On November 3, GitHub expanded its Copilot usage metrics API with local BYOK and enterprise custom-model fields. Local BYOK includes user-managed providers and locally served models; enterprise custom models are configured centrally and observed server-side. GitHub notes that local client telemetry can undercount usage and earlier report days are not backfilled.
For platform leaders, AI usage reporting is becoming a governance interface. A mixed-model strategy needs trustworthy attribution by provider, user, runtime and cost owner—but a reporting API is not itself a complete audit or billing system.
Source: GitHub ↗
The following is original Binary Solutions editorial analysis. It explores the engineering implications of this topic; it does not reproduce the source article or claim independent verification of its results.
Distinguish model routing from model ownership
Two model calls with the same displayed name may be billed, hosted and governed differently. Enterprise reporting must preserve whether the call used a Copilot-hosted model, a developer-managed key, a local runtime or a centrally configured custom model.
Treat telemetry completeness as a quality attribute
GitHub explicitly warns that local BYOK metrics rely on client telemetry and may undercount when disabled. Token accounting may also be incomplete for some local providers. Dashboards should mark those limitations rather than present false precision.
Keep authorization separate from measurement
Knowing that a developer used a local model does not establish whether the action complied with policy. Organizations still need approved-provider lists, data classification rules, endpoint controls and a documented exception path.
Build a consistent cost model
Provider-billed usage and local compute consumption require different accounting methods. Connect usage records to infrastructure cost, model pricing, project ownership and completed engineering outcomes.
Use the API as a starting point
Export the one-day and 28-day metrics, annotate missing telemetry and track the release boundary because historical custom-model fields were not backfilled. Compare against provider bills and approved model inventories before relying on totals for policy decisions.
Follow the evidence
Original publisher announcement: [BYOK and enterprise custom models in the Copilot usage API](https://github.blog/changelog/2026-11-03-byok-and-enterprise-custom-models-in-the-copilot-usage-api/). This article distinguishes publisher facts from Binary Solutions editorial interpretation. The source date is not the Binary publication date.
Follow the evidence
Use the original publication for the author’s full argument, methodology and updates. This perspective is a starting point for a conversation about your own context.
Open original publication ↗← Return to the archive