Administrators used to get the general assistant, and the only view of what the assistant was costing the workspace was the AI spend page. That page is still there, with the full breakdown, the six-month trend and the month picker. What is new is that you can now ask about it.

Opening the assistant as an administrator puts you in the govern register. The side panel carries the month’s spend so far as a meter, and the line under it says what the bar is a share of, how many days of the month have gone, the daily rate that follows, and the day that rate passes the figure it is measured against. There is a link straight through to the full spend page beside it.

Two questions are answered in the conversation:

• “Are we going over?” — the same meter, with the register’s reading of it and the assumption it thinks you should attack first.
• “What is each agent costing us?” — a table with one row per agent: what it spent, its share of the month, how many turns it ran, how much tool work those turns did, and how many of them failed or were refused. Underneath, the register separates a cost symptom from a quality cause — an agent can be expensive because it is asked more often, because each turn does more work, or because its turns fail and get asked again.

There is no spend limit anywhere on the platform, and the register never pretends otherwise. The percentage is measured against your own most recent month with spend, and every figure names that month rather than calling it a budget. A workspace in its first month of assistant use gets no meter at all — there is nothing yet to measure against, and the register says so instead of drawing a bar.

The register reads and nothing more. It cannot raise a limit, turn an agent off or change what one does; asked to, it says so and points at where that is done. Only administrators of a workspace see any of it — the meter, the questions and the link are all absent for everyone else, exactly as the spend page has always been.