When AI Steps Out of the Chat Box: three.ws Begins Shaping Bodies for Agents.
Written by: KarenZ, Foresight News
A collaboration announcement with IBM catapulted Solana ecosystem project three.ws into the market spotlight, and its token three saw a 50x surge.
The direct catalyst for the price surge came from IBM. Between the evening of June 1 and the early hours of June 2, IBM's official Twitter account responded twice to content about the collaboration posted by three.ws.
As the news spread, three.ws' token three rose rapidly. According to GMGN data, three's market cap jumped from around $300,000 before IBM's response to a high of $16.38 million on June 4, a surge of up to 53x. Currently, three's market cap is fluctuating around $13 million.
It's clear this isn't a rally that has been building since the token's launch. three was issued on the Solana chain as early as late April, with most of its gains concentrated around IBM's public response and the release of the collaboration announcement.
However, if we only view three.ws as a Solana AI project endorsed by IBM, we miss the real problem it aims to solve: most AI Agents today are still hidden in chat boxes and backend programs—users can't see them, and it's hard to identify, own, or invoke them across different websites, devices, and on-chain environments.
three.ws hopes to equip AI Agents with bodies, memories, identities, wallets, and distribution channels, turning them into 3D digital characters that can appear on web pages, perform actions, and conduct transactions.
Moving AI from Chat Boxes to Web Pages
three.ws defines itself as the "3D Agent Layer" of the internet. Founder @nichxbt currently has over 20,000 followers on Twitter and is verified with a blue checkmark.
The project is currently available on AWS Marketplace and Alibaba Cloud International, and has joined the Google Cloud for Web3 Startups program. three.ws is also listed in Anthropic's official MCP Registry, is a W3C contributor, and a participant in the Solana Frontier Hackathon.
From the project's existing code and documentation, three.ws's core capabilities include loading, inspecting, and displaying 3D models on the web, followed by integration with large language models, memory, voice, skills, on-chain identity, and payment functions.
Simply put, developers can create a 3D character on the platform, connect it to large language models, memory systems, voice, and skills, then embed it into a website via an
For example, businesses can deploy a 3D shopping guide Agent on product pages to introduce products, answer questions, and demonstrate product features through actions. Developers can also build digital customer service agents, virtual teachers, game characters, or personal AI assistants.
This process is somewhat similar to embedding a YouTube video. Developers don't need to build complex 3D pages separately; after adding the component and Agent ID, users can see and interact with the Agent in their browser.
three.ws offers multiple ways to create characters. Users can upload a selfie to generate an animation-capable 3D avatar in about 60 seconds; they can also generate models via text or images, upload their own GLB or glTF files, or use the character editor to create one.
After generating the character, developers can configure it with different large language models, voices, and skills.
three.ws also integrates on-chain capabilities for Agents. Agents can have Solana wallets, pay for paid interfaces using USDC via the x402 protocol, and register their identities as Metaplex Core assets on Solana or via ERC-8004 on EVM chains. It's important to distinguish between identity and funds here: on-chain identity proves who the Agent belongs to and where its data points to, while the wallet handles payments and transaction execution.
How Does a 3D Agent Work?
three.ws consists of four independent, replaceable technical layers. Developers can use all four layers in combination or only some of them.
The bottom layer is the Viewer layer, or rendering layer.
This layer is responsible for loading and displaying 3D models in the browser, including lighting, cameras, materials, and animations. It is built on three.js and does not know whether there is AI, a wallet, or on-chain identity behind the model. Therefore, even without integrating an Agent, the Viewer can act as a standalone 3D model viewer.
The second layer is the Agent layer, which is the character's brain and behavior system.
After a user inputs content, the large language model makes a judgment based on the character's settings, historical memory, and installed skills. If the user asks the character to wave, the model calls the corresponding tool, and the scene controller plays the waving animation; if the character needs to remember something, the memory module saves the relevant information.
This layer also includes an emotion system. The character can change its expression, gaze, and actions based on events—for example, appearing happier after completing a task or worried when an operation fails.
The third layer is the Identity layer, which is an optional module.
The Identity layer is responsible for maintaining the Agent's consistent identity across different websites, devices, and sessions. The Agent's data, memory mode, and resource addresses can be written into a Manifest file and stored via IPFS or the platform's servers.
According to three.ws's official documentation, its Solana Agent supports on-chain identity registration via Metaplex Core, but the on-chain reputation registration and verification registry related to ERC-8004 are currently only available on the EVM side, and the verification registry is still in the testnet phase.
The fourth layer is the Embedding and Distribution layer.
This layer is responsible for bringing the Agent to users. Developers can add the character to websites, apps, and enterprise interfaces via web components, iframes, Widgets, or SDKs.
To put it simply, the rendering layer handles the body and actions, the Agent layer provides the brain, memory, and skills, the identity layer offers an optional digital passport, and the embedding layer is responsible for deploying the character to websites and apps.
three.ws's pricing system should be understood from two dimensions: through which channels do users use or purchase services? And what fees are incurred after using the platform?
Regarding purchase/usage channels: users can subscribe directly, choosing between the free version, Pro ($49/month), or enterprise plan. AWS Marketplace is an enterprise procurement channel.
After subscribing to three.ws, developers can use x402 to set prices for the Agent's chat, content generation, or API calls, and the caller pays per use in USDC. Platform fees are deducted from the developer's revenue: the free version has a 0% platform fee during public testing (the rate after testing has not been announced); the Pro version has a 2.9% platform fee; and the enterprise version's platform fee is customized according to the agreement.
IBM Completes Enterprise Capabilities, AWS and Other Platforms Handle Distribution
For 3D AI Agent projects, creating a demoable digital character is not difficult. The real challenge lies in getting the product into the enterprise procurement system and meeting requirements such as billing, deployment, identity verification, and AI governance.
three.ws is supplementing these aspects through platforms like IBM and AWS Marketplace.
On May 27, three.ws announced it had become a member of the AWS Partner Network (APN) and subsequently launched on AWS Marketplace. This means enterprise customers can purchase three.ws services through their existing AWS accounts.
Later, three.ws published a technical article about SaaS product billing on the AWS Builder Center blog. This solution connects AWS Marketplace's customer verification, usage-based billing, and subscription management with the on-chain x402 payment interface.
Regarding the IBM collaboration: three.ws plans to integrate its 3D Agent technology with IBM's enterprise AI, hybrid cloud, and market channels, and connect to IBM's Granite series models for scenarios such as conversational AI, image understanding, semantic matching, market forecasting, and enterprise governance.
AWS Marketplace helps three.ws enter the enterprise procurement and billing system, while IBM provides enterprise AI technology and business channels. Both partnerships aim at the same goal: turning the 3D AI Agent in the browser from an eye-catching demo into a service that enterprises can purchase, deploy, and manage.
In a bear market, IBM's public response gave three.ws scarce market attention.
However, after the hype, the project still needs to answer more practical questions: do enterprises and developers really need an AI Agent with a body, skills, and digital identity? And what role will three ultimately play in this system?