Poka

Trustless agent-to-agent communication protocol with ERC-8004

Comment

Inspiration

The rise of autonomous AI agents promises to revolutionize how we interact with blockchain systems, but trustless coordination between agents remains a major challenge. We wanted to create a protocol where AI agents can discover, verify, and collaborate with each other without centralized intermediaries, making multi-agent workflows transparent, auditable, and secure. Our ERC-8004 implementation is the answer to enabling trustless agent economies.

What it does

Our ERC-8004 Multi-Agent System is a blockchain-powered coordination protocol that:

  • Provides portable agent identities using ERC-721 NFTs, making agents discoverable and transferable.
  • Enables intent-based interactions where users authorize agents to act on their behalf with cryptographic signatures.
  • Implements a reputation system for rating agent performance and building trust over time.
  • Supports pluggable validation models (stake-secured, zkML, TEE attestation) for verifying agent work.
  • Facilitates agent-to-agent communication via HTTP with on-chain coordination and audit trails.
  • Uses LangChain and AI for autonomous decision-making and tool execution.

How we built it

We built the system using:

  • Four Solidity smart contracts (Identity, Reputation, Validation, Intent Coordinator) following the ERC-8004 standard.
  • Viem for blockchain interactions with custom chain configuration.
  • Hono for lightweight HTTP servers that expose agent capabilities and actions.
  • LangChain with Google Gemini for AI-powered agent reasoning and tool calling.
  • Intent-based architecture with user signatures (EIP-191) and revocation locks for secure coordination.
  • In-memory storage for execution results with verification endpoints for validators.

Challenges we ran into

  • Intent lifecycle management: Coordinating the complex flow of create → lock → execute → validate across multiple agents and blockchain calls.
  • Signature verification: Implementing proper EIP-191 message signing and recovery for intent authorization.
  • Tool execution routing: Getting LangGraph to properly route between the agent node and tools node for function calling.
  • Cross-agent discovery: Fetching and caching agent metadata from blockchain while maintaining real-time HTTP communication.
  • Error handling: Managing transaction failures, expired intents, and lock timeouts gracefully.

Accomplishments that we're proud of

  • Built a complete ERC-8004 implementation with all four registries working together.
  • Created autonomous AI agents that can discover and call other agents' capabilities via blockchain + HTTP.
  • Implemented user-controlled authorization through cryptographic signatures with revocation safety.
  • Achieved full transparency with on-chain audit trails for every agent interaction.
  • Integrated LangChain with blockchain tools for truly autonomous agent decision-making.

What we learned

  • Intent-based architectures are powerful for user authorization but require careful lifecycle management (nonces, expiry, locks).
  • Hybrid on-chain/off-chain coordination (blockchain for trust, HTTP for execution) is the optimal approach for agent economies.
  • Reputation systems need time and multiple interactions to build meaningful trust signals.
  • Tool calling with LLMs requires proper workflow setup (conditional edges, tool nodes) to execute correctly.
  • ERC-8004's modular design (separate registries) makes it easy to deploy only what you need.

What's next for ERC-8004 Multi-Agent System

  • IPFS integration for decentralized storage of execution results and validation data.
  • Persistent database (PostgreSQL) to replace in-memory storage for production use.
  • Automated validator service that monitors IntentExecuted events and performs verification.
  • zkML proof generation for cryptographic verification of agent computations.
  • Multi-chain support to enable cross-chain agent interactions.
  • Agent marketplace UI for discovering, rating, and deploying agents.
  • Delegation mechanisms for long-lived user authorizations with granular permissions.
  • Economic incentives for validators through stake slashing and reward distribution.

Built With

Share this project:

Updates