Bringing local models and sandboxed tools to Windows and GitHub Copilot

Microsoft announced that a model called MAI Code 1.1 Flash, described as a 130‑billion‑parameter local AI, will be integrated into Windows 11. The same announcement says the model will be made available through GitHub Copilot, extending the service with sandboxed tools that run on the user’s device. According to the Microsoft release, the integration is intended to let Windows applications call the model directly, while developers can invoke the model from the Copilot command‑line interface. Coverage of the announcement appears across several outlets. KuCoin’s report foregrounds the 130‑billion‑parameter size of MAI Code 1.1 Flash and its placement in Windows 11.

BigGo Finance emphasizes the “deeper Windows access” granted to Copilot and labels the move a “Hybrid Intelligence Push.” Neowin focuses on the developer angle, noting that local AI models will soon be usable inside GitHub Copilot. The GitHub Blog details how the new capability can be discovered in the Copilot CLI, and Microsoft’s own blog frames the rollout as a broader effort to bring local models and sandboxed tools to both Windows and Copilot. The shift reflects a broader industry trend toward on‑device AI, where computation is performed locally to address latency, privacy, and connectivity concerns. Previously, GitHub Copilot relied primarily on cloud‑hosted models, and Windows has offered limited built‑in AI services. By embedding a large local model and providing sandboxed execution, Microsoft aims to combine the speed of on‑device inference with the flexibility of cloud‑augmented “hybrid intelligence.” The announcement positions Windows 11 as a platform for advanced AI workloads without mandatory cloud calls.

Future coverage will need to confirm rollout timing, as the announcements do not specify a release schedule. Watch for updates from Microsoft and GitHub regarding the availability of MAI Code 1.1 Flash in Windows update channels and the activation of local model support in the Copilot CLI. Additional guidance on configuring sandboxed tools and any subsequent hybrid‑intelligence features is likely to follow in technical documentation and blog posts.