because the endpoint behind it never sent the figures the panel was written to read. It does now: calls this month, tokens consumed, what those calls cost, and a per-model table showing which models did the work, busiest first. The same response also had two storage limits going out as text rather than numbers, which is the kind of difference that turns a size comparison into nonsense; both are numbers now. An API key with the profile scope can read all of it, so you can pull your own usage from a script, and the developer guide shows the exact shape.",
"tags": [
"fix",
"improvement"
],
"link": "/dashboard/settings"
},
{
"date": "2026-08-15",
"title": "A typo in a vault request now tells you which field is wrong instead of returning a server error",
"summary": "The vault routes take ids and amounts straight from the caller, and anything that was not a real id or a real number reached the database or the number parser and blew up there: depositing, redeeming, trading, or claiming fees with a mistyped vault id answered a generic \"internal error\" with a support reference, which told you nothing about what to fix. Every one of those now answers a plain 400 naming the field, before anything is looked up and long before any key is touched. Opening a vault with a malformed agent id used to hand back the raw database error code; it now reads as an ordinary validation message. The owner's terms editor gained the same care: a fee or drawdown value that is not a number is refused rather than stored as nothing, a per-trade ceiling raised above the daily budget is refused against the budget you already have set, and a negative spend limit is refused instead of being ignored while the page reports success.",
"tags": [
"fix",
"improvement"
],
"link": "/vaults"
},
{
"date": "2026-08-15",
"title": "Your walk keeps syncing across devices, and a remotely driven avatar reports where it really is",
"summary": "Four fixes across the walking avatar. Signed-in walkers sync their walk (avatar, environment, camera, position) between devices, and the page saves that snapshot on a half-minute heartbeat, but the endpoint only allowed thirty writes an hour and counted every read against the same budget: about a quarter hour into a walk, syncing quietly stopped, and everyone sharing an office or cafe connection shared one allowance. Writes are now measured per account, sized to how often the page actually saves, and reading your snapshot back costs nothing. When an outside program drives a live walk page through the control API, the page reports its position, heading, motion and scene on every poll; one unexpected value in that report used to discard the whole thing, so the controller's status read could show a stale avatar with nothing to explain it. Each field is now taken on its own merits. A command sent with an invalid control token is refused before its contents are inspected. And the leaderboard now shows an unranked walker with the same handle, profile link and picture as everyone else on the board, always drawn from a single avatar rather than pairing one avatar's name with another's picture.",
"tags": [
"fix"
],
"link": "/walk"
},
{
"date": "2026-08-15",
"title": "A vanity bounty or a sealed drop now goes live only after its payment truly clears",
"summary": "Posting a grind bounty and creating a sealed wallet drop are both paid with x402, which checks a payment and then settles it as two separate steps. Until now the bounty was published, and the drop wallet funded on-chain, before that second step finished. A payment that passed the check but failed to settle therefore left behind a live bounty a worker could be paid out for, or a funded gift wallet a recipient could still open, with nothing collected for either. Both now stay in a pending state that nobody can claim, refund, or reveal until the payment has actually settled: if it never settles, the bounty is voided and the drop's funding is swept straight back to the platform wallet. Separately, the certificate registry answers a certificate-id lookup with wording written for that lookup instead of the address version, and accepts an id pasted in any letter case, and the rarity share card is now rate limited.",
"tags": [
"security",
"fix"
],
"link": "/vanity/bounties"
},
{
"date": "2026-08-15",
"title": "Saving a big build no longer re-downloads the whole world first",
"summary": "Every time you saved a world, the service fetched and re-parsed the entire previous version of that world just to check who owns it. On a large build that meant a full extra download and parse on each autosave, for a single ownership field. The check now reads only the small index record, so saves on big worlds are noticeably quicker and the storage layer does half the work. Two error replies also got honest labels: a build that exceeds the size cap now answers \"payload too large\" and a request sent with the wrong content type answers \"unsupported media type\", instead of both arriving as a generic validation error that gave a client nothing to act on.",
"tags": [
"improvement",
"fix"
]
},
{
"date": "2026-08-15",
"title": "Widget analytics can tell you where your visitors are again, and a shared widget link previews correctly everywhere",
"summary": "The country column in your widget stats has been empty since the platform moved to its current host in July. A visitor's country was read from a header only the old host set, so once it was gone every view was recorded with no country at all and the top-countries panel stayed blank no matter how much traffic a widget got. Country is now read from the headers our current edge provides, and it is checked against the shape of a real two-letter country code before it is stored, so nobody can push junk into your analytics by editing a request. Chat sessions on talking-agent widgets get the same fix. Separately, the oEmbed endpoint that Slack, Discord, WordPress, Ghost and Notion call to unfurl a shared widget link now says plainly when a consumer asks for a format we do not publish, instead of returning JSON labelled as the format that was asked for and leaving the embed broken with nothing to explain why.",
"tags": [
"fix",
"improvement"
],
"link": "/widgets"
},
{
"date": "2026-08-15",
"title": "The trades feed stopped labelling a winning exit with a losing multiple",
"summary": "When an agent sells part of a position to take its initial stake back and lets the rest ride, only that final sell is recorded against the position, while the profit percentage covers both sells. The feed was working out the “×” badge from the final sell alone, so a trade that returned +88% could show up as “0.56× +88%” on the same row. The multiple is now worked out from the trade's actual return, so the two numbers always agree. A coin's page also asked the feed for every trade on that coin and was quietly given only the ones above +10%, hiding the rest of that coin's history; it now gets what it asked for. The feed also answers a malformed page cursor or coin address by saying so, instead of returning results for the wrong coin, and a database problem no longer looks like a platform where nobody has ever turned a profit.",
"tags": [
"fix"
],
"link": "/trades"
},
{
"date": "2026-08-15",
"title": "A profile’s follower count now matches the people the follower list can actually show you",
"summary": "The follower and following numbers on a profile counted every follow ever recorded, including ones from accounts that were later deleted or that never claimed a username. The list behind those numbers only shows people you can open a profile for, so a profile could advertise followers and then open to an empty list. Both numbers are now counted over exactly the people the list renders, so the badge and the list always agree. The list itself also stopped dropping those accounts only after it had already paged, which could hand you a blank page of followers with no way to reach the real ones behind it. Separately, the Creations tab used to answer a corrupted page cursor with a silent empty page that read as \"nothing more to load\"; it now says the cursor is malformed, and an unknown filter is rejected instead of quietly returning everything.",
"tags": [
"fix"
]
},
{
"date": "2026-08-15",
"title": "The Tutor’s running tab and end-of-session invoice now show what you actually paid",
"summary": "The Tutor charges one cent per answer and keeps an itemized tab you can reload later and close for a signed invoice. That ledger only ever knew how to store itself in a key-value service this deployment does not run, so nothing was ever written down: reopening the page never restored your history, and ending a session always produced an invoice reading \"no questions\" and a total of zero, however many answers you had paid for. The tab is now stored in the platform’s own database, so it survives a reload, a new browser tab, and any server restart, and the invoice lists every answer with its charge and a verification hash over the line items. Closing a session twice returns the identical invoice instead of a freshly stamped one, closing a session id that was never used no longer makes that id unusable for its real owner, and if the ledger itself is ever unreachable the page says so rather than quietly reporting that you owe nothing. Asking a question still succeeds even if the ledger write fails, because a bookkeeping outage should never cost you the answer you paid for.",
"tags": [
"fix"
],
"link": "/tutor"
},
{
"date": "2026-08-15",
"title": "Asking for a voice that does not exist now tells you so, instead of quietly speaking in a different one",
"summary": "The voice API accepts a provider and a voice id from the voice catalog. If the id was not one that provider actually has, nothing said so: the request was answered with audio rendered in that lane's default voice, and on the metered lanes it was billed, so a single typo in a voice id came back as a stranger's voice at full price with a success status. A well-formed id for a voice that had been removed took the opposite route and failed as a server error, which reads as our fault and tells a client to retry something that will never work. Both now return a plain rejection that names the id you sent and the catalog address to pick a real one from, and that check runs before anything is charged. Leaving the voice id out still gets you the lane default on purpose. Asking the catalog for a provider that does not exist used to return an empty list that was indistinguishable from a provider being down; it now says which providers exist. ElevenLabs voices, whose catalog grows every time you clone or add one, are still decided by ElevenLabs, but an id it rejects is reported as your request being wrong rather than ours, with any charge refunded first. Voice cloning also now asks who you are before reporting anything about how this server is configured. In the same pass, a key ElevenLabs turns down when you list your voices or browse the public voice library now comes back as your key being rejected, so the page can ask you for a new one, instead of a gateway error that invites a retry which can never succeed.",
"tags": [
"fix",
"improvement"
],
"link": "/docs/api-reference"
},
{
"date": "2026-08-15",
"title": "Looking up an on-chain agent by its CAIP reference works again",
"summary": "The public resolver that turns an ERC-8004 agent reference into its owner, token URI and verified agent card could not be reached at all. An agent reference ends in a slash before its token id, and the documentation told callers to percent-encode the whole reference, which turns that slash into %2F. The API refuses any address segment holding an encoded separator, because that is the same trick used to escape a path and reach files that were never meant to be served, so every correctly-formed lookup was turned away as an unknown address before it ever reached the resolver. The verification badge you can embed on your own site built its request in exactly that shape, so it never showed a result either. The resolver now accepts the reference with its slash as a normal part of the address, the badge asks for it that way, and the documentation shows the form that works. The protection against encoded separators is untouched. New tests resolve the address through the same routing the live site uses, so this cannot quietly break again.",
"tags": [
