GitHub announced on 3 November 2026 that the Copilot usage metrics API now reports bring-your-own-key (BYOK) usage and enterprise custom model usage separately from Copilot-hosted models.
What the changelog says
- Local BYOK means a developer points Copilot CLI or the GitHub Copilot app at their own provider key or a local model (Ollama, Foundry Local, LM Studio and llama-server are named). New fields count users, sessions, requests and tokens for it.
- Enterprise custom models are set up by admins in GitHub settings. A new flag,
is_custom_enterprise_model, marks them in the reports. - Reports are for enterprise and organisation owners, billing managers, and custom roles with the View Copilot Metrics permission. The metrics policy must be switched on.
- Local BYOK figures are a lower bound: they rely on client telemetry and undercount when it is disabled. Earlier days are not backfilled.
The interpretation
AI coding tools are no longer one subscription with one bill. Developers can plug in their own model keys, and companies can add their own approved models. That is flexibility, but it also means your AI spend and your data exposure can sit outside the tool you thought you were buying. GitHub is adding the reporting to catch up, and even then it admits some usage will go uncounted.
So what
If your software is built with AI assistance, ask who decides which models see your code, who pays for them, and how usage is reported to you. A good partner can answer in a sentence. BuildApps works with AI-assisted development under clear ownership and review rules. See AI-assisted development or start a project.