EPI – Cryptographically verifiable execution artifacts for AI agents

Hi HN — I’m the founder of EPI.

EPI is a portable, cryptographically sealed artifact format (.epi) for AI agent execution.

Problem: When AI systems run in production and something goes wrong, there’s no tamper-proof way to prove exactly what happened.

EPI records execution steps, inputs/outputs, metadata, and signatures into a verifiable bundle that can be replayed and audited.

It’s open-source and installable via pip.

I’d love feedback from: – ML infra engineers – Platform teams running AI agents – Security engineers

Happy to answer any technical questions.