"fix"
],
"link": "/docs/api-reference"
},
{
"date": "2026-08-15",
"title": "A vault download rate-limit hit now tells you when to retry, and a browse outage stops reading as an empty listing",
"summary": "Two fixes on the BNB vault endpoints. Downloading a purchased vault file is deliberately rate limited, but when you hit that limit the reply came back without a retry-after time and without the limiter's own reason, so a client had no idea how long to wait, and a limiter outage (which deliberately refuses requests rather than letting them through unchecked) was indistinguishable from you genuinely spending your own quota. Both now come back with the refusal, the same way every other vault endpoint already answered. Separately, when browsing the vault, a listing whose stored file details could not be looked up was always reported as \"listed on-chain, details missing\". That is the honest answer when the storage network really has nothing for it, but the same answer was given when the lookup itself was down, quietly turning an outage into what looks like an incomplete listing. A genuine miss still shows as a listing without details; an outage now says the storage lookup is unavailable. Both endpoints also gained full test coverage, including proof that a download link issued to one buyer cannot fetch another buyer's file.",
"tags": [
"fix"
],
"link": "/vault"
},
{
"date": "2026-08-15",
"title": "A retried agent spend can no longer hand its transaction to somebody else, and an unsupported chain says so instead of blaming the network",
"summary": "When an agent redeems a permission you granted it, it can send a retry key so a request whose answer it never saw is not paid for twice. That key was filed under the permission alone rather than under the agent that used it, and it was read before the check that the caller owns the permission at all, so a different agent replaying the same permission and key was handed the original transaction hash and receipt instead of being turned away. The key is now filed per caller: a retry only ever returns your own result, and anyone else gets the same not-found answer they should always have got. Two more fixes in the same endpoint. A permission recorded on a chain three.ws has no delegation contract on could never settle, but the attempt came back as a network error, which reads as a temporary outage worth retrying rather than the permanent chain mismatch it is; it is now named plainly and refused before it consumes any of the permission's spending allowance. And the relayer kept its own private one-address-per-chain list of blockchain endpoints, separate from the failover chain the rest of the platform uses, so it would have refused to work on the next supported chain even where the shared endpoints reach it fine. It now uses the shared list, with any operator override still taking priority. The permissions guide gained a section on how the retry key behaves and how long it lasts.",
"tags": [
"security",
"fix"
],
"link": "/docs/permissions"
},
{
"date": "2026-08-15",
"title": "Agent .sol name lookups work again, and the avatar manifest schema now answers at its own address",
"summary": "The avatar-agent MCP server resolves .eth and .sol names to wallet addresses. Its .sol side went through a hosted name-lookup service that has since been retired: every request came back as \"not found\", which reads exactly like an unregistered domain, so a working name and a nonexistent one were indistinguishable. Solana names are now read straight from the chain, from the same name-service accounts the three.ws backend already reads, so the lookup depends on nothing but a Solana RPC. Two things were fixed alongside it. The per-lookup time budget was under five seconds, which a cold Ethereum lookup routinely exceeds, so real names were being reported as timeouts; it is now fifteen seconds and adjustable. And a lookup that genuinely fails now says why instead of quietly reporting \"not found\". Separately, the avatar manifest schema names https://three.ws/schema/avatar.v1.json as its own identifier, but nothing was served there, so any validator that followed the identifier got a 404. The schema is now published at that address, with a test that keeps the served copy identical to the one shipped in the package.",
"tags": [
"fix",
"sdk"
]
},
{
"date": "2026-08-15",
"title": "The trading and MCP docs now match what the code and the live API actually do",
"summary": "An audit of eight documentation pages ran every command they print and checked every claim against the code behind it, then corrected what had drifted. The biggest one was in the trading tutorial: the command it gave you for verifying a fleet's fills on-chain asked for a field the leaderboard does not publish, so it returned an empty explorer link for every agent and the verification step could not be completed at all. It now uses the fields that exist, linking both legs of each trade straight to Solscan, and it skips agents whose wallet has not been provisioned yet instead of building a URL that goes nowhere. The MCP tool safety page undercounted the tools that share a name across two servers, which matters because that collision is exactly why a trust policy has to key on the namespaced id; it now lists all nine and prints a command that re-derives the list so it cannot go stale again. The MCP tool catalog itself had drifted from its source and was serving outdated descriptions for three tools, on a page whose whole promise is that it is generated from source and cannot drift; it has been regenerated. The package extraction guide had every version in its roster out of date along with its build and test columns, and still described a publisher field as missing that was set some time ago. The 10 SOL trading experiment page stated a risk rule in absolute terms that contradicted the setup script it tells you to run, which read as a bug in one of the two; the rule now carries the scope the code actually applies.",
"tags": [
"docs",
"fix"
],
"link": "/docs/mcp-safety"
},
{
"date": "2026-08-15",
"title": "The AgenC task room explains itself now, instead of showing a wall of raw error text",
"summary": "The task room renders every task a wallet has posted on the AgenC coordination protocol as a 3D agent driven by its live on-chain state. It has been rebuilt around the moments where it used to leave you stuck. An unreadable wallet address, a Solana outage, and a dropped connection all used to print the raw server reply into an otherwise empty page; each now gets its own explanation, a Retry button, and a link to the docs on posting a task, and the room keeps retrying by itself when the fault is upstream and stops when it is something only you can fix. A wallet with no tasks says so and offers to check the other cluster instead of leaving you guessing. While a load is in flight you get skeleton cards rather than the word loading, and a line above the grid tells you how many tasks are showing, on which cluster, and when they were last refreshed. Card status colors now match the states the chain actually reports, so a task in progress or awaiting validation is no longer left uncolored, and each card links to its account on Solana Explorer. On a phone the wallet field no longer runs off the edge of the screen, the title no longer sits under the site logo, both form fields are labelled for screen readers, and every control has a visible focus ring. The room also stops running a 3D viewer for every card at once: avatars load as their cards come into view, up to eight at a time, and are released when you scroll past, which keeps a large board from stalling the browser and stops off-screen cards polling the chain for nothing.",
"tags": [
"improvement",
"fix"
],
"link": "/agenc/room"
},
{
"date": "2026-08-15",
"title": "The embed and sharing tutorials now describe the surfaces that actually exist",
"summary": "An audit of eight beginner tutorials found several describing features the platform does not have. The avatar-swap tutorial sent readers to Widget Studio for a body swap it cannot do, through a gallery with filters, preview buttons, and a 'Bust cache' save dialog that were never built; it now documents the real one-click swap in the agent editor's Outfit panel, and explains why nothing needs a cache purge. The sharing tutorial claimed rich social previews came from the plain agent URL, which actually serves a generic card, and described a headless renderer, a 24-hour image cache, a 'Refresh preview image' button, and a private-profile toggle that do not exist; it now points at the real share URL that carries the agent's name, description, and card, and explains what genuinely controls discoverability. The AR tutorial's iframe example used an address that returns 404 for an avatar id. The embed quickstart showed an agent ID in the wrong format, overstated the bundle size, and had the effect of the shadow DOM backwards. The docs-world tutorial gained the search palette it never mentioned, and the greeting tutorial now warns that the editor's Greeting field is not what makes an embed talk. Five page descriptions were corrected too, including one promising knobs the tutorial never covered.",
"tags": [
"docs",
"fix"
],
"link": "/tutorials/swap-avatar-in-studio"
},
{
"date": "2026-08-15",
"title": "Two tutorials stop shipping code samples that quietly do nothing",
"summary": "An audit of eight developer tutorials found two whose main code sample looked right and silently failed. The custom-skill weather handler split the weather API's reply on a plus sign, but the temperature field arrives with a leading plus of its own, so every field after the condition landed one slot off and the temperature always came back empty; it also mapped the fahrenheit option to a format code that means feels-like rather than Fahrenheit, so the celsius and fahrenheit paths returned the same units. Both are fixed and the sample was run against the live API in both unit modes. The gesture tutorial's AI lane read an emotion-breakdown field the Granite analysis tool has never returned, so the strength check never passed and the agent never played a single animation; it now reads the score and risk flags the tool actually returns, and grades the gesture across four bands instead of a binary. Separately, twelve tutorial hero buttons pointed at anchors on the docs hub that stopped existing when the hub split into per-topic pages, and now open the real pages; a placeholder URL in the Pump.fun tutorial was being parsed as HTML, which broke a link and injected stray elements into the page; and the self-host guide was corrected throughout, including storage variable names that no longer matched the code, a database step that left most tables uncreated, an audit-logging example importing a function that does not exist, a Cloudflare worker setup for a proxy the platform does not run, and a license claim that described the repository as open source when it is not.",
"tags": [
"docs",
"fix"
],
"link": "/tutorials/custom-skill"
},
{
"date": "2026-08-15",
"title": "A shared link to a docs page now opens that page instead of stranding you",
"summary": "Every cross-reference inside a docs page turns into a link like /docs/agent-wallets#financial-controls. Following one worked, but the URL you copied afterwards did not: on a fresh load the viewer only recognised the second half if that doc happened to be listed in the sidebar, and silently left you on the page you arrived from otherwise. Links to Financial controls, the x402 payment tutorial, and the wallet access and proof-of-custody chapters all failed that way. The viewer now asks the server whether the doc exists before giving up, so those links land where they say they will, and the browser tab takes the destination's name. Separately, /docs/3d-api was serving the 3D API landing page rather than the reference doc that the API's own discovery response, the OpenAPI license link, and the docs sidebar all point at; it now serves the reference doc, and the landing page keeps its home at /3d.",
"tags": [
"fix",
"docs"
],
"link": "/docs/agent-wallets"
},
{
"date": "2026-08-15",
"title": "The 3D API and agent manifest references now match what the endpoints actually return",
