Windows AI

A new era of AI has arrived at Microsoft and Windows operating system is at the forefront of supporting AI developers.

Overview

Microsoft AI Learning Hub

What's new

Microsoft Copilot documentation

Concept

Microsoft Responsible AI

video

Microsoft AI Show Live

Enhance your Windows apps with AI through local APIs and Machine Learning models

Find guidance on how to responsibly integrate AI into your Windows apps using APIs and ML models.

Get started using AI in Windows

Learn more about what you can do with AI on Windows.

• Overview of Windows Copilot Runtime
• Responsible Generative AI Development on Windows
• FAQs about using AI on Windows
• AI on Windows Sample Gallery

Use AI-backed APIs

Add AI features to your Windows app using APIs

• Get started using AI-backed APIs locally or in the cloud
• Phi Silica API - Small Language Model with reasoning and language understanding
• Text Recognition API (Optical Character Recognition / OCR)
• Recall API - Retrieval of past activities for user to continue where they left off
• Studio Effects - NPU-accelerated camera and microphone AI effects

Use Machine Learning models

Find guidance on how to find, optimize, fine-tune, and run ML models locally for use in your Windows app.

• Get started using ML models in your app
• Find pre-built ML models
• AI Toolkit for Visual Studio Code overview

DirectML for hardware-accelerated Machine Learning

Integrate Machine Learning (ML) inferencing workloads into your game, engine, middleware, backend, or other applications. DirectML is a low-level API supported by all DirectX 12-compatible hardware.

Overview

Guidance to help you discern whether DirectML is appropriate for your project, the workflow, and how to get started.

• DirectML Overview
• Get started with DirectML
• DirectML Version History
• DirectML developer tools
• Sample applications

DirectML Programming Guide

DirectML executes layers of your inference model as operators. Learn more about programming DirectML.

• Binding in DirectML
• UAV barriers and resource state barriers
• Resource lifetime and synchronization
• Using strides to express padding, memory layout
• Using fused operators to improve performance
• Troubleshooting - using the debug layer
• Troubleshooting - Handling errors and device removal

DirectML support for PyTorch or WebNN

Guidance on how to use DirectML with PyTorch or WebNN.

• Enable PyTorch with DirectML on Windows
• Enable PyTorch with DirectML on WSL 2
• Web Neural Network API (WebNN) Overview
• Get started with WebNN with Windows

API Reference

API Reference for DirectML.

• DirectML interfaces
• DirectML functions
• DirectML structures
• DirectML enumerations
• DirectML constants