Microsoft announced on Wednesday, October 7, that Copilot on Copilot+ PCs will gain three capabilities that bring it closer to a full desktop agent: local context, to understand files and recent activity on the PC with the user’s permission; local actions, to organize files, review device diagnostics, troubleshoot problems, and run workflows directly on the machine; and local models, to run tasks using AI installed on the PC itself and turn to the cloud only when needed. The announcement, signed by Pavan Davuluri, executive vice president of Windows and Devices, on the Windows Experience blog, frames the approach as “hybrid intelligence” and anticipates a gradual rollout over the coming months, with timing varying by device, market, and chip platform.
These capabilities are integrated into the three experiences of the new Copilot introduced on September 25: Home will be able to bring together the files a user is working on and turn them into material ready to share; Code will let users create native Windows apps from a single prompt and run code with agent containment; and Autopilot, the persistent, proactive agent, will add local context and actions on the PC to its workflow.
MXC sets boundaries on what an agent can do
Microsoft Execution Containers (MXC) became generally available on Windows 11 on October 7. It lets users define which files and network destinations an agent can use through policies applied at runtime. Microsoft lists options ranging from process and session isolation to virtual machines. These are mechanisms for containing actions and managing permissions, not a guarantee that every delegated task will be correct or safe.
Agents that already support MXC include OpenAI’s Codex, GitHub Copilot, Replit, LM Studio, NVIDIA’s OpenShell, and Unsloth; Anthropic’s Claude Code, Manus, and Perplexity are among those that will integrate it. That same day, GitHub made GitHub Copilot’s local sandboxes—available in its CLI, app, and VS Code—generally available. According to GitHub’s official changelog, they are built specifically on MXC.
Hardware for running models at home
The October Windows and Surface event opened preorders for PCs with NVIDIA RTX Spark from ASUS, Dell, HP, Lenovo, MSI, and Surface itself. The Surface Laptop Ultra, with up to 128 GB of unified memory, can run models with more than 120,000 million parameters locally, according to Microsoft, and will be available on October 16, while the Surface RTX Spark Dev Box will ship to customers in the United States in November. Windows desktop PCs with DGX Station, for larger-scale AI workloads, will arrive by the end of 2026, according to Microsoft’s roadmap.
What’s arriving now and what’s still pending
MXC is already generally available, but Copilot’s new local capabilities are scheduled to roll out to Copilot+ PCs over the coming months. The company warns that timing may vary by device, market, and chip platform. Having Windows 11 does not mean all the announced features are active starting today, or that every PC can run the same models with the same performance.
It is also important to distinguish RTX Spark, the hardware announced for this generation of PCs, from DGX Spark, the AI system that evovo has covered before. The Windows announcement does not make the two names refer to the same product. When choosing hardware, it is best to start with the model you want to run, its memory requirements, and the compatible application—not just the parameter count mentioned in a presentation.