"summary": "The agent manifest page told you to validate a manifest with assertValid from @three-ws/avatar-schema. That package validates avatar manifests (mesh, skeleton, owner), not agent manifests, so the sample always threw. It now shows loadManifest from the published bundle, running against a live agent, and says plainly which document the avatar schema covers. On the 3D API page, the sample responses showed a model URL shape the service never returns, the inspect example was a stale capture of a model that has since been re-exported, and the Forge Pro tier table claimed draft and standard generations come back with no textures when every tier bakes them (1K, 2K, and 4K plus PBR at high). All three are corrected against live calls, and the page now explains why browser code should load a generated GLB through the first-party CDN path rather than the object store URL directly. The prompt art-director note is confirmed accurate and the code comment that contradicted it was fixed.",
"tags": [
"fix",
"docs"
],
"link": "/docs/3d-api"
},
{
"date": "2026-08-15",
"title": "The embodied AgenC demo hands you a live task, and reacts to the state a claimed task is actually in",
"summary": "The /agenc/embodied page binds a 3D agent to a real AgenC task on Solana, and until now it opened on an empty input box: it needed a task account address and offered no way anywhere on the page to find one. It now lists real tasks three.ws itself posted, claimed, and settled on-chain, newest first, so the demo works the moment it loads; a new public read, /api/agenc/recent-tasks, serves that list. Two things it showed were also wrong. The scene coloured itself for a state named Claimed, which the chain has never used: a claimed task is In Progress on-chain, so the halo, the badge, and the per-state animation switch all sat inert through the one transition the demo exists to show. And the page named a program address that no read has touched since the protocol's interface moved; it now shows the program each live read actually ran against. Pasting an address that holds something other than a task used to report a server error, as though the platform were broken, rather than saying no such task exists there; that is now a plain not-found, with the page telling you to check the address or switch cluster. The panel has designed loading, empty, and error states throughout, the form fields have real labels, the layout no longer scrolls sideways on a phone, and a task that cannot be read is retried with a backoff instead of being re-requested every four seconds forever.",
"tags": [
"fix",
"improvement",
"feature"
],
"link": "/agenc/embodied"
},
{
"date": "2026-08-15",
"title": "Pinning an avatar to IPFS works for real avatars, not just tiny ones",
"summary": "Minting an avatar from the Avatar Studio pins the model and its metadata to IPFS, and the whole file is sent through the platform so the storage credential never has to reach your browser. The endpoint said it accepted files up to 10 MB, but it was actually rejecting anything over roughly 730 KB, which is smaller than almost any real avatar. The error it gave back was a bare 'payload too large' that named no limit, so there was nothing to act on. The cause was two size limits that had never been reconciled: files are sent as text, which makes them a third larger in transit, and the transport limit was left at its default while the documented limit was set eight times higher. The limits now agree. Inline uploads are accepted up to 5 MB, which is what the request size can actually carry, and anything larger has a clear path: upload it to storage first and pin it by URL, a lane that has always accepted 50 MB. Over-sized uploads now say which ceiling was crossed. Two smaller fixes came with it: the filename you send is now the filename filed with the storage provider, instead of every avatar being stored as 'avatar.glb' and every metadata file as 'manifest.json', which made a pin list impossible to read; and a request with a genuine mistake in it, such as a file from somewhere the platform does not own, now gets an error explaining the mistake rather than a generic 'service unavailable' that hid it. The endpoint is now documented in the API reference.",
"tags": [
"fix",
"docs"
]
},
{
"date": "2026-08-15",
"title": "Eight build tutorials now match what the platform actually does",
"summary": "An audit of eight tutorials found instructions that could not work. The AI-brain tutorial described a model line-up that had been superseded: it never mentioned the Claude 5 family (Sonnet 5, Opus 5, Fable 5), which are live and selectable, and it recommended Sonnet 4.6 as the default. It also sent you to screens that do not exist, telling you to add API keys under a gear icon in My Agents and to change models, sampling knobs, and streaming under a Brain menu. Provider keys actually live in the AI Provider Keys panel on your account page, the model is a field on the agent's embed policy, and streaming has no off-switch. The promised usage tab with token charts and threshold emails did not exist either; the tutorial now documents the real per-agent usage read and points at the monthly quota and per-minute rate limit, which are the controls actually enforced before your money is spent. It also now explains that a visitor can never escalate your agent onto a paid model. The custom-GLB tutorial was worse for anyone following along: it linked a Studio address that no longer resolves, and four of its commands invoked gltf-transform subcommands that have never existed, so converting an FBX, renaming animation clips, and the full Mixamo walkthrough all failed at the first step. Those are replaced with the converter the platform itself uses and a clip-renaming script that was run against a real avatar to confirm it works. Its Draco flag was wrong, its upload path pointed at the wrong page, and it still warned that an avatar without an idle clip would stand frozen in a T-pose, which stopped being true when animation retargeting became universal. Finally, the page-events tutorial listed the wrong fallback order for emote animations.",
"tags": [
"docs",
"fix"
],
"link": "/tutorials/connect-ai-brain"
},
{
"date": "2026-08-15",
"title": "The Agent Exchange stops offering a purchase its wallet cannot pay for",
"summary": "On /agent-exchange two 3D agents buy and sell live crypto intel, and the buyer settles $0.01 of real USDC on Solana for every signal. The page already warned you when the buyer's wallet was too empty to settle, but it left the Buy intel button live anyway, so the only thing a click could produce was a failed payment. The button now turns itself off and says why, and the warning carries a Check again button, so a wallet you fund while the page is open recovers without a reload. The balance is also re-read after every purchase, so a settlement that empties the wallet is reflected immediately instead of one click too late. When the balance check itself does not answer, the page used to say nothing at all and quietly log a browser error; it now tells you the balance is unknown and offers the same retry. Two smaller fixes came with it: the page title and the buy button were being rewritten at load time from the translation catalog, which had drifted from the page and still carried the old punctuation, now corrected in all 85 languages; and the two avatar frames were keyboard stops that did nothing and showed no focus outline, so they are out of the tab order and the topic chips and buy button are two tabs closer for anyone navigating by keyboard.",
"tags": [
"fix",
"improvement"
],
"link": "/agent-exchange"
},
{
"date": "2026-08-15",
"title": "The 3D pipeline docs stop telling you to run a command that does not exist",
"summary": "Every paid stage on the 3D asset pipeline page (rig, remesh, game-ready, stylize, background removal) showed you how to pay for it by running npx @x402/fetch. There is no such command: that package publishes a library, not a CLI, so anyone who copied a sample got a could-not-determine-executable error and no asset. The samples are now the real thing: an unpaid curl that returns the live 402 challenge and costs nothing, plus one Node snippet that wraps fetch with our published @three-ws/x402-fetch buyer and actually settles. The end-to-end generate, rig and game-ready walkthrough was rewritten the same way and now polls the forge job with the fields the API really returns. The page also stops promising a specific settlement network, because the 402 challenge names the networks itself. Alongside it: the changelog page CDN pinning table listed four alias URLs that all 404 and pointed at an integrity file for a version that no longer publishes one; the AgenC page opening sentence linked the word AgenC to a different protocol page; and the vanity protocol spec showed a stale copy of the service key document, which now matches what the live endpoint serves.",
"tags": [
"docs",
"fix"
],
"link": "/docs/3d-pipeline"
},
{
"date": "2026-08-15",
"title": "The pump.fun trade history works again, notification settings show every category, and a bad agent id stops reading as a server crash",
"summary": "Three fixes across the public MCP packages and the API behind them. Asking for a token's recent buys and sells (get_token_trades on the free pump.fun MCP server) had stopped working entirely: the platform reads several transactions in one batched call, and the free Solana providers we route through cap those batches at one call each, so they rejected the whole request and the tool returned a raw provider error instead of trades. The platform now recognises that refusal, re-sends the reads one at a time to the same provider, and remembers the cap so later calls skip the wasted attempt; every batched on-chain read benefits, including wallet and agent activity history. Separately, the notifications MCP server told agents there were six notification categories when there are seven, so an agent could never change how it hears about a finished 3D generation; all seven are now listed, along with the security notices that always reach your bell inbox. And the agent provenance log answered a non-uuid agent id with an internal error and a support reference; it now says plainly that the id must be a uuid.",
"tags": [
"fix",
"sdk"
]
},
{
"date": "2026-08-15",
"title": "Avatars can move their lips again: the Audio2Face lane is back",
"summary": "Every call that turns speech into facial animation had been failing with an upstream not-found error, so a talking avatar spoke with a frozen face and the one-shot say() path returned nothing usable. The hosted model id the lane pointed at had stopped resolving for our NVIDIA account. It now points at the hosted Audio2Face NIM that is actually live, verified end to end: a synthesized clip drives a 30 fps ARKit blendshape track whose jaw genuinely opens and closes with the words. The discovery helper that finds these model ids was missing the name NVIDIA publishes them under, which is why the outage looked like a configuration mystery; it now finds them.",
"tags": [
"fix"
]
},
{
"date": "2026-08-15",
"title": "Voice and vision SDKs ship runnable examples, and the x402 seller SDK reads its Solana sponsor from the environment",
"summary": "@three-ws/voice and @three-ws/vision-mcp now include example scripts you can run as-is against the live free lanes: the voice one speaks a sentence, animates a face from those exact bytes, and transcribes it back; the vision one probes the lane, writes alt text for an image, and reads the text off it. @three-ws/x402-server used to demand the Solana sponsor account as a code option even though its own error message told you to set X402_FEE_PAYER_SOLANA, so a correctly configured deployment still got missing_fee_payer; it now reads that environment variable when the option is omitted. The VS Code x402 extension's endpoint scaffolder was generating Base-only endpoints and now generates Solana-first ones, matching every paid route on the platform, and it ships with a test suite covering which payment requirement it selects.",
"tags": [
"sdk",
"fix"
]
},
{
"date": "2026-08-15",
"title": "The embed tutorials now name the attribute that actually hides the built-in chat",
