Quick Answer
Microsoft says Windows hybrid intelligence will let Copilot+ PCs combine on-device models with cloud AI, use local PC context with permission, and take selected actions in Windows. The rollout begins over coming months, not all at once. Review Copilot permissions before enabling file-aware features, especially on work or shared computers where local data may be sensitive.
Key Takeaways
- Microsoft announced Windows hybrid intelligence on October 7, 2026.
- Copilot may use files and recent activity only with user permission.
- Microsoft says Copilot can organize files, assess diagnostics, troubleshoot, code, and run workflows.
- Local models are intended for appropriate tasks, while cloud AI supplies broader capability.
- GitHub local sandboxing has separate Windows 11 version and update requirements.
What is Windows hybrid intelligence?
Windows hybrid intelligence is Microsoft’s name for a Copilot+ PC approach that combines AI models running on the device with cloud-based AI services. Microsoft announced the strategy on October 7, 2026, and says related features will begin rolling out over the coming months. Microsoft’s Windows announcement describes the goal as using local AI models for appropriate tasks while combining them with cloud intelligence.
The practical distinction is where a task is handled. A local model runs on the PC’s available hardware, while cloud intelligence relies on Microsoft’s online systems. Windows hybrid intelligence is designed to select between those approaches rather than requiring every Copilot request to follow one path.
Windows hybrid intelligence matters because local processing can support features that need awareness of the computer’s current state, such as a device diagnostic or a workflow involving files. Cloud AI can still be useful when a request needs broader model capability. Microsoft has not said that every Copilot feature will work locally, so users should treat the announcement as a platform direction and wait for individual feature availability.
Copilot+ PC owners should watch Windows Update and Copilot notices for feature access rather than assuming that the October 7 announcement immediately changes their PC. Users who are also tracking taskbar-based controls can compare the changes with Windows Search actions, which cover a separate Windows interface feature.
Will Copilot use files and recent activity on a Windows PC?
Copilot will be able to use local context, including files and recent activity, only with permission, Microsoft says. The permission requirement is the most important privacy boundary in the announcement because local context can contain documents, work material, personal records, and traces of recent computer use.
Microsoft has not described the announcement as unrestricted access to every file on a Windows PC. The stated capability is permission-based local context, which means the available information depends on the controls Microsoft presents and the choices the PC owner makes. Readers should examine the specific prompt, setting, and account context before allowing Copilot to use local material.
Local context can make an assistant more useful because an assistant can respond to the files and work already open on the computer rather than relying only on a typed prompt. The same feature creates a higher privacy standard than a general web question, particularly on a household computer, a school device, or an employer-managed system.
The sensible approach is to grant the narrowest permission that supports the task. Avoid enabling file-aware assistance while signed into another person’s Windows account, and do not use the feature to process confidential workplace documents unless your organization permits that use. Users who want to reduce AI exposure in other services can also review how to turn off Gemini in Gmail before sharing sensitive content across multiple tools.
What actions can Copilot take directly in Windows?
Copilot will be able to take local actions in Windows, including organizing files, assessing device diagnostics, troubleshooting, coding, and carrying out workflows, Microsoft says. The announced scope moves Copilot beyond answering questions and toward helping users complete operating-system tasks.
File organization could involve arranging material into a more usable structure, while device diagnostics and troubleshooting could help interpret an issue that Windows reports. Coding and workflow support could help users move through multi-step work on a PC. Microsoft has not provided a complete list of every action, permission screen, or recovery control that will ship with the rollout.
Windows actions require more caution than ordinary chat responses because an action can change a file, alter a setting, or affect a work process. The benefit is reduced manual effort. The limitation is that an AI assistant can misunderstand an instruction, select an unwanted file, or suggest a change that does not match a particular PC configuration.
Review every proposed action before approving it, especially when a task affects folders, account settings, installed software, or diagnostic data. Back up important files before using an assistant for broad file reorganization. Stop and contact your PC maker or workplace IT support if Copilot suggests changing firmware, BIOS or UEFI settings, storage partitions, or security controls.
| Announced Copilot capability | Potential benefit | Main user check |
|---|---|---|
| Local file context | More relevant help based on PC material | Confirm which files and permissions are involved |
| Recent activity context | Help that reflects the current task | Use it only on a personal or authorized account |
| File organization | Less manual sorting work | Review the destination and keep a backup |
| Device diagnostics | Faster interpretation of Windows issues | Do not approve risky system changes without understanding them |
| Workflow assistance | Support for multi-step PC tasks | Check each action before it runs |
How does local AI differ from cloud AI on Copilot+ PCs?
Local AI runs models on the Windows PC, while cloud AI uses remote computing systems to handle a request. Microsoft says Windows hybrid intelligence combines both approaches, using local models for appropriate tasks rather than treating on-device AI as a full replacement for cloud services.
The distinction matters because a PC can use local processing for tasks tied closely to its files, settings, or hardware state. Cloud processing can support requests that need larger-scale AI capability. Microsoft’s announcement does not state which future Copilot tasks will always remain local, which will use the cloud, or whether users will be able to choose the processing path for every request.
Copilot+ PC hardware is central to Microsoft’s strategy because the company is building features around on-device AI capability. Users should not infer from the announcement that every Windows 11 PC receives the same tools at the same time. The rollout applies to Copilot+ PCs, and Microsoft says the features will arrive over coming months.
The practical response is to read feature descriptions closely when a new Copilot option appears. A description should clarify what data is used, whether the request needs an internet connection, and whether the feature can make changes on the device. Users who prefer offline-first AI can also consider how offline AI note-taking changes the data-handling tradeoff for a narrower task.
