Windows Goes Hybrid: Copilot Gets Local Models, MXC Sandboxes Ship
Microsoft makes MXC agent sandboxes generally available on Windows 11 and previews hybrid intelligence Copilot features that use local models and files.
Microsoft is turning Windows into a platform where AI agents run locally when it makes sense and reach the cloud when they need to. The company calls it hybrid intelligence, and it announced the strategy on October 7 in a blog post by Pavan Davuluri, Executive Vice President of Windows + Devices.
The headline pieces: Microsoft Execution Containers (MXC), the sandboxing layer for agents, is now generally available on Windows 11, and a new generation of Copilot features that can read your files, take actions on your PC and use local models is expected to begin rolling out on Copilot+ PCs in the coming months.
Agents get a sandbox, and a list of launch partners
MXC lets organizations define which files and networks an agent can access, with those policies enforced at runtime. According to Microsoft, leading agents already support it, including Codex from OpenAI, GitHub Copilot, OpenClaw, Replit, LM Studio, OpenShell from NVIDIA and Unsloth AI. Anthropic Claude Code, Box, Egnyte, Heidi Health, Hermes Agent by Nous Research, Manus, Perplexity, Raycast and Simular are among those Microsoft says will add support.
Meta’s personal AI agent, Muse for Windows, is coming soon as a native Windows app with MXC integration, Microsoft said.
The Register’s Tobias Mann noted the security pitch comes with familiar caveats: local agents will run in MXC sandboxes with fine-grained permissions, but the publication was skeptical of Davuluri’s claim that Windows is now “the most secure platform for agents.”
Local models, routed by GitHub HydraFusion
Microsoft is bringing MAI Code 1.1 Flash — 137 billion total parameters, 6.8 billion active — to local PCs using 3-bit precision that cuts model size by nearly 80% while keeping a 256K context window. Also headed local: an upcoming Nemotron model from NVIDIA with over 70 billion parameters, quantized in 2-bit to use just over 20GB of memory, plus DeepSeek V4 Flash, a 284B parameter model.
GitHub’s HydraFusion router, which already picks the right cloud model per task, is being extended to tap models running on device. Microsoft says hybrid intelligence powered by HydraFusion is coming to the GitHub Copilot app, GitHub Copilot CLI and Visual Studio Code in experimental preview later in October. The Register reports the GitHub Copilot rollout starts next week, with the Copilot app following over the next few months.
For developers, llama.cpp support is coming to Windows ML, Microsoft’s runtime for deploying models across GPU, NPU and CPU.
What Copilot will actually do on your PC
On Copilot+ PCs, Copilot will be able to use local context such as files and recent activity with your permission, take actions like organizing files, assessing device diagnostics and troubleshooting, and tap into local models. In Home, it can pull in files you have been working on and turn them into collaboration-ready artifacts; in Code, it can help build native Windows apps from a single prompt and run code more securely with MXC; in Autopilot, it brings local context and PC actions into the agent flow.
Microsoft insists it will only look through your documents if you ask it to. In a demo, per The Register, Microsoft showed Copilot finding and organizing tax documents, then drafting an email to an accountant with those files attached.
Windows Search on the taskbar is also getting new ways to take action, and mini desktop PCs get a simplified out-of-box setup for popular agents in a few clicks, starting with a native Windows gateway for OpenClaw users.
Pre-orders open for RTX Spark machines
Surface Laptop Ultra, powered by NVIDIA RTX Spark, is available for pre-order today alongside new devices from ASUS, Dell, HP, Lenovo and MSI. It combines up to 128GB of unified memory and can run AI models exceeding 120 billion parameters locally. Surface Laptop Ultra ships beginning October 16, while the Surface RTX Spark Dev Box ships to U.S. customers in November.
Microsoft claims Windows PCs with RTX Spark deliver up to 2.1x faster time to first token, 4.3x faster AI image generation and 6.2x faster AI video generation compared with an Apple MacBook Pro 16-inch with M5 Pro.
Copilot features powered by hybrid intelligence are expected to begin rolling out on Copilot+ PCs in the coming months. If you own a Copilot+ PC, that update is the one to watch.
Related reading
Sources
- Building Windows for hybrid intelligence — Windows Blog
- Microsoft is about to let Copilot loose on your file system — The Register
Some links on this page may be affiliate links. If you buy through them we may earn a commission at no extra cost to you. See our affiliate disclosure.
