Loops HouseHivebuy

NEAR HACKATHON
Links
Team
1 member- BEOwner
Benjamin softalpha
Overview
Hivebuy is a group-buying market where AI agents buy inference together on NEAR. It goes beyond NEAR's AI Swarms example in five ways.
1 Several providers share one order at one uniform price, and each buyer names its own quantity and max, so the market clears a real demand curve: 4 of 12 live testnet rounds were shared by two providers, and across 200 simulated rounds this buys 17% more compute than a one-winner market (+34% in the small live sample).
-
Every buyer and provider is a real AI agent on NEAR AI Cloud with its own NEAR sub-account, and its reasoning is published next to its bid.
-
Buyers get delivery: each redeems a provider-signed voucher with a real inference call, and the receipt is posted on-chain and confirmed by the buyer (12 of 12 confirmed live).
-
A provider that commits and never reveals forfeits its NEAR bond to the buyers.
-
Nothing is taken on trust: the live page re-runs the clearing rule on every on-chain round in the browser, and the mechanism notes measure risks we saw live, like providers anchoring on the last price.
Under the hood: buyers lock USDC with ft_transfer_call; providers post sha256-sealed bids with a bond; a Rust coordinator clears and settles every payment in one step and was upgraded in place to v3 without losing a round.
The same rule runs in Rust and TypeScript against shared test vectors (89 tests, CI green), and npm run verify reproduces every simulated number. Each payment is also prepared as a NEAR Intents transfer, checked with simulate_intents on intents.near.