MAAT

Demo

About

The problem it solves Building autonomous AI agents that interact with external APIs is painful. Every integration is manual, there's no standard way to charge for tool usage, and once an agent runs, there's no transparent record of what it did or what it spent. This makes it nearly impossible to deploy AI agents at scale with any real accountability or fairness to the people who own the tools being used. Our product, Maat, makes that entire process automatic and trustless. Any API or smart contract can be converted into a structured, AI-ready tool in minutes, with pricing, permissions, and rules defined upfront and registered on-chain. Developers no longer have to hand-wire integrations or build custom billing logic, it's handled at the infrastructure level. AI agents get access to a growing catalog of verified tools they can call autonomously, paying per use via automatic micropayments that settle on-chain before the tool even responds. For enterprises and developers deploying agents, every action is publicly recorded on the blockchain, who called what, when, and what was paid. This creates a real audit trail for the first time, making it possible to run AI agents in environments that require accountability and compliance. Operators can set rate limits, spend caps, and restrict which tools each agent can access, so there's always a human-defined boundary around what the agent is allowed to do. For API and smart contract owners, this opens up an entirely new monetization channel. Instead of only serving human users, their endpoints can now serve AI agents directly, with automatic payments, no chargebacks, no disputes, and no middlemen. Every call is a settled transaction on-chain. For the broader AI ecosystem, this solves one of the most fundamental problems with agentic AI today: there's no economic or accountability layer between what an agent does and the real-world systems it touches. This project builds that layer, making AI actions safer, fairer, and verifiable for everyone involved. Use of AI tools and agents Maat is built from the ground up to be used by AI agents, not just humans. When a developer registers an API, the system automatically converts it into the OpenAI function-calling format, which is the standard way AI models like GPT-4 understand what tools are available and how to call them. This means any AI agent that supports function calling can discover and use tools on the platform without any custom integration work. Each agent gets its own on-chain identity, its own wallet balance for paying for tools, and its own reputation score that builds up over time based on how reliably it executes. When an AI agent decides to call a tool, it signs the request cryptographically, the system checks its balance, makes the API call on its behalf, records the result, and settles the payment on-chain automatically. The agent itself just sends a request and gets a result back. The more advanced use case is agent-to-agent commerce. One AI agent can browse the tool catalog, see what another agent is offering, send a request to use that agent as a service, and pay for it directly. All of this is coordinated through a messaging protocol that runs over the blockchain, with no human stepping in to facilitate the transaction. Agents discover each other, negotiate, transact, and build reputation entirely on their own. The system is essentially the economic and accountability infrastructure that lets AI agents operate in the real world, spending money, accessing services, and being held to a verifiable track record of what they did.

How it's made

Challenges we ran into The hardest part was making payments and tool execution autonomous. If an agent calls a tool and something fails halfway through, you can't leave the money in limbo. We built a system where funds are set aside the moment a tool is called, before anything actually runs. If the tool completes successfully, the charge goes through and gets recorded on-chain. If anything fails, the money is returned to exactly where it was before the call started. We also had to make sure that if multiple agents were using the system at the same time, their balances could never accidentally interfere with each other. Every agent's funds are tracked separately, so a problem with one agent has no effect on another. Getting the order of all these steps right, so there was never a moment where money could go missing or be charged twice, was the part that took the most work to get right.