Choose a time range
The toolbar offers four ranges and picks the bucket size from the range you choose:
When no records are available the page shows an empty state; it never treats an unknown value as zero.
Filters and grouping
You can combine five filters: Agent, Model provider, Model, Project, Session. Results can be grouped along five dimensions, which are the tabs above the table: Time, Agent, Project, Model provider, Model.The Project dimension maps to projects in the sidebar, and one project may contain several workspaces. Session is a filter only and cannot be used as a grouping dimension.
Read the metrics
The summary area shows six metrics:
The grouped table has these columns: the dimension name, Requests, Total, Input, Output, Cache, Hit rate, Last activity, Coverage.
What Coverage means
Coverage tells you how much to trust the numbers in that row:
Hovering reveals a finer breakdown (complete / partial / baseline only / unreported / unsupported).
Trend views
While refreshing, the page keeps the previous result and marks it stale. If loading fails, it prefers the previous authoritative result and shows Refresh failed; showing previous data.
Data boundaries
- Usage comes from Agent executions recorded by the authoritative runtime and from adapter reports. The authoritative runtime is the authoritative source of these statistics.
- Usage Statistics does not read full prompts, file contents, terminal output, or provider keys.
- Mobile does not show usage detail — the mobile usage page only shows session counts and says “Usage details are available on the desktop host.”
- A single query has a row limit (100000 rows); exceeding it returns
agent_usage_query_too_large.