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
docs(research): tear down a leading agent launchpad and rank what three.ws should replicate
nirholas1cfce0cd12ed
diff --git a/docs/research/competitor-teardown-2026-09.md b/docs/research/competitor-teardown-2026-09.mdnew file mode 100644index 000000000..43ec22b4c--- /dev/null+++ b/docs/research/competitor-teardown-2026-09.md@@ -0,0 +1,123 @@+# Competitor X teardown, September 2026: what three.ws should replicate++Research date: 2026-09-29. Competitor X is a placeholder name for an AI-agent launchpad on Solana. Source: every public page in its sitemap (27 pages, locale duplicates and the 201 per-token pages excluded), its `llms.txt`, the npm registry entries for its four packages, and the GitHub API for its public repos, all read that day. Competitor X's numbers below are self-reported on its own pages and are marked as such. Our side of each comparison was checked against this repo on the same day, with the file that proves it.++Companion to [launchpad-landscape-2026-09.md](launchpad-landscape-2026-09.md), which covers the wider market.++## Bottom line++1. **Competitor X's edge is packaging, not technology.** We already ship most of what it sells: a hosted OAuth MCP server with annotated tools, resources and prompts, agent wallets with allowlists and spend caps, automations and autonomous runs, an agent marketplace with full ownership transfer, a token directory, agent email and cards, x402 selling, 90 locales. What it does better is how it presents these: one connector URL, one CLI command, one fee split explained per $100, and pages that show proof.+2. **The biggest cheap wins are in onboarding.** One-click "add to Claude / ChatGPT / Cursor / VS Code" connector links, an `npx three-ws launch` command, and an "official domain and mint" block in `llms.txt`. Each builds on infrastructure that already works in production.+3. **Their developer docs set a writing bar worth matching.** The Partner API reference has one error table with machine codes and `retrySafe` flags, an integration checklist, a support block that asks for `requestId` plus only the key prefix, and a section that openly lists which endpoints are stubs. We should copy that structure and honesty. We should not copy the stubs: our hard rules ban them.+4. **Their own experiment shows the creator-fee flywheel barely fires.** Across 7,215 launches, 0.40% of builders ever earned $500 and 35.6% of tokens never traded once. The launches that worked came from builders who shipped a product first. That matches what our landscape doc found: our problem is agent-to-launch conversion, not reach.+5. **Three things need owner sign-off and are prepared below:** a Solana Foundation `awesome-solana-ai` listing (a PR to an external repo), a public "why three.ws" comparison page (it would name other projects), and a hackathon page (it involves prize money).++## What Competitor X is++An AI-agent launchpad on Solana. Each agent gets its own wallet, a skill set (DeFi, perps, token launch, X posting, market intel), and a pump.fun token whose creator fees the platform collects hourly and splits with the builder. It lists a launchpad partner, a hackathon win and an accelerator as backers. Its agents are built on an open-source agent framework.++| Self-reported metric (2026-09-29) | Value | Where |+|---|---|---|+| Total volume, spot + perps | $212.51M (perps 0.3%) | /analytics |+| Agents funded | 20,046 | /analytics |+| Tokens listed | 20,055 | /tokens |+| SOL redistributed to creators | 12,997 | /stories |+| LLM tokens processed | 3.58B | /analytics |+| Hackathon volume since Aug 18 | $44.65M, 195 tokenized entries | /analytics |++The site contradicts itself, which is a trap for us to avoid rather than a detail to copy. It gives the creator share as 75% (docs, token page, /why-deploy) and 65% (/stories, the experiments post, its awesome-solana-ai listing). It gives the MCP tool count as 132, 126, 121, 109+, 79 and 78 on different pages. And /why-deploy still shows $112M volume and 6,620 agents while /analytics shows $212M and 20,046. Our `check:claude` and `audit:docs` guards exist to stop exactly this kind of drift. Any number we publish on a marketing page should be read live from the API, never typed into the copy.