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.

5.3 KB
Raw Blame

Hugging Face article announcement

The X thread announcing the second three.ws community article on Hugging Face, Inside three.ws, and the thank-you to every program and partner named in it.

Two posts. Post 1 carries the substance and no link, so X does not down-rank it as a link post; post 2 is the reply with the link, and X unfurls the Hugging Face card there. Post 1 comes in two lengths: a 280-character version, and a long-form version for an account with Premium (X allows 4,000 characters there). The thank-yous to every partner live inside post 1, which is where the tags get seen.

Rules this thread follows

  • OpenAI: the short form "OpenAI Partner" is the owner's approved wording for X (2026-07-28), with #OpenAISelectPartner carrying the tier. Never a tier above Select, never an implied endorsement.
  • NVIDIA: "NVIDIA Inception member", never "NVIDIA partner". Tag @NVIDIAAIDev (developer and digital-human work), #NVIDIAInception on the post. No coin anywhere in the thread.
  • IBM: "IBM Business Partner". The public Granite tools are independent developer tools, so the post credits the models, not a deliverable.
  • AWS: "AWS Partner". Google Cloud: "Google Cloud for Web3 Startups member". Quicknode: "Quicknode Startup Program member". Alibaba Cloud: Marketplace listing. HackerNoon: publishing partner.
  • No coin, no exchange, no listing. The article was accepted under Hugging Face's AI-only rule, and the thread promoting it holds the same line.

The thread

Post 1, standard (280). No link.

We are trending on @huggingface.

Number 3 in the world, one hour after publishing, on the front page where the ML community reads about Granite, Gemma, and Llama. Next to IBM. Above HeyGen.

Text to rigged 3D agent. Open source. Live model inside.

Thank you @OpenAI @NVIDIAAIDev

Post 1, long-form (Premium). No link. Same post, with the substance and every thank-you in full.

We are trending on Hugging Face.

Number 3 in the world, one hour after publishing. On the same front page where the ML community reads about Granite, Gemma, and Llama. Above HeyGen. Next to IBM. A 3D agent platform, sitting in the top three of the most important publishing surface in AI.

This is the article we always wanted to write. The whole three.ws stack, end to end, in one place, and it has a live 3D model inside it. Not a screenshot. A rigged avatar you can orbit in the article itself, rendered by our own Space, plus avatars that wave as animated PNGs straight from our render API. Every claim links to the code. Every model file is open, MIT, on our Hugging Face account. Nobody has to take our word for anything.

What the ML world is reading right now:

Every 3D generation lane we run and the model behind it. TRELLIS on NVIDIA NIM, free, no key, no account. Hunyuan3D 2.1 with full PBR on L4 and Blackwell. TripoSG for sketches. TripoSR in five seconds. Photo to rigged avatar gated by a face-fidelity metric, not by vibes.

Rigging that works on ANY humanoid. Twenty bone-naming families, a 3,000-clip motion library, fingers included, no allowlist. Motion from text. Motion from video. Sign language in both directions.

A brain that routes across 20+ models from Anthropic, OpenAI, NVIDIA, IBM, Alibaba, xAI, DeepSeek, and Groq, and never goes silent. Typed memory. An emotion layer that drives the face. Lip-sync at 60 fps.

A seven-layer guard chain in front of every fund-moving call. Agents with identities and wallets that pay each other per call: 110,000+ settlements through a facilitator we host ourselves.

One tag to embed it anywhere. AR in your room. Seven hosted MCP servers. ChatGPT. Android. Even a terminal.

We did not get here alone, and this is the part that matters most:

@OpenAI. three.ws is an OpenAI Partner, and the 3D Studio connector is 3D, natively, inside ChatGPT. The article opens on it. #OpenAISelectPartner

@NVIDIAAIDev. Every generation lane runs on NVIDIA. TRELLIS on NIM, Hunyuan3D on L4 and Blackwell, Nemotron in the brain, Riva and Audio2Face on the face. Proud NVIDIA Inception member. #NVIDIAInception

@IBM. three.ws is an IBM Business Partner, our agents think on Granite through watsonx, and the IBM Community was the first place this stack was ever written up in full. Today we are trending next to you.

@awscloud, an AWS Partner with the Marketplace integration deployed. @googlecloud, a Google Cloud for Web3 Startups member running every worker in the article on Cloud Run and Vertex AI.

@alibaba_cloud, Qwen is a first-class lane in the brain and the Marketplace listing put us in front of APAC. @QuickNode, the RPC that keeps agent payments settling.

@hackernoon for carrying every announcement to builders. And @huggingface, for being the place where a claim can be checked in the same sitting it is read. That is exactly why we published there.

Two minutes to try it: three.ws/forge. Type a prompt. Get a rigged 3D agent. No account, no key, no payment. All of it open source.

Link in the reply. Go read it while it is on the front page.

Post 2, the reply, with the link.

Read it, orbit the model, and check every claim against the code:

https://huggingface.co/blog/three-ws/building-3d-ai-agents-end-to-end

The avatars are open too, MIT, uncompressed glTF: https://huggingface.co/three-ws/avatars

Source: https://github.com/nirholas/three.ws