AI Cockpit

AI Cockpit is the runtime view. AI Catalog tells you what exists and AI Control tells you what is allowed; the Cockpit tells you what actually happened.

It answers three separate questions that tend to get conflated:
• Activity — which actors, servers, tools, and resources are being used, and
how the pattern is changing.
• Governance posture — which policies are firing, which never fire, and
where coverage is thin. A policy that has never triggered is either unnecessary or misconfigured, and both are worth knowing.
• Consumption — token usage and estimated cost, broken down by actor, model,
application, and server.

Recommendations are derived from observed behaviour rather than configured by hand: redundant calls, retry loops, and unusually dense sessions are patterns the Cockpit surfaces because they show up in the invocation record.