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.
Build an animated agent in 10 minutes
This guide begins with the working character already visible in the starter. Each step changes one thing and keeps the page runnable.
1. Understand the two tags
Open index.html. The module script registers the agent-3d web component. The element below
it supplies a model URL and a display name. No framework or bundler is involved.
2. Change the agent's identity
Change the element's name to a role that belongs on your site. Update the introduction in
app.js to match. Refresh the page and press Introduce yourself.
3. Create a model
From the terminal, describe one full-body character in a neutral pose:
npm create @three-ws/agent "a friendly museum guide in a navy uniform"
Generation and rigging normally take one to three minutes. The command reports real elapsed time, downloads the finished GLB, and prints its hosted URL.
4. Put the model on the page
Replace the body value in index.html with that hosted URL. To serve the downloaded model
from the starter instead, copy agent.glb beside index.html and use:
<agent-3d body="./agent.glb" name="Museum Guide"></agent-3d>
5. Add behavior
The starter waits for the component to load before enabling controls. Add another button and call one documented method, such as:
await agent.playClip('idle', { userInitiated: true });
agent.speak('What would you like to explore?');
Use userInitiated: true only in response to a real user gesture. Ambient animation should
respect reduced-motion preferences.
6. Ship it
The starter is static. Any host that serves index.html, app.js, styles.css, and the model
can run it. Before publishing, run:
npm test
Share the finished URL in the three.ws Show and tell discussion.