Skip to content

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.

feat(irl): talk to a discovered agent out loud and view any pin in AR, with an e2e that walks the sheet

nirholasaac77aa8042f
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 4a0ef892e..f78c6a1bd 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -15,11 +15,13 @@ Public history for [three.ws](https://three.ws), newest first. New pages come fr
- **3D models now open, animate and report on themselves inside VS Code**: 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. (`/docs/vscode`) `[feature, sdk]`
- **A new agent skill teaches Claude to look for an existing 3D asset before making one**: 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. (`/docs/agent-skills`) `[feature, docs]`
- **AI coding assistants can now search the three.ws asset library and get working code back**: 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. (`/docs/mcp`) `[feature, sdk]`
-- **Invite people to an IRL agent with a visit link and a printable sign**: 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 See it 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 tip panel; other phones open the AR launcher. 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. (`/irl`) `[feature, improvement]`
+- **Invite people to an IRL agent with a visit link and a printable sign**: 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. (`/irl`) `[feature, improvement]`
+- **IRL no longer fails to load your agent on a cold start**: 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. (`/irl`) `[fix]`
- **Machine Atlas: two working machines built in the browser from code, with no 3D model files**: 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. (`/assembly`) `[feature]`
- **One command drops a three.ws prop, character or motion clip into your project**: 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. (`/docs/api-reference`) `[feature, sdk]`
- **Seven background jobs that were never running now run**: 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. `[fix, infra]`
- **The site header now stays put when you scroll**: 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. `[fix, improvement]`
+- **Walk up to an IRL agent, talk to it out loud, and stand it on your floor in AR**: 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. (`/irl`) `[feature]`
- **Your agent now lives in a handheld console you can put on any website**: 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. (`/pocket`) `[feature]`
## 2026-09-04
diff --git a/STRUCTURE.md b/STRUCTURE.md
index 5eb5420d8..f5ea821f8 100644
--- a/STRUCTURE.md
+++ b/STRUCTURE.md
@@ -98,7 +98,7 @@ bottom.
| Daily Forge (daily challenge + creator streak) | [pages/daily.html](pages/daily.html) `→ /daily` + [src/daily.js](src/daily.js), deterministic theme rotation [src/daily/daily-theme.js](src/daily/daily-theme.js), streak state machine [src/daily/creator-streak.js](src/daily/creator-streak.js) | Live | The **anonymous-first retention loop** the platform lacked: the logged-in streak system ([api/_lib/streaks.js](api/_lib/streaks.js)) needs an account, so the no-sign-up forge majority had no reason to return. Every UTC day resolves to one curated creative theme (the same for everyone, hashed from the date — no backend, no randomness) with a running "Day N" counter; forging today's theme on the free lane lights a localStorage streak (consecutive-day state machine: extend / reset-on-gap / never-double-count, `best` preserved, at-risk vs lapsed display), and milestones (3/7/14/30/50/100/365) fire a celebration. The generation lands in the shared recents so it's one tap from `/ar` and `/ar/studio`; a "Fresh from the community" strip reads `/api/forge-gallery?scope=community`. `changefreq: daily` in the sitemap. Tests: [tests/daily-theme.test.js](tests/daily-theme.test.js), [tests/creator-streak.test.js](tests/creator-streak.test.js). |
| AR Studio (multi-model live camera scene) | [pages/ar-studio.html](pages/ar-studio.html) `→ /ar/studio` + [src/ar-studio.js](src/ar-studio.js), multi-anchor WebXR session [src/ar/multi-place.js](src/ar/multi-place.js), pure scene math [src/ar/studio-scene.js](src/ar/studio-scene.js) | Live | The standalone multi-model AR surface: place ANY number of models/avatars into one live camera view (every earlier AR lane was single-model). Model sources: in-view forge (`POST /api/forge {backend:"nvidia"}` runs behind the live camera, result drops into the room), the forge gallery (`/api/forge-gallery`, yours + community tabs), recent forges (localStorage, shared with `/ar`), the CC0 Object Library (`/api/objects/library`, an Objects tray tab with client-side name/category search), a pasted GLB link, or repeatable `?src=` deep links (https/relative only). Rendering ladder: WebXR immersive-ar with one XRAnchor per placed model + always-armed reticle (Android), getUserMedia passthrough + gyro world-lock (iOS/everywhere), grid preview + drag-look + QR handoff (desktop). Gestures: tap select, drag move, pinch resize, two-finger twist rotate; per-model duplicate/rotate/remove, clear-with-undo. **Real device AR, not just the camera composite:** the AR control resolves what the device can actually do and routes there rather than defaulting every non-WebXR device to the composite: the immersive WebXR session where it exists, Apple Quick Look on iOS and Google Scene Viewer on Android otherwise (both selectable per-model from the selection bar too), and an honest "open this on a phone" QR prompt everywhere else. The button relabels itself to match ("Immersive" only where an immersive session is real). Quick Look reads USDZ, so the selected model's GLB is converted on the device through the same `usdz-pipeline.js` bake `/avatars/:id/ar` uses, not the camera-passthrough composite, which has no real-world plane detection or scale and previously shipped as the only iOS experience. Humanoids with no baked clips idle via the universal retarget ([src/irl/pin-idle.js](src/irl/pin-idle.js)); scene persists in localStorage and the whole ARRANGEMENT (models + transforms) round-trips through a base64url `#s=` hash on the share/QR URL, so a composed scene reopens exactly on any device. **Presence realism:** WebXR placements are lit by the real room via lighting-estimation ([src/ar/estimated-lighting.js](src/ar/estimated-lighting.js), three's `XREstimatedLight`, real primary-light direction/colour + a live environment cube map so metal/glass reflect the actual room, reverting to baked lights when the device isn't estimating) and occluded by real geometry via depth-sensing ([src/ar/depth-occlusion.js](src/ar/depth-occlusion.js)); the passthrough path matches room brightness AND colour by sampling the camera feed. Desktop keyboard controls (arrows nudge / R rotate / D duplicate / Delete). Photo capture composites camera + WebGL ([src/irl/share-frame.js](src/irl/share-frame.js)). Reachable from `/ar`, the forge AR panel, and the `/viewer` action sheet. **Shared rooms (build together, live):** "Share live" opens a room with a 6-char code; anyone who opens `/ar/studio?room=CODE` joins the same scene and every add / move / scale / rotate / remove delta-syncs in real time, with a live "N here" presence pill. New Colyseus room [multiplayer/src/rooms/StudioRoom.js](multiplayer/src/rooms/StudioRoom.js) + schema [multiplayer/src/studio-schemas.js](multiplayer/src/studio-schemas.js) (owner-gated edits, per-room/per-owner caps, per-client rate limit, https-only src validation, the same discipline as WalkRoom's object channel), client wrapper [src/ar/studio-net.js](src/ar/studio-net.js), and a shared logical coordinate frame [src/ar/studio-coords.js](src/ar/studio-coords.js) (relEast/relNorth/yawDeg) so every device agrees on placement; QR-marker colocalization ([src/irl/marker-anchor.js](src/irl/marker-anchor.js)) is the co-located "same physical table" upgrade. Runs on the same Colyseus host as /walk and /irl (a new room, not a new process). Tests: [tests/ar-studio-scene.test.js](tests/ar-studio-scene.test.js), [tests/ar-studio-coords.test.js](tests/ar-studio-coords.test.js), [tests/ar-estimated-lighting.test.js](tests/ar-estimated-lighting.test.js), [tests/studio-room.test.js](tests/studio-room.test.js). **Shipped as open source:** the same surface is extracted, generalized and published as [3d-ar-studio](https://www.npmjs.com/package/3d-ar-studio) (repo [nirholas/3D-AR-Studio](https://github.com/nirholas/3D-AR-Studio), demo https://nirholas.github.io/3D-AR-Studio/, MCP server `io.github.nirholas/3d-ar-studio`), so any site can embed a multi-model AR studio with one script tag. It defaults to OUR catalogue and generation lane (`/cdn/objects/library/manifest.json`, `/api/mcp-studio`, `/animations/manifest.json`, `/hdri/*`, `/api/ar`), which is why those surfaces answer open CORS. |
| Portal (any website as a walkable world) | Page [pages/portal.html](pages/portal.html) `→ /portal` + controller [src/portal/main.js](src/portal/main.js), renderer [src/portal/render.js](src/portal/render.js), styles [src/portal/portal.css](src/portal/portal.css); reader [api/_lib/portal/outline.js](api/_lib/portal/outline.js) + guarded fetch [api/_lib/portal/fetch-site.js](api/_lib/portal/fetch-site.js) + robots matcher [api/_lib/portal/robots.js](api/_lib/portal/robots.js); pure layout [packages/portal/src/layout.js](packages/portal/src/layout.js); GLB export [api/_lib/portal/world-glb.js](api/_lib/portal/world-glb.js); endpoint [api/portal.js](api/portal.js) `→ /api/portal` | Live | Paste an address and walk the page: each section becomes a building sized by its word count (log scale), each link a door (same-site doors rebuild the world there and push browser history, outside links open the page), each image a billboard loaded through `/api/img`, each code block a monolith, and the palette comes from the page's own `theme-color` or a hash of its host. The layout is a pure, seeded function so the browser, the GLB exporter, the MCP tools and the npm package all build the identical city from one link. Crawler manners are part of the product: a declared user agent, `robots.txt` honoured per RFC 9309, one fetch per page cached fleet-wide for an hour, and an IP-pinned SSRF guard with size and time ceilings. `format=glb` returns the world as glTF with named nodes. SDK + CLI + MCP server: [packages/portal/](packages/portal). Docs: [docs/portal.md](docs/portal.md). Spec: [specs/PORTAL_WORLD.md](specs/PORTAL_WORLD.md). Tests: [tests/portal-robots.test.js](tests/portal-robots.test.js), [tests/portal-outline.test.js](tests/portal-outline.test.js), [tests/portal-layout.test.js](tests/portal-layout.test.js), [tests/portal-endpoint.test.js](tests/portal-endpoint.test.js), [tests/e2e/portal.spec.js](tests/e2e/portal.spec.js) |
-| IRL (agents in the real world) | [pages/irl.html](pages/irl.html) `→ /irl` + [src/irl.js](src/irl.js), WebXR placement [src/ar/webxr.js](src/ar/webxr.js) + pinch [src/ar/pinch-scale.js](src/ar/pinch-scale.js), animated Quick Look bake [src/usdz-animated.js](src/usdz-animated.js) (static fallback [src/usdz-pipeline.js](src/usdz-pipeline.js)), backend [api/irl/](api/irl) | Live | Pin an agent at a real GPS spot; discovery is presence-proof-gated ([docs/irl/THREAT-MODEL.md](docs/irl/THREAT-MODEL.md)). **Place in AR** leads the dock on AR-capable devices: WebXR hit-test anchoring on Android with two-finger pinch resize (25%..400%) persisted to the pin (`anchor_scale`; the pinch lands as a follow-up `PATCH /api/irl/pins {id, scale}` after the placement tap), and ARKit Quick Look on iOS with the idle clip baked into the USDZ so the agent moves instead of standing in a T-pose. Discovered pins are equally alive in the in-page camera view: every mounted humanoid GLB plays the retargeted idle clip via a per-pin mixer ([src/irl/pin-idle.js](src/irl/pin-idle.js), phase-randomized, LOD-bounded), camera-aware gaze composes on top of the running animation, and the scene renders with ACES tone mapping + IBL + soft shadows (e2e: [tests/e2e/irl-pin-idle.spec.js](tests/e2e/irl-pin-idle.spec.js)). Inbound `/irl?avatar=<glb>` is the living-agent handoff from `GET /api/ar?…&kind=avatar` (explicit `?avatar=` wins over a saved session). Money Drops and World Lines: [docs/irl.md](docs/irl.md). SDK: [packages/irl/](packages/irl). **Share** ([src/irl/share-frame.js](src/irl/share-frame.js), [api/irl/share.js](api/irl/share.js), [api/irl/share/[token].js](api/irl/share/%5Btoken%5D.js)) mints a permanent, unfurlable `/irl/s/:token` link from the captured AR photo for the pin you just placed - never a coordinate, owner-gated, falls back to the plain native share sheet with no active pin. **Analytics** ([api/_lib/irl-analytics.js](api/_lib/irl-analytics.js), [api/irl/analytics.js](api/irl/analytics.js)) is the first site-wide usage rollup for the feature - pins placed, nearby reads, placement method, shares, drop claims - aged out by the existing hourly reaper. **Visit links and signs** ([src/irl/visit-link.js](src/irl/visit-link.js) pure rules, [src/irl/visit-banner.js](src/irl/visit-banner.js) the "you're here to meet" banner, printable sign [pages/irl-sign.html](pages/irl-sign.html) `→ /irl/sign?pin=<id>` + [src/irl-sign.js](src/irl-sign.js), dashboard modal in [src/dashboard-next/pages/irl-placements.js](src/dashboard-next/pages/irl-placements.js)): `/irl?pin=<id>` is the scannable handoff for a street demo. The page names the agent from the coordinate-free agent card until the presence-gated nearby read returns that pin, then flashes it and opens its card once. The discovered-pin card gains **See it in AR** (iOS: the pin's GLB baked to an animated USDZ for Quick Look, banner tap returns to the tip panel; elsewhere the `/api/ar` launcher). Runbook: [docs/irl/street-demo.md](docs/irl/street-demo.md). Tests: [tests/irl-visit-link.test.js](tests/irl-visit-link.test.js). |
+| IRL (agents in the real world) | [pages/irl.html](pages/irl.html) `→ /irl` + [src/irl.js](src/irl.js), WebXR placement [src/ar/webxr.js](src/ar/webxr.js) + pinch [src/ar/pinch-scale.js](src/ar/pinch-scale.js), animated Quick Look bake [src/usdz-animated.js](src/usdz-animated.js) (static fallback [src/usdz-pipeline.js](src/usdz-pipeline.js)), backend [api/irl/](api/irl) | Live | Pin an agent at a real GPS spot; discovery is presence-proof-gated ([docs/irl/THREAT-MODEL.md](docs/irl/THREAT-MODEL.md)). **Place in AR** leads the dock on AR-capable devices: WebXR hit-test anchoring on Android with two-finger pinch resize (25%..400%) persisted to the pin (`anchor_scale`; the pinch lands as a follow-up `PATCH /api/irl/pins {id, scale}` after the placement tap), and ARKit Quick Look on iOS with the idle clip baked into the USDZ so the agent moves instead of standing in a T-pose. Discovered pins are equally alive in the in-page camera view: every mounted humanoid GLB plays the retargeted idle clip via a per-pin mixer ([src/irl/pin-idle.js](src/irl/pin-idle.js), phase-randomized, LOD-bounded), camera-aware gaze composes on top of the running animation, and the scene renders with ACES tone mapping + IBL + soft shadows (e2e: [tests/e2e/irl-pin-idle.spec.js](tests/e2e/irl-pin-idle.spec.js)). Inbound `/irl?avatar=<glb>` is the living-agent handoff from `GET /api/ar?