GitHub published three Copilot changes across June 23 and June 24 that point in two directions at once: simpler default routing for some users, and more control for teams that need their own model providers.
For Copilot Free and Student plans, GitHub says auto model selection is now the default and only model selection experience. Auto chooses a model dynamically for each task, subject to plan restrictions. GitHub is also retiring the (Preview) label from Microsoft-released models because routing and model improvements are happening behind the scenes.
At the same time, the GitHub Copilot app now supports bring your own key, or BYOK. Users can connect model providers including OpenAI, Azure OpenAI, Microsoft Foundry, Anthropic, LM Studio, Ollama, and OpenAI-compatible endpoints.
That combination says a lot about where Copilot is going.
Defaults are becoming opinionated
The Free and Student change reduces model choice. For those plans, GitHub wants users to ask Copilot for help and let the service route the request.
That makes sense for lower-friction use. Many users do not want to think about which model family fits a refactor, explanation, test, or small agent task. Auto routing can hide provider churn, preview labels, and model retirement cycles.
The trade-off is visibility. If users cannot choose a model manually, they also have less control over the reasoning behind a result, the speed/cost trade-off, and whether a specific model behaves better for their workflow. GitHub is effectively saying that, for these plans, routing quality should be a platform responsibility.
This is becoming a broader pattern in AI products. Consumer and entry-level tiers get simpler controls. Enterprise and power-user surfaces get more policy, provider, and routing options.
BYOK moves the Copilot app toward enterprise boundaries
The Copilot app BYOK change sits on the other side of that split. GitHub says users can add a provider in Settings with an endpoint and API key, or just a host for LM Studio or Ollama. The provider’s models then appear in the model picker alongside Copilot-hosted models.
GitHub says keys are stored in the local OS keychain and are not read back by the UI. It also notes that Business or Enterprise access to the Copilot app depends on admins enabling Copilot CLI in policy settings.
The value is not only model variety. BYOK lets teams route inference through an existing cloud account, internal gateway, local model, or self-hosted stack. That matters for regulated environments, quota management, regional controls, and data-boundary requirements.
It also gives developers a practical split: use a frontier hosted model for complex reasoning, and a local or tenant-controlled model for execution, narrower tasks, or data-sensitive work.
The CLI is becoming a work surface
GitHub also made the redesigned Copilot CLI terminal interface generally available. The new interface adds tabs for sessions, gists, issues, and pull requests when the CLI is run inside a GitHub repository.
The workflow is small but useful. A developer can highlight an issue or pull request, drop a reference into the prompt, and ask Copilot to investigate, fix, comment, or review it. GitHub also moved tool configuration into the CLI, including /mcp add, /mcp search, /skills, /plugin, and /settings.
That makes the terminal a fuller agent surface. Copilot is not only answering prompts. It is browsing work items, connecting tools, and operating closer to repository context.
The platform split is the story
These changes should be read together. GitHub is simplifying Copilot for users who should not need to manage model selection, while expanding control for users and organizations that need provider choice, local models, key ownership, and terminal-native workflows.
The next question is governance. BYOK makes model routing more flexible, but it also creates more combinations to support and audit. Auto mode simplifies the user experience, but it asks users to trust GitHub’s routing decisions.
For teams already tracking Copilot AI credits, the next useful layer is visibility across this routing stack: which models ran, under which policy, through which provider, and with what outcome. GitHub is giving users more paths. Enterprises will want the map.
