Back to rankings

GitHub Copilot Fusion: Route Between Local and Cloud Models (Experimental)

GitHub CopilotView original changelog

GitHub Copilot Fusion can now route requests between on-device models and cloud models in Copilot CLI, the GitHub Copilot app and VS Code, as an opt-in experimental preview. Windows users can set up MAI Code 1.1 Flash through the Windows ML provider, while others can bring MAI Code 1.1 Flash or Qwen3.8 Flash Next via a local OpenAI-compatible server. Routing decisions still happen in the cloud, so local inference does not make a session offline.

Key Takeaways

  • Fusion can now route between local and cloud models, extending its multi-model routing to on-device inference.
  • Windows users get a managed setup for MAI Code 1.1 Flash via the Windows ML provider.
  • Any platform can bring its own model (MAI Code 1.1 Flash or Qwen3.8 Flash Next) through a local OpenAI-compatible server.
  • It spans Copilot CLI, the Copilot app and VS Code, though the app's intelligent routing supports only managed Windows ML models.
  • Local inference is not offline, since routing decisions still happen in the cloud.
  • Local models may have different safety protections than cloud models, and BYOK models are not automatically eligible for routing.

Sources & Mentions

1 external resource covering this update

GitHub Copilot Routes Tasks Between Local and Cloud Using MAI Code 1.1 Flash

AlphaSignal

Local models join Fusion routing

GitHub Copilot Fusion, which routes and orchestrates models for each task, can now include on-device models in its choices. Fusion can send a request to a supported local model or to a cloud model, depending on the task. The capability is an opt-in experimental preview and works in GitHub Copilot CLI, the GitHub Copilot app and Visual Studio Code.

Managed setup on Windows

On supported Windows devices, users can download and configure MAI Code 1.1 Flash through the Windows ML provider. Setup differs by client:

  • In Copilot CLI, enable experimental mode and use the /model command.
  • In the Copilot app, configure the provider and enable local model auto routing.
  • In VS Code, enable the chat.copilot.localModel.enabled setting.

Bring your own local model

Users on Windows, macOS or Linux can run MAI Code 1.1 Flash or Qwen3.8 Flash Next through a local OpenAI-compatible server. In this path the user manages the download and the runtime. In Copilot CLI, local routing works with these models. In the Copilot app, intelligent routing supports only the managed Windows ML models.

Caveats

GitHub flagged several limits. Bring-your-own-key models are not automatically eligible for Fusion routing. Local inference does not make a session offline, because routing decisions are still made in the cloud. Local models may also have different safety protections than cloud models, which teams with strict policies should weigh before enabling the feature.

Mentioned onAlphaSignal