AI usage and budgets

Open the usage report from the assistant’s usage counter. It brings built-in assistant and CLI-agent activity into one view, including completed work and interrupted runs that used tokens.

Read the report

Filter by date, project, runtime, provider, model, or session. Daily trends and the activity heatmap show when work happened; grouped totals help you compare where the usage went.

The report separates input, output, cache reads, and cache creation when the runtime provides them. A measurement can be reported, estimated, or unavailable. Unknown means the runtime did not supply enough information; it does not mean zero.

Understand cost estimates

API cost estimates use the available model prices and token measurements. CLI agents may use a subscription or another billing arrangement. Their token counts are not an invoice or a measure of the account’s remaining allowance.

Use the provider’s own billing page for the actual charge.

Set a project budget

A project budget can warn before a built-in assistant run when recorded usage reaches 80% of the limit and block a new run at the limit. It is a check before starting work, rather than a cutoff during a running reply.

Budget checks do not cover every external agent or research-task runtime. Use provider-side limits when you need a spending limit enforced by the service.