What does AI Cockpit show?

Aggregate AI activity across the estate: how much is happening and where, which actors and tools account for it, how often calls are refused, which policy rules are firing and which never fire, and a filterable chronological feed of events. It is the view over all activity; the record of any single decision lives in AI Control.

Where does the data come from?

From the same invocation and decision record the platform already produces when it governs a call. There is no separate analytics database and no second pipeline to keep in sync, so what the aggregate views show and what the decision record says cannot disagree.

Can it show policies that are not doing anything?

Yes. Policy activity shows the breakdown of allow, deny, warn and audit outcomes, which rules fire most, and which rules have never fired at all. Rules that never fire are worth attention: they are usually either redundant or written against something that no longer matches.

Take a Test Drive Talk to Us

AI Cockpit

What your AI is doing, across the whole estate.

A single decision is answered in AI Control. Cockpit answers the other question — the shape of the whole: what is busy, what is being refused, which rules carry the load, and what nobody is governing yet.

The question it answers

Is our AI behaving the way we said it would?

That question is not answered by any single decision. It is answered by the shape of many — how much is happening, who is doing it, how often it is refused, and whether the rules you wrote are the rules doing the work.

Cockpit reads from the same record the platform produces when it governs a call. There is no separate analytics store, so the aggregate view and the decision record cannot disagree.

The surfaces

Six views over the same record.

All of it comes from the same decision and invocation record. There is no separate analytics store to keep in sync, and no second version of the truth.

Overview

Activity volume, denial rate, and the actors and tools that account for most of it.

Actors

Per-actor activity: what each agent called, how often, what was denied, and when it was last seen.

MCP servers

Per-server activity and enforcement mode, including the breakdown of who is calling it.

Tools and resources

Which operations are actually used, ranked, and which are governed by no policy at all.

Policy activity

Allow, deny, warn and audit outcomes, the rules firing most, and the rules never firing.

Activity feed

The chronological stream, filterable, expanding into the detail behind any single event.

Governance gaps

The useful part is often what is missing.

Tools no policy covers

Operations in active use that no rule governs. Usually the fastest thing to act on.

Rules that never fire

A rule that has never matched is either redundant or written against something that no longer exists.

Callers you do not recognize

Activity from actors that were never reviewed — the signal that something is running that nobody registered.

Turn a gap into a policy

Talk to us

See this against your own traffic.

These views are most useful with real activity behind them. We are glad to show what that looks like for the systems you run.

See the Cockpit Yourself Talk to Us