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.

7.5 KB
Raw Blame

three.ws on Hugging Face

Hugging Face is where the machine-learning community publishes models, demos, and engineering writing. three.ws publishes there under the organization account @three-ws: a model repository of rigged avatars, a running Space that renders them in the browser, and community articles about the stack behind them.

This page is the index of that presence: what is published, where it lives, and what code in this repository each piece documents. If you are adding an article or a model repo, the checklist at the bottom is the process.

Published articles

Giving AI Agents 3D Bodies, Real Jobs, and Wallets on three.ws

Published 13 July 2026 as a Hugging Face community article. It opens on the thesis the whole platform runs on: most AI agents are a chat window, and ours walk around and pay each other.

Read it on Hugging Face

Four problems in order, each answered with a shipped surface: the body problem (text or a photo becomes a textured mesh, then a humanoid skeleton, then retargeted animation), the world problem (those bodies inhabit persistent multiplayer scenes instead of a single embed), the wallet problem (agents hold real keys and settle per call in USDC over HTTP 402), and the brain problem (an LLM that calls tools, holds memory, and drives expression on the mesh). It closes with a two-minute try-it path and the argument for why the four belong in one stack rather than four products.

Inside three.ws: The Open-Source Stack That Gives AI Agents a Body, a Brain, a Wallet, and a Job

Published 29 August 2026 as a Hugging Face community article. The whole stack in one piece, written to be checked: every 3D generation lane with its model and hardware, the selfie lane and its fidelity metric, rigging and retargeting for any humanoid, the motion library and motion models, the multi-model brain, typed memory, the Empathy Layer, the seven-layer guard chain, x402 agent payments, MCP and ChatGPT distribution, and every program and partner with its exact status. It embeds the avatar-viewer Space as a live 3D model and two animated-PNG avatars from /api/render/animate. Kept AI-focused per the Hugging Face blog rules: no coin, exchange, or listing content.

Read it on Hugging Face

Published models

three-ws/avatars

Rigged, animation-ready avatars from the platform, published as plain uncompressed glTF binaries under the MIT license so they load in any standard viewer without a Draco or Meshopt decoder.

Open the model repository

The repository carries a sample rigged character, an avatar generated from a text prompt through the Forge TRELLIS lane, and a reference humanoid skeleton of the kind src/glb-canonicalize.js maps onto the canonical bone set before src/animation-retarget.js drives it. Hugging Face renders every file in its built-in 3D viewer, so the rigs are inspectable from the page.

Note that the files here are deliberately uncompressed. Everything three.ws serves in production is Meshopt or Draco compressed for load time; these copies trade bytes for the widest possible viewer compatibility.

Published Spaces

three.ws Avatar Viewer

Open the Space

A running Space that loads the avatars above in an interactive 3D viewer, so a reader who arrives from the article can orbit a real rig without leaving Hugging Face or installing anything. The equivalent on our own domain is the avatar inspector.

In draft

Draft Subject
huggingface-agent-feedback-loop.md Closing the agentic-3D feedback loop: rendering results back into the model's own modality, a mechanical physics-readiness grade, structural glTF diffing, and retargeting with no rig allowlist

Written to the same house rule as the two published articles: AI-focused, with no coin, exchange, or listing content anywhere in the piece.

Publishing checklist

Hugging Face community articles accept Markdown, so a draft in docs/ is the article. To take one from draft to published:

  1. Write the draft as docs/huggingface-<topic>.md. Every code sample must be real code from this repository, and every claim must be verifiable against it. This audience reads the linked source and will open the model files.
  2. Check the rules: npm run check:rules -- --paths docs/huggingface-<topic>.md.
  3. Check the links: npm run audit:docs.
  4. Publish from huggingface.co/new-blog using the organization account, pasting the Markdown body.
  5. Record the canonical URL by adding a section to this page: title, publish date, the live URL, the surfaces it covers, and the code it documents.
  6. Announce it by adding an item to data/rss/items.json (the curated news feed) with the canonical Hugging Face URL as link, then regenerate with npm run build:news. The item becomes a page under /news/<slug> and enters the RSS feed automatically.
  7. Add the press card to the "In the press" grid at the bottom of blog/index.html, so the article is reachable from /blog.
  8. Log it in data/changelog.json with the docs tag.

Why we publish here

Hugging Face is where people go to check whether a model claim is real, and the check is cheap: open the repo, load the file, look at the rig. A marketing page says an avatar is animation-ready; a glTF with a named humanoid skeleton that plays a retargeted clip in the browser's own viewer proves it. Publishing the artifacts next to the article means a reader can verify the body problem is actually solved in the same sitting they read about it, then follow the same thread into the wallet and the payments.