# Microsoft's Windows AI launch spans available tools and future Copilot features

> Microsoft's October 7 Windows announcements cover MXC's general availability, experimental local model tools, a later preview, future Copilot+ features and RTX Spark preorders.

By BIG CHANGE Editorial

Published: 2026-10-08T04:14:55.343Z
Updated: 2026-10-08T04:14:55.343Z
Canonical: https://bigchange.ai/blog/microsoft-windows-hybrid-intelligence-release-stages

![Conceptual charcoal illustration of a desktop PC connected by cable to a wall jack, with a server rack visible through a doorway.](https://bigchange.ai/api/media/file/windows-hybrid-local-remote-hero-v1.png)
AI-generated conceptual illustration by BIG CHANGE.

Microsoft has set out a plan for Windows to run AI work on a PC, in the cloud, or across both. Its October 7 announcements cover agent containment, local model tools, GitHub Copilot routing and new PCs. They have different release dates and requirements. For someone deciding what to try or buy, the distinction matters more than the shared name, “hybrid intelligence.”

The clearest current release is [Microsoft Execution Containers (MXC)](https://blogs.windows.com/windowsdeveloper/2026/10/07/microsoft-execution-containers-policy-driven-containment-for-ai-agents/), which Microsoft says is generally available. Developers can use it to put an agent's tools or generated code inside a policy that defines file, network and interface access. The new local routing in GitHub Copilot is an experimental preview due later in October. Copilot's broader use of PC context, actions and local models is expected to start rolling out on Copilot+ PCs over the coming months.

## The big change

- **What changed:** Microsoft is connecting controls over what an agent can do with more ways to run models on a Windows PC. MXC is available now; the announced Copilot experiences have later or experimental release stages.
- **Who it affects:** Developers can examine MXC's SDK and policies now. Copilot users and organizations need to check device requirements and rollout status before planning around local routing or PC actions.
- **What follows:** The practical test will be whether these separate parts work together in released products. Microsoft's announcements describe that direction, but they do not establish availability of every feature on every Windows PC.

## A release map for the Windows announcements

| Part of the announcement | Status on October 7 | Who can use it or what remains open |
| --- | --- | --- |
| Microsoft Execution Containers | Microsoft says MXC is generally available, including Windows 11 and Windows 365 support. | Agent developers integrate its SDK and define a workload policy. Backends differ by operating system; some remain experimental. |
| Windows ML support for GGUF models through llama.cpp | Available as an experimental developer integration. | Developers supply a model and use the new text generation API or local endpoint. This is not a general Copilot rollout. |
| GitHub Copilot local models and HydraFusion routing | Experimental preview planned for later in October in the Copilot app, CLI and VS Code. | Microsoft describes automatic local/cloud routing and explicit local model selection. The launch does not make these features generally available. |
| Copilot's local context, actions and models | Expected to begin rolling out over coming months. | Microsoft specifies Copilot+ PCs and says availability can vary by device, market and silicon. Local context and actions are described as permission based. |
| RTX Spark Windows PCs | Surface Laptop Ultra and several partner PCs opened for preorder. | Microsoft says those laptops begin shipping October 16; the Surface RTX Spark Dev Box is slated to ship to US customers in November. Neither preorder is a hands-on result. |

The dates and conditions come from [Microsoft's Windows launch post](https://blogs.windows.com/windowsexperience/2026/10/07/building-windows-for-hybrid-intelligence/), its [MXC developer post](https://blogs.windows.com/windowsdeveloper/2026/10/07/microsoft-execution-containers-policy-driven-containment-for-ai-agents/) and its [Windows ML developer post](https://devblogs.microsoft.com/foundry-on-windows/build-on-winml-oct-7-26/). Microsoft also announced Search actions for Windows Insiders in an experimental channel, a separate rollout from the AI system described here.

## MXC gives developers a policy boundary for agent tools

An AI agent may need to run commands, read a project folder or reach a service. MXC lets the developer specify those resources in a JSON request and execute the workload within a selected containment backend. Microsoft's [public repository](https://github.com/microsoft/mxc) provides Rust, .NET and Node SDKs, samples and a versioned schema. The repository describes MXC as an SDK dependency within an application, not a switch that automatically confines every agent on a PC.

Microsoft says the generally available options include a process container on Windows 11, macOS and Linux, as well as a Windows 11 session container and WSL container. Its repository labels Windows Sandbox, MicroVM and Hyperlight backends experimental; the [developer announcement](https://blogs.windows.com/windowsdeveloper/2026/10/07/microsoft-execution-containers-policy-driven-containment-for-ai-agents/) likewise marks MicroVM experimental. A developer choosing a backend still has to check the specific platform and security properties. Microsoft's account of the enforcement model is a product claim, not an independent security evaluation by BIG CHANGE.

There is also an important distinction inside Microsoft's “containment, identity and manageability” framing. MXC provides the containment layer now. The developer post says Entra attribution of agent activity and Intune policy for MXC process containers are coming soon. IT teams should evaluate those controls on their own release status rather than treating the whole set as generally available.

