"AITeCXO team on Vultr: Agent Rush Hackathon Hackathon"
Team Idea
Receipts: an agent that can't say "done" without proof. Agents rarely fail hard questions. They fail chores, like reporting "backup verified" because a log says success while the archive is truncated. Receipts runs every agent action in an isolated, disposable Vultr sandbox (no secrets, resource caps, destroyed after each task) and records each command's output, exit code, and files as a receipt. A verifier checks every claim against those receipts before it reaches the user. Unsupported claims are blocked or flagged. We measure it with agent-trap-evals, our open-source trap set built from real agent incidents: same model, verifier off vs. on. Built on Vultr Serverless Inference, sandboxed containers on Vultr, and a FastAPI control layer. Team lead: Tim Andrus, Chief AI Officer at AITeCXO, 15+ years leading cloud and platform engineering, AWS-certified and CKA.
Submission
Team Leader hasn't made a submission yet