"summary": "Several tutorials told you to set avatar-chat=\"off\" to hide an agent's built-in chat input. It does not do that: it moves the avatar out of the chat column and lays the chat out as a plain bottom bar, leaving the input on screen. The multi-agent walkthrough leaned on it to keep its behind-the-scenes stage-director messages hidden, so anyone following it verbatim saw those messages leak into the visible chat. The affected tutorials now use chat=\"off\", the real bare-avatar switch, and note that the brain still boots behind it so say(), ask(), and the event stream keep working. The same accuracy pass corrected the walking-avatar leaderboard example, which read fields the API never returns and so always rendered an empty top-walkers list, and the JavaScript API event table, which named the wrong key for the skill a tool-call event reports.",
"tags": [
"fix",
"docs"
],
"link": "/tutorials/web-component-end-to-end"
},
{
"date": "2026-08-15",
"title": "The vanity SDK now uses the payment wallet you hand it, and the in-chat 3D avatar keeps its tuned lighting",
"summary": "Three fixes in the developer packages. The @three-ws/vanity SDK documented a fetch option on its paid address-grinding lane but silently ignored it, so a developer who passed a payment-enabled fetch got a payment-required error back instead of the keypair they paid for; that option now works, as do a custom API host and key. The avatar MCP server's in-chat 3D viewer had been hand-tuned with softer shadows, brighter exposure and a neutral environment light, but the script that rebuilds it before every release still produced the old flat version, so the next release would have quietly undone the tuning; the script now produces the tuned viewer, and the same settings apply to the fallback view for chat apps that cannot run the interactive one. The tool SDK's quick-start example pointed at an API that does not exist and skipped the wrapper an MCP server needs to return anything at all, so following it produced a tool that answered with nothing; it now runs end to end against a real free three.ws endpoint.",
"tags": [
"fix",
"sdk",
"docs"
]
},
{
"date": "2026-08-15",
"title": "Seeding an agent from GitHub is locked to your own click, and a mistyped pick no longer costs you six hours",
"summary": "Three fixes to the GitHub memory lane, the one that reads the profile, repositories, and READMEs you tick and distills them into what your agent knows about your work. First: writing or deleting those memories now requires the one-time token the site attaches to your own actions, so another site cannot make your browser re-seed or wipe an agent behind your back. Every neighbouring memory action already worked this way; this one did not. Second: seeding is allowed once every six hours per agent, and that window used to be spent by attempts that never read anything, so ticking nothing and pressing Seed, or picking a repository that had been renamed since the list loaded, locked you out for six hours over a correctable mistake. The window is now taken only once your selection checks out, immediately before anything is actually read. Third: when every model provider is busy at once, the seed used to fail with a bare internal error and a support reference; it now says plainly that nothing was read, that your existing memories are untouched, and roughly how long until you can try again.",
"tags": [
"security",
"fix"
],
"link": "/docs/github-memory-seeding"
},
{
"date": "2026-08-15",
"title": "Three docs pages stop pointing readers at a dead page, a missing section, and a snippet that does nothing",
"summary": "The Vanity page told you to grind a long address yourself in the browser at /vanity, but that path has never existed and answers 404; the in-browser grinder lives at /vanity-wallet, and both links on the page now go there. In Troubleshooting, the fix for a Quick Look model that opens and instantly closes linked to a model-optimization section that is not on the AR page, so the link dropped you at the top with nothing to read; it now opens the section that actually covers it. The same page told you to add a responsive attribute to size an inline embed, with a snippet that set the width in CSS. Neither half worked: responsive sizing is already on by default, and the element reads the width attribute rather than the computed style, so the documented snippet never produced the 3:4 ratio it promised. The instructions now show the form that does work, and name the two things that switch it off.",
"tags": [
"fix",
"docs"
],
"link": "/docs/troubleshooting"
},
{
"date": "2026-08-15",
"title": "The agent economy volume dashboard stops sending you to profiles that are not there",
"summary": "Top earners and top spenders on the agent-to-agent volume page linked every ranked agent to a profile, including agents whose identity had been deleted. Those rows sent you to an empty profile shell for an agent that no longer exists. Both leaderboards now name a departed agent without linking it, the same rule the settled-hire feed below them already followed, and a row whose stored avatar has been pruned falls back to its initials tile instead of a broken-image icon. The page also cleared its accessibility floor: the small print under each stat, the footnote, and the retry button were all below the WCAG AA contrast minimum in dark mode, and in light mode the cyan links and the green live badge were effectively unreadable. Links inside the empty-state sentences now carry a standing underline, so they are not marked by colour alone.",
"tags": [
"fix",
"improvement"
],
"link": "/agent-economy-volume"
},
{
"date": "2026-08-15",
"title": "The guard and gesture tutorials stop teaching steps that fail, and one that wiped your clip list",
"summary": "Two tutorials had drifted away from the code they document, and following either one exactly produced a bad outcome rather than a confusing one. The gesture tutorial's API example padded its request body with an empty clip array, and told readers that field was required; it has not been required for a while, and an empty array is a real value, so anyone who copied that snippet to change one gesture slot silently erased every custom clip on their agent. The endpoint's actual rule is now stated: every field is optional, present replaces it, absent leaves it alone. That page also listed four gestures on the /walk quick keys that are not on them, so pressing 1 through 8 never produced what it described; the eight keys are now named in order, with the wheel and the code call given for the four that have no key. The guard tutorial's registry entry left out the proof block, which the registry auditor has since made mandatory, so a reader who followed every step to the end hit a failure the page never mentioned. It now includes a working proof, shows the exact error you get without one, and adds a step for the command that runs it, which is the difference between a guard that is wired up and a guard anyone has shown still catches anything. Its summary line also described a completely different guard from the one it builds.",
"tags": [
"docs",
"fix"
],
"link": "/docs/tutorials/write-a-guard"
},
{
"date": "2026-08-15",
"title": "Arming an Oracle agent for live trading no longer rounds your trade size up",
"summary": "The arm page clamps the risk knobs on the server so a bad number can never reach the trading loop, and those clamps have a floor: the smallest trade the loop will place is 0.001 SOL. That floor was being applied in the wrong direction. Ask to arm a live agent with a per-trade size of zero, and instead of being told no, you were armed for real spending at 0.001 SOL a trade with a half-SOL daily ceiling you never asked for. The check meant to catch this looked at the number after the clamp had already raised it, so it could never fire. Arming live now refuses any per-trade size below the floor, and any daily cap smaller than a single trade, and tells you which number to fix; nothing is armed until you do. Clamping a too-large number back down to the ceiling is unchanged, and simulate runs keep the same forgiving behaviour as before. The arm page also shows the reason a save was rejected instead of one generic line, so an ownership problem no longer reads the same as a bad trade size, and a failed Telegram test now prints what Telegram actually said.",
"tags": [
"fix",
"security"
],
"link": "/oracle/arm"
},
{
"date": "2026-08-15",
"title": "The vanity receipt verifier stops calling honest receipts fraudulent, and the avatar CLI takes a Solana address",
"summary": "Two documented commands did not do what their page said. The command-line verifier for a provably-fair vanity address receipt scored the difficulty claim with the retired flat model, the one that treats every Base58 character as equally likely. Every receipt issued since the leading-character correction landed therefore failed its difficulty check and was reported as untrustworthy, even though the in-browser verifier at /vanity/verify passed the same receipt. The verifier now reads the model each receipt names in its own signed difficulty block, exactly as the page and the SDK do, so current receipts verify and older ones keep verifying under the model they were issued under. Separately, the Rig Doctor page showed an avatar-manifest command taking a Solana address as the owner, which the CLI rejected outright: it understood Ethereum addresses and full CAIP-10 identifiers and nothing else. Since Solana is the home chain, a bare Solana address is now accepted and recorded as solana:mainnet-beta. Alongside those, eight tutorial and reference pages were audited against the code they describe: the Vercel self-hosting guide had a stale copy of the config file and pointed at a migrations folder that has moved, the MCP guide listed three placeholder OAuth scopes where the live server publishes twelve real per-capability ones, and the Avatar CLI page pinned a version number that npm has never had.",
"tags": [
"fix",
"docs",
"sdk"
],
"link": "/docs/rig-doctor"
},
{
"date": "2026-08-15",
"title": "The agent-payment and MCP docs now show the code that actually ships",
"summary": "Three developer docs had drifted from the code they document, in ways that cost a reader real time. The Payment Sessions walkthrough, which is the guide for giving an agent a budget instead of a wallet key, quoted five snippets that no longer match the source: the branch that refuses a payment when the budget will not cover it, the host-allowlist check (which has moved into a shared module so the dry-run simulator and the live enforcement can never disagree), the expiry sweep that refunds unspent budget, the dollars-to-atomic-units helper, and a column missing from the table definition. All five now match the shipped code line for line, and the page links the two source files it had not been pointing at. The MCP article listed the Solana vanity grinder at up to $0.50 scaled by the work done; it is a flat $0.05 per successful grind, and its setup example set a wallet variable the package does not read, so anyone copying it would have configured a payout address that was silently ignored. That example now shows only the variable that works, and says plainly that settlement is USDC on Solana mainnet. The IBM Granite forecasting example showed three data points for a tool that requires at least sixty-four, so a copied sample was guaranteed to be rejected; it now says so.",
"tags": [
"docs",
"fix"
],
"link": "/docs/aws-builder-center-agent-payment-sessions"
},
{
"date": "2026-08-15",
"title": "The Alibaba Cloud MCP server starts now, and five MCP packages document how to import them",
"summary": "The @three-ws/alibaba-cloud-mcp package, which lets an agent use Qwen chat, embeddings, and model discovery on your own Alibaba Cloud key, could not start at all. Two faults, both on the first line of use: the check for whether the file was launched directly crashed before the server ever opened, and the three tools were registered with a schema format the MCP library rejects. Both are fixed, so the server now boots, advertises its tools, and answers a client. The payment-sessions MCP package had a documentation fault with the same cost: its quick start told you to set your session variable to the whole cookie including the name, which the server then wrapped in the cookie name again, so every session-management call came back unauthorized. It now says to use the cookie value, and ships its two runnable examples inside the published package instead of only in the repository. Five packages (payment sessions, Agora, alerts, the assistant widget, and Alibaba Cloud) gained a Programmatic use section showing the exact code to import their tools into a host of your own, and the sniper package's HTTP table now lists every endpoint it actually serves, including the paid close-position route it had left out.",
