Branch token usage

Open Usage in the header services menu. It reports work executed on the selected Thread branch, grouped by model, provider, work kind and measurement status. This includes conversation, context work, planning, MCP sampling and recorded tool inference.

Rows show input, output, cache-read, cache-write and reasoning tokens when reported. Provider definitions can overlap, so do not add every field into a total or assume input always excludes cached tokens.

Actual, estimated, missing and pending usage remain distinct. An unknown count is not zero. Requests with no recorded dispatch are identified separately. Child Threads and standalone utility work have their own usage; this is not an all-branches total or a context-capacity gauge. No price estimate is made.

If updates disconnect, the panel labels the last recorded snapshot. Use Refresh to reconnect. See Context and Continuing goals for their separate controls.