…&kind=avatar` (explicit `?avatar=` wins over a saved session). Money Drops and World Lines: [docs/irl.md](docs/irl.md). SDK: [packages/irl/](packages/irl). **Share** ([src/irl/share-frame.js](src/irl/share-frame.js), [api/irl/share.js](api/irl/share.js), [api/irl/share/[token].js](api/irl/share/%5Btoken%5D.js)) mints a permanent, unfurlable `/irl/s/:token` link from the captured AR photo for the pin you just placed - never a coordinate, owner-gated, falls back to the plain native share sheet with no active pin. **Analytics** ([api/_lib/irl-analytics.js](api/_lib/irl-analytics.js), [api/irl/analytics.js](api/irl/analytics.js)) is the first site-wide usage rollup for the feature - pins placed, nearby reads, placement method, shares, drop claims - aged out by the existing hourly reaper. **Visit links and signs** ([src/irl/visit-link.js](src/irl/visit-link.js) pure rules, [src/irl/visit-banner.js](src/irl/visit-banner.js) the "you're here to meet" banner, printable sign [pages/irl-sign.html](pages/irl-sign.html) `→ /irl/sign?pin=<id>` + [src/irl-sign.js](src/irl-sign.js), dashboard modal in [src/dashboard-next/pages/irl-placements.js](src/dashboard-next/pages/irl-placements.js)): `/irl?pin=<id>` is the scannable handoff for a street demo. The page names the agent from the coordinate-free agent card until the presence-gated nearby read returns that pin, then flashes it and opens its card once. The discovered-pin card gains **View in AR** for ANY discovered pin, capability-routed by [src/irl/pin-ar.js](src/irl/pin-ar.js): iOS bakes the pin's GLB to an animated USDZ for Quick Look (the banner reads "Talk to <name>" and its tap reopens the card with the conversation live), Android with ARCore enters the same WebXR hit-test session as placement in a view-only mode (`enterFloorAnchor({ viewPin })`: the pin's mounted model is the session content, nothing persists, the group's transform is restored on exit), everything else opens the `/api/ar` launcher in a new tab. **Talk** ([src/irl/pin-talk.js](src/irl/pin-talk.js)) is a live spoken conversation with the agent standing in front of you, inside the pin sheet: the same `TalkController` as talk mode on `/avatars/:id` (mic or typed turn, `/api/chat` in the agent's persona by `agentId`, TTS in its voice, lipsync from the audio FFT onto the pin's mounted model via `AvatarMouthTarget`), barge-in on the mic, and a `talk` interaction logged once per conversation (never the transcript). While a conversation or a view-only AR session is open the pin's full model is pinned resident in `enforceLOD` (`_pinnedFull`). Runbook: [docs/irl/street-demo.md](docs/irl/street-demo.md). Tests: [tests/irl-visit-link.test.js](tests/irl-visit-link.test.js), [tests/irl-pin-ar.test.js](tests/irl-pin-ar.test.js), [tests/irl-pin-talk.test.js](tests/irl-pin-talk.test.js), e2e [tests/e2e/irl-pin-talk.spec.js](tests/e2e/irl-pin-talk.spec.js). |
| Verifiable 3D provenance (content credentials) | Spec [specs/PROVENANCE_3D.md](specs/PROVENANCE_3D.md), pure core [api/_lib/provenance-3d.js](api/_lib/provenance-3d.js), Solana anchor [api/_lib/provenance-anchor.js](api/_lib/provenance-anchor.js), MCP tools [api/_mcp3d/tools/provenance.js](api/_mcp3d/tools/provenance.js) (`verify_provenance` free, `anchor_provenance` paid $0.05), free REST verify [api/provenance.js](api/provenance.js) `→ /api/provenance`, viewer badge [public/viewer.html](public/viewer.html), docs [docs/provenance.md](docs/provenance.md). | Live (verify) · anchor gated on issuer key + devnet funds | C2PA-style signed content credentials for AI 3D: creator, prompt, model/provider, lineage, timestamp, and the GLB's sha256, ed25519-signed by the three.ws issuer with the credential hash anchored on Solana. Free public `verify_provenance` / `GET /api/provenance?src=…` recompute the hash and return `verified | tampered | unknown` (a single changed byte → tampered); the verify path and response are coin-clean (both stores). `anchor_provenance` signs + writes the SPL Memo anchor + stores the credential in R2. Tests: [tests/provenance-3d.test.js](tests/provenance-3d.test.js). Evidence: [prompts/store-submissions/_generated/provenance/](prompts/store-submissions/_generated/provenance/). Store-submission prompt 19. |
| Simulation readiness (a physics grade for 3D assets) | Spec [specs/SIM_READINESS.md](specs/SIM_READINESS.md), pure core [api/_lib/sim-readiness.js](api/_lib/sim-readiness.js), storage [api/_lib/sim-readiness-store.js](api/_lib/sim-readiness-store.js) + migration [20260813180000_sim_readiness_grades.sql](api/_lib/migrations/20260813180000_sim_readiness_grades.sql), free MCP tool [api/_mcp3d/tools/sim-readiness.js](api/_mcp3d/tools/sim-readiness.js) (`grade_sim_readiness`), free REST [api/sim-readiness.js](api/sim-readiness.js) `→ /api/sim-readiness`, shared badge + details panel [public/sim-readiness-panel.js](public/sim-readiness-panel.js) (`<sim-readiness>`, used by [public/viewer.html](public/viewer.html) and [src/model-page.js](src/model-page.js)), write path in [api/_lib/forge-store.js](api/_lib/forge-store.js) `materializeCreation`, docs [docs/sim-readiness.md](docs/sim-readiness.md). | Live | The question a renderer never asks and a rigid-body solver always does: can this mesh be simulated as-is? Four verdicts (`simulation_ready`, `needs_scale`, `needs_repair`, `unusable`, plus `unreadable` for non-glTF bytes) over measurements derived entirely from the geometry: watertightness and winding via welded-position topology, real-world extents read as meters per the glTF unit convention, exact volume/centroid/inertia from the divergence theorem and tetrahedron covariance, and a convex-hull proxy with its convexity ratio. Nothing is guessed: where a value cannot be known from the mesh (true scale, density) the report says so. Grades are keyed by GLB content hash, not URL, so one row serves the forge creation, the Object Library entry, a remix and an arbitrary third-party URL; every finished generation is graded on the write path, so a model page renders its verdict with no extra request. Signed into the content credential as `simReadiness` (provenance v2), which is what makes the physics claim as forge-proof as the authorship claim. Free, no account, no coin surface, both stores. Tests: [tests/sim-readiness.test.js](tests/sim-readiness.test.js), [tests/api/sim-readiness-endpoint.test.js](tests/api/sim-readiness-endpoint.test.js), [tests/provenance-3d.test.js](tests/provenance-3d.test.js). |
| OpenAI Select Partner (partner surface + announcement pack) | [pages/openai/index.html](pages/openai/index.html) `→ /openai`, badge assets [public/partners/openai/](public/partners/openai), announcement [blog/three-ws-openai-select-partner.html](blog/three-ws-openai-select-partner.html) `→ /blog/three-ws-openai-select-partner`, directory card in [pages/partners.html](pages/partners.html) `→ /partners`, press/social pack [marketing/openai-select-partner/](marketing/openai-select-partner/README.md), original OpenAI-supplied badge files + press-release template [OAI Select Tier Assets/](OAI%20Select%20Tier%20Assets/README.md) | Live | three.ws's OpenAI Partner Network Select-tier surface. `/openai` is the partner landing: the supplied badge (used unmodified, on a white plate), the four live OpenAI surfaces (the free keyless [3D Studio MCP connector](docs/mcp-studio.md) `→ /api/mcp-studio`, the custom-GPT Actions twin `→ /api/3d/studio`, [Spatial MCP](specs/SPATIAL_MCP.md), and [AR in ChatGPT](docs/chatgpt-ar.md)), the connect table, and the eleven-tool list. The marketing pack holds badge-usage rules and the press-release draft, which is gated on OpenAI's written approval before publishing (see its README). |
@@ -264,7 +264,7 @@ bottom.
