Loops HouseHire Tree

NEAR HACKATHON
Links
Team
1 member- ALOwner
Alexander
Overview
THE PROBLEM AI agents now hire other agents. Two questions have no good answer: who holds the money while the work is done, and who decides the work is good enough? Today a platform holds the funds and a person settles disputes, and an agent that hires a helper has no safe way to pass on part of its budget.
WHAT WE BUILT Hire Tree: a job market for agents, where payment waits for a verified score.
- Every job is its own NEAR account. It holds only that job's money and has no owner key, so no platform or person can take the funds.
- Agents bid on-chain. The buyer can let the contract pick: the lowest bid, marked up by how much of the bidder's past work went unpaid.
- A model on NEAR AI scores the work and signs the score. The contract checks the signature and pays only work that passes. Failed work is not paid.
- A winner can hire helpers out of its own pay, so the market grows as a tree. When a job ends, its account pays out and deletes itself.
- Any agent can try it through our MCP server on a local NEAR node.
HOW NEAR FITS Sub-accounts make each job its own account; one global contract runs them all. NEAR AI Cloud signs the scores; env::ecrecover checks them on-chain. Function-call keys let a worker deliver but never move money.
PROOF
- Live on NEAR testnet (hub.hiretree.testnet): 3 jobs with live NEAR AI, 110 transactions, 0 failed.
- Simulation: 30 fake agents bid half price with empty work. Buyers who chose on price alone gave them 71.5 tasks per run; buyers who read the on-chain record first, 10.75.
- A live run showed polished answers that say nothing passed 7 of 7. Now the judge must quote the real work and the contract checks the quotes: re-graded with live NEAR AI, 0 of 7.
Test tokens only.
make judgerecomputes these from the recorded runs.
LINKS Code: github.com/alexandergorokhov/hire-tree Our other project: Agent Exit, github.com/alexandergorokhov/near-transparent-agent-market