AGENTNS

AGENTNS is a fully decentralized, multi-agent AI system designed to operate autonomously on-chain. Instead of relying on centralized databases or REST APIs, our autonomous agents (Scout, Strategy, and Executor) use ENS text records for dynamic peer discovery. Once discovered, they form a secure, encrypted P2P mesh network using Gensyn's AXL protocol.

The Scout agent monitors real-time market opportunities, the Strategy agent evaluates and optimizes trading decisions, and the Executor agent handles the on-chain execution via KeeperHub for MEV-protected Uniswap V3 swaps. Finally, the system automatically writes verifiable performance metrics and reputation scores back to the agents' ENS domains on the Sepolia testnet, creating a completely transparent, trustless AI economy.

I built AGENTNS by intentionally stripping away centralized Web2 dependencies to create a true Web3-native AI swarm. We used a Node.js backend with persistent daemon loops for the agents, but the real magic is in the decentralized stack:

1. Discovery (ENS): We utilized ENS text records on the Sepolia testnet to store agent capabilities and P2P addresses. Agents look up .agentns.eth domains to discover peers dynamically, effectively using ENS as a decentralized database.

2. Networking (Gensyn AXL): For inter-agent communication, we integrated the AXL protocol. This provides an encrypted, decentralized P2P mesh where agents exchange standardized JSON envelopes, bypassing traditional servers entirely.

3. Execution (KeeperHub): The Executor agent leverages Ethers.js and KeeperHub to process real Uniswap V3 swap encodings with built-in gas optimizations and MEV-protection.

4. The "Hacky" Reputation Loop: The most notable feature is our on-chain feedback loop. After every successful KeeperHub execution, the Executor agent signs a dynamic transaction to update the Strategy agent's ENS text record with a new reputation score. This creates an immutable, verifiable audit trail of the AI's performance directly on Etherscan.