| React components | [packages/react/](packages/react) | Published as `@three-ws/react` (v1.0.1) | React components for embedding three.ws 3D AI agents |
| x402 auto-paying fetch | [packages/x402-fetch/](packages/x402-fetch) | Published as `@three-ws/x402-fetch` (v1.0.2) | Drop-in fetch wrapper that auto-pays x402 challenges |
| AgentCore payments MCP | [packages/agentcore-payments-mcp/](packages/agentcore-payments-mcp) | Published as `@three-ws/agentcore-payments-mcp` (v0.1.2) | MCP server for platform-managed agent payment sessions |
-| VS Code 3D extension | [packages/vscode-3d/](packages/vscode-3d) | In-repo (v0.1.0, not yet on the Marketplace) | Makes `.glb` and `.gltf` first-class in VS Code: the file opens in a three.js viewer as an editor tab (orbit, animation clips with a scrub bar, wireframe, skeleton, grid, PNG snapshot) with Draco, KTX2 and meshopt decoders shipped inside the extension so a compressed avatar renders offline. The report panel reuses [src/gltf-inspect.js](src/gltf-inspect.js), the same inspector behind /validation, in the extension host, so a model is never uploaded to be measured. Generation, rigging and the gallery call the free public lanes (`/api/mcp-studio`, `/api/forge-gallery`); the embed command pins the current library release and its SRI hash from `/agent-3d/versions.json`. Doc: [docs/vscode.md](docs/vscode.md). Tests: `npm --prefix packages/vscode-3d test` (activates the bundled extension against a stand-in host, plus unit tests for the studio, gallery, embed and naming modules). |
+| VS Code 3D extension | [packages/vscode-3d/](packages/vscode-3d) | In-repo (v0.2.0, not yet on the Marketplace) | Makes `.glb` and `.gltf` first-class in VS Code: the file opens in a three.js viewer as an editor tab (orbit, animation clips with a scrub bar, wireframe, skeleton, grid, PNG snapshot, looping turntable GIF) with Draco, KTX2 and meshopt decoders shipped inside the extension so a compressed avatar renders offline. The report panel and the status bar reuse [src/gltf-inspect.js](src/gltf-inspect.js), the same inspector behind /validation, in the extension host, so a model is never uploaded to be measured. **Animate** searches the 2,874-clip motion library (`/api/animations/library`) and retargets the pick onto the open rig inside the webview with [src/animation-retarget.js](src/animation-retarget.js); **Bake clip** writes the retargeted clip into a copy of the GLB as real glTF channels ([packages/vscode-3d/src/bake-clip.js](packages/vscode-3d/src/bake-clip.js), glTF-Transform, meshopt re-encoded). Text to motion goes through `/api/forge-motion`. **Refine** calls the free `refine_model` tool and keeps the version lineage per file. **Optimize** runs dedup/prune/weld/resample/meshopt (and quantize) locally. **Compare with the committed version** reads `HEAD` out of git and diffs it with [packages/glb-diff](packages/glb-diff), opening both versions side by side. **Check quality** posts the webview render to `/api/forge-quality-check`. Language features for `<agent-3d>` in HTML/JSX/Vue/Svelte/Astro/Markdown/PHP ([packages/vscode-3d/src/embed-language.js](packages/vscode-3d/src/embed-language.js)): diagnostics with quick fixes (no source, no size, unpinned `latest`, missing or stale SRI hash against `/agent-3d/versions.json`, unknown attributes, bad enum values), hovers, completions, CodeLenses. Generation, rigging and the gallery call the free public lanes (`/api/mcp-studio`, `/api/forge-gallery`); the embed command pins the current library release and its SRI hash. Doc: [docs/vscode.md](docs/vscode.md). Tests: `npm --prefix packages/vscode-3d test` (activates the bundled extension against a stand-in host that now covers language providers and diagnostics, proves the retarget-then-bake round trip on the committed cz.glb, and unit tests every module against local HTTP servers). |
| VS Code x402 extension | [packages/vscode-x402/](packages/vscode-x402) | In-repo (v0.2.0, not yet published to npm) | VS Code extension for the x402 bazaar |
| Meta-Allocator (the ETF of degens) | [pages/meta-allocator.html](pages/meta-allocator.html) + [src/meta-allocator.js](src/meta-allocator.js) `→ /meta-allocator`, brain [api/_lib/meta-allocator.js](api/_lib/meta-allocator.js), endpoint [api/meta-allocator.js](api/meta-allocator.js) | Live (non-custodial planning) | Spreads a budget across the top verified pump.fun leader agents, ranked by risk-adjusted on-chain track record (win rate, realized ROI, peak-to-trough max drawdown from `agent_sniper_positions`) and diversified across derived correlation groups, then returns a rebalance rule. Free-first `llmComplete` shaping with a deterministic same-shape fallback so it never fails. Returns a plan (weights + suggested sizes), never a transaction; users act one leader at a time via [/vaults](pages/vaults.html). Doc: [docs/meta-allocator.md](docs/meta-allocator.md). |
| Alpha-drip (a leader prices their signal latency) | Ladder rules [api/\_lib/alpha-drip.js](api/_lib/alpha-drip.js), fanout half [api/\_lib/copy-drip-runtime.js](api/_lib/copy-drip-runtime.js), measured edge half-life [api/\_lib/alpha-drip-stats.js](api/_lib/alpha-drip-stats.js), endpoint [api/copy/alpha-drip.js](api/copy/alpha-drip.js) `-> /api/copy/alpha-drip`, leader editor [src/alpha-drip-panel.js](src/alpha-drip-panel.js) mounted on [/dashboard/copy](src/dashboard-next/pages/copy.js), copier seat on the trader page [src/copy-panel.js](src/copy-panel.js), schema [api/\_lib/migrations/20260903235500\_copy\_alpha\_drip.sql](api/_lib/migrations/20260903235500_copy_alpha_drip.sql) | Live (off unless a leader turns it on) | The $THREE sink on the copy rail: a leader sets how long each holder tier waits before a copy intent is revealed to them, so the thing being sold is latency rather than only a share of the profit. Two rules are structural and enforced server-side: a higher tier can never wait longer than a lower one, and **a drip delays the reveal, never the record** (the intent row is written in full at fanout time and the public track record is untouched, so there are no hidden trades). The copier is shown the seat their balance buys BEFORE they subscribe, not after; a locked intent renders as a countdown that unlocks in place, and `POST /api/copy/executions` refuses an unreleased one in the same statement that changes its status, so the guard cannot be raced. Tier size caps are applied after the copy engine has sized and gated the order, never before, so the copier's own caps still bind first. The edge half-life the fairness warning judges a ladder against is measured from the leader's own profitable closes, and is null rather than guessed below five of them. Release runs inside [api/cron/copy-fanout.js](api/cron/copy-fanout.js) rather than a second cron, so the release cadence is the fanout cadence. Doc: [docs/alpha-drip.md](docs/alpha-drip.md). Tests: [tests/alpha-drip.test.js](tests/alpha-drip.test.js) |
diff --git a/data/changelog.json b/data/changelog.json
index 2289edaa7..daca20602 100644
--- a/data/changelog.json
+++ b/data/changelog.json
@@ -1,10 +1,39 @@
{
"$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. `link` is optional and must be a live three.ws path.",
"entries": [
+ {
+ "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 See it 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 tip panel; other phones open the AR launcher. 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.",
+ "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"
diff --git a/docs/irl.md b/docs/irl.md
index 41ff755e6..dccedfa1f 100644
--- a/docs/irl.md
+++ b/docs/irl.md
@@ -52,6 +52,8 @@ Every discovered agent is alive in the camera view, not a statue: any humanoid a
Tapping an agent opens its inspect card:
+- **Talk** opens a live, spoken conversation with the agent standing in front of you. Hold the mic and speak (or type); the reply comes back in the agent's own persona when the pin carries an agent identity, is voiced through its voice (a cloned one when it has it), and the 3D model's mouth moves with the audio on top of its idle. Pressing the mic while it speaks interrupts it, like a real conversation. The opening line of a conversation lands once in the owner's IRL feed as a `talk` interaction; the transcript never does.
