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
/**
* Bring-it-alive launcher for Studio — Pillar 2.
*
* Turns anything Studio produces (a forged avatar, a Lab GLB) into a live,
* talking agent by handing it to the platform's real conversational stack:
* openTalkMode() → TalkController → /api/chat (LLM, SSE) → /api/tts/speak
* (NVIDIA Magpie → OpenAI) → lip-sync driver on the avatar's mouth.
* Per-agent memory rides along through /api/agent-memory when an agent_id is
* present. No mocks — this is the same runtime the rest of the site uses.
*
* The avatar-object mapping is a pure function (buildTalkAvatar) so it can be
* tested in isolation; launchTalk just resolves the real opener and calls it.
* setTalkOpener() exists purely as a test seam.
*/
const UUID_RE = /^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$/i;
const isUuid = (v) => typeof v === 'string' && UUID_RE.test(v);
let _opener = null;
/** Test seam — inject a fake openTalkMode. */
export function setTalkOpener(fn) {
_opener = fn;
}
function defaultPersona(name, kind) {
if (kind === 'object') {
return [
`You are "${name}", a 3D object that has just been brought to life in three.ws Studio.`,
'Play along as this object with warmth and a light sense of humor.',
'Keep replies to one or two short spoken sentences — this is a voice conversation.',
].join(' ');
}
return [
`You are "${name}", a 3D avatar a creator just made in three.ws Studio.`,
'Speak in first person, with personality and warmth. You have a body and can move and emote.',
'Keep replies to one or two short spoken sentences — this is a voice conversation.',
].join(' ');
}
/**
* Build the {avatar, systemPromptFn} payload openTalkMode expects.
* @param {{name?:string, glbUrl?:string, glbBlob?:Blob, id?:string, agentId?:string, persona?:string, kind?:'avatar'|'object'}} opts
*/
export function buildTalkAvatar(opts = {}) {
const { name, glbUrl, glbBlob, id, agentId, persona, kind = 'avatar' } = opts;
if (!glbUrl && !glbBlob) {
throw new Error('Nothing to talk to yet — generate a model first.');
}
const displayName = name || (kind === 'object' ? 'Your creation' : 'Your avatar');
const systemPrompt = persona || defaultPersona(displayName, kind);
const avatar = { name: displayName };
if (glbUrl) { avatar.model_url = glbUrl; avatar.url = glbUrl; }
if (glbBlob) avatar.glbBlob = glbBlob;
if (isUuid(id)) avatar.id = id;
if (agentId) avatar.agent_id = agentId;
return { avatar, systemPromptFn: () => systemPrompt };
}
/**
* Resolve the real talk-mode opener (lazy — the stack is heavy) and launch.
* @returns the active talk session, or null if the opener declines.
*/
export async function launchTalk(opts) {
const built = buildTalkAvatar(opts);
const open = _opener || (await import('../voice/talk-mode.js')).openTalkMode;
return open(built);
}