"tags": [
"fix",
"sdk",
"docs"
]
},
{
"date": "2026-08-15",
"title": "The Agent Economy demo works again when you just pick a service, and it fits a phone screen",
"summary": "On the Agent Economy page, two 3D agents trade with each other for real SOL: Nova buys a market read, an on-chain insight, or a risk score from Oracle. Picking a service and typing nothing in the optional topic box, which is what almost everyone does, sent the request off in a shape the endpoint refused, so the demo answered with a validation error instead of a trade. It now sends the request correctly, and the endpoint accepts an empty topic besides. On a phone the page was worse than broken: the transaction feed and the status line, the entire output of the demo, were cut off the bottom of the screen with no way to scroll to them, and the floating language button sat on top of the footer links. The centre column now scrolls, the feed keeps a readable height, and the footer stays clickable. Three failure states also stopped lying: a payment that could not go through no longer ends on Transaction complete, a request that never reached the server now says so and offers to try again, and unreadable wallet balances now say they are unavailable and offer a retry instead of spinning forever.",
"tags": [
"fix",
"improvement"
],
"link": "/agent-economy"
},
{
"date": "2026-08-15",
"title": "Oracle stops answering \"nothing is happening\" when the database is the thing that is down",
"summary": "Several Oracle reads shared a habit: if the database was unreachable for a few seconds, they answered normally with an empty result. The movers panel said nothing was moving, the stats hero said no coin had ever been scored and no agent was armed, a coin's conviction sparkline came back flat, and the machine-readable signal endpoint told autonomous agents there were no plays, which is exactly the cue for an agent to stand down. Every one of those answers was then cached at the edge for a minute or two, so a blip lasting seconds kept showing a dead market long after the database was back. Those endpoints now tell the difference between a connectivity failure and an ordinary query fault: the first is reported honestly as a temporary outage with a retry hint and is never cached, the second still degrades to the empty answer it always did. The social ingest got the sharper end of the same bug: it used to reply that a batch had been accepted while reporting zero mints written, a receipt for a write that never landed. It now fails loudly so the sender retries. Shared conviction cards behave too: one built during an outage is no longer pinned to a coin's link preview for 15 minutes.",
"tags": [
"fix",
"improvement"
],
"link": "/oracle"
},
{
"date": "2026-08-15",
"title": "Docs pages stop opening with a wall of raw metadata, and the AWS pages say where the Marketplace listing actually stands",
"summary": "Nine documentation pages are written as article drafts, so each one starts with a small block of publishing metadata: the title, where it is meant to run, its tags. That block is meant for the tooling, not for you, but the docs reader had no idea what it was and rendered it as an enormous heading at the very top of the article, above the actual writing. It is now recognised and hidden, on the page and in the Copy page text, so those articles begin where they should. Reading the same pages closely turned up claims that had drifted: several said three.ws was already live on AWS Marketplace and billing usage to your AWS invoice. The integration is built and deployed, but the listing itself has not been published yet and the pricing decision is a free subscription with no AWS metering, so the AWS pages now say exactly that and point at the page that tracks the status. The MCP article's vanity-address tool was also listed at a metered price of up to fifty cents; it is a flat five cents per grind, which is what the code has always charged.",
"tags": [
"docs",
"fix"
],
"link": "/docs/aws-marketplace"
},
{
"date": "2026-08-15",
"title": "A silent outage in the agent reasoning ledger now reports itself",
"summary": "Every agent decision is committed to a hash chain, and that chain head is written to Solana on a schedule so anyone can check the history was never edited. The wallet that pays for those on-chain writes ran empty, and the scheduled job could not tell that apart from the harmless case where no signing key is set up at all: it answered a healthy 200 every ten minutes while re-sending the same doomed transaction for twelve agents in a row, and not one head reached the chain. A rejected write is now reported as a rejected write, the rest of the run stops instead of repeating a failure that cannot succeed, and it raises an alert so the wallet gets refilled. Verifying a ledger already worked without the on-chain leg and still does: the recomputed hash chain is the tamper proof, the on-chain write is the independent timestamp on top of it. Separately, the once-a-minute economy heartbeat now refuses to drive production from a machine that is not production, so a developer running it locally can no longer fire thirty-six live money engines at the real site.",
"tags": [
"fix",
"infra"
],
"link": "/docs/reasoning-ledger"
},
{
"date": "2026-08-15",
"title": "The plugin marketplace stops hiding a plugin when you page through it",
"summary": "Browsing plugins a page at a time could show you the same plugin twice and never show another one at all. The catalogue was sorted by install count and publish date, and the built-in plugins were all published in the same instant with zero installs, so that sort left them tied and the database was free to order the tied rows differently on every request. One plugin therefore landed on two consecutive pages and one fell through the gap. The order is now decided all the way down, so paging through the catalogue returns every plugin exactly once. Searching also takes your text literally: typing a percent sign or an underscore used to act as a database wildcard and match rows it should not. Publishing got stricter at the same time, because a plugin manifest is handed to the model as a tool definition by whoever installs it: a tool name now has to be plain letters, digits, underscores, or hyphens, every tool needs a real description, and a manifest has to fit the same 64KB ceiling that importing one by URL already enforced. And pointing the importer at a repository web page instead of the raw manifest file now says exactly that, rather than showing a raw parser error.",
"tags": [
"fix",
"improvement"
],
"link": "/marketplace"
},
{
"date": "2026-08-15",
"title": "Plan checkout: a paid upgrade can only ever settle on real mainnet, and a late payment is no longer stranded",
"summary": "Paying for a plan creates a quote, you send the funds, then the server checks the chain before it upgrades you. Three things in that path were wrong. The checkout side has always refused test-network quotes in production, because test-network dollars are free from a faucet, but the confirm side still let a caller name the test network itself, and it is the confirm side that decides which chain it looks at and which dollar it accepts. Both ends now read the same setting, so a production upgrade can only ever be settled by real funds on mainnet. Second, an Ethereum-side payment has to wait twelve blocks before it counts, which meant funds sent near the end of a session could land on-chain and then be refused for arriving late, with the money already gone; that lane now honors the same one-hour grace window the Solana lane has, so an on-chain payment is always claimable. Third, confirming an Ethereum payment could reach a Solana quote and mark it expired, and asking to buy a skill from a malformed agent id answered a generic server error instead of telling you what was wrong with the request.",
"tags": [
"fix",
"security"
],
"link": "/pricing"
},
{
"date": "2026-08-15",
"title": "The persona builder stops reporting a model outage as if you had been signed out",
"summary": "The two routes behind the persona builder (the one that turns your interview answers or a sample of your own writing into a structured persona, and the one that replies to you in that persona's voice) used to hand you whatever status code the model provider returned. If a server-side key expired, the provider answered 401 and so did we, which reads to a browser as an expired session on a request where nothing was wrong with your sign-in. Both now report a provider failure as exactly what it is: a server-side fault, with a plain message and no implication that you need to log back in. Separately, when the platform's daily model-spend ceiling is reached, the builder now says so and tells you when it resets, instead of asking you to retry something that could not have succeeded for the rest of the day. Both routes also now record their model usage against your account, so the spend they generate is visible and capped rather than untracked, and the daily extraction budget now holds across every server instance instead of being counted separately on each one.",
"tags": [
"fix",
"improvement"
]
},
{
"date": "2026-08-15",
"title": "Self-hosting docs no longer hand you a command that wipes your environment",
"summary": "The deployment guide told self-hosters to set Cloud Run environment variables with `gcloud run services update --set-env-vars`. That flag replaces the service environment wholesale, so following the instruction for a single key silently deletes every other variable the service had. The guide now uses `--update-env-vars` and says plainly why. Four more corrections landed in the same sweep across the deployment, configuration and contributing pages: the Avaturn signup link and the GitHub Discussions links were dead (Discussions are not enabled on the repository, so bug reports and design questions both go to Issues); the in-browser avatar builder was documented as needing its own hosted origin in production when the normal build already serves it same-origin at /avatar-studio; and three counts that had drifted from the code (cron jobs, MCP server packages, contract tests) were replaced with pointers to the files that hold the real numbers.",
"tags": [
"docs",
"fix"
],
"link": "/docs/deployment"
},
{
"date": "2026-08-15",
"title": "A broken paid endpoint is no longer reported as a free one",
"summary": "Payment sessions let you hand an agent a budget instead of a key, and two parts of that path judged an endpoint by one rule: if it did not ask for payment, the call was free. That is only true when the endpoint actually succeeded. A paid API answering 404, 400, or 500 was being scored as a free call, and the consequences ran in the wrong direction in both places. The dry run at /pay/simulator, whose whole job is to tell you in advance whether a budget and an allowlist will carry a workload, reported a plan as feasible at a cost of nothing when in fact every endpoint in it was erroring: free calls are never priced, never allowlist-checked, and never counted, so a wall of failures looked like a clean run. And an agent spending through a session got back a success envelope saying no payment was needed for an endpoint that had just failed on it. Both now separate the two cases: a real success with no payment challenge is still free and still costs nothing, while an error status is reported as an endpoint failure carrying the status the endpoint actually returned. No session is charged either way. The simulator lists each failing endpoint by name in its problems list and no longer calls the plan feasible.",
"tags": [
"fix"
],
"link": "/pay/simulator"
},
{
"date": "2026-08-15",
"title": "Agent Screen is usable on a phone, and reads correctly to a screen reader",
