Unlocking AI Innovation: Windows App SDK Experimental Release with Copilot Runtime

Microsoft is once again stirring the waters of desktop innovation by rolling out an experimental release of the Windows App SDK, now armed with artificial intelligence APIs. In this latest preview, dubbed the Windows Copilot Runtime, developers and tech enthusiasts alike are getting an early taste of how AI can be seamlessly integrated right into Windows applications. If you've ever wondered how your PC might soon double as a smart assistant without being tethered to the cloud, you'll want to sit up and pay attention.

The Evolution of the Windows App SDK​

The Windows App SDK has been a critical element in empowering developers to build cutting-edge Windows applications. Traditionally, the SDK has been offered in three distinct channels:

• Stable Channel: Currently at Version 1.6.4, optimized for Microsoft Store publication.
• Preview Channel: For those who want a sneak peek at upcoming features.
• Experimental Channel: Where the newest, most groundbreaking features make their debut.

The newly released version, 1.7.0-experimental3, is the first to incorporate support for the Windows Copilot Runtime APIs. This marks a monumental shift, laying the groundwork for AI-powered capabilities to operate directly on your laptop or PC without relentless calls to cloud servers.

At the Heart of the Copilot Runtime​

This experimental release introduces support for a neural processing unit (NPU)-optimized small language model (SLM) known as Phi Silica. Much like OpenAI’s GPT models—but with a smaller footprint—Phi Silica is designed to deliver robust text generation, summarization, and content formatting capabilities while consuming significantly less power. Here’s what makes Phi Silica stand out:

• Text Generation and Summaries: Responds to prompts by generating human-like text.
• Content Reformatting: Can transform and format content, for instance, converting messy data into neat tables.
• Content Moderation: Offers built-in safeguards to minimize unwanted outputs.

Microsoft is effectively democratizing the AI models that power its flagship Copilot+ PCs by making these capabilities accessible to third-party developers through the Windows App SDK.

Getting Started: A Bit of a Labyrinth​

While the promise of on-device AI is enticing, getting set up isn’t exactly a walk in the park. To experiment with these new features, you’ll need:

• A Copilot+ PC: Running Windows 11, Version 24H2, with access to the Windows Insider Beta or Dev channels.
• An Updated Visual Studio: Configured for .NET desktop application development alongside the Windows 10 SDK.
• SDK Requirements: Ensure that the Windows App SDK C# Templates are uninstalled prior to installation. Also, don’t forget to enable support for preview releases.

Even after installation, building your first Windows Copilot Runtime applications can be a challenge. Developers might find themselves juggling specific builds and SDK versions, leading to occasional compilation issues or unexpected behavior—issues that are common in early experimental phases. However, these bumps on the road are expected to smooth out when the SDK transitions to a stable release later this year.

Under the Hood: Using Phi Silica​

Interacting with Phi Silica is done through the Microsoft.Windows.AI.Generative namespace. The procedure is fairly straightforward:

• Availability Check: Use the isAvailable method to verify that the target system includes the necessary AI model.
• Asynchronous Connections: Establish a connection with Phi Silica through asynchronous calls. Send a prompt string to the model and wait for a generated response.
• Content Moderation: Customize moderation settings to control how strict the filtering should be, ensuring safe outputs and a polished user experience.

Despite some initial hurdles—such as the model being more challenging to invoke correctly from a console application—the visual impact is undeniable. Task Manager reports show the ONNX models loading and processing efficiently, especially when compared to similar tasks performed entirely on the CPU.

Experimenting with AI on the Edge​

One of the most revolutionary elements of the Windows Copilot Runtime is how it shifts AI inferencing from the cloud to your local machine. With capabilities such as optical character recognition (OCR), image resizing, and computer vision tasks, developers can now build applications that run AI processing locally. This not only reduces dependency on data centers but also inherently enhances privacy and responsiveness.
In addition to the runtime APIs, Microsoft has introduced an AI Dev Gallery as part of the rollout. This gallery serves as a showcase for Windows’ AI tools, complete with models optimized for text operations, computer vision, and more. Through the gallery, developers can explore sample projects that integrate AI into common Windows controls—imagine a combo box that not only presents data but contextualizes it with semantic intelligence.

What Does This Mean for the Future of Windows AI?​

The experimental nature of the Windows App SDK release means that while there are still some rough edges to iron out, the potential is immense. Microsoft’s vision for a Windows environment where AI functions are built directly into applications could revolutionize how we interact with our devices. By enabling both consumer and enterprise applications to operate with on-device intelligence, the reliance on extensive cloud infrastructure could be significantly reduced. This represents a significant stride towards a more decentralized, privacy-conscious AI ecosystem.

Key Takeaways​

• Microsoft is pushing forward with an experimental Windows App SDK that integrates on-device AI APIs through the Windows Copilot Runtime.
• The introduction of Phi Silica, an NPU-optimized small language model, promises efficient text generation, content moderation, and much more.
• Developers are encouraged to experiment despite initial setup challenges—think of it as a beta phase of a groundbreaking change in how Windows applications are built.
• The shift to on-device processing could lead to more privacy-focused, responsive, and resource-efficient AI applications on Windows.

As Windows developers begin to dig into these experimental tools, the ecosystem is poised for a wave of innovative applications that blend traditional desktop functionality with next-generation AI. Whether you’re a developer eager to push the boundaries or a tech enthusiast curious about what the future holds, the Windows Copilot Runtime is definitely a space to watch.
So fire up your Copilot+ PC, update that Visual Studio install, and start exploring—there’s a whole new world of on-device AI waiting to be unlocked on your Windows machine.

---

Source: InfoWorld Diving into the Windows Copilot Runtime