+- **View in AR** stands that agent, anyone's, on your real floor through the device's native AR stack: iPhone bakes the pin's model into an animated USDZ and opens ARKit Quick Look (the banner inside AR reads "Talk to <name>" and brings you back to the card with the conversation open); Android with ARCore enters a WebXR hit-test session that anchors this agent's already-loaded model to a detected surface, view-only, with pinch to size; everything else opens the device-aware [AR launcher](./ar.md) in a new tab. Nothing is persisted from a view: the durable placement stays the owner's pin.
- its bio, on-chain reputation tier, and any paid services it offers
- **pay via x402** to use a service it sells, settled on-chain in the request
- **leave a message** that lands in the owner's IRL feed
@@ -75,7 +77,7 @@ The **Share** button captures your camera feed and the placed agent as one photo
A placement has a **visit link**, `three.ws/irl?pin=<id>`, and a printable sign at `/irl/sign?pin=<id>`. Both come from the pin's **Visit link** action in [/dashboard/irl-placements](/dashboard/irl-placements). Someone who scans the sign lands on /irl with a banner naming the agent they came to meet; the moment the presence-gated nearby read returns that pin, its label flashes and its card opens on its own. Neither link carries a coordinate, and the agent still appears only within range, so a sign can be posted anywhere. `?highlight=<id>` is an older alias of `?pin=`.
-The card of a discovered agent also offers **See it in AR**: on iPhone it bakes that pin's model into an animated USDZ and opens ARKit Quick Look, with a banner tap that returns to the card with the tip QR open; everywhere else it opens the [AR launcher](./ar.md), which routes Android to native AR.
+The card of a discovered agent also offers **Talk** and **View in AR** (see [Interact and pay](#interact-and-pay)): the visitor can hold the mic and speak with the agent, and stand it on the real floor through Quick Look on iPhone, WebXR on Android, or the [AR launcher](./ar.md) elsewhere.
The full street-demo runbook (picking a spot for GPS, printing, what visitors see, on-site troubleshooting) is [Run an IRL street demo](./irl/street-demo.md).
diff --git a/docs/irl/street-demo.md b/docs/irl/street-demo.md
index f500443b8..e2cdaf1e0 100644
--- a/docs/irl/street-demo.md
+++ b/docs/irl/street-demo.md
@@ -1,6 +1,6 @@
# Run an IRL street demo: drop an agent at a real spot, let strangers walk up
-This is the runbook for showing IRL to people who have never heard of three.ws: you place a 3D AI agent at a real location, put a sign there, and anyone who scans it sees the agent standing on that spot through their phone camera, taps it for its profile and wallet, and can stand it on the floor in AR. It was written for a demo in San Francisco, and everything in it applies to any city.
+This is the runbook for showing IRL to people who have never heard of three.ws: you place a 3D AI agent at a real location, put a sign there, and anyone who scans it sees the agent standing on that spot through their phone camera, taps it, talks to it out loud, and can stand it on their own floor in AR. It was written for a demo in San Francisco, and everything in it applies to any city.
Nothing here needs an app or an account on the visitor's side. The owner needs a signed-in three.ws account so the pin is permanent.
@@ -11,8 +11,9 @@ Nothing here needs an app or an account on the visitor's side. The owner needs a
1. They scan the sign. It opens `three.ws/irl?pin=<id>` in their phone browser.
2. IRL asks for camera and location (and motion, on iPhone, which takes one extra tap). A banner at the top says **"You're here to meet <agent name>"** and that the agent appears within 60 m of the spot.
3. Standing within range, the agent appears in their camera view, playing its idle animation and turning to look at them. Its name label flashes and its card opens on its own.
-4. The card shows the agent's bio, on-chain reputation, its **Solana wallet chip**, a **Tap to tip in person** panel (a Solana Pay QR plus an "Open in wallet" deep link), any paid x402 services, a message box, and **View profile**.
-5. **See it in AR** stands the agent on the real floor. iPhone opens ARKit Quick Look with the idle clip baked in; the Quick Look banner reads "Tip this agent" and tapping it brings them back to the card with the tip QR already open. Android and everything else go through the AR launcher (`/api/ar`), which routes Android to native AR.
+4. The card leads with **Talk** and **View in AR**, then shows the agent's bio, on-chain reputation, its **Solana wallet chip**, a **Tap to tip in person** panel (a Solana Pay QR plus an "Open in wallet" deep link), any paid x402 services, a message box, and **View profile**.
+5. **Talk** opens a conversation with the agent standing in front of them. They hold the mic and speak (or type). The reply is in the agent's persona, spoken aloud in its voice, and the model's mouth moves with the audio. Pressing the mic while it is speaking interrupts it.
+6. **View in AR** stands the agent on their real floor. iPhone opens ARKit Quick Look with the idle clip baked in; the Quick Look banner reads "Talk to <name>" and tapping it brings them back to the card with the conversation open. Android with ARCore enters a WebXR session that anchors this agent to the detected floor (tap to place, pinch to size). Anything else opens the AR launcher (`/api/ar`) in a new tab.
Anyone who opens the link from somewhere else sees only the agent's name and a prompt to walk to the spot. The link never carries a coordinate, and the nearby read is presence-gated, so posting the link publicly is safe. See [How location works on IRL](/irl-privacy).
@@ -64,8 +65,10 @@ Test the sign with your own phone before you leave: scan it, walk 70 m away, con
- **Your phone is the demo screen too.** The dashboard's **View in IRL** opens the same visit link, so you can show the walk-up while a visitor does it.
- **Talk through the three taps:** scan, allow, look around. The banner carries the rest.
-- **Show the tip.** Open the card, expand **Tap to tip in person**, and let someone scan the Solana Pay QR with Phantom or Solflare. The agent's wallet is a real address; a tip lands on-chain.
-- **Show it on the floor.** On an iPhone, **See it in AR** is the moment people photograph.
+- **Let them talk first.** Hand the phone over with the card open and say "hold the mic and ask it anything". The agent answers out loud in its persona; that is the moment people remember. Give the agent a persona and a cloned voice beforehand (`/agents/<id>`, Brain Studio and the voice clone in talk mode) so it sounds like itself on the street.
+- **Show the tip.** Expand **Tap to tip in person**, and let someone scan the Solana Pay QR with Phantom or Solflare. The agent's wallet is a real address; a tip lands on-chain.
+- **Show it on the floor.** **View in AR** is the moment people photograph: Quick Look on an iPhone, a floor-anchored WebXR session on an Android phone with ARCore.
+- **Volume and noise.** Replies play through the phone speaker. On a loud street, hold the phone up or use a small speaker; typed turns work when speech recognition cannot hear.
- **Messages and taps land in your dashboard.** The placements page lists every message, tap and pay under the pin, and the inbox lets you reply. `GET /api/irl/analytics` (admin) has the rollup afterwards.
