What is new?
On 23 October 2026, GitHub added the totals_by_client field to the Copilot usage metrics API. It captures model usage from tools that use Copilot as their model provider, including tools such as OpenCode that were previously missing from reports. The field is available in the 1-day and 28-day reports for enterprises, organizations and individual users.
The background, according to GitHub: more and more developers use Copilot as the model provider behind tools outside the editor. These often send no editor telemetry and were therefore missing from totals_by_ide. The new count happens on the server and closes this gap.
What data does totals_by_client provide?
Each entry represents one tool. Unrecognized tools are grouped under other. GitHub names opencode, claude_code, codex, zed, copilot_cli, github_app, cloud_agent, code_review, vscode and jetbrains as examples.
- request_count: number of successful model requests, not prompts
- session_count: number of distinct sessions
- token_usage: prompt and output tokens plus average tokens per request
- totals_by_model: the same values per model, with model_selection_mode (user_selected, auto, product_selected)
What do reports show about custom extensions?
Since 22 October 2026, the totals_by_skill, totals_by_custom_agent, totals_by_mcp, totals_by_slash_cmd and totals_by_plugin lists show the configured name of custom entries instead of other or custom. According to GitHub, each list now contains up to 25 entries instead of five; the rest is combined under other.
The names come from Copilot CLI, the GitHub Copilot app, ACP integrations such as JetBrains and the Copilot SDK. This requires Copilot CLI 1.0.97 or later and Copilot SDK 1.0.21 or later. No names are reported from an IDE's built-in chat. Names appear only for report days after the release.
Where are the limits of the numbers?
GitHub names several limitations. request_count counts model calls; one prompt can trigger several calls, so the field cannot be compared directly with prompt counts. Tools are identified by the integration ID they send, which is a best-effort grouping, not a verified identity. For code review and the cloud agent, the product chooses the model, and code review models appear as unknown.
The names of custom extensions come from client telemetry and, according to GitHub, can undercount, for example with outdated clients or blocked telemetry. GitHub recommends using them for trends and adoption, not exact accounting. Names should also be reviewed before sharing reports outside the organization.
Why does this matter for businesses?
Many teams now use several AI tools in parallel. When Copilot serves as the model provider behind Claude Code, Codex or OpenCode, this usage was hardly visible until now. With totals_by_client, businesses can see which tools drive requests and tokens and base decisions about supported tools and policies on actual usage, as GitHub describes.
For works councils and data protection: reports are also available per user. Clarify in advance who may see the data and what it is used for. Monitoring the performance of individuals is a sensitive issue and should be covered in your AI policy.
What should admins do now?
Check whether the Copilot usage metrics policy is enabled, because without it there are no reports. Grant the View Copilot Metrics permission selectively through custom roles. Evaluate totals_by_client over 28 days and compare the list with the tools you have approved. Update Copilot CLI and the SDK so that names of custom skills and MCP servers appear.
Do existing fields change? No, according to GitHub the change is additive, and totals_by_client is an empty array when there is nothing to report. The free AI System Check helps organize tools, approvals and metrics across your team. These notes are suggestions, not customer results.
Keep it verifiable
