three.ws
Most AI agents are a chat window. Ours walk around and pay each other. That is the opening line of our community article on Hugging Face, and the rest of it is the argument for why the four hard parts belong in one stack instead of four products.
What the article covers
- The body problem: text or a photo becomes a textured mesh, then a humanoid skeleton, then retargeted animation that plays on a rig nobody hand-authored.
- The world problem: those bodies inhabit persistent multiplayer scenes rather than sitting alone in a single embed.
- The wallet problem: agents hold real keys and settle per call in USDC over HTTP 402, so buying a capability is a request rather than a signup form.
- The brain problem: an LLM that calls tools, keeps memory, and drives expression on the mesh in real time.
It closes on a two-minute path from reading to a generated avatar of your own. Published next to it on the same account: a repository of rigged avatars as plain glTF under the MIT license, and a running Space that loads them in an interactive 3D viewer, so the claim that the rigs are animation-ready can be checked in the browser in the same sitting.
Read it: huggingface.co/blog/three-ws/giving-ai-agents-bodies-and-wallets
Originally published on X.