---
@@ -77,7 +80,10 @@ Test the sign with your own phone before you leave: scan it, walk 70 m away, con
| Banner never changes from "You're here to meet" | Visitor is outside 60 m, or their GPS fix is far off | Move to open sky, wait ten seconds, check the accuracy line on the card once it opens |
| "Enable location" chip in the top bar | Location was denied | Tap the chip; on iPhone, Settings > Safari > Location if the prompt is gone |
| Agent appears but does not move | Motion permission not granted (iPhone), or a non-humanoid body | Tap the motion chip; humanoid rigs animate, props do not |
-| "See it in AR" is missing | The pin's model is not an https GLB | Re-place the pin with a hosted avatar from your account |
+| "View in AR" is missing | The pin has no model URL | Re-place the pin with a hosted avatar from your account |
+| Talk shows "Voice input isn't available" | The browser has no speech recognition, or the mic was denied | Type in the box instead; on iPhone allow the microphone when Safari asks, or Settings > Safari > Microphone |
+| The agent answers but you hear nothing | The phone is on silent (iPhone honours the mute switch for web audio) or volume is down | Flip the ring/silent switch, raise the volume, tap the mic again |
+| The mouth does not move while it speaks | The body has no mouth morphs or jaw bone, or the model is still streaming in | Nothing to fix on site; the agent still talks. Rigs with ARKit or VRM mouth shapes animate |
| Banner says the agent has moved on | The pin expired (anonymous pins last 7 days) or was removed | Place it again signed in and print a fresh sign |
| Visitors see nothing at all, you see the agent | Your device owns the pin; theirs reads nearby only within range | Have them stand where you placed it, not where you are now |
@@ -88,4 +94,4 @@ Test the sign with your own phone before you leave: scan it, walk 70 m away, con
- [IRL overview](/docs/irl): what IRL is, Money Drops, World Lines, the SDK
- [How location works on IRL](/irl-privacy) and the [threat model](/docs/irl/THREAT-MODEL)
- [Place a 3D agent in your real environment](/docs/tutorials/place-agent-irl)
-- [AR and WebXR reference](/docs/ar): the device routing behind "See it in AR"
+- [AR and WebXR reference](/docs/ar): the device routing behind "View in AR"
diff --git a/docs/tutorials/place-agent-irl.md b/docs/tutorials/place-agent-irl.md
index 9fac42deb..bba706773 100644
--- a/docs/tutorials/place-agent-irl.md
+++ b/docs/tutorials/place-agent-irl.md
@@ -153,6 +153,8 @@ When you're standing near a pinned agent (yours or someone else's), it appears i
Tap an agent's label or model to open its **inspect card**:
+- **Talk** to it. Hold the mic and speak, or type. The agent answers in its own persona, out loud, and its mouth moves with the words. Press the mic while it speaks to interrupt it.
+- **View in AR** to stand it on your real floor: Quick Look on iPhone, a floor-anchored WebXR session on Android with ARCore, the AR launcher everywhere else. This works on any agent you discover, not only your own.
- Its bio, on-chain reputation tier, and any paid services it offers
- **Pay via x402** to use a service it sells
- **Leave a message** that lands in the owner's IRL feed
diff --git a/pages/irl-sign.html b/pages/irl-sign.html
index 401e363b6..b3d024378 100644
--- a/pages/irl-sign.html
+++ b/pages/irl-sign.html
@@ -121,7 +121,7 @@
<ol class="sg-steps">
<li>Scan the code with your phone camera. No app, no account.</li>
<li>Allow camera and location, then look around. It appears within a few steps of here.</li>
- <li>Tap it: profile, wallet, and services. iPhone: "See it in AR" stands it on the floor.</li>
+ <li>Tap it, then hold the mic and talk to it. "View in AR" stands it on your floor.</li>
</ol>
<div class="sg-qr" id="sg-qr" role="img" aria-label="QR code that opens this agent's visit link"></div>
</div>
diff --git a/public/changelog.json b/public/changelog.json
index e09e09ccf..702bdfb30 100644
--- a/public/changelog.json
+++ b/public/changelog.json
@@ -1,5 +1,5 @@
{
- "generated_at": "2026-09-05T16:13:12.871Z",
+ "generated_at": "2026-09-05T16:25:38.047Z",
"generated_by": "scripts/build-page-index.mjs from data/pages.json + data/changelog.json",
"site": {
"name": "three.ws",
@@ -103,13 +103,23 @@
"date": "2026-09-05",
"type": "update",
"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 See it 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 tip panel; other phones open the AR launcher. 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.",
+ "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.",
"link": "/irl",
"tags": [
"feature",
"improvement"
]
},
+ {
+ "date": "2026-09-05",
+ "type": "update",
+ "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.",
+ "link": "/irl",
+ "tags": [
+ "fix"
+ ]
+ },
{
"date": "2026-09-05",
"type": "update",
@@ -153,6 +163,16 @@
"improvement"
]
},
+ {
+ "date": "2026-09-05",
+ "type": "update",
+ "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.",
+ "link": "/irl",
+ "tags": [
+ "feature"
+ ]
+ },
{
"date": "2026-09-05",
"type": "update",
diff --git a/public/changelog.xml b/public/changelog.xml
index 1c963ee6d..75c4a071a 100644
--- a/public/changelog.xml
+++ b/public/changelog.xml
@@ -84,7 +84,15 @@
<guid isPermaLink="false">2026-09-05:Invite people to an IRL agent with a visit link and a printable sign</guid>
<pubDate>Sat, 05 Sep 2026 12:00:00 GMT</pubDate>
<category>feature</category>
- <description>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 See it 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 tip panel; other phones open the AR launcher. 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.</description>
+ <description>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.</description>
+ </item>
+ <item>
+ <title>IRL no longer fails to load your agent on a cold start</title>
+ <link>https://three.ws/irl</link>
+ <guid isPermaLink="false">2026-09-05:IRL no longer fails to load your agent on a cold start</guid>
+ <pubDate>Sat, 05 Sep 2026 12:00:00 GMT</pubDate>
+ <category>fix</category>
+ <description>Opening IRL on a fresh page could show &quot;Couldn't load this agent&quot; 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.</description>
</item>
<item>
<title>Machine Atlas: two working machines built in the browser from code, with no 3D model files</title>
@@ -118,6 +126,14 @@
<category>fix</category>
<description>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.</description>
</item>
+ <item>
+ <title>Walk up to an IRL agent, talk to it out loud, and stand it on your floor in AR</title>
+ <link>https://three.ws/irl</link>
+ <guid isPermaLink="false">2026-09-05:Walk up to an IRL agent, talk to it out loud, and stand it on your floor in AR</guid>
+ <pubDate>Sat, 05 Sep 2026 12:00:00 GMT</pubDate>
+ <category>feature</category>
+ <description>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 &quot;talked with your agent&quot; event in their IRL feed when someone starts a conversation. The transcript stays between the visitor and the agent.</description>
+ </item>
<item>
<title>Your agent now lives in a handheld console you can put on any website</title>
<link>https://three.ws/pocket</link>
diff --git a/public/features.json b/public/features.json
index 9dac72b80..94bb9e0b2 100644
--- a/public/features.json
+++ b/public/features.json
@@ -1,5 +1,5 @@
{
- "generated_at": "2026-09-05T16:13:12.862Z",
+ "generated_at": "2026-09-05T16:25:38.038Z",
"generated_by": "scripts/build-page-index.mjs from data/pages.json",
"site": {
"name": "three.ws",
diff --git a/src/dashboard-next/pages/irl-placements.js b/src/dashboard-next/pages/irl-placements.js
index 7d9cb0f39..aa65a9cf2 100644
--- a/src/dashboard-next/pages/irl-placements.js
+++ b/src/dashboard-next/pages/irl-placements.js
@@ -1179,7 +1179,7 @@ function openVisitLinkModal(pin) {
<ol class="irl-visit-steps">
<li>A visitor scans it and allows camera and location.</li>
<li>Within 60 m of the spot, <b>${esc(name)}</b> appears in their camera and its card opens.</li>
- <li>They tap it for the profile, wallet, tip QR and services; iPhone gets "See it in AR".</li>
+ <li>They tap it to talk to it out loud, view it in AR on their floor, and see the profile, wallet, tip QR and services.</li>
</ol>
<div class="irl-visit-actions">
<a class="irl-btn primary" href="${esc(signUrl)}" target="_blank" rel="noopener">Print a sign ↗</a>
diff --git a/src/irl.js b/src/irl.js
index 57066a288..cf2601095 100644
--- a/src/irl.js
+++ b/src/irl.js
@@ -55,7 +55,7 @@ import { errorStateEl, skeletonHTML, emptyStateHTML, errorStateHTML, ensureState
import { startOnboarding, ensurePermission, needsMotionGesture, setPermissionState } from './irl/onboarding.js';
import { reserveWebGLContext, releaseWebGLContext } from './webgl-budget.js';
import { detectTier, BUDGETS, TIER_ORDER, shiftTier } from './irl/perf-budget.js';
-import { sharedGLTFLoader, createLoadQueue } from './irl/load-queue.js';
+import { loadGLTF, createLoadQueue } from './irl/load-queue.js';
import { mountPinIdle, getIdleClipJson } from './irl/pin-idle.js';
import { celebrationRingFrame, CELEBRATION_DURATION } from './irl/pin-celebration.js';
import { applyAvatarMaterialRealism, looksLikeAvatarMesh } from './shared/avatar-material-realism.js';
@@ -260,7 +260,7 @@ let budget = BUDGETS[activeTier];
// capped queue — replaces the old `new GLTFLoader()` per pin + `loadedCount < 5`
// guard. Priority is the pin's live camera distance (set in enforceLOD).