"summary": "Watching an agent work on a narrow screen was close to impossible. The row of viewer controls along the top (fit, screenshot, picture-in-picture, fullscreen, ambient, zen, and the link back to the agent) ran off the edge of the page with no way to reach them, the agent name and its live badge painted over the buttons underneath, and the two panels that open by default covered the live screen and the avatar cam completely: the whole point of the page, hidden. The control row now scrolls sideways so every control stays reachable at any width, the name keeps its space, and on a phone the page opens on a bare stage with the panels one tap away. Three fixes for anyone using a keyboard or a screen reader landed with it: the page now has a real heading and a main region for the skip link to jump to, the hidden Skip to content link stopped rendering as visible text at the top of the page, and every text field on the page gained a proper label. Finally, if the agent list fails to load for any reason other than being signed out, the page says what went wrong and offers a retry instead of telling a signed-in owner to go sign in.",
"tags": [
"fix",
"improvement"
],
"link": "/agent-screen"
},
{
"date": "2026-08-15",
"title": "Claim Your Wallet stopped telling real traders they had no track record",
"summary": "Pasting a wallet into Claim Your Wallet could answer \"not indexed yet\" to a wallet with thousands of indexed pump.fun trades. The page only accepted a wallet the reputation scorer had already graded, and that scorer runs on a rollup that lags the trade indexer by design, so a wallet the platform had watched trade for days was still turned away. Any wallet with a real on-chain record now renders its full ledger and window aggregates immediately, with the reputation panel honestly reading unproven until the grade lands. Three more fixes shipped with it. Win and loss counts now cover closed positions only, so a trader still holding sixty fresh bags no longer reads as sixty losses, and a new realized profit-and-loss figure sits beside the existing net figure that charges an open bag's entry as if it were written off. Asking for devnet returned the wallet's mainnet trades; both halves of the read are now scoped to the network you asked for. And if the database is unreachable, the page now says the record is temporarily unavailable and offers a retry, instead of publishing an empty scorecard that the CDN would then serve to everyone for the next few minutes.",
"tags": [
"fix"
],
"link": "/claim-wallet"
},
{
"date": "2026-08-15",
"title": "Premium Vanity Inventory shows the whole shelf, and is readable in light mode",
"summary": "The premium vanity shelf listed only the first hundred pre-ground addresses out of the hundreds actually in stock, with nothing on the page saying so: every address past that point was unbuyable because no one could see it. The page now loads sixty at a time, states plainly how many of the total you are looking at, and pages through the rest on demand. In light mode the page was close to unusable, because the addresses themselves, the buy dialog, and the custody notice were all painted in colours meant for the black background; the whole page now switches palette properly. Four smaller fixes shipped with it: the retry button on a failed load rendered as raw browser chrome and threw away your filter, so it is now a real button that simply re-runs the search you had; filtering down to no matches now offers a one-click way to clear the filter instead of just suggesting it; the word x402 in the intro linked to the vanity gallery rather than the x402 page; and keyboard users get a visible focus ring on the search and sort controls, which previously had theirs suppressed.",
"tags": [
"fix",
"improvement"
],
"link": "/vanity/premium"
},
{
"date": "2026-08-15",
"title": "Solana Vanity Wallet: one footer, a working toast, and the gift fee actually shows up",
"summary": "The sealed-gift composer opened with \"Loading pricing…\" and, depending on which finished first, could sit on that line forever: the translation pass reapplied the placeholder over the fee the page had already fetched, so nobody learned what a gift costs. The same collision could wipe the grind time estimate. Both now hold their own text. Confirmation toasts (\"Claim link copied\", \"Funds reclaimed\") were built with one class name and styled under another, so instead of floating over the page they dropped in as unstyled text at the very bottom, out of sight; that applies to the gift claim page too. Assigning a freshly ground wallet to an agent succeeded on the server and then reported \"Network error\" anyway, because the success message referenced an address it could not see. The page also shipped two footers stacked on top of each other, one of them a stale copy with broken markup and a Docs button wearing the X logo; the duplicate is gone. Rounding it out: the sign-in prompt pointed at a legacy /login.html URL, the gift form fields had no labels for screen readers, the expiry and asset choices now follow what the server actually accepts, keyboard focus is visible on every control (the text fields had their ring suppressed), buttons have press states, a failed agent list offers a retry instead of a raw error, and copying an address still works when the browser denies clipboard access.",
"tags": [
"fix",
"improvement"
],
"link": "/vanity-wallet"
},
{
"date": "2026-08-15",
"title": "What is three.ws: a living sample avatar, a fallback that still shows one, and a readable light mode",
"summary": "The sample avatar in the hero stood in a rigid T-pose, because the character ships an idle animation that nothing was playing. It now idles, faces the reader, and is framed to fit the pane at every screen size. Before it loaded, the pane was simply an empty rectangle with a caption promising an avatar you could drag; it now shows a real still of that same character straight away, and if the 3D viewer never arrives (a blocked CDN, a device without 3D, a slow network that stalls) the still stays put and the caption says so instead of inviting you to drag nothing. In light mode the page was unreadable: the background stayed black while every word switched to near-black ink, so the headline sat at roughly 1.2 to 1 contrast. The whole page now follows the chosen theme, and the Start free button no longer prints black text on the light-mode blue. The Common questions section was rebuilt too: each question is now a real heading, its answer is announced with the question it belongs to, the four questions and answers are translated along with the rest of the page rather than staying English, the search-result markup finally lists the four questions actually on the page, and opening an answer eases in instead of snapping.",
"tags": [
"fix",
"improvement"
],
"link": "/what-is"
},
{
"date": "2026-08-14",
"title": "The video studio stops loading an avatar it is about to throw away",
"summary": "Open /create/video signed out and the page did two things at once: it started downloading the default avatar model, and it checked whether you were signed in so it could send you to the login page. The redirect always won, which killed the model download halfway through and left an error in the browser console for the abandoned file, plus one more for every texture the viewer had started unpacking. Nothing was broken on screen, but a signed-out visitor was made to pull down three quarters of a megabyte of 3D model on their way to a login form, on whatever connection they happened to be on. The page now waits until it knows who you are before it loads anything: signed in, the avatar appears exactly as before, picked from your own avatars with the bundled one as the fallback; signed out, nothing is fetched and the console stays clean.",
"tags": [
"fix",
"improvement"
],
"link": "/create/video"
},
{
"date": "2026-08-14",
"title": "Eight more docs pages audited, and the embed snippet that would have failed on paste is fixed",
"summary": "A line-by-line sweep of the walk leaderboard, Atlas, repository guards, wallet sign-in, spending envelope, avatar image render, character library embed, and skill trial pages, with every command in them run against the live API. The embed tutorial pinned a Subresource Integrity hash that no longer matched the script it names, so a reader who pasted it got a blank box and a browser console error instead of a character; it now carries the hash the site itself publishes. The skill trial page printed a start-trial response the endpoint never returns, and now shows the real one plus the two refusals a seller has to handle. The walk leaderboard page listed the walk documentation set with two pages missing. The Atlas page said the slash key always opens search, when a page that claims that key keeps it. The spending envelope page quoted a free-call note with the wrong wording and now matches the API, and the render tutorial's optimizer figures were stale, so it now tells you to read the byte headers instead of the printed numbers.",
"tags": [
"docs",
"fix"
],
"link": "/docs/tutorials/character-library-to-embed"
},
{
"date": "2026-08-14",
"title": "The privacy policy's data table no longer pushes the page sideways on a phone",
"summary": "The privacy policy lists what three.ws collects in a three-column table: the category, an example, and why we hold it. On a narrow phone screen that table was wider than the screen, and because it sat in the normal flow of the page it dragged the entire document sideways with it, so every paragraph and heading in the policy scrolled left and right too. The table now scrolls on its own inside a box the width of the screen, and the rest of the page stays put. The box is reachable with the keyboard and announced to screen readers, so the table can be read across without a mouse. Checked at phone, tablet, and desktop widths.",
"tags": [
"fix"
],
"link": "/legal/privacy"
},
{
"date": "2026-08-14",
"title": "Eleven links that led nowhere now land where they promise",
"summary": "The audit that guards every page against dead ends had a hole in it: the rule that serves the not-found page matched every address on the site, so the audit read every link as reachable and reported a clean bill of health it could not vouch for. With that fixed, eleven links turned out to lead to pages that were never built. Sign in to save on the choreography and gesture pages pointed at an address that does not exist, so the one thing those pages ask you to do could not be done. The extension terms page linked its own terms of service to a missing address, twice. The motion studio offered to show you how to capture an avatar from a selfie and led nowhere, as did the pricing page's link to your account, the memory tutorial's link to your library, and two entries in the dashboard trading menu. Pressing Message on a friend in the 3D world walked you into a blank page instead of the conversation, and now opens the thread with that person. Three outbound links to other projects' documentation had rotted and now point at the pages that replaced them. The audit itself was rebuilt to be trustworthy: it no longer calls a live site dead because the site answers a robot differently than a browser, and it no longer treats the blog and docs it ships, or the app manifest it generates, as missing files.",
"tags": [
"fix",
"improvement"
]
},
{
"date": "2026-08-14",
"title": "A generation that quietly switched engines no longer counts as a failure",
"summary": "When a 3D generation lane loses your job, the platform resubmits it to another lane and you still get your model. Until now the abandoned attempt was recorded as a plain failure, with nothing tying it to the run that actually succeeded, so the platform's own health readout treated its recovery machinery as an outage: over the past week, 13 of the 14 lost jobs were rescued within seconds and all 14 still counted against generation health. Each recovered attempt is now linked to the run that replaced it, so the status page and our failure reports show what you experienced instead of what the machinery did.",
"tags": [
"improvement",
"infra"
],
"link": "/status"
},
{
"date": "2026-08-14",
"title": "Cookbook posters came back, and a docs link you copy now reopens the page you were reading",
