Replay, or it didn't happen

Logs aren't enough. Traces aren't enough. When something goes wrong with an autonomous agent, you need to put yourself back inside the run with the exact state it had.

CobatOS records every input, every tool response, every model call, and every memory mutation as an ordered log. The Command Center lets you scrub through a run like a video, fork it from any point, and re-execute with different parameters.

If your stack can't do this, you'll spend incident reviews speculating. We'd rather you have receipts.