IntentMesh

IntentMesh is a system that turns high-level user intent into real execution through coordinated AI agents. Instead of users figuring out complex workflows, they simply express what they want, and the system handles planning, data gathering, decision-making, and execution end-to-end. Each agent signs its output, and every step is verified before moving forward, creating a transparent chain of trust. The final result is anchored on-chain as a proof, making the entire process auditable and tamper-resistant. The goal is to make automation both simple to use and reliable enough for real-world use cases.

IntentMesh is built as a modular system that combines a Python-based multi-agent backend with an EVM smart contract layer. The backend is structured around four agents: a planner, data agent, strategy agent, and execution agent. Each agent is implemented as a separate module and communicates through structured JSON, with every output being cryptographically signed using ECDSA via libraries like eth-account. A verification layer ensures that each agent validates the previous step before proceeding, forming a chain of trust.

For data, the system integrates external APIs like CoinGecko for real-time price feeds, combined with mocked DeFi yield data to keep the system lightweight for a hackathon setting. The execution agent computes hashes of inputs and outputs using keccak256 and submits a proof to an on-chain registry contract deployed via Foundry. Interaction with the blockchain is handled using Web3.py.

On the frontend side, a React + TypeScript app uses libraries like React Flow and Framer Motion to visualize agent coordination in real time, making the system more intuitive and interactive.

One hacky but effective design choice was keeping agents as simple function-based services instead of over-engineering a full distributed system. This allowed quick iteration while still demonstrating real multi-agent coordination, signing, and verification. Overall, the system balances simplicity with a clear demonstration of verifiable, autonomous execution.