Loops HouseNomad

NEAR HACKATHON
Links
Team
1 member- CHOwner
Charles M S
Overview
The problem: AI agents and wallets across chains hold assets that others want, but trading between them requires a middleman, either a market maker who takes a spread or a solver who must be trusted. Removing that middleman is hard. Strangers who don't trust each other and sign with different wallet standards must settle a multi-party trade in which everyone is paid or nothing moves. No one should be able to alter a signed leg, spend one offer twice across competing deals, or drain an agent's funds, and no one should need permission to complete the trade.
What we built: Nomad is a permissionless, multi-party settlement layer in which AI agents and wallets trade in rings of two to five parties. A ring finder searches for cycles and sizes every leg so each token nets to zero after fees and every party gets at least its limit price. Our rings contract validates each proposed ring on chain. Each party signs only its own leg, as a NEAR account (NEP-413), an Ethereum wallet (ERC-191) or a Solana-style key (raw ed25519). Any stranger can then close the ring and collect the signed tips. One shared nonce per offer guarantees that at most one of an agent's competing rings can ever execute. Agents can run under on-chain mandates (allowed tokens, per-leg cap, daily budget and minimum price) that sign through Chain Signatures, with no hot key. Ethereum wallets can submit ERC-7683 orders, which the contract verifies with ecrecover and enforces on every fill. Every settlement records ERC-8004-style reputation. A live swarm of 12 agents and competing matchers settled 27 rings with zero failures, and all seven attack classes were rejected on chain.
How NEAR fits in: NEAR Intents' execute_intents verifies legs signed under three wallet standards and settles them atomically in one transaction, with no bridge or escrow. Chain Signatures let a contract control an agent's Ethereum address. It all runs on testnet, against a Verifier with mainnet intents.near's exact code.