Show HN: Gryph – Audit Trail for AI Coding Agents (Claude Code, Cursor, Gemini)

1 points by abhisek a month ago · 0 comments · 2 min read

Hi everyone

I am the author of Gryph.

I have been using AI coding agents daily and realized I had no idea what they were actually doing across sessions. Sure, I could check git diff, but that doesn't show:

- Files the agent read but didn't change

- Commands it ran

- The sequence of actions in a session

- What happened last week when something broke

So I built Gryph - a CLI tool that maintains an audit log of all AI agent actions.

How it works:

- Installs hooks into Claude Code, Cursor, Gemini CLI (and other supported coding agents)

- Logs every action to a local SQLite database

- Provides rich querying: filter by time, agent, file path, action type

Quick demo:

$ gryph install Discovering agents... [ok] Claude Code v2.1.15 [ok] Cursor v2.4.21 Installation complete.

$ gryph logs --today 14:32 claude-code session 7f3a2b1c ├─ 14:32:12 read src/index.ts ├─ 14:32:18 write src/utils/helper.ts +12 -3 └─ 14:32:22 exec npm test exit:0

$ gryph query --file ".env" --since "7d" # See if any agent touched sensitive files

Privacy-first:

- 100% local - no cloud, no telemetry

- Sensitive file patterns are protected (actions logged, content never stored)

- Configurable verbosity

GitHub: https://github.com/safedep/gryph

Built with Go. Would love feedback from others using AI coding tools!

Previous post by someone else: https://news.ycombinator.com/item?id=46846849

No comments yet.