Microsoft is adding AI tools to Windows that let agents use models on a PC alongside cloud services, with new controls over their access to files and networks. Microsoft Execution Containers (MXC) is now generally available on Windows 11, while local model routing for GitHub Copilot is scheduled for experimental preview later in October. Microsoft says the broader Copilot capabilities are planned for Copilot+ PCs.
The routing system, GitHub HydraFusion, will choose between models running on the device and those in the cloud. The preview will cover Visual Studio Code, GitHub Copilot CLI and the GitHub Copilot app. Microsoft says using local computing for suitable tasks should help customers stretch their AI token budgets while retaining access to cloud intelligence.
Read more: Microsoft Starts Routing Excel and Outlook AI Prompts to Its Own MAI Models
MXC lets organizations set boundaries for the files and networks an agent can reach, then enforce those rules as it operates. Windows supports several containment options, including virtual machines and isolation at the process or session level. GitHub Copilot, LM Studio and OpenAI’s Codex already support MXC, while Anthropic Claude Code and other agents are expected to add support. Microsoft also says Meta’s Muse for Windows will arrive soon as a native app with MXC integration.
For Copilot+ PCs, Microsoft plans to let Copilot draw on files and recent activity with the user’s permission. Planned actions include sorting files, checking device diagnostics and troubleshooting problems directly on the PC. These consumer features remain prospective, even as MXC has reached general availability.
Microsoft is bringing MAI Code 1.1 Flash to devices with 3-bit precision, saying that reduces its size by nearly 80% while maintaining coding quality. The model has 137 billion total parameters, 6.8 billion active parameters and a local context window of 256K. Windows ML is also gaining llama.cpp support, giving developers another route to running open-source models across CPU, GPU and NPU hardware.
Surface Laptop Ultra, which uses NVIDIA RTX Spark™, opened for pre-orders on October 7. Microsoft also described hardware ranging from mini desktop PCs designed to stay on continuously to deskside systems powered by DGX Station for Windows.
Microsoft’s earlier AI hardware work included Maia accelerators, first unveiled in late 2023. Maia 100 shifted OpenAI’s GPT-3 workloads away from GPUs, freeing GPU capacity for other uses.
