Michael JohnsonAs of November 3, 2026, organization and enterprise administrators can see the use of bring-your-own-key setups and custom enterprise models separately in GitHub Copilot reports. The Copilot usage metrics API update helps distinguish requests to GitHub-hosted models from BYOK (bring your own key) usage or a model configured by the organization.
For companies, this is a practical change to usage tracking: costs for an external provider and local deployments can be separated from use of models served by GitHub. Managers and administrators get a more useful view for tracking which AI workflows teams are using. GitHub says the new fields break out BYOK from existing totals without changing the counters themselves.
Which scenarios the API reports separately
GitHub distinguishes two BYOK scenarios. With local BYOK, a developer configures their own key or a local model in Copilot CLI or the GitHub Copilot app; requests go directly to the provider or local server. New fields for this activity include the user-level used_local_byok field, the totals_by_local_byok aggregates, and counts of active users with local BYOK.
GitHub lists Ollama, Foundry Local, LM Studio, and llama-server among local models and tools. The new local BYOK breakdown fields apply to Copilot CLI and the Copilot app: they break down sessions, requests, prompts, and tokens by BYOK type and usage surface when the provider supplies the relevant data. Local BYOK usage is not broken down by model name, so the report shows whether and how much this activity occurred, but not which specific local model was used.
The second scenario is a custom enterprise model configured by an administrator in organization or enterprise settings, with requests routed through GitHub to the provider. GitHub added the Boolean is_custom_enterprise_model field to model entries in totals_by_client. Because these requests are observed server-side, the field covers different clients, including IDEs.
What changes when working with reports
Previously, BYOK usage could be mixed in with metrics for Copilot-hosted models, especially when models had similar names. The new fields make it possible to separate these categories when checking internal policies and analyzing usage. GitHub says the data is available in one-day and 28-day reports, at the user level and in organization or enterprise aggregates.
Administrators should start by asking what question the report needs to answer: who is using their own key, what share of requests goes to local models, and how this affects cost estimates. They can then compare the new fields with existing metrics and internal BYOK policies. This provides a more concrete basis for discussing model choice and payment than a general Copilot activity count.
Access to the metrics depends on the user’s role: the API is intended for enterprise owners, enterprise billing managers, organization owners, and users with the View Copilot Metrics permission. The Copilot metrics policy must also be enabled. These data show recorded activity, not the quality of responses or a model’s suitability for a particular task.
Why the two new metrics have different levels of accuracy
Local BYOK has a specific tracking limitation: GitHub calculates its fields from client telemetry. If telemetry is disabled, the number of these requests will be lower than the actual number; some local providers also do not report token counts. These data are therefore useful for identifying usage and estimating its scale, but are incomplete for precise financial reconciliation.
Custom enterprise models are tracked server-side and, according to GitHub, are not subject to this telemetry limitation. The same model may appear on two separate rows if it is used both as a Copilot-hosted model and as a custom enterprise model. This separation prevents different channels from being combined just because they share a name.
There is also a timing caveat: GitHub does not add the is_custom_enterprise_model field to reports for days before the update’s release. Historical data cannot automatically be assumed to be comparable with the new categories; when analyzing trends, note the transition date and account for the schema change when comparing periods.
The update helps organizations classify Copilot usage involving their own or local models more accurately. Budgeting for local BYOK still requires accounting for gaps in client telemetry and provider data, while custom enterprise models now have a separate server-side indicator in reports.
