papa-raw/slabclaw-acquisition-desk

SlabClaw Acquisition Desk is an experimental project built for a hackathon that explores what happens when an AI agent is allowed to buy and resell physical collectible items on its own, with only one required human approval step. You tell it what item to look for and what budget to use, and a group of always on AI worker agents takes it from there. The agent searches for a graded trading card listed for less than its estimated fair value, buys it using a digital currency called USDC on a blockchain called Base, and confirms the card's grading certificate against the grading company's public records. It then resells the card to a real buyer through Stripe's checkout system and records the profit from the price difference in its own internal ledger. Every worker agent involved only gets paid once its part of the process is verified. The one thing the agent cannot do on its own is actually spend money to make a purchase. That step always waits for a person to approve it, and there is a hard spending limit the agent itself cannot raise. The project is honest about what parts are fully real versus staged for the demo: the underlying contracts are deployed and tested on a test blockchain network called Base Sepolia, and the AI decision making uses a real NVIDIA Nemotron model. However, Stripe is running in test mode only, so no real payments occur, and the items sold in the demo are placeholder collectibles rather than actual graded cards, since selling a real card to a real buyer takes weeks. Technically, the project is written in TypeScript and organized into three parts: smart contracts tested with Foundry, an engine that plans routes and detects price differences, and a frontend web interface built with Node.js and Vite. Setting it up locally requires Foundry and Node.js 22 or newer, and the contract and engine tests can run without any API keys. The project is released under the Apache 2.0 license.