Redigeeri

AI-powered features

This section provides guidance for using the Windows App SDK to create apps that leverage AI-powered features in Windows.

AI-powered Windows features

Topic Description
Click to Do Learn how to use the Click to Do feature to take action based on content seen in apps.
Copilot hardware key Learn how to use the Copilot hardware key feature to quickly access Copilot.
Recall Learn how to use the Recall feature to help find content.
Studio Effects Learn how to use Studio Effects to enhance video calls with AI-powered background effects.

Other AI features

Topic Description
App Content Search Learn how to use App Content Search to enable AI-powered search capabilities.
Phi Silica Learn how to use Phi Silica to enhance app performance with AI acceleration.
Imaging SDK Learn how to use the Imaging SDK to add AI-powered image processing capabilities to your apps.
Image generation API Learn how to use the Image Generation API to create images from text prompts.
Text Recognition API Learn how to use the Text Recognition API to extract text from images.
Video super resolution Learn how to use AI to enhance video playback quality.

On-device model deployment

For running ML models directly on a user's device (without cloud calls), Windows supports several approaches depending on your scenario:

Approach Best for Key link
ONNX Runtime General-purpose inference (vision, NLP, classification) with hardware acceleration via DirectML ONNX Runtime
ONNX Runtime GenAI Running generative AI models (SLMs like Phi) locally with optimized text generation ONNX Runtime GenAI
Windows ML WinRT-based inference using the built-in Windows ML runtime (DirectML backend) Windows ML
DirectML Low-level GPU-accelerated ML operators for custom inference engines DirectML

Note

ONNX Runtime with the DirectML execution provider is the recommended approach for most desktop apps. It supports the widest range of models from the ONNX Model Zoo and Hugging Face, and runs on any DirectX 12 GPU. For generative AI scenarios (chatbots, text completion, summarization), use ONNX Runtime GenAI, which adds optimized decoding loops on top of ONNX Runtime.

  • Last updated on