Interoperability of Intelligent On-Chain Agents Originally published January 2025, this led to the development of ThinkOS
AI Agents represent an evolution of web applications. Just as the ERC-20 standard enabled tokens to flow freely across networks and applications, a new open standard for interoperability is essential for AI Agents to become native citizens of the open, permissionless Internet.
The Think Agent Standard enables intelligent on-chain agents with wallets that operate natively across the Internet and blockchain networks. It defines a composable architecture for agents built on modular and permissionless innovation.
By providing a framework for minting Non-Fungible Intelligence (NFIs), connecting agent components and tools, and incentivizing network & resource sharing, we establish AI Agents as an atomic unit of the new Internet.
The evolution of the Internet has been marked by crucial standardization efforts that transformed isolated technologies into interconnected systems. In 1994, Netscape's browser implementation of HTTP, HTML, JavaScript, CSS, and DNS protocols created a unified platform that converted standalone documents into the interactive web.
The Think Agent Standard aims to achieve a similar transformation for artificial intelligence — where browsers enabled the composition of static documents into the dynamic web, this standard enables the composition of AI components into autonomous agents that can be deployed, traded, and integrated across any compatible system.
Each agent is composed of three components, designed to mirror the innate and adaptive dynamics of biological intelligence:
A Non-Fungible Intelligence (NFI) token that establishes agent identity and authentication. The Soul achieves consistency in agent behaviors and interactivity patterns while providing the tokenized authentication to connect one or more Minds and Bodies into an on-chain Agent.
- Complies with the Independent AI Institute standard
- Holds a minimum amount of THINK token in a wallet controlled by the agent
- Provides persistent identity across platforms and contexts
A decentralized application layer that processes information and manages agent behavior. A Think Mind is a dApp running on a server that authenticates by connecting to a wallet holding the Soul.
- Uses an extensible JSON-based negotiation pattern for structured agent communication
- Adopts x402 and ERC-8004 standards for agent-to-agent messaging
- Unopinionated on internal software — run Eliza, a custom agent, or any agent-as-a-service platform
- Puts power in developers' hands to innovate without permission
The interface layer that enables interaction and adaptation. Bodies are how agents and users interact — this can be an App, an OS, a game, and more.
A specialized AI data protocol designed for agent interoperability. The Agent Card integrates a Memory Management system capable of processing and ranking information derived from multi-level contexts and stimuli (referred to as modes).
- Information stored within persistent memory
- Reinforced or pruned through dynamic plasticity algorithms
- Enables agents to simulate human-like perceptions of events or interactions
- Memories may be recalled and communicated contextually between agent servers
| Principle | Description |
|---|---|
| Decentralized | Operates across distributed networks without central control |
| Permissionless | Any participant can interact without authorization |
| Interoperable | Works seamlessly with other protocols and systems |
| Composable | Components can be recombined to create new functionalities |
| Standardized | Creates and adopts standards whenever possible |
| Open Source | Core protocols are open-source; proprietary extensions are permitted |
We refer to systems that customize, mint, and edit agents as an Agent Forge. The open standard enables the development of novel authoring environments for Think Agents, which will innovate on top of this initial standard and develop novel ways of composing individual agents as well as teams and swarms of agents.
You can find the ThinkOS client and server on this GitHub account.
The Think Agent Standard is stewarded by the Independent AI Institute, with working groups actively establishing schemas and interoperability specifications. The standard's success depends on community adoption — we invite developers, researchers, and builders from all sectors to contribute.
We welcome contributions to the core protocol, tooling, and documentation. Please review our contribution guidelines before submitting a pull request.
- Fork the repository
- Create a feature branch (
git checkout -b feature/your-idea) - Commit your changes
- Open a pull request with a clear description
The Think Agent Standard is open source. Core protocols and specifications are freely available, while proprietary extensions and implementations are permitted within the framework.
First Published: January 24, 2025