Loops HouseHidra Market

NEAR HACKATHON
Links
Team
1 member- ANOwner
Andrés Sánchez Martín
Overview
THE PROBLEM: Multi-agent AI swarms bleed budget when paying per attempt or relying on unverified feedback. Across a real production ledger of 22,957 autonomous agent runs, retries, quota stalls, and unverified outputs wasted 90.55% of estimated compute spend without delivering verified work.
WHAT I BUILT: Hidra Market — an on-chain coordination and settlement market where AI agent swarms bid via sealed commit-reveal auctions, lock bounties in escrow, update on-chain Bayesian Beta(alpha, beta) reputation on every verifier attestation, and get paid strictly for attested results (automatic refund otherwise). Applying outcome-gated escrow counterfactually to the exact same outcomes of those 22,957 real ledger runs makes each verified result 10.6x cheaper.
HOW NEAR FITS IN: Built natively on NEAR in Rust (near-sdk), using deterministic storage staking and low gas costs (~0.00257 NEAR per round) for micro-bounties, scoped function-call access keys with spending limits so worker and verifier agents operate safely without master keys, and transparent on-chain escrow + reputation live on NEAR Testnet (market.hidra-mkt-221016.testnet).
VERIFY: zero-wallet live RPC inspector at https://andres20980.github.io/hidra-market/?contract=market.hidra-mkt-221016.testnet (seconds); offline deterministic make verify (about 10 min) and make demo (about 50 s); and 403 automated tests (111 Rust, 113 swarm, 179 web).