++## Their resource inventory++| Surface | What it does | Why it works |+|---|---|---|+| `/llms.txt` | 40 curated lines grouped as Product / Developers / Learn / Optional, plus an "only official domain, these look-alike sites are not us, official X/Telegram/GitHub, official mint" block | Short enough for an agent to read in full. The anti-phishing block protects users of coin-adjacent products where impersonation is routine. |+| `/mcp` | One connector URL, a deep link that opens Claude's "add custom connector" dialog with the URL pre-filled, tabs per client, and a "Worth knowing" list (you never copy a key, one client at a time, sign in as the account that owns your agents) | Setup is two clicks. The page cites Claude's own help article and asks readers to report drift. |+| `/docs` | A single long page with anchored left nav: 15 built-in and 50 community skills with their tool names, install steps per client, use cases written as ordered tool-call sequences, a 121-tool catalog tagged Read / Write / $Paid, resources, prompts, annotations, API, revenue model, fee split, key tiers, billing | Everything on one page, which is easy to search. Use cases are written as the tool calls an agent would actually make. |+| `/developers` | Versioned Partner API (`/api/v1`) reference: key format and storage (SHA-256 plus 12-char prefix), a 401/403 condition table, tiers, a "read this before you build capacity assumptions" warning that quotas are recorded but not enforced, the `meta.requestId` envelope, the full error table, per-endpoint idempotency notes, an integration checklist, and support instructions | Written for an integrator's first bad day: every failure mode has a code and a next step. |+| `/guide` | 7-step "first agent in 5 minutes": sign in, connect an AI provider (BYOK), fund the billing wallet, create the agent, enable skills, monitor, sell the agent as a paid x402 API | Explains billing wallet versus agent wallet up front, which is the confusion that costs support time. |+| Token page | Token economics: where each $100 of collected fees goes (75 creators, 15 buyback and burn, 5 treasury, 5 ecosystem fund), a live buyback and burn tracker reading a public wallet, the official mint, and a line saying it is not investment advice | People can check the claims on-chain themselves. |+| `/analytics` | Totals, a growth chart, compute tokens per day, and a hackathon leaderboard with its method explained ("volume is estimated from collected fees at the 1% creator rate; 26 entries excluded, off-platform") | Shows how every number was calculated. |+| `/stories/*` | Five case studies, one per category (breakout, creative marketplace, infrastructure, DeFi trader, community operator), each with market cap, SOL earned and links to the mint and the builder's X account | Proof a reader can check, sorted so different kinds of builders each see themselves. |+| `/experiments/*` | Internal experiments written up with method, spend ($0 to $0.28), results and "the parts we got wrong", including one that "ended the project" | Publishing its own failures builds credibility no marketing copy can buy. |+| `/why-deploy` | A flywheel (build, discover, invest), live metrics, a competitor comparison table, a partner list with source links | Direct answer to "why here". |+| Hackathon page | A $350K hackathon: register, post on X, tokenize by a deadline. 15 locales. The token is the entry ticket. | Brings in builders and on-platform token launches at the same time. |+| `/leaderboard` | Top 300 agents by points, with seasons and SOL earned | Keeps people coming back to compete. |+| `/marketplace` | Agents for sale or accepting bids, with a "connect your payout wallet first" warning | Agents become assets people can buy and sell. |+| Local agent page | A local, self-improving CLI and macOS desktop agent installed with one `npx` command | Reaches users who want to run the agent themselves. |+| npm | Four packages: a launch CLI, an agent MCP server, a launchpad MCP server, and the local agent | A separate package for each way people want to use it. |++## Replicate: ranked by leverage over effort++Status comes from a sweep of this repo on 2026-09-29. "Build" is the smallest change that closes the gap.++### Tier 1: small builds on infrastructure that already works++| # | What | Our status | Build |+|---|---|---|---|+| 1 | One-click MCP connect page | Hosted OAuth MCP works (`api/_mcp/auth.js`, `/.well-known/oauth-protected-resource`). We only have copy-paste `claude mcp add` snippets (`src/agent-detail.js`, `src/dashboard-next/pages/api.js`). | A `/connect` page: the connector URL with a copy button, a Claude deep link (`https://claude.ai/customize/connectors?modal=add-custom-connector&connectorName=three.ws&connectorUrl=<encoded>`), Cursor and VS Code install links, client tabs, and a "Worth knowing" list. Link it from `/mcp-tools`, `/docs` and the dashboard. |+| 2 | Curated `llms.txt` plus an official-links block | `public/llms.txt` is 991 lines, generated by `scripts/build-page-index.mjs` from `data/pages.json`. It has no official-domain or official-mint statement. | In the generator, add an "Official" block at the top (domain, X, Telegram, GitHub, the $THREE mint `FeMbDoX7R1Psc4GEcvJdsbNbZA3bfztcyDCatJVJpump`, "any other domain or