"summary": "Eight developer pages were read against the code they describe and exercised in a real browser: the MCP 3D tool and CI asset gate recipes, and the payment sessions, developer platform, media API, agent-forged gallery, generation watch, and KOL tracker guides. Two things were broken for readers. Every cookbook recipe that shows a poster image was rendering without it: the figure system accepted three kinds of illustration and silently deleted the fourth, which is the kind the recipe posters use, with nothing in the page or the console to say a picture had gone missing. All of them render again. Separately, the previous and next buttons and the sidebar on a documentation page produced a link that only worked while you stayed on the page: copy it, share it, or reload it, and you landed back on the page you started from instead of the one the link named. Those links now open the page they promise, correct the address bar to that page's own permanent link, and rename the browser tab to match, while in-page section links still jump to the section as before. The pages themselves were corrected where they had drifted from the code: the gallery's sample receipt quoted an old price for the tier it describes, the media API's two library samples reported counts no live response would return, and the KOL tracker described all four of its endpoints as sharing one dispatcher when the trade feed is served by its own file. Every command and code sample printed on the eight pages was run, and every link on them resolves.",
"tags": [
"fix",
"docs"
]
},
{
"date": "2026-08-14",
"title": "Anyone can run an inference node: the operator client now ships CPU and GPU images, and proves itself end to end",
"summary": "The open inference network is the part of the roadmap where the compute stops being ours. The node operator client is how you join it: your machine generates a Solana keypair, registers itself with the platform, claims inference jobs, runs them locally with a real open model, and returns every answer with a signature that binds it to your key. There is no shared password anywhere in that flow, so nothing can be leaked and reused to impersonate your node, and the platform re-derives each signature from its own record of the job before it accepts the work. This release makes the graphics-card story honest. The old image claimed it would use an NVIDIA card automatically when one was present, and it never could: the runtime it shipped had no GPU support compiled in at all. There are now two real images, one for CPU and one for NVIDIA cards, and the client tells you which device actually ran each job. Ask for the GPU on a machine that cannot reach one and it stops immediately with the reason, instead of quietly earning at CPU speed for a month. Three faults in the job pipeline were also fixed: results submitted by a node hit a server error every time, registered nodes read back as unregistered, and every finished job leaked two silent background errors. A single command (npm run e2e) now registers a node, sends it a job, runs the model, and verifies the signed answer against the real platform code, and it fails loudly if any step regresses.",
"tags": [
"feature",
"fix",
"infra"
],
"link": "/docs/inference-node-operator"
},
{
"date": "2026-08-14",
"title": "The legal pages now say what is in them, and call the terms by their real name",
"summary": "The six pages under three.ws/legal (the index, privacy policy, terms of service, end-user license agreement, risk disclosure, and content policy) were audited in a real browser at phone and desktop widths. Every page loads, every link on them resolves, and the layout holds at 320 pixels wide. Three things were wrong. The terms of service was filed under the name Terms of Use in the site's page index, so a search result or a shared link showed one name while the document itself showed another; it is now Terms of Service everywhere. The privacy policy, the terms, and the EULA each carried a search-result description that only restated their title, so a search engine or a chat preview told you nothing about what was inside; each now describes what the document actually covers. And several of the policies linked to each other by their raw file names rather than their real addresses, which split the same document across two web addresses; they now all use the address shown in the browser bar. The content policy was also missing its publication date in the page index, which is what the changelog and sitemap read.",
"tags": [
"fix",
"docs"
],
"link": "/legal"
},
{
"date": "2026-08-14",
"title": "The security headers every three.ws page serves are now checked against the live site, not just the config",
"summary": "Every page three.ws serves carries a set of browser security rules: which scripts are allowed to run, that the connection must stay encrypted, that browsers must not second-guess a file's type, how much of the current address is passed on when you follow a link, and who is allowed to put a three.ws page inside a frame. Those rules are written in one config file, but until now nothing checked that the site actually answered with them. The sweep that loads real pages in a real browser now also compares each response against that config, and it can be pointed at the live site to check the whole catalogue in about a minute. Running it against production found no gap: all 25 of the most script-heavy pages, and the rest of the catalogue with them, serve exactly the rules the config declares. Two fixes came out of building it. Files that are not web pages, like robots.txt and the machine-readable service descriptions, were being reported as wrong when they were correct. And a page whose file changed while the server was running could be sent with the script rules of the previous version, which would have blocked its own scripts and shown a blank page; the server now notices the change and recomputes them.",
"tags": [
"security",
"infra"
]
},
{
"date": "2026-08-14",
"title": "The 3D viewer no longer starts up blank when its box has not been measured yet",
"summary": "Every page that shows a 3D avatar builds its viewer the moment the page's script runs. On pages where the area holding the avatar has not been given its size yet, because the layout lands a frame later or the panel starts hidden, that meant the viewer was built at zero pixels wide and zero pixels tall. Graphics drivers reject an image with no width or height, so the viewer came up against a surface the driver had refused to create: the browser console filled with rejected-texture and incomplete-surface errors, and on the mocap studio the avatar stage was the one that hit it. The viewer now starts at a minimum size and resizes itself to the real box on the first frame that has one, which it was already set up to do for every later resize. Nothing changes for a page whose avatar area is measured up front, which is most of them.",
"tags": [
"fix"
],
"link": "/mocap-studio"
},
{
"date": "2026-08-14",
"title": "One cookbook recipe stopped introducing itself as a generic page in search results and shared links",
"summary": "All five cookbook recipes are drawn by one shared page template, and the server fills in each recipe's own title, description, and social card as the page is served. One recipe was skipping that step. The template carried a leftover marker naming a specific recipe as its own address, and the server reads that marker as this page already knows who it is, so it left that one page alone: text to 3D from the command line went out with the placeholder title every recipe shares and a one-line description that said nothing about it. Search results, and the preview card in any chat where the link was pasted, showed the placeholder instead of the recipe. The template now names the cookbook index rather than any single recipe, matching how the tutorial and walkthrough templates have always been set up, so all five recipes present themselves properly. A test now walks the real page templates and fails if any of them ever claims one of its own recipes again, which is what let this sit unnoticed.",
"tags": [
"fix"
],
"link": "/cookbook/text-to-3d-cli"
},
{
"date": "2026-08-14",
"title": "The Forge gallery no longer goes quiet for hours after a model provider blips",
"summary": "A job runs every minute to keep new models flowing into the Forge gallery, and it steps back when the upstream generator is failing so it is not hammering a dead lane. That step-back had no ceiling: each further failed attempt lengthened the pause, so a provider outage that lasted through a few retries could park the job for hours, and by the time the provider recovered nothing was left running to notice. The pause is now capped at an hour, so the job keeps checking back on its own and the gallery starts filling again as soon as generation works.",
"tags": [
"fix",
"infra"
],
"link": "/forge"
},
{
"date": "2026-08-14",
"title": "The 3D generation SDK can finally pay per call, and its bring-your-own-key option actually does something",
"summary": "The npm package that turns a prompt into a 3D model documented two things it did not do. It listed a per-call option for supplying your own Meshy or Tripo key, and that option was read from nowhere: the key was silently dropped, so anyone following the README's own example was told to add a key that the generator then said was missing. It also documented paying per call in USDC as the default billing lane, when the flag it described was ignored by the endpoint it was sent to. Both are now real. A key passed on the call is used, overrides the one set on the client, and is resent on every status check, which the generator requires and the old code never did, so a job started with your own key could never be polled to completion. The pay-per-call lane now routes to the endpoint that actually accepts payment, quotes Solana USDC, and hands back a job you poll for free, and it refuses upfront the options it cannot honour rather than accepting them and ignoring them. The package now also exposes a way to read one job's state without waiting for it, so a generation survives a restart. Alongside that, the reference pages listed a price and a set of engines that had both moved on, named an error the service never returns, and pointed at three addresses that no longer exist: the sign-up link for API keys in the copy-trading and LLM router packages went to a dead page, as did the link to the 3D studio tool server. All of them now resolve.",
"tags": [
"sdk",
"fix",
"docs"
]
},
{
"date": "2026-08-14",
"title": "The AR docs stopped handing readers three commands the tool rejects or quietly ruins",
"summary": "Eight pages were read line by line against the code and the live site: the agent abilities dossier, news, the changelog, AR viewing, BNB payments, Agent Sniper, earned autonomy, and the agent personality tutorial. Every command printed on them was actually run. The AR page was wrong three times, and each one cost a reader real time. Its model optimization step used a flag the glTF tool does not accept, so the command simply failed. The line beside it promised a lossless result, when that tool compresses and simplifies by default: the file it produced could not be opened by the iPhone and Android AR viewers the page is about, and nothing said so. Its troubleshooting step for exactly that failure repeated the same rejected flag, so the fix for the problem hit the problem again. All three now carry flags the tool accepts, verified by running them on a real model and reading back what came out, with a note about what the defaults do. The AR page also linked to a placeholder address rather than a page, and now points at a real one. On the earned autonomy page, the list of settings a trading arm unlocks once it has proven itself was one short, missing the size of the position it keeps riding after taking its stake back. Everything else held up, including all 4,854 links across the eight pages.",
"tags": [
"docs",
"fix"
],
"link": "/docs/ar"
},
{
"date": "2026-08-14",
"title": "Links inside the tutorials now go where they say, and the AWS page names the right region",
"summary": "Every tutorial on the site is written as a markdown file that lives beside the other documentation, and the links inside it were written against that location. The page that renders them was reading those links as if they pointed at site addresses instead, so a tutorial that sent you to a related guide, or cited the exact source file a step describes, sent you to an address that does not exist. Links that reach across to another document, that point at a specific section of one, or that cite a file in the source tree are now each resolved to the right destination, with source files opening on GitHub. The scene tutorial's own list of related tutorials pointed at a secondary address that rendered them under a generic title, and now uses the real one. Separately, the AWS page claimed the platform's production deployment runs in AWS us-east-1. It does not: production runs on Google Cloud Run, and us-east-1 is where the AWS Marketplace billing and entitlement APIs are called. The page now says that instead.",
