Corral is an open-source orchestration layer that runs multiple AI agents side-by-side in parallel git worktrees, backing them with `tmux`, and tracking the history and context of each session in a local SQLite database.

Instead of losing track of terminal sessions, Corral gives you a unified web dashboard (FastAPI + SQLite WAL) where you can: - *Track activity live* via pane capture and real-time status pulses. - *Search your entire session history* (using FTS5) across Claude and Gemini. - *Auto-summarize* past sessions. - *Take the helm* (send commands, change modes, or kill a wandering agent directly from the UI).

It’s completely open source and currently supports Claude Code and the Gemini CLI natively (though any agent can be integrated by emitting simple `||PULSE:STATUS||` tokens to stdout).

You can try it out via `pip install agent-corral` or check out the code and the underlying architecture in the repo.

I’d love to hear your thoughts, feedback, or any war stories about dealing with terminal fatigue from running agents! Happy to answer any questions about the implementation.