ranwhat
Open-source command-line tool that reads AI coding agent transcripts to find credential leaks and irreversible actions.
- Based in
- Zagreb, Croatia
- Founded
- 2026
- Ownership
- Private
About ranwhat
ranwhat is an open-source command-line tool for securing AI coding agents. It reads the history that agents like Claude Code and eleven others have already written to disk, showing what commands were run and which credentials leaked into their transcripts. The tool works locally, so nothing is sent anywhere unless the user asks. It does not block actions while an agent works. The company sells a free local tool and a paid subscription called Plus. Plus costs €12 per month or €120 per year per organisation and keeps a capability catalogue current. The tool is aimed at developers and organisations that use AI coding agents and need to know what those agents did and what their credentials still allow. It runs after the fact, with no wrapper or proxy in the agent's path. What makes ranwhat different is its focus on evidence rather than detection. It produces a record of irreversible actions that a third party can check. It reports what it does not know instead of assuming safety. The code is MIT licensed and has no runtime dependencies.
Products
- ranwhat watchReads what agents already ran and highlights irreversible actions worth knowing about.
- ranwhat cleanFinds plaintext credentials in agent transcripts, says what to rotate, and masks them on request.
- ranwhat scanScores the authority of an agent's credentials and shows which permissions were never used.
- ranwhat checkRuns watch and clean in one read-only pass without changing anything.
- PlusA paid subscription that keeps the capability catalogue current.