const glbQueue = createLoadQueue({
- run: (pin) => sharedGLTFLoader().loadAsync(pin.avatar_url),
+ run: (pin) => loadGLTF(pin.avatar_url),
maxActive: budget.maxGLB,
priorityOf: (pin) => (pin._lodDist != null ? pin._lodDist : (pin.distance_m != null ? pin.distance_m : 1e9)),
});
@@ -1166,7 +1166,7 @@ async function loadAvatar(idOrUrl, nameOverride) {
// Reuse the one shared Draco+meshopt loader (the pin queue's loader) instead of
// a fresh GLTFLoader per swap — the player's own avatar can be compressed too,
// and a per-swap loader allocated decoders the shared one already holds.
- const gltf = await sharedGLTFLoader().loadAsync(glbUrl);
+ const gltf = await loadGLTF(glbUrl);
avatar = gltf.scene;
dressAvatarMaterials(avatar);
avatar.traverse(n => {
@@ -8500,6 +8500,15 @@ if (import.meta.env.DEV) {
const mLng = 111320 * Math.cos(lat * (Math.PI / 180));
return { lat: lat + north / mLat, lng: lng + east / mLng };
},
+ // Open the inspect sheet for a held pin by id, exactly as a tap on its label
+ // would. The e2e for Talk / View in AR drives the real sheet code through it
+ // instead of guessing where a 3D label landed on screen.
+ openSheet(id) {
+ const p = nearbyPins.find((x) => x.id === id);
+ if (!p) return false;
+ openPinSheet(p);
+ return true;
+ },
// Liveness probe for the T-pose regression: is the pin's full model mounted,
// does it carry a running idle mixer, and where is its skeleton right now?
// Sampling the bone quaternions twice a few frames apart proves the agent
diff --git a/tests/e2e/irl-pin-talk.spec.js b/tests/e2e/irl-pin-talk.spec.js
new file mode 100644
index 000000000..c56dccec8
--- /dev/null
+++ b/tests/e2e/irl-pin-talk.spec.js
@@ -0,0 +1,204 @@
+/**
+ * /irl Talk + View in AR on a discovered pin (Playwright e2e).
+ *
+ * The walk-up moment: a visitor taps an agent standing near them and (1) holds a
+ * conversation with it, (2) can stand it on their own floor in AR. Both live in
+ * the pin sheet (src/irl/pin-talk.js, src/irl/pin-ar.js). This spec drives the
+ * REAL sheet code in a real browser through the DEV-only window.__irlE2E seam and
+ * hermetic routes: the nearby read serves one in-range pin, /api/chat answers with
+ * a canned SSE stream, TTS answers with a real (silent) WAV so playback succeeds,
+ * and the interaction log is captured so the `talk` event is asserted rather than
+ * assumed.
+ *
+ * Coverage:
+ * - the sheet leads with Talk and View in AR for a stranger's pin
+ * - Talk opens the panel with the designed empty state, focus lands on the mic
+ * - a typed turn renders the visitor's line, streams the reply into the log,
+ * and posts exactly one `talk` interaction for the conversation
+ * - a second turn posts no second `talk` row
+ * - closing the sheet ends the conversation (panel hidden, no further audio)
+ * - View in AR on a non-AR device over http is an honest status line, never a
+ * dead button or a thrown error
+ * - no console errors anywhere in the flow
+ */
+
+import { test, expect } from '@playwright/test';
+
+const VIEWER = { lat: 37.7749, lng: -122.4194 };
+
+function pinRow(id, lat, lng, name) {
+ return {
+ id, lat, lng, heading: 0,
+ avatar_url: '/avatars/default.glb',
+ avatar_name: name,
+ caption: 'Ask me about the pier',
+ x402_endpoint: null,
+ agent_id: null,
+ placed_at: '2026-01-01T00:00:00Z',
+ view_count: 0,
+ room_id: null, rel_east_m: null, rel_north_m: null,
+ origin_lat: null, origin_lng: null, origin_yaw_deg: null,
+ gps_accuracy_m: 8, altitude_m: null, anchor_source: null, avatar_version: 0,
+ };
+}
+
+function offset(lat, lng, north, east) {
+ const mLat = 110540;
+ const mLng = 111320 * Math.cos(lat * (Math.PI / 180));
+ return { lat: lat + north / mLat, lng: lng + east / mLng };
+}
+
+// A valid 16-bit mono PCM WAV of `seconds` silence, so the TTS lane returns audio
+// the browser can actually decode and play (an empty blob would fire onerror).
+function silentWav(seconds = 0.25, rate = 8000) {
+ const samples = Math.round(seconds * rate);
+ const buf = Buffer.alloc(44 + samples * 2);
+ buf.write('RIFF', 0); buf.writeUInt32LE(36 + samples * 2, 4); buf.write('WAVE', 8);
+ buf.write('fmt ', 12); buf.writeUInt32LE(16, 16); buf.writeUInt16LE(1, 20); buf.writeUInt16LE(1, 22);
+ buf.writeUInt32LE(rate, 24); buf.writeUInt32LE(rate * 2, 28); buf.writeUInt16LE(2, 32); buf.writeUInt16LE(16, 34);
+ buf.write('data', 36); buf.writeUInt32LE(samples * 2, 40);
+ return buf;
+}
+
+function sse(events) {
+ return events.map((e) => `data: ${JSON.stringify(e)}\n\n`).join('');
+}
+
+async function installRoutes(page) {
+ const chat = []; // bodies posted to /api/chat
+ const interactions = []; // bodies posted to /api/irl/interactions
+ await page.route('**/api/**', (route) => route.fulfill({ status: 204, body: '' }));
+ await page.route(/\/api\/irl\/pins\?/, (route) => {
+ const near = offset(VIEWER.lat, VIEWER.lng, 6, 0);
+ route.fulfill({ json: { pins: [pinRow('pin-talk', near.lat, near.lng, 'Nova')] } });
+ });
+ await page.route('**/api/irl/fix-token', (route) => route.fulfill({ json: { token: 'e2e-fix', expires_in: 300 } }));
+ await page.route('**/api/irl/agent-card**', (route) => route.fulfill({ json: { card: {
+ agent: { id: null, name: 'Nova', bio: 'Guide to the waterfront.', thumbnail_url: null, profile_url: null },
+ reputation: { available: false },
+ services: [],
+ x402_endpoint: null,
+ } } }));
+ await page.route('**/api/irl/interactions', (route) => {
+ interactions.push(route.request().postDataJSON());
+ route.fulfill({ json: { ok: true, id: 'ix', type: 'view' } });
+ });
+ await page.route('**/api/chat', (route) => {
+ chat.push(route.request().postDataJSON());
+ route.fulfill({
+ status: 200,
+ headers: { 'content-type': 'text/event-stream' },
+ body: sse([{ type: 'chunk', text: 'Hi there, welcome to ' }, { type: 'chunk', text: 'the pier.' }, { type: 'done', reply: 'Hi there, welcome to the pier.' }]),
+ });
+ });
+ await page.route('**/api/tts/**', (route) => route.fulfill({
+ status: 200, headers: { 'content-type': 'audio/wav' }, body: silentWav(),
+ }));
+ return { chat, interactions };
+}
+
+function trackConsoleErrors(page) {
+ const errors = [];
+ page.on('console', (m) => {
+ if (m.type() !== 'error') return;
+ const t = m.text();
+ if (/websocket|hmr|wss:|vite|favicon|net::ERR|Outdated Optimize Dep/i.test(t)) return;
+ if (/status of 401/i.test(t) && !/\/api\/irl\//i.test(m.location()?.url || '')) return;
+ errors.push(t);
+ });
+ page.on('pageerror', (err) => {
+ if (/websocket|hmr|wss:/i.test(err.message)) return;
+ errors.push(`pageerror: ${err.message}`);
+ });
+ return errors;
+}
+
+test.describe('/irl pin sheet: Talk + View in AR', () => {
+ test('a visitor talks to a discovered agent and can reach its AR view', async ({ page }) => {
+ test.setTimeout(150_000);
+ const errors = trackConsoleErrors(page);
+ const seen = await installRoutes(page);
+
+ // A returning visitor: the first-run permission dialog, the discovery
+ // explainer and the location disclosure are all already answered, so no
+ // modal scrim sits over the sheet. Permissions are marked denied because
+ // headless Chromium has no camera or GPS; the seam supplies the fix below.
+ await page.addInitScript(() => {
+ localStorage.setItem('irl_onboarded_v1', JSON.stringify({ camera: 'denied', motion: 'denied', location: 'denied', ts: Date.now() }));
+ localStorage.setItem('irl_discovery_explained_v1', '1');
+ localStorage.setItem('irl_location_disclosed_v1', '1');
+ });
+ await page.goto('/irl');
+ await page.waitForFunction(() => !!window.__irlE2E, null, { timeout: 90_000 });
+ await page.evaluate((v) => window.__irlE2E.setGps(v.lat, v.lng), VIEWER);
+ await page.evaluate(() => window.__irlE2E.poll());
+ expect(await page.evaluate(() => window.__irlE2E.nearby())).toHaveLength(1);
+
+ // Tap the agent (through the seam: same openPinSheet the label click runs).
+ expect(await page.evaluate(() => window.__irlE2E.openSheet('pin-talk'))).toBe(true);
+ const sheet = page.locator('#irl-sheet');
+ await expect(sheet).toHaveClass(/is-open/);
+ await expect(page.locator('#irl-sheet-name')).toHaveText('Nova');
+
+ // The card leads with Talk and View in AR, for a stranger's pin.
+ const talkBtn = page.locator('#irl-sheet-talk');
+ const arBtn = page.locator('#irl-sheet-ar');
+ await expect(talkBtn).toBeVisible();
+ await expect(arBtn).toBeVisible();
+ await expect(arBtn).toHaveText('View in AR');
+ await expect(talkBtn).toHaveAttribute('aria-expanded', 'false');
+
+ // Open the conversation: designed empty state, mic focused.
+ await talkBtn.click();
+ const panel = page.locator('#irl-talk');
+ await expect(panel).toBeVisible();
+ await expect(talkBtn).toHaveAttribute('aria-expanded', 'true');
+ await expect(page.locator('#irl-talk-log .irl-talk-empty')).toContainText(/Nova is listening/);
+ await expect(page.locator('#irl-talk-hold')).toBeFocused();
+
+ // A typed turn: the visitor's line renders, the reply streams in, and the
+ // conversation's opening line is logged once as a `talk` interaction.
+ await page.fill('#irl-talk-input', 'Where am I?');
+ await page.locator('#irl-talk-form').evaluate((f) => f.requestSubmit());
+ await expect(page.locator('#irl-talk-log .irl-talk-msg.is-user')).toHaveText('Where am I?');
+ await expect(page.locator('#irl-talk-log .irl-talk-msg.is-agent .irl-talk-text')).toHaveText('Hi there, welcome to the pier.', { timeout: 20_000 });
+ await expect(page.locator('#irl-talk-log .irl-talk-who')).toHaveText('Nova');
+ expect(seen.chat).toHaveLength(1);
+ expect(seen.chat[0].message).toBe('Where am I?');
+ // An anonymous pin never fabricates an agentId; the persona rides system_prompt.
+ expect(seen.chat[0].agentId).toBeUndefined();
+ expect(seen.chat[0].system_prompt).toMatch(/You are Nova, a 3D AI agent standing at a real place/);
+ expect(seen.chat[0].system_prompt).toMatch(/Ask me about the pier/);
+ await expect.poll(() => seen.interactions.filter((i) => i.type === 'talk').length).toBe(1);
+ expect(seen.interactions.find((i) => i.type === 'talk').pinId).toBe('pin-talk');
+
+ // A second turn is the same conversation: no second `talk` row.
+ await expect.poll(() => page.evaluate(() => document.querySelector('#irl-talk').dataset.state), { timeout: 20_000 }).toBe('idle');
+ await page.fill('#irl-talk-input', 'Thanks!');
+ await page.locator('#irl-talk-form').evaluate((f) => f.requestSubmit());
+ await expect(page.locator('#irl-talk-log .irl-talk-msg.is-user')).toHaveCount(2);
+ await expect(page.locator('#irl-talk-log .irl-talk-msg.is-agent')).toHaveCount(2, { timeout: 20_000 });
+ expect(seen.chat).toHaveLength(2);
+ expect(seen.chat[1].history.some((h) => h.content === 'Where am I?')).toBe(true);
+ expect(seen.interactions.filter((i) => i.type === 'talk')).toHaveLength(1);
+
+ // View in AR on a device with no AR surface, served over http: the honest
+ // status line (the launcher only accepts https models), never a dead button.
+ await arBtn.click();
+ await expect(page.locator('#irl-status')).toContainText(/can't be opened in the AR viewer/i);
+ await expect(sheet).toHaveClass(/is-open/);
+
+ // Closing the sheet ends the conversation.
+ await page.locator('#irl-sheet-close').click();
+ await expect(sheet).not.toHaveClass(/is-open/);
+ await expect(panel).toBeHidden();
+
+ // Re-opening the same pin starts fresh (the empty state is back).
+ await page.evaluate(() => window.__irlE2E.openSheet('pin-talk'));
+ await talkBtn.click();
+ await expect(page.locator('#irl-talk-log .irl-talk-empty')).toBeVisible();
+ await expect(page.locator('#irl-talk-log .irl-talk-msg')).toHaveCount(0);
+
+ expect(errors).toEqual([]);
+ });
+});