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.

6.1 KB
Raw Blame

Integrations: drop-in 3D agents for any site

Integrations is the catalog of ways to put a three.ws agent on a website you do not control the runtime of. One script tag gives any page a 3D avatar viewer, a talking chat agent, a walk companion that roams the corner of the screen, a page narrator, a live token market widget, or an MCP server that lets Claude and other AI tools drive your agents. Every option is a paste-in snippet with no backend to run.

Page: /integrations

Why it exists

The platform's agents are only as useful as the places they can appear. Integrations is the distribution layer: it turns an agent, a saved widget, or a live token feed into something a marketer can paste into a CMS, a developer can wire into a framework, or an AI assistant can call as a tool. It is the front page of embedding; the deep per-method references live in Embedding and Widgets, and this doc is the map that points you to the right one.

How it works

Every integration is a static, self-contained snippet loaded from three.ws. The avatar, chat, and token widgets share one loader (/embed.js) that upgrades custom elements or data-widget script tags into live iframes or web components. The walk companion and page narrator ship their own small SDKs. Nothing needs an API server on your side; the agent runtime, rendering, and (for chat) the model calls all run on three.ws.

The catalog covers six integration types:

  • Avatar Viewer: a <threews-avatar> web component that renders a rigged 3D avatar with optional JS control (play('wave'), poses, hide-chrome).
  • Talking Chat Agent: a saved widget that renders a 3D agent which listens, answers, and speaks, driven by a data-widget id from Widget Studio.
  • Walk Companion: a floating 3D character fixed to a corner of the page that walks, gestures, and speaks, loaded from /walk-embed-sdk.js.
  • Page Narrator: an agent that reads and narrates the page it lives on, shipped as the @three-ws/page-agent package.
  • Token Market Widget: a live pump.fun-style token feed rendered as a widget, driven by a data-widget id and the pumpfun-feed type.
  • MCP Integration: the @three-ws/avatar-agent server (from the packages/avatar-agent-mcp workspace), so Claude, Cursor, or any MCP-compatible assistant can render, speak, gesture, and emote your agents. See MCP.

Walkthrough

  1. Open /integrations. Filter the catalog by type (avatar, chat, companion, narrator, token, MCP).
  2. Try the live demo. Each card has a working preview and a Try Live button that opens the real embed in a modal.
  3. Copy the snippet. Every card exposes its exact paste-in code.
  4. Swap in your ids. Replace the demo widget or avatar id with your own from Widget Studio.
  5. Paste it into your site. Drop the snippet anywhere in your HTML, CMS block, or framework component.

Examples

Avatar Viewer (web component):

<!-- Load once -->
<script src="https://three.ws/embed.js" defer></script>

<!-- Drop anywhere in your HTML -->
<threews-avatar avatar-id="a4bad2f5-8a07-43cf-82e5-b6ba1314441e" hide-chrome pose="idle"></threews-avatar>

<!-- Optional: JS control (type="module" so top-level await is allowed) -->
<script type="module">
  const el = document.querySelector('threews-avatar');
  await el.ready;
  el.play('wave');
</script>

Talking Chat Agent (saved widget):

<script
  src="https://three.ws/embed.js"
  data-widget="wdgt_demo_talking"
  data-type="talking-agent"
  defer
></script>
<!-- Replace wdgt_demo_talking with your widget ID from three.ws/widgets -->

Walk Companion:

<script
  src="https://three.ws/walk-embed-sdk.js"
  data-avatar="a4bad2f5-8a07-43cf-82e5-b6ba1314441e"
  data-position="bottom-right"
  data-env="studio"
  defer
></script>

Page Narrator:

<script
  src="https://cdn.jsdelivr.net/npm/@three-ws/page-agent/dist/page-agent.global.js"
  data-page-agent
  data-avatar="nova"
  data-auto-narrate
  defer
></script>

Token Market Widget:

<script
  src="https://three.ws/embed.js"
  data-widget="wdgt_demo_pumpfun"
  data-type="pumpfun-feed"
  data-width="420"
  data-height="600"
  defer
></script>

MCP Integration (claude_desktop_config.json):

{
  "mcpServers": {
    "avatar-agent": {
      "command": "npx",
      "args": ["-y", "@three-ws/avatar-agent"]
    }
  }
}

The core 3D tools (inspect_glb, validate_glb, optimize_glb, thumbnail_glb, viewer_url) and pump_snapshot work with no environment variables. Voice, generation, and signing tools take their own provider keys via env (for example SOLANA_RPC_URL, OPENAI_API_KEY, REPLICATE_API_TOKEN); see the package README for the full table.

States and limits

  • Get your own ids. Demo ids in the snippets are live examples. Create your own avatars and widgets in Widget Studio and swap them in.
  • Self-contained by design. The embeds run entirely off three.ws; you host no backend. The chat and token widgets make their model and data calls server-side.
  • One loader, many widgets. /embed.js can be loaded once and drive multiple avatar, chat, and token embeds on the same page.
  • MCP keys are per-tool. The @three-ws/avatar-agent server needs no key for its core 3D tools; voice, generation, and signing tools each read their own provider key from the environment. The embeds are public and need no key.
  • Deeper control lives elsewhere. This page is the catalog. For the full web component API, iframe and oEmbed options, CSS, and events, use the references below rather than duplicating them here.

Related