Start with solution-centric Windows AI APIs - now expanding