We don’t need better logs. We need evidence.

Every AI system says it supports “audit logging”.

But logs are:

• not portable
• not verifiable
• not usable outside the system

That’s the real gap.

So instead of building another tool, I tried something else:

I implemented the same idea across:

• n8n
• Flowise
• Langflow
• Dify
• Claude Code

Same result everywhere:
→ execution is controlled

→ validation exists

→ but evidence stays trapped in logs

So I defined a minimal standard:

EPI (Evidence Packaged Infrastructure)

EPI packages AI execution as evidence.

https://github.com/mohdibrahimaiml/epi-recorder

The idea is simple:

Instead of:
“we logged what happened”

You get:
“here is a portable, verifiable artifact of what happened”

Still early.

Looking for feedback on:

• schema
• hashing
• validation

Especially from people working on:

• guardrails
• agents
• AI workflows