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
// Declarative scanner: reads `data-agent-*` attributes from a root element
// and wires the corresponding behaviors. Lets a static HTML page describe its
// avatar interactions without writing JS.
//
// Supported actions:
// data-agent-action="stand-on" — park the avatar here on ready
// data-agent-action="track-typing" — focus walks to + caret tracking
// data-agent-action="privacy-mode" — focus walks to + covereyes (once+hold)
// data-agent-action="navigate-on-click" — runOff before following href
// data-agent-action="react-on-click" — play a clip on click
//
// Optional per-element attributes:
// data-agent-direction="left|right" — runOff direction (default right)
// data-agent-clip="<clipName>" — clip to play
// data-agent-anchor="<anchorName>" — anchor for walkTo target
// data-agent-delay="<ms>" — pre-navigation delay
// data-agent-loop="true|false" — loop flag for react-on-click clip
// data-agent-hold="true|false" — clampWhenFinished for react clip
import { startCaretTracking, caretScreenX } from './caret.js';
export function scan(root, agent) {
const cleanups = [];
let activeInput = null;
let stopTracking = null;
function clearTracking() {
stopTracking?.();
stopTracking = null;
activeInput = null;
}
const els = root.querySelectorAll('[data-agent-action]');
els.forEach((el) => {
const action = el.dataset.agentAction;
const dir = el.dataset.agentDirection === 'left' ? 'left' : 'right';
const clip = el.dataset.agentClip;
const anchor = el.dataset.agentAnchor;
const delay = el.dataset.agentDelay ? parseInt(el.dataset.agentDelay, 10) : undefined;
const loop = el.dataset.agentLoop === 'true';
const hold = el.dataset.agentHold === 'true';
switch (action) {
case 'stand-on': {
agent.whenReady(() => agent.standOn(el, anchor ? { anchor } : undefined));
break;
}
case 'track-typing': {
const onFocus = async () => {
clearTracking();
await agent.walkTo(el, anchor ? { anchor } : undefined);
agent.play('lookdown', { loop: true });
activeInput = el;
stopTracking = startCaretTracking(el, (cx) => agent.lookAt(cx), () => activeInput);
};
const onBlur = () => { clearTracking(); agent.faceFront(); };
const reaim = () => { if (activeInput === el) agent.lookAt(caretScreenX(el)); };
el.addEventListener('focus', onFocus);
el.addEventListener('blur', onBlur);
el.addEventListener('input', reaim);
el.addEventListener('keyup', reaim);
el.addEventListener('click', reaim);
cleanups.push(() => {
el.removeEventListener('focus', onFocus);
el.removeEventListener('blur', onBlur);
el.removeEventListener('input', reaim);
el.removeEventListener('keyup', reaim);
el.removeEventListener('click', reaim);
});
break;
}
case 'privacy-mode': {
const onFocus = async () => {
clearTracking();
await agent.walkTo(el, anchor ? { anchor } : undefined);
agent.play(clip || 'covereyes', { loop: false, hold: true });
agent.faceFront();
};
const onBlur = () => agent.play('idle', { loop: true });
el.addEventListener('focus', onFocus);
el.addEventListener('blur', onBlur);
cleanups.push(() => {
el.removeEventListener('focus', onFocus);
el.removeEventListener('blur', onBlur);
});
break;
}
case 'navigate-on-click': {
cleanups.push(agent.interceptNavigation(el, { direction: dir, delay }));
break;
}
case 'react-on-click': {
const onClick = () => {
if (!clip) return;
agent.play(clip, { loop, hold });
};
el.addEventListener('click', onClick);
cleanups.push(() => el.removeEventListener('click', onClick));
break;
}
}
});
return () => { cleanups.forEach((fn) => fn()); clearTracking(); };
}