Hey Product Hunt 👋

I'm Atnabon Deressa, founder of Goderash.

We built this because of a wall I kept hitting in production: AI agents that work perfectly in the demo, then die in compliance review.

The reason is simple. Every audit trail today is a logging trail. Logs answer "what happened." Auditors want something different — chain-of-custody, tamper-evidence, and the ability to replay history under alternate policies. LLM trace dumps don't survive that conversation.

So we built Goderash.

One decorator wraps any tool call, LLM call, or policy decision and writes a typed, SHA-256 hash-chained event into a per-tenant Postgres ledger. Mutate one byte → the chain breaks → /v1/verify catches it. The auditor verifies it themselves over a single HTTP call. No trust required.

What ships in v0.1.0:
🔗 Hash-chained event ledger (Postgres, append-only)
🛡️ Runtime guards — permission modes, velocity limits, fraud guards, biometric confirm
🔄 What-If projector — replay history under alternate policies
📦 5 evidence packs — SOC 2, HIPAA, FFIEC, FINRA, SEC Rule 17a-4
🧩 Adapters for LangGraph, OpenAI Assistants, Anthropic, Claude SDK, AutoGen, LangChain
🔓 Apache 2.0 — self-host free, or use our hosted control plane

Origin: We built this for an in-app financial AI assistant. To get past a Tier-1 risk-and-compliance committee, we had to build the whole stack. Goderash is that work, made framework-agnostic and open-sourced.

Traction:
→ 4 PyPI + 4 npm packages
→ 1,636 downloads in the first week
→ Production-tested in regulated banking

If you're building agents in fintech, banking, healthtech, insurance, or legal try it and tell us what breaks. I'll be in the comments all day.

🐙 GitHub: github.com/goderash/goderash
📚 Docs: ai.goderash.com

Built in Addis Ababa 🇪🇹

Goderash

The audit layer for regulated AI agents

The audit layer for regulated AI agents