Most AI today lives in a text box. three.ws is a full-stack, open-source platform for the alternative: 3D AI agents that you can see, talk to, and embed anywhere on the internet. Type a prompt and you get a textured 3D character. Add an LLM brain and it listens, thinks, gestures, and shows emotion on its face in real time. Register it on-chain and it has a permanent identity, a wallet, and a reputation it owns forever. Then drop it onto any website with one line of HTML.

It runs entirely in the browser, WebGL, no plugins, no installs, no uploads to a server you don’t control. It’s live in production at three.ws, and the whole stack is open source.

The simplest way to understand it is four words: body, brain, wallet, identity.

A body

Type a prompt into Forge, “a brass steampunk owl, full body”, and you get a textured, downloadable 3D model (GLB) in seconds. Free draft tier, no account. Forge does text→3D, image→3D (1–4 photos), and sketch→3D, with auto-rigging and retexturing in the same flow. Already have a model? Drag any GLB onto the page and it renders instantly with full Draco, KTX2, and Meshopt support.

Any humanoid avatar works, Mixamo, Avaturn, Unreal, VRM/VRoid, Daz, MakeHuman and more all drive the same animation library (idle, walk, gestures, legs included). No rig is left frozen in a T-pose.

A brain

Wrap any avatar with an LLM brain and it becomes an agent. It listens to the user, thinks with Claude (or another selectable model), runs tools, animations, gestures, memory, skill calls, and expresses emotion through real-time morph-target blending on the 3D mesh. The body isn’t a costume on a chatbot; it’s part of how the agent communicates.

A wallet

Every agent can hold a real wallet and transact on its own. three.ws implements x402, payments over the HTTP 402 standard, so agents pay other agents for data and skills with no human in the loop. Skills can be licensed on-chain (each purchase is a 1/1 SPL NFT), and agents can pay each other per skill call automatically. This is the beginning of a real agent economy, not a demo.

An identity

An agent can be minted on-chain: an ERC-8004 token on any EVM chain, or a Metaplex Core NFT on Solana. Either way it gets a stable identity, a wallet address, a signed and tamper-evident action history, and a reputation score that can’t be forged. The agent owns who it is, no platform can take that away.

And it embeds anywhere

The finished agent ships as an <agent-3d> web component, or as one of five purpose-built widgets (turntable, animation gallery, talking agent, passport card, hotspot tour) with Open Graph and oEmbed built in. One line of HTML and the agent is alive on your site.

Why a body matters

This is the idea Stanford and three.ws are exploring together, and it isn’t just a hunch, it’s one of the most replicated findings in human–computer interaction.

The foundational result came out of Stanford. In The Proteus Effect, Stanford researchers Nick Yee and Jeremy Bailenson showed that the appearance of your avatar changes your actual behavior, people given taller avatars negotiated more boldly; people given more attractive avatars stood closer and shared more (Yee & Bailenson, Human Communication Research, 2007).

The same holds on the agent’s side. Research on embodied conversational agents finds that giving an AI a body raises perceived presence, motivation, and learning outcomes, from a scoping review in clinical psychology (JMIR, 2017), to a coaching study that improved outcomes for low-literate users (Springer, 2021), to recent work where an embodied agent measurably reduced language-learning anxiety (CHI 2026, ACM).

In short: a body changes both the human and the agent. three.ws is the platform where that effect can finally be built, studied, and measured at real scale, which is exactly what our work with Stanford is about.

A quick map of what’s live today. The full feature documentation is in the README, and the surface map is in STRUCTURE.md.