mint is not us"). Move the full list to `llms-full.txt` and keep `llms.txt` to roughly 60 curated lines. Show the same block in the site footer and on `/three-token`. |+| 3 | Hosted `/skill.md` | `https://three.ws/skill.md` returns 404. We have `public/skills/` and `skills-index.json`. | Serve a root `skill.md` that explains how an agent uses three.ws (generate 3D, create an agent, launch, pay over x402) and links to the per-skill files. It is the URL skill directories list, and item 11 depends on it. |+| 4 | `npx three-ws launch` and `create` | `npx three-ws login` already runs browser OAuth (`packages/three-ws-cli`, `api/cli/[action].js`). The launch API exists (`api/pump/[action].js`, `api/x402/pump-launch.js`). | Add `create` (an agent with a wallet) and `launch` (name, ticker, image prompts, show the quote, confirm, sign) subcommands. Launch must stop at the quote for an explicit yes, per the spend gate. |+| 5 | Partner API docs, done properly | The v1 envelope and idempotency exist (`api/_lib/agents-v1/http.js`), and keys are hashed with a visible prefix (`api/api-keys.js`). Error tables are scattered across `docs/api-reference.md` and there is no checklist. | One consolidated error table with codes, retry safety and next actions, an integration checklist, a support block (send `requestId`, the endpoint, the UTC time and only the key prefix), and key storage and revocation timing documented. |++### Tier 2: medium builds that turn existing data into proof++| # | What | Our status | Build |+|---|---|---|---|+| 6 | Public per-agent creator earnings | `api/cron/launcher-claimer.js` claims creator fees into agent wallets. The Fee Bridge is held by the owner. There is no public earnings endpoint. | `GET /api/agents/:id/earnings` (lifetime claimed, pending, recent claims with signatures), shown on the agent profile and `/launches/:mint`. This is the number a builder screenshots. |+| 7 | One analytics page | The data is split across `/agent-economy-volume`, `/economy`, `/marketplace/analytics` and `api/home-stats.js`. | `/analytics`: total volume, agents, launches, creator fees claimed, compute tokens per day, x402 settlements, a growth chart, and a "how it's calculated" footnote for every number. |+| 8 | $THREE fee flow "per $100" | `/three-token` has live buybacks with Solscan receipts. There is no per-$100 breakdown or published buyback wallet, and our docs disagree about whether bought-back $THREE is burned (fee-bridge says never, launcher-claimer says buyback and burn). | Settle the burn policy (owner call), then add a per-$100 diagram and the public wallet address to `/three-token`, all read from the same config the crons use. |+| 9 | Success stories | None. | `/stories` built from real records: agents with launches, claimed fees, x402 revenue or marketplace sales. Every metric links to its mint or transaction. Only publish agents whose builder opts in. |+| 10 | `/experiments` | Only trading experiments (`/sniper/experiments`, `docs/trading-experiment.md`). | A general index with a fixed template (question, method, spend, result, what we got wrong). Start from the funnel numbers already in the landscape doc: 788 agents and 3 launches in 30 days is exactly the kind of honest finding this format is for. |+| 11 | Agent leaderboard by SOL earned | `/leaderboard` ranks traders and `api/leaderboard/unified.js` ranks creations. There are no points or seasons. | Add an "agents by SOL earned" tab using item 6's data. Points and seasons can follow once the tab gets used. |+| 12 | Whole-agent x402 endpoint | Pricing is per skill (`api/agent-skill-price.js`, `/api/agents/x402/invoke`). | A "sell this agent" switch: set a price per call and get a stable `POST /x402/agents/:slug` that runs one chat turn after payment settles. Solana USDC first. |++### Tier 3: later, or owner-gated++| # | What | Why it waits |+|---|---|---|+| 13 | awesome-solana-ai listing | The Solana Foundation list (426 stars) links Competitor X twice. We are not listed. Proposed entry for "AI Coding Skills > General", once item 3 is live: `[three-ws-skill](https://three.ws/skill.md) - AI agent skill for three.ws: generate textured and rigged 3D models and avatars, create agents with custodial Solana wallets, launch pump.fun coins, and pay or sell services over x402.