cot

cot records what coding agents do in each session: their reasoning, the commands they run, and the tokens they spend. It runs entirely on your machine, with no cloud service and no account. I contribute to cot. Learn more at cot.run.

Source code: cot-intelligence/cot

How it works

• One install command starts the collector in Docker, bound to localhost, and wires hooks into the agents you pick: Claude Code, Cursor, or Codex.
• Each agent hook calls a local bridge, which posts the event to the collector.
• The FastAPI collector normalizes each event's category, phase, and token counts into SQLite.
• A React dashboard shows sessions, a trace timeline, insights, overview metrics, and command search.
• If the collector is down, events queue on disk and replay the next time it is reachable.