GitHub's Copilot usage API now counts model traffic from tools outside the IDE

Victoria Shinder26 October 2026, 14:481 min read

GitHub's Copilot usage metrics API added a totals_by_client array on 23 October, reporting model usage broken down by the tool that sent each request. It appears in the enterprise, organization, enterprise-user and organization-user reports, in both the 1-day and 28-day versions. Each entry names a client — opencode, claude_code, codex, zed, copilot_cli, github_app, cloud_agent, code_review, vscode, jetbrains, with anything unrecognised grouped under other — and reports request_count, session_count and token usage, with a further breakdown by model that records whether the model was user-selected, auto-chosen or product-selected.

The change is additive: existing fields are untouched and the new array is empty when there is no usage. GitHub cautions that request_count counts model calls rather than prompts, and that clients are identified by the integration ID they send, so the grouping is best-effort, not a verified identity.

What it means

The honest admission buried here is that a lot of Copilot usage was invisible. Tools that drive the model from outside an editor — CLIs, agents, third-party clients — don't send editor telemetry, so they never showed up in the IDE-based totals. Counting requests at the server closes that gap, and for anyone doing capacity or governance work it turns "we think most usage is in VS Code" into a number. The caveat about best-effort identity is the one to carry into any report built on this: a client string is a claim the integration makes about itself, not something GitHub verifies, so treat the breakdown as a strong signal rather than an audit trail.