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.

2.1 KB
Raw Blame

Coach Leo

A complete, minimal example of a three.ws agent defined entirely as files: a manifest pointing at a 3D body, a brain, a voice, a memory policy, and one installed skill. Coach Leo is a football coach who waves when you greet him, comments on your form, and remembers what you tell him across sessions.

Use this directory as the template for your own agent: copy it, swap the GLB, rewrite the instructions, and pick your skills.

Files

File Role
manifest.json The agent definition (agent-manifest/0.1): body GLB, brain model, voice, skills, memory, tools.
instructions.md The system prompt. Frontmatter (name, model, temperature) mirrors the manifest; the body defines personality and tool-usage rules.
SKILL.md Top-level capability summary surfaced to hosts that index agents by skill.

How the manifest is wired

  • Body: /avatars/cz.glb, a Mixamo-rigged GLB served by the main app (boundingBoxHeight: 1.78).
  • Brain: Anthropic claude-opus-4-6 with instructions.md as the prompt, temperature 0.8.
  • Voice: browser TTS and STT (en-US), speech rate 1.05.
  • Skills: one entry, the starter wave skill at ../skills/wave/. The instructions tell the model to call wave() on greetings.
  • Memory: mode: "local" with an 8192-token budget, indexed at memory/MEMORY.md.
  • Tools: wave, lookAt, play_clip, setExpression, speak, remember.

Run it

From the repo root:

npm run dev

Then load the manifest into any agent surface, for example the web component:

<agent-3d manifest="/examples/coach-leo/manifest.json"></agent-3d>

Greet him. Per instructions.md he calls wave(), sets a focused expression while explaining drills, and calls remember() when you share your position, goals, injuries, or schedule.

Related