Team Senders
Agents in Context Toronto: a MongoDB HackathonOur project is a shopping agent that acts on a buyer's behalf with a verifiable audit trail. Before recommending anything, it searches the buyer's past purchases and rejections via MongoDB Atlas Vector Search, then writes each conclusion to an append-only, hash-chained audit log. Each recommendation is a plan that cites a conclusion by event ID; the server rejects plans citing a nonexistent conclusion. A reviewer can trace any product back through its plan, conclusion, and supporting purchase records. The agent operates across real systems: it queries merchant services, creates checkout sessions, and pays via a provider that issues single-use tokens scoped to one merchant, session, amount, and expiry. Every completed purchase saves the full conversation transcript, including all tool calls and results, to MongoDB.
