GitHub Copilot Will Soon Switch Between Local and Cloud AI

GitHub Copilot will soon be able to automatically choose between local and cloud AI models on Windows, giving developers more flexibility over where AI workloads run.

We already know that Microsoft is “supercharging” Copilot on Windows with hybrid intelligence, and the company is now extending that approach to GitHub Copilot.

GitHub Copilot will choose between local and cloud AI

Microsoft says GitHub Copilot will be able to automatically choose between cloud and local AI models by the end of October.

The system will handle local and cloud inference automatically, so developers will not need to manage the underlying infrastructure themselves.

Developers who want more control can manually select a local model instead of relying on automatic orchestration. This option should help with workflows that require a particular model, provider, or endpoint.

Local AI support will initially extend across GitHub Copilot CLI, the Copilot app, and Visual Studio Code.

Microsoft’s approach allows Copilot to use local hardware when it makes sense while retaining access to larger cloud models for workloads that require them.

MAI Code 1.1 Flash will run through Windows ML

One of the models available for local Copilot workloads will be MAI Code 1.1 Flash.

Microsoft recently introduced the on-device mixture-of-experts model with 137 billion total parameters and 6.8 billion active parameters.

Developers will be able to select MAI Code 1.1 Flash through the Windows ML provider.

The model will initially ship on the new Surface Laptop Ultra, which uses NVIDIA RTX Spark hardware to handle demanding local AI workloads.

The integration gives Microsoft a way to combine its own local coding model with GitHub Copilot without forcing developers to move requests through cloud infrastructure.

GitHub Copilot will support other local models

Microsoft is not limiting the system to MAI Code 1.1 Flash.

GitHub Copilot will also support OpenAI-compatible local endpoints and the models exposed through them.

Developers can explicitly select a provider, model, or endpoint when a workflow depends on a specific configuration.

This means teams running their own local AI infrastructure can connect those models to Copilot while keeping more inference on their own devices or systems.

Automatic orchestration remains available for developers who do not want to choose a model manually. Copilot can decide whether to use local or cloud inference depending on the workload and available resources.

Microsoft Execution Containers will sandbox Copilot tools

Microsoft is also adding security controls around tools executed by GitHub Copilot.

Microsoft Execution Containers will provide sandboxing and translate account policies into operating system-native security controls.

On Windows, the sandbox will use ProcessContainer’s BaseContainer tier. Linux systems will use bubblewrap, while macOS will rely on Seatbelt.

This approach limits what AI-powered tools can access while they execute commands or perform development tasks.

Microsoft also plans to expand the isolation model in future versions. The company says later implementations could support separate virtual machines or container images for different workloads.

By combining automatic model selection, local AI support, and sandboxed execution, Microsoft is moving GitHub Copilot toward a hybrid model where developers can use on-device AI without giving up access to cloud models.