aadhaa

Agent Marketplace is a requester-controlled network for delegating AI work to shared compute. Enterprises need shared private compute that protects intellectual property without high-end hardware for every employee; consumers need occasional demanding AI without buying an expensive laptop. Compatible compute owners could eventually earn from contributed capacity. Our public-work MVP proves the delivery mechanics with a fixed expense-report application. Meta proposes dependencies and complexity for three prepared module contracts. Separate Python worker processes register, bid, win scoped leases and call Meta to generate code. The requester independently tests artifacts, requests repair when needed, settles simulated credits once per accepted task and separately checks the complete application. A clickable dashboard shows components, placement, progress and backend traces, including each Meta API call. Accepted modules can process new public CSV locally without another model call. One recorded live run accepted all three modules, repaired one real rejection and passed integration in 79.7 seconds at about $0.0534 by the local guard. The demo runs on one laptop with hosted Meta inference. Credits are simulated; physical multi-machine scaling, arbitrary application decomposition and protected private execution are outside this recording. Confidential-worker code/research is separate and the hardened protected path still needs approved trust references and live validation. The sanitized public demo source is MIT licensed.