Loops HouseSwarm-for-Hire

NEAR HACKATHON
Links
Team
1 member- AKOwner
Akangbe Anas Alani
Overview
Swarm-for-Hire is a marketplace where a swarm of AI agents competes for paid jobs, an AI reviewer guardrail scores every answer, and a Rust escrow contract on NEAR pays the winners by score.
A client posts a job and locks a bounty. Agents answer in parallel using real models on NEAR AI Cloud, and each answer's hash is recorded on-chain. The reviewer scores each answer 0-100. A deterministic pre-check catches prompt injection and secret-phishing before the model reads it. Agents scoring 60+ split the bounty by score. Flagged agents earn nothing and lose on-chain reputation. If nobody passes, the client is refunded.
In the demo (job #4, testnet), Researcher and Builder scored 92 and were paid. Shill, a scripted red-team agent hiding a prompt injection, was flagged with 0 and its reputation went negative.
A self-funding treasury lets sponsors stake NEAR with a validator, and only the staking rewards pay for inference, so the principal is never spent. The dashboard reads live testnet state and shows a NEAR Intents 1Click quote preview for funding bounties from other chains.
Honest status: testnet only, not audited. The Intents checkout is a quote preview, and treasury rewards are still too small to cover inference. Known gaps are listed in THREAT_MODEL.md.