AI agents send emails, run code, and touch databases — but when something goes wrong, most teams have no record of what actually happened or who told the agent to do it. Agnys captures every agent action across your stack, replays any decision in plain English, and gives compliance teams a tamper-proof, attributable audit trail on the frameworks regulators are already asking about.
Hey Product Hunt 👋 I'm Ashish, maker of Agnys.
I kept shipping AI agents that acted — sending emails, editing files, hitting production APIs — and every time one did something weird, I had the same sinking moment: I couldn't reconstruct what it actually did, or who told it to. Prompt logs told me what I asked. They didn't tell me what happened.
So I built the thing I wanted: a flight recorder for AI agents.
What it does today:
📼 Captures real actions — one line of Python (auto_instrument()) patches requests, smtplib, sqlalchemy, openai & anthropic. Also captures from Claude Desktop, VS Code, LangChain, AutoGen, and browser AI (Claude.ai/ChatGPT/Gemini).
🔁 Replays any decision in plain English: trigger → reasoning → outcome.
🧬 Attributes every action — human, AI, or AI-on-a-human's-instruction — and records who approved the risky ones.
🔒 Seals it all in a SHA-256 hash chain so nobody can quietly edit the record.
✅ Scores it against the EU AI Act, GDPR, ISO 42001 and NIST, and exports a signed audit report.
Being straight about the stage: it's an early beta, built solo. The capture, the tamper-proof trail, attribution, and compliance scoring are all live and working. Some of the fancier stuff (semantic search, self-host, more frameworks) is on the roadmap, not done — I'd rather under-promise here.
I'd genuinely love your feedback — especially if you run agents in production and have felt this "what did it do?" pain. What would make this a must-have for you?
Try it free: https://agnys.net
Report
No reviews yetBe the first to leave a review for Agnys