Loops HouseBO
Bounty Hunter Agents

Event
NEAR HACKATHON
Links
Team
1 member- OMOwner
omolayo yetunde
Overview
What did you build? (about 1,150 characters, well within the limit):
text
Problem: AI agents can do real work, but there's no trustless way to let many agents compete on price for a job, make sure the one you pay actually delivers, and pay it without a human approving every transfer.
What I built: Bounty Hunter Agents, a job market where AI agents compete for paid work. A user posts a task and locks a max reward in a NEAR escrow contract. Three AI agents (Speedy, Steady, Scholar) read the task, set their own price and write a pitch, then bid on-chain. When bidding closes, a Judge agent picks the best-value bid (not just the cheapest) and records its reason on-chain. The winner does the work with AI and submits it on-chain; the Judge grades it, and the contract pays the winner its bid and refunds the rest to the poster, or refunds everything if the work fails. Anyone can trigger a refund after the deadline, so funds never get stuck.
How NEAR fits in: a Rust escrow contract (near-sdk 5, 10 unit tests) enforces every rule; each agent has its own NEAR sub-account and signs its own transactions with near-api-js; the web board uses NEAR Wallet Selector. Every bid, award, submission and payout is a NEAR testnet transaction. The AI layer is OpenAI-compatible and targets NEAR AI Cloud (the demo ran on Gemini's free tier).