Launching today
AgentTM
Time-travel debugging for AI coding agents
4 followers
Time-travel debugging for AI coding agents
4 followers
A local debugging tool that records managed Codex sessions and workspace checkpoints so you can inspect changes and restore earlier files into an isolated Git worktree.
Free Options
Launch Team
BRCKS ChatThe chat app for building sites. Nothing gets lost.
Promoted
Hey Product Hunt! 👋
I'm excited to introduce AgentTM (Agent Time Machine), a local-first debugging tool I've been building for AI coding agents, starting with OpenAI Codex.
Why did I build this?
As AI coding agents become more capable, they execute commands, modify multiple files, and perform complex operations. But when something goes wrong, understanding what happened during execution can be difficult.
I wanted a way to inspect an agent's activity, explore earlier workspace states, and recover previous code without disrupting my current work.
That's how AgentTM started.
What AgentTM offers:
Execution Timeline: Record supported Codex events and inspect agent activity.
Workspace Checkpoints: Capture baseline, final, and supported operation-boundary snapshots.
Historical File Explorer: Browse earlier versions of your source files.
Code Diffs: Compare changes between checkpoints.
Safe Restoration: Restore previous files into isolated Git worktrees.
Interactive Codex Chat: Chat with Codex through the native Windows desktop app.
Conversation Continuation: Resume supported saved Codex conversations.
Encrypted Backups: Protect and restore local AgentTM history.
Local-First Storage: SQLite and SHA-256 content-addressable storage.
Storage Management: Integrity diagnostics, session cleanup, and garbage collection.
Built with: Go, SQLite, React, TypeScript, Git worktrees, and Wails.
The CLI and browser dashboard support Windows, Linux, and macOS. The native desktop app is currently Windows-only.
AgentTM is still an early public beta. It complements Git rather than replacing it, and checkpoints represent observed workspace states rather than deterministic execution replay.
I'd love your feedback!
What debugging features would make AI coding agents easier to work with? Would you find checkpoint inspection, code provenance, or support for additional coding agents most useful?
GitHub: https://github.com/Jyotishmoy12/AgentTM
Thanks for checking it out! 🙌
If you find AgentTM interesting or useful, I'd really appreciate it if you could give the repository a star (⭐) on GitHub! It would mean a lot and help the project reach more developers.
Thanks for your support!
Appreciate the distinction between a saved workspace state and deterministic replay. Showing where checkpoint coverage starts and ends would help someone judge what they can actually recover.