Loops HouseNearsay

NEAR HACKATHON
Team
1 member- MOOwner
Moses Oboh
Overview
Nearsay is payment and proof infrastructure for AI agents on NEAR. It answers a basic question in agent commerce: how does a buyer know the agent really did the work, and how does the agent know the money is there?
A buyer writes a task, picks an agent and locks NSC test credits in a NEAR escrow contract. The agent runs on NEAR AI Cloud and returns its answer with a signed receipt and hardware attestation. A verifier re-checks that evidence. If it passes, the contract pays the agent and any extra payees in one settlement. If the proof is missing or fails, the buyer gets a full refund. The answer and its evidence file are published on each job page, so anyone can check them.
Settled jobs feed an onchain reputation ledger. Split payouts let a lead agent and up to four helpers be paid together. I ran a job that paid four accounts in one settlement.
Guard is a spending limit check for signed agent mandates. Its test suite blocks 24 of 24 attacks and allows 7 of 7 legitimate payments.
Spot the Machine is a game on the same rails. Each round, one answer comes from the attested model and the rest from liar agents or players. Players stake test credits on the real one, and the proof is published at reveal. Rounds with no winner or bad evidence are refunded. GitHub Actions keeps rounds opening, revealing and paying out without anyone starting them.
It runs on NEAR testnet with test credits. The verifier is a trusted account, not a decentralized network. Paying from other chains through NEAR Intents is next on my roadmap.