A missing part of the usage picture
GitHub announced on October 23 that its Copilot usage-metrics API can separate model traffic by client. The new totals_by_client array appears in supported one-day and 28-day enterprise, organization and user reports. It can identify integrations such as OpenCode, Codex, Claude Code, Zed, Copilot CLI, GitHub’s app, cloud agent and code review, while unfamiliar identifiers are grouped as other.
Each client entry includes successful model-request and session counts, token use and a model breakdown. This differs from agent-app activity, which counts jobs. GitHub cautions that one prompt may cause several model calls, so request_count is not a prompt or user-interaction total.
Access and data limits still matter
The reports require the Copilot usage-metrics policy and an eligible owner, billing manager or custom role with the relevant view permission. GitHub’s API documentation provides daily and rolling 28-day reports; the announcement does not identify a separate price or country-specific rollout for the new field.
Client labels depend on the integration ID supplied with a request and should be treated as approximate groupings. GitHub also documents differences among dashboards, APIs and exports, along with delayed telemetry that can leave recent days incomplete for as many as three full UTC days. CGH has not tested these reports.
CGH perspective
Use the new breakdown to find tools that were absent from IDE-centered reporting, then compare stable reporting windows rather than the newest partial day. Record policy changes and major client rollouts before interpreting a jump in requests or tokens.
Treat volume as an operating signal. It can support licensing and governance reviews, but it does not show whether the resulting code was correct, secure or useful. Pair client-level traffic with delivery outcomes, review findings and cost data before changing tool policy.