## Local models and routing have different starting points

[Windows ML's new GGUF path](https://devblogs.microsoft.com/foundry-on-windows/build-on-winml-oct-7-26/) gives developers an experimental way to run a supplied model through its Text Generation API; the same post describes an OpenAI-compatible local endpoint. Microsoft says a separate Speech Recognition API accepts an ONNX Whisper model. These are development surfaces, with model files and suitable compute still required. Microsoft has not turned every Windows PC into the announced Copilot experience by shipping these APIs.

For GitHub Copilot, the intended choice is between explicit local model selection and HydraFusion's automatic routing across local and cloud models. Microsoft's [technical explanation](https://commandline.microsoft.com/local-models-sandboxed-tools-github-windows/) says the preview will reach the Copilot app, CLI and VS Code by the end of October. It names MAI Code 1.1 Flash on the Windows ML provider and OpenAI-compatible local endpoints as options for explicit selection. It also distinguishes model inference from tool execution: running a model locally does not itself restrict the commands an agent launches, which is why the MXC integration matters.

That explanation reports Microsoft's model footprint, throughput and benchmark measurements on specific preproduction hardware and configurations. Those numbers do not establish how an individual user's full coding session will perform. We have not tested the device, the model or the routing preview.

## The new PCs are an access path, not a prerequisite for MXC

Microsoft positions RTX Spark PCs as machines for larger local models. It opened preorders for Surface Laptop Ultra and partner laptops from ASUS, Dell, HP, Lenovo and MSI, with October 16 as the stated shipping start. Surface RTX Spark Dev Box was also offered for preorder, with US shipment expected in November. Microsoft says Surface Laptop Ultra can be configured with up to 128 GB of unified memory. Its performance comparisons are company or partner tests, not measurements by BIG CHANGE.

Those devices provide a different route from the MXC SDK, which Microsoft's repository lists across Windows 11, Linux and macOS. A Windows user interested in Copilot's local context or actions should instead watch the Copilot+ PC rollout and the feature's availability on their device and in their market. A developer interested in containment can inspect the current [MXC samples and backend documentation](https://github.com/microsoft/mxc) now. Anyone planning around HydraFusion routing needs to wait for the experimental preview and check its supported app, model and hardware conditions when it arrives.

## Sources & further reading

- [Microsoft, “Building Windows for hybrid intelligence” (October 7, 2026)](https://blogs.windows.com/windowsexperience/2026/10/07/building-windows-for-hybrid-intelligence/) is the launch overview and source for device preorder and shipping dates, Copilot+ PC rollout language, and HydraFusion timing. Its performance figures are Microsoft's claims based on specified tests, not independently verified here.
- [Microsoft, “Microsoft Execution Containers: Policy-driven containment for AI agents” (October 7, 2026)](https://blogs.windows.com/windowsdeveloper/2026/10/07/microsoft-execution-containers-policy-driven-containment-for-ai-agents/) explains the announced MXC availability, policy model, backend examples and separate future identity and Intune controls. It describes intended security properties; this article does not independently validate them.
- [Microsoft's MXC repository](https://github.com/microsoft/mxc), reviewed October 8, provides the current SDK, schema, samples and backend status. Its README is more specific than the launch post about experimental Windows Sandbox, MicroVM and Hyperlight backends; implementations and prerequisites can change.
- [Microsoft, “AI Development on Windows: from PyTorch and llama.cpp to Windows ML” (October 7, 2026)](https://devblogs.microsoft.com/foundry-on-windows/build-on-winml-oct-7-26/) documents the experimental GGUF/llama.cpp path and developer APIs. It does not say that Copilot's future local features have already rolled out.
- [Microsoft and GitHub, “Bringing local models and sandboxed tools to Windows and GitHub Copilot” (October 7, 2026)](https://commandline.microsoft.com/local-models-sandboxed-tools-github-windows/) details planned routing and direct selection, tool sandbox boundaries, and Microsoft-reported model measurements. The announced routing preview is due later in October.

## Sources

- [Building Windows for hybrid intelligence](https://blogs.windows.com/windowsexperience/2026/10/07/building-windows-for-hybrid-intelligence/) — Launch overview: feature stages, PC preorders, shipping estimates and Copilot rollout caveats.
- [Microsoft Execution Containers: Policy-driven containment for AI agents](https://blogs.windows.com/windowsdeveloper/2026/10/07/microsoft-execution-containers-policy-driven-containment-for-ai-agents/) — MXC GA claim, policy surfaces and separate forthcoming identity and Intune controls.
- [Microsoft MXC repository](https://github.com/microsoft/mxc) — SDKs, platform matrix and experimental backend labels. Living repository; recheck before publication.
- [AI Development on Windows: from PyTorch and llama.cpp to Windows ML](https://devblogs.microsoft.com/foundry-on-windows/build-on-winml-oct-7-26/) — Experimental GGUF integration and local developer APIs.
- [Bringing local models and sandboxed tools to Windows and GitHub Copilot](https://commandline.microsoft.com/local-models-sandboxed-tools-github-windows/) — Future routing preview, explicit model selection and tool execution boundary distinctions.