Last Updated on June 11, 2025 by Arnav Sharma
AI on your device, tools in your flow
Build 2025 wasn’t just about the cloud. Windows showed up strong with a clear focus on local AI development, smarter app tooling, and native support for AI models. Microsoft is positioning Windows as the go-to platform for building, testing, and running AI experiences, even without an internet connection.
Here’s what made headlines for Windows this year.
Windows AI Foundry brings AI right to your desktop
The big news is Windows AI Foundry. It’s a new unified platform that helps developers work with AI models on their devices. You can select models, fine-tune them, optimize performance, and run everything locally using your hardware.
Windows AI Foundry supports CPUs, GPUs, and NPUs and builds on top of Windows ML and ONNX for a smooth developer experience.
Foundry Local makes offline AI practical
Foundry Local lets developers run models and agents directly on Windows 11 or macOS devices. This is a huge win for privacy-focused applications and for teams working in secure environments.
Whether you’re building something for healthcare, finance, or just want your app to run anywhere, Foundry Local offers flexibility without sacrificing performance.
Models and APIs ready to use out of the box
Microsoft is shipping prebuilt AI models with Windows. You can use them for tasks like text generation, image recognition, object removal, and advanced search.
APIs for semantic and lexical search are also included. These support retrieval augmented generation workflows using your own data, all processed locally on your device.
Developer tools get a major quality-of-life boost
Windows is streamlining the developer workflow in a big way. New features include:
- WinGet Configuration to replicate development environments in one step
- Command Palette in PowerToys for fast navigation and control
- A new default text editor called edit, which is open source and accessible right from the terminal
- A fully open-sourced Windows Subsystem for Linux
Everything is focused on helping developers stay in flow and avoid unnecessary switching between tools.
Windows apps can now speak agent
With native support for Model Context Protocol, Windows apps can now expose their capabilities to AI agents. This makes it easier for agents to discover and interact with local applications.
App Actions allow developers to define what their apps can do in a standardized format. It improves discoverability and opens the door for more integrated AI experiences on Windows.
Privacy and security take priority
Two key changes are coming to give users more control and help developers follow best practices.
- Access to the camera, microphone, and location will default to off unless the user gives explicit permission
- Admin-level actions will now require user validation through Windows Hello
- Apps can declare package identities to streamline permissions and protect sensitive actions
These changes aim to make privacy a default rather than an afterthought.
Better support for React Native and the Windows App SDK
React Native for Windows now includes support for the new React architecture. Developers using the Windows App SDK will also benefit from its new modular packaging system. You only bring in the features your app needs.
There’s also experimental support for LoRA fine-tuning on device, especially on Snapdragon NPUs. This means even small or custom AI models can run locally with excellent performance.
Final thoughts
Windows is not just a place to consume AI. It’s turning into a platform to build, test, and run smart experiences wherever you are.
Whether you’re building a native app, tuning an AI model, or testing how your agent responds to live data, Windows gives you the environment to do it all without leaving your desk or relying on cloud services.
I help organisations secure their cloud infrastructure and stay ahead of evolving cyber threats. Microsoft MVP and Certified Trainer, author of Mastering Azure Security, and founder of arnav.au — a platform for practical Cloud, Cybersecurity, DevOps and AI content.
Frequently Asked Questions
Windows AI Foundry is a unified platform that enables developers to work with AI models directly on their devices. It allows you to select models, fine-tune them, optimize performance, and run everything locally using your hardware, with support for CPUs, GPUs, and NPUs built on Windows ML and ONNX.
Yes, Foundry Local allows developers to run AI models and agents directly on Windows 11 or macOS devices without requiring internet connectivity. This is particularly valuable for privacy-focused applications, secure environments like healthcare and finance, and scenarios where offline capability is essential.
Microsoft is shipping prebuilt AI models with Windows for tasks like text generation, image recognition, object removal, and advanced search. The platform also includes APIs for semantic and lexical search that support retrieval augmented generation workflows using your own data, all processed locally on your device.
Native support for Model Context Protocol enables Windows apps to expose their capabilities to AI agents, making it easier for agents to discover and interact with local applications. Developers can define what their apps can do using App Actions in a standardized format, enabling more integrated and discoverable AI experiences.
Windows is implementing several privacy-first changes: camera, microphone, and location access will default to off requiring explicit user permission, admin-level actions will require Windows Hello validation, and apps can declare package identities to streamline permissions and protect sensitive actions. These changes make privacy a default rather than an afterthought.