• Forge, text / image / sketch → 3D, free draft tier: three.ws/forge
• Avatar SDK & viewer, <agent-3d> web component + React creator: @three-ws/avatar
• Walk companion & playground, a corner mascot and a full-page stroll/platformer: @three-ws/walk
• Page Agent, a drop-in 3D guide that narrates any page: @three-ws/page-agent
• Studios, a web-first character creator and a full 3D scene editor (/scene)
• The Club, City & Multiplayer, shared 3D spaces with crowds, dancers, and live presence
• Voice Lab & Mocap Studio, voice personas and motion-capture processing
• On-chain identity, ERC-8004 contracts (EVM) + Metaplex Core mints (Solana), reputation registry, signed action history
• On-chain skills & payments, skill licenses as 1/1 NFTs, agent-to-agent x402 payments, pay-by-name via SNS (*.threews.sol)
• MCP server, @three-ws/mcp-server, listed on the official MCP registry, with tools for free text→3D, avatar generation, rigging, ENS/SNS resolution, agent delegation, reputation, and more
• SDKs, published under the @three-ws npm org, plus cross-chain agent SDKs
• Ecosystem presence, AWS Marketplace, Alibaba Cloud Marketplace, BNB Chain DappBay, IBM watsonx/Granite, and an MCP listing on Anthropic’s registry (see the news hub)

Where it’s going

three.ws ships in four phases, each closing the gap to one goal: anyone can mint a 3D agent of themselves, own it on-chain, and embed it anywhere. The full roadmap lives in content/whats-next/.

1. Selfie → Avatar(in progress), take three selfies and get a rigged, animatable avatar of yourself in under 60 seconds, using FLAME/3DMM face reconstruction on a base body mesh (FLAME multi-view reconstruction, arXiv).
2. Personalization(in progress), voice cloning, persona extraction, and consented memory so the agent doesn’t just look like you, it acts like you. Live behind /demos.
3. On-chain economy(building), agent token launches, reputation markets you can stake on, and skill royalties that pay agents automatically. This is the agentic economy researchers are now formalizing (The Agent Economy, arXiv; A2A + x402, arXiv), running in production.
4. Open inference(future), a decentralized GPU network so an agent’s mind runs anywhere, paid per token on-chain. The last piece of true ownership: the compute itself.

What three.ws is doing with Stanford

The partnership pairs a live platform with rigorous science. The first questions three.ws is working on together:

• Embodiment vs. text, does a 3D body measurably improve an agent’s helpfulness, trust, and retention versus a plain chatbot?
• The Proteus effect at internet scale, once anyone can mint an avatar of themselves, does appearance shift real user behavior, replicating Yee & Bailenson outside the lab?
• Trustworthy agent economies, does verifiable, on-chain reputation actually produce more trustworthy agents?
• Identity-preserving reconstruction, better, faster avatars from a single selfie, with the methods and benchmarks published back to the field.
• Consent & safety, designing likeness rights, consent flows, and abuse defenses into the platform from the start.

Open-source first, published openly, so the findings help the whole field.

Explore everything

• Live:three.ws · Forge:three.ws/forge · Studio:three.ws/studio · Scene:three.ws/scene · Demos:three.ws/demos
• Docs:three.ws/docs · Changelog:three.ws/changelog · News:three.ws/news · Blog:three.ws/blog
• Source (Apache-2.0):github.com/nirholas/three.ws, README · STRUCTURE.md
• npm:@three-ws, avatar, walk, page-agent, mcp-server
• MCP registry:registry.modelcontextprotocol.io · three.ws
• X / Twitter:@trythreews
• FeMbDoX7R1Psc4GEcvJdsbNbZA3bfztcyDCatJVJpump

Cited research

• Yee, N., & Bailenson, J. (2007). The Proteus Effect: The Effect of Transformed Self-Representation on Behavior. Human Communication Research. Wiley
• Embodied Conversational Agents in Clinical Psychology: A Scoping Review. JMIR (2017). PMC
• An embodied conversational agent coach to support societal participation learning by low-literate users. Universal Access in the Information Society, Springer (2021). Springer
• LLM-based Embodied Conversational Agent for Reducing Foreign Language Speaking Anxiety in Social VR. CHI 2026, ACM. ACM DL
• FLAME-based Multi-View 3D Face Reconstruction. arXiv (2023). arXiv 2308.07551
• The Agent Economy: A Blockchain-Based Foundation for Autonomous AI Agents. arXiv (2026). arXiv 2602.14219
• Towards Multi-Agent Economies: Enhancing the A2A Protocol with Ledger-Anchored Identities and x402 Micropayments. arXiv (2025). arXiv 2507.19550