"tags": [
"docs",
"fix"
]
},
{
"date": "2026-08-14",
"title": "The commit feed went quiet for thirteen hours, and the reason it could not restart itself is fixed",
"summary": "Every commit that lands on the platform is posted to the holders' Telegram channel as it happens. That feed stopped at 06:45 UTC and stayed silent for the rest of the day, and the cause was in how it found its place again after each run. It looked back over only the thirty most recent commits for the last one it had announced. On a day when work lands in bursts of twenty or thirty at a time, that marker regularly fell out of view, and when it did, the feed quietly skipped to the newest commit and posted nothing at all. Every commit in the gap was passed over with no record that anything had been missed, and because each run repeated the same skip, the feed had no way back on its own. It now pages back as far as five hundred commits to find its place, and when it is behind it works forward from the oldest commit it still owes rather than jumping to the front, so a backlog is delivered in order over the following runs instead of being discarded. Anything more than three days old is still left alone, so catching up can never flood the channel with history. The 117 commits missed during today's outage are not recoverable and will not appear.",
"tags": [
"fix"
]
},
{
"date": "2026-08-14",
"title": "A dedicated channel for first-time pump.fun creator fee claims",
"summary": "The first time a coin creator pulls their accrued fees out of the vault is an irreversible on-chain signal that they are actually engaged with what they launched, and it was only visible by polling an API. There is now a delivery lane that posts each of those first claims to its own Telegram channel as it lands, one message per claim, with the creator, the amount in SOL, and links to the transaction and the coin. It keeps its own state so switching it on never dumps a backlog into the channel, it drains a backlog in order rather than skipping ahead if it falls behind, and it recognises claims by transaction rather than by timestamp, so two claims landing in the same second both get posted instead of one being silently dropped. The feed stays quiet until a claims indexer is configured: the direct on-chain fallback samples under a second of activity on a program this busy, which is not enough to tell a first claim from a repeat one.",
"tags": [
"feature",
"infra"
]
},
{
"date": "2026-08-14",
"title": "The choreography docs no longer teach an example that quietly does nothing",
"summary": "A routine step can pin one specific animation clip instead of letting the gesture slot pick one, and the choreography reference showed how, with a shareable link that pinned the dance slot to the rumba clip. That example never pinned anything. Rumba is already the platform default for the dance slot, and the runtime deliberately drops a clip that matches the slot's own default, so that a shared routine still adapts to an agent that has remapped its gestures. Anyone copying the example got a step that resolved normally instead of the pinned motion they asked for, with nothing to indicate why. The example now pins a clip that is not the default, and the field reference states the drop rule outright. The selfie-to-avatar page also had a summary line quoting an older, lower head-coverage number than the measurement reported further down the same page, and the avatar-thumbnails page printed a sample coverage readout from a time when three quarters of avatars still had no preview image. Both now show figures taken from the running system.",
"tags": [
"docs",
"fix"
],
"link": "/docs/choreography"
},
{
"date": "2026-08-14",
"title": "The marketplace, the directory and the home page open with a lot less work behind them",
"summary": "A Lighthouse sweep of eight pages against the live site found four costs that every visitor was paying for content they could not see. The marketplace was laying out its entire catalogue on first paint, a grid measured at over 31,000 pixels tall of which one screenful is visible, and the directory did the same at smaller scale; both now skip the offscreen rows until you scroll near them. The analytics library was loading in the page head on every route, which cost around a second of browser time before anything was interactive, and now waits for the first idle moment instead, with nothing dropped in between. The embeddable avatar bundle on our CDN was shipping unminified because of a build-tool rule meant for packages that get re-bundled, which does not apply to a file browsers download directly; it is now 2.77 MB instead of 4.24 MB, and 792 KB instead of 1.03 MB over the wire. And the thumbnails in both grids now decode off the critical path. The rules behind all four, and how to measure before changing any of them, are written up so the next change does not quietly undo one.",
"tags": [
"improvement",
"infra"
],
"link": "/marketplace"
},
{
"date": "2026-08-14",
"title": "Three pieces of docs text that were too faint to read comfortably are now legible",
"summary": "An accessibility sweep of the documentation found three places where text sat below the readable-contrast standard against the surface behind it: the comment lines inside every code sample, which is where most of our samples put the explanation, the See what changed link on the page-freshness banner, and the Live label on runnable code blocks. All three came from the same habit of dimming text that was already dark, and all three are now above the standard in both the dark and light themes. The status page KPI labels got the same treatment. Nothing moved and no wording changed, the affected text is simply readable now.",
"tags": [
"improvement",
"docs"
],
"link": "/docs/custody"
},
{
"date": "2026-08-14",
"title": "The example responses in four API guides now match what the API actually returns",
"summary": "The docs for ghost-copy, Trader Passport, the wallet portfolio, and the airdrop checker each show a sample JSON response so you can see the shape before you write any code. Those samples had fallen behind the endpoints: they were missing whole fields that the live API returns, including the timestamp on an airdrop scan, the window start and per-trade detail on a ghost-copy replay, and the issuer note, verification instructions, and generation time on a passport. Anyone building against the samples would have concluded those fields did not exist. Every sample was re-checked field by field against a live call to the running endpoint and corrected, and the places where a sample is deliberately shortened are now marked as shortened instead of looking complete.",
"tags": [
"docs"
],
"link": "/docs/trader-passport"
},
{
"date": "2026-08-14",
"title": "Your agent's memory graph was always empty. It is not any more",
"summary": "Every memory your agent stores is scanned for the things it mentions: tokens, tickers, wallets, people, and the trading rules you told it to follow. Those become the nodes of a knowledge graph you can browse, so you can ask what your agent remembers about one specific token and read back every memory that touched it. The scan ran correctly and then failed at the last step, when it tried to save what it found, because of a type mismatch on one database column. The failure was recorded quietly rather than shown, so nothing ever looked broken: memories saved fine, search worked fine, and the graph simply came back empty for every agent, forever. The save now works, memories that were skipped get scanned the next time you open the graph, and a test covers the exact shape that broke.",
"tags": [
"fix"
],
"link": "/docs/memory"
},
{
"date": "2026-08-14",
"title": "A failed database migration can no longer freeze the schema in silence",
"summary": "Every deploy carries database migrations that must run before the new code does. When one of them failed, the runner printed a single line with the database's own error and stopped, naming neither the file that failed nor the migrations queued behind it. Those stayed unapplied and unmentioned, so the site ended up serving code that reads columns the database never got, and the first visible symptom was an unrelated-looking error on a live page. The runner now names the failing file, how many ran before it, and every migration still waiting, and the deploy step that catches a database running behind the code is now part of the documented deploy path instead of something you had to remember to run.",
"tags": [
"infra",
"fix"
]
},
{
"date": "2026-08-14",
"title": "Your agent's memory graph was never being built, and now it is",
"summary": "Memory Studio promises a knowledge graph: every fact your agent stores is scanned for the mints, tickers, wallets, people, and strategies it mentions, and those become nodes you can click through to ask what your agent remembers about any one of them. That scan was failing on a type mismatch the moment it tried to save a node, and because the failure was caught quietly, nothing ever looked broken: the graph simply came back empty forever, for every agent, and the memories were never marked as scanned. The save is fixed, so the graph now fills in on your next visit, including for everything your agents have stored up to now. Recall and search were never affected.",
"tags": [
"fix"
],
"link": "/memory"
},
{
"date": "2026-08-14",
"title": "AgenC task lists no longer hang when a Solana provider goes quiet",
"summary": "Browsing tasks on the AgenC coordination protocol reads directly from Solana, and the platform keeps a chain of providers so a bad one gets skipped. That skipping only worked when a provider answered with an error. A provider that accepted the request and then went silent held the page open for minutes with nothing to react to, which is what happened to the task list, the heaviest read of the set. Each provider now gets a fixed time to answer before the next one is tried, so a silent provider costs seconds instead of stalling the page.",
"tags": [
"fix"
],
"link": "/agenc"
},
{
"date": "2026-08-14",
"title": "Agent-runtime and agent-guards docs now match what the code actually returns",
"summary": "Two developer packages shipped quick-start examples that described results the code does not produce: the runtime's guard example annotated the wrong decision and named a field on the ledger checker that does not exist, and the guards package documented only its hosted half while the local, no-network guard functions went unmentioned. Both are corrected against real executed output, with the local guards now documented end to end including a runnable example and the exact reason codes they return.",
"tags": [
"docs",
"sdk"
]
},
{
"date": "2026-08-14",
"title": "Agents that discover us the documented way can now reach the wallet tools",
"summary": "An AI client that finds three.ws on its own follows a trail: it calls a tool, gets told it needs to sign in, and reads a small public document that lists the permissions it is allowed to ask for. That document had fallen behind. It advertised eight permissions while the sign-up endpoint actually grants twelve, and the three it left out were exactly the ones that unlock the agent wallet tools, so a client doing everything correctly could never ask to check a balance, pay for a service, provision a wallet, or put an endpoint up for sale. The published list and the list the server honors are now read from one place, with a test that fails the build if they ever drift apart again. Separately, the paid-call example an agent receives with the payment request pointed at a sample model file on a domain that does not exist, so the very first call a paying agent copied failed on a dead address; it now points at a real model served from three.ws. Every discovery endpoint was re-checked end to end afterwards.",
"tags": [
"fix",
"security"
],
"link": "/docs/mcp-agent"
},
{
"date": "2026-08-14",
"title": "The activity floor stops reporting a flat week when nothing has closed yet",
"summary": "The heaFile truncated at 1 MB.