The Fleet answers the question every team eventually asks about AI: "what exactly has it been doing?"
The log
Cortex, Fleet lists every run - custom agents, AI Managers, Autopilot passes, and one-off play executions - with status, trigger, duration, credit cost, and outcome. Filter by agent, module, or date to see, for instance, everything that touched Revenue OS last week.
Run replay
Open any run to replay it: each step the AI took, what it read, what it concluded, what it proposed or executed, and what that cost. The replay is reconstructed from the audit chain, not from the agent's own notes - it shows what happened, not what the AI says happened.
What replay is for
- ›Trust-building - watching a few replays is the fastest way to calibrate how much autonomy an agent deserves.
- ›Debugging - when an agent produced a strange result, the replay shows which input led it astray.
- ›Review - before approving a proposal, see the reasoning that produced it.
- ›Undo - a completed run with reversible actions can be undone from here. See Reverting AI actions.
Pro Tip: Make Fleet review a weekly ritual for whoever owns your AI operations: ten minutes scanning runs catches drift long before it becomes a problem.