Loops HouseDay One: AI Purchasing for Acquired Manufacturing Plants

NEAR HACKATHON
Links
Team
1 member- ANOwner
Anthony Langford
Overview
The problem: small manufacturers, 98% of American manufacturers by the SBA's count, still run production on paper batch records and spreadsheets. Shortages are caught by hand and orders are chased by email. When such a plant is acquired, its purchasing must follow the new owner's rules from day one, long before any ERP migration.
The product: Day One gives the plant an AI purchasing and inventory agent, Kevin. He reads scanned batch records, the inventory spreadsheet and the purchasing mailbox, and a person confirms each reading before anything posts. He keeps inventory across the company's plants, catches shortages against the production plan, gets quotes from supplier agents and places orders. Staff work on ordinary screens and never see a wallet, a token or gas.
How NEAR fits: a spend-policy contract on NEAR testnet holds the plant's budget and enforces the limits HQ sets. In the recorded demo it refused Kevin's $5,150 order against his $2,500 limit even though his model had signed APPROVE inside a NEAR AI Cloud enclave. The ops director's approval let Kevin fund escrow, and the supplier was paid only after receiving and QC released the lot. Supplier agents quote from their own NEAR accounts, two plants pool one order, a relayer pays staff gas, and batch records and lot histories are anchored on chain. The contract account has no keys, so its code cannot change.
The demo uses fictional companies, a mock ERP, a simulated text reply, server-held staff keys and test tokens. Try it without installing at https://day-one-kevin.vercel.app, or run npm ci && npm run prove for a keyless scorecard of every claim.