Local models and sandboxed tools coming soon to GitHub Copilot (commandline.microsoft.com)

GitHub has announced the upcoming integration of local models and sandboxed tools into GitHub Copilot, set to roll out by the end of the month. This update will enable developers to choose between utilizing powerful on-device intelligence or leveraging cloud-scale models for various coding tasks, bolstering the flexibility and autonomy offered by the platform. With new intelligent orchestration features, GitHub Copilot will automatically manage local and cloud inference based on task needs, enhancing performance while reducing the burden of infrastructure management for developers. Significantly, this enhancement also introduces Microsoft Execution Containers (MXC) to provide security for interactive coding sessions. These containers will enforce strict access policies, ensuring that agents operate within defined boundaries, safeguarding files, networks, and system capabilities. The new MAI Code 1.1 Flash model—a quantized, locally optimized version with sophisticated decoding techniques—promises improved responsiveness and memory efficiency while maintaining code quality. With adaptive options for local model use and an emphasis on security through sandboxing, these developments are set to empower developers with greater control and confidence in their coding processes.