New Copilot Usage API Fields Expose BYOK and Enterprise Custom Model Activity

The Copilot usage API now returns explicit BYOK (bring‑your‑own‑key) and enterprise custom model data, letting you distinguish those requests from standard Copilot‑hosted usage. The change matters because it surfaces previously blended activity, enabling clearer policy enforcement, spend analysis, and cross‑tool reporting.

What Changed in the API

Two new reporting dimensions were introduced:

• used_local_byok – flags a user who made at least one local BYOK request in a day via the Copilot CLI or app.
• totals_by_local_byok – breaks down session, request, prompt, and token counts by byok_kind (e.g., local_managed, local_user, local_auto, remote_user) and by surface (copilot_cli or copilot_app).
• Aggregate user activity counters: daily_active_local_byok_users, weekly_active_local_byok_users, and monthly_active_local_byok_users.
• A boolean is_custom_enterprise_model on each totals_by_client[].totals_by_model[] entry, set to true when the request used a model configured at the enterprise or organization level.

Both BYOK and custom‑model entries appear in 1‑day and 28‑day reports at user and aggregate levels, and they are included in enterprise and organization scopes.

Why It Matters to Engineers and Ops

Visibility into BYOK activity lets you enforce internal policies that restrict or audit the use of external provider keys, local model servers (e.g., Ollama, LM Studio), or custom enterprise models. Cost attribution becomes more accurate because usage billed to your own providers or run on free local models no longer inflates the Copilot‑hosted totals. The unified view across IDEs, CLI, and the Copilot app simplifies cross‑tool efficiency analysis and helps you compare BYOK users against standard Copilot consumers.

Architectural and Operational Implications

Local BYOK metrics rely on client‑side telemetry. If telemetry is disabled, counts will be lower than actual activity, and token counts may be missing for providers that do not report them. In contrast, enterprise custom model usage is observed server‑side, so its counts are complete. Existing aggregate totals remain unchanged; BYOK data is a subset of the existing totals_by_cli and totals_by_copilot_app metrics, while custom model data already lives in totals_by_client.

Access to the new fields is limited to roles that can view Copilot metrics: enterprise owners, billing managers, organization owners, and any custom role granted the “View Copilot Metrics” permission. The Copilot usage metrics policy must be enabled for the data to be returned.

Because historical data is not backfilled, reports for days prior to this release will lack the is_custom_enterprise_model flag. Model names for custom models follow the existing normalization rules; unknown names appear as unknown, and local BYOK usage is not broken out by model name.

Related CloudNinjas coverage: AI engineering.

What This Means For Practitioners

• Enable the Copilot usage metrics policy if you haven’t already, and verify that the appropriate roles have the “View Copilot Metrics” permission.
• Ingest the new BYOK fields into your existing monitoring or cost‑allocation pipelines to separate external provider spend from GitHub‑hosted usage.
• Audit telemetry settings on developer machines; consider enforcing telemetry for accurate BYOK reporting or account for under‑counting in your analyses.
• Update any internal dashboards that aggregate Copilot usage to include the new user‑level BYOK flags and the is_custom_enterprise_model boolean, ensuring you do not double‑count requests that appear both as hosted and custom.
• Review BYOK governance policies against the newly visible activity, and adjust access controls or usage guidelines as needed.

By integrating these metrics, you gain a clearer operational picture of AI model consumption across your organization, supporting better governance, budgeting, and performance tuning.