What does the GitHub HydraFusion update add?
GitHub HydraFusion will gain access to on-device models on Windows, with an experimental preview planned later in October for the GitHub Copilot app, Copilot CLI, and Visual Studio Code, Microsoft says. The preview connects Microsoft’s wider local-model work with tools used by developers.
Microsoft also says MAI Code 1.1 Flash has 137 billion total parameters and 6.8 billion active parameters. The company says a local 3-bit version is designed to reduce model size by nearly 80% while retaining a 256K context window. Those are Microsoft statements about its own model design, not independent performance measurements.
The claimed reduction matters because local model size affects whether a model can be practical on a PC. A large context window can also help a coding tool consider more material in one request. At the same time, model parameters, compression, and context capacity do not establish that a local tool will produce reliable code or understand every project correctly.
Developers should use the experimental preview on noncritical work first and review generated code before running it. Do not place passwords, private keys, customer data, or production secrets into a coding prompt simply because a tool supports local models. A local workflow can reduce some exposure, but safe software development still requires access controls and careful review.
What Windows versions are required for GitHub local sandboxing?
GitHub local sandboxing on Windows requires Windows 11 version 25H2 with KB5124010 or later, or Windows 11 version 26H1 with KB5124006 or later. GitHub’s local-sandboxing documentation lists those prerequisites for the Windows configuration.
Local sandboxing is a separate technical requirement from the broader Windows hybrid intelligence announcement. A sandbox is an isolated environment intended to limit how a tool interacts with the broader system. The isolation can reduce risk during development work, but a sandbox does not make generated code inherently safe or eliminate the need to inspect commands and dependencies.
Windows version requirements matter because a missing update can prevent a supported feature from working as expected. Check the Windows version under Settings > System > About, then install the required Windows updates through Settings > Windows Update if the device is eligible. Restart the PC when Windows requests it so the update can complete properly.
Back up active work before installing major system updates or changing development environments. Stop and contact workplace IT support if the computer is managed, if an update is blocked by policy, or if a sandboxing setup asks for administrative changes that you do not understand. Users considering broader system repair should follow a documented Windows 11 reset process rather than treating a factory reset as a first troubleshooting step.
What does Windows ML support mean for local AI models?
Windows ML now supports llama.cpp, Microsoft says, expanding support for open-source local AI models across GPU, NPU, and CPU hardware. Microsoft’s technical announcement presents the support as part of the company’s work on local models and sandboxed tools for Windows development.
A GPU is a graphics processor that can accelerate certain computing tasks, an NPU is a processor designed for neural-network workloads, and a CPU is the PC’s general-purpose processor. Supporting more than one hardware type can give developers more deployment options, particularly when a PC does not have the same AI-focused hardware as a newer Copilot+ device.
Windows ML support does not guarantee that every open-source model will run quickly, fit in memory, or deliver the same output across different PCs. Performance depends on the selected model, the available hardware, the software configuration, and the task. Microsoft has not provided a universal performance figure for llama.cpp models across Windows hardware in the announcement.
Developers should begin with a supported model and a controlled test workload, then measure performance and output quality on the target PC. Keep model files and development tools updated from trusted sources. Stop using an unfamiliar package if Windows security tools flag it, and verify the publisher before bypassing a warning.
How should Windows users prepare for Copilot actions?
Windows users should prepare for Copilot actions by reviewing privacy settings, backing up important files, and deciding which categories of local data an AI assistant may access. Microsoft’s announcement makes permission a condition for local context, so the feature’s usefulness and privacy impact will depend on the choices made when it reaches a PC.
Start with lower-risk tasks, such as asking Copilot to explain a device diagnostic or suggest a file organization plan. Review the proposed result before allowing any change. A staged approach gives users a clearer view of how the feature behaves on their own PC without exposing an entire document collection or allowing broad system changes immediately.
- Check Windows Update for eligible Copilot+ PC features as they become available.
- Review the Copilot permission request and identify the files or activity context involved.
- Back up important documents before approving file organization or workflow actions.
- Approve narrow tasks first, then verify the result in File Explorer or the relevant Windows setting.
- Contact Microsoft support, your PC maker, or workplace IT support before approving firmware, disk, or security-policy changes.
Windows hybrid intelligence may make Copilot more practical for everyday PC tasks, but automation does not transfer responsibility for the outcome. The safest use is supervised use: provide limited context, inspect proposed actions, and keep control of consequential changes.
FAQ
Will Copilot read all of my files automatically?
Copilot will not use local files automatically under the announced approach because Microsoft says local context requires permission. Review each permission request carefully, particularly on work PCs and shared family computers.
Do all Windows 11 PCs get hybrid intelligence features?
Windows hybrid intelligence features are being announced for Copilot+ PCs, and Microsoft says the rollout will occur over coming months. Availability can therefore vary by PC eligibility and the specific feature’s release status.
Can Copilot change Windows settings without approval?
Copilot can be designed to take local actions, but users should review each proposed action before allowing a change. Do not approve changes involving security, storage, firmware, or accounts unless you understand the result.
What is GitHub local sandboxing?
GitHub local sandboxing is an isolated Windows environment for supported Copilot workflows. The isolation can limit broader system interaction, but developers still need to review generated code and commands before running them.
What should I back up before using Copilot file actions?
Windows users should back up important documents, photos, project folders, and any file collection that Copilot may reorganize. A backup gives you a recovery path if a file is moved, renamed, or changed in an unintended way.