AI usage

Actual tokens reported by the model API, reset each month.

Cached input is shown separately and does not count toward the configured hard limit.