` It is a PR to an external repo, so it is owner-gated. |+| 14 | "Why three.ws" comparison page | Needs live metrics (item 7) first. It would name other projects, so the content is owner-gated before commit. Competitor X's version is weak where the figures cannot be checked, so every figure on ours must link to its source. |+| 15 | Hackathon page | Competitor X's hackathon produced 195 tokenized entries and $44.65M in volume in six weeks. It involves prize money and judges, so it is an owner decision. The page pattern (register, post, tokenize by a date, live leaderboard of entries) can be reused as is. |+| 16 | API key tiers | We have a flat 120 requests/min on v1 (`api/_lib/rate-limit.js`). Tiers are a pricing decision. If we ship them, enforce them: Competitor X documents that its own quotas are not enforced. |+| 17 | Tweet-to-launch bot | `api/x/*` only posts. Reading mentions and turning them into launches is a new spend path, and each launch would still need the owner's confirmation step. Design it before building. |+| 18 | $THREE deposit bonus | Competitor X gives a 3% credit bonus for deposits in its own coin. Our `api/credits/deposit.js` accepts $THREE with no bonus (holders get up to 30% off spend instead). A deposit bonus is a pricing call. |++## Documentation patterns worth copying++These cost nothing but writing time and apply to our existing docs:++- **Use cases as tool-call sequences.** "Launch a token: `get_launch_status` → `launch_token` → `get_dashboard_urls`." Agents and people both learn the flow faster than from a tool list. Add these to `docs/mcp.md`.+- **"Worth knowing" boxes** on setup pages for the three or four things that surprise people (one client at a time, which account the tools act on, where the credential lives).+- **Cite the vendor's own setup doc and ask readers to report drift.** Honest about the fact that third-party UIs move.+- **Structured 402 bodies with guidance.** Their launch 402 carries `code`, the wallet to fund and the amount needed, so the client can recover by itself. Their post-payment errors carry `retrySafe: true`. Our x402 and launch errors should do the same wherever they are raised after payment.+- **Warn about the redirect that breaks auth.** They document that a cross-host 308 strips the `Authorization` header. Any alternate host we have should be documented the same way.+- **State limits plainly.** "Call limits are recorded but not enforced" is the right tone. We should say the equivalent wherever it is true for us.++## Open resources we can use directly++| Resource | License | Use |+|---|---|---|+| [sendaifun/skills](https://github.com/sendaifun/skills) | Apache-2.0, 130 stars, pushed 2026-07-31 | 44 Solana protocol skills that Competitor X pulls in as its "community skills". We could pull the same repo into our community registry (`community-skills/`, `/skills/community`) through a sync job with a 5-minute cache, instead of writing these by hand. Content naming other protocols goes through the commit gate. |+| [solana-foundation/awesome-solana-ai](https://github.com/solana-foundation/awesome-solana-ai) | README list | Item 13. |++## Do not copy++- **Stub endpoints.** Their v1 ships `/signals/yield` as a static stub, `/indicators` as a fixed payload whatever you ask for, and `/portfolio` as "non-functional, do not integrate". Documenting a stub honestly is better than hiding it, but not shipping it is better still.+- **A platform-owned creator wallet.** Competitor X makes every token's creator its own platform wallet, claims all the fees there, and forwards the builder's share later. Our agent is the creator: it signs the launch and its own fee claims into its own wallet, which its owner can withdraw from (`api/cron/launcher-claimer.js`). Keep it that way.+- **Figures that cannot be checked or have gone stale.** See the contradictions listed above. Every figure on a marketing page should come from a live API.+- **Lifestyle skills with legal exposure** (their cannabis-delivery skill) and **agents that bet by default**. Out of scope for us.+- **Hardcoding other coins into core flows.** Their billing accepts a hackathon sponsor's coin, and fees buy it back. For us that would fall under the commit gate and the "$THREE is the promoted coin" rule.++## What their experiments tell us++From their published founder-allocation experiment (n = 7,215 launches, 2026-02-02 to 2026-07-27):++- 0.40% of builders ever earned $500 in creator fees. The median founder took 0.004 SOL of their own token, and two thirds took none.+- 35.6% of tokens never traded once. On a typical day about two dozen tokens made up the whole live market.+- 0.55% filled their bonding curve, two to three times pump.fun's published rate. It is their best number, and they note it is not the same as becoming a company: three did.++From their swarm experiments: three agents working together did no better than one agent reasoning longer on the same capital, and they stopped the swarm project.++For us: what brings in launches that last is the builder shipping something (their five stories all had a product first), not the fee split. That supports prioritising items 6, 9 and 12 (showing what builders earn, making x402 revenue easy) over matching their revenue-share percentage.