Proofline Keeper — Transaction Firewall for AI Agents

Enjoy the videos and music you love, upload original content, and share it all with friends, family, and the world on YouTube.

PROBLEM

AI agents can reason correctly and still execute the wrong blockchain transaction when parameters change after simulation, an approval is interpreted too broadly, or a retry creates a duplicate write. The execution boundary needs to fail closed before signing or broadcast.

APPROACH

Proofline Keeper binds human approval to the exact simulated intent, amount cap, expiry, simulation hash, and idempotency key. A transaction can proceed only when the deterministic decision is READY and the approval matches the same evidence. Stale, reverting, future-dated, changed, or duplicate intents are blocked.

DELIVERY

• Deterministic safety core with no private keys and no network calls • KeeperHub Direct Execution API adapter with simulation-first flow • Hash-bound approval and stable idempotency protection • Secret-free evidence packet with execution ID, transaction hash, block receipt, and explorer link • 22 focused automated tests, all passing • Verified Base Sepolia transaction using testnet funds and KeeperHub-sponsored gas • Public repository and end-to-end demo

VERIFIED RESULT

The live Base Sepolia execution completed successfully. The public receipt is independently verifiable in block 45139083, and the evidence packet records the exact decision trail without exposing credentials. Reviewers can reproduce why an action was allowed or blocked instead of trusting model-generated prose.

Repository: https://github.com/ceodaradigu/proofline-keeper Verified transaction: https://sepolia.basescan.org/tx/0x1772f39f5beb7bfeb6813124bccd3854bf31d956b65ac9becf8541c83867e040

Built by RELAUNCH DEPT. for a public hackathon. AI-assisted, human-reviewed. No prize or production deployment is claimed.