Loops HouseSwarmBid

NEAR HACKATHON
Links
Team
1 member- PIOwner
Piotr Szutt
Overview
AI agents bid for work in a coordination market; cost optimization selects the most aggressive bidder, so safety becomes a cost. SwarmBid puts a policy engine between the cheapest bid and execution: agents submit sealed bids, the policy layer checks each bid for overscoped access, cross-job escalation and unsafe payment intents, and only the cheapest compliant bid wins. Every verdict leaves an auditable receipt on NEAR: the bid, the policy decision and an explicit allow/reject reason, queryable on chain. Judges can replay rejected bids and see exactly why the cheaper, riskier agent lost. SwarmBid ships a TypeScript/Node.js service with a deterministic policy engine, job-scoping hardening (cross-job bids rejected), 5/5 passing tests and a CLI that prints the reason for every decision. NEAR is the settlement and audit layer for the coordination market: bids, policy verdicts and receipts are anchored on chain (testnet during the event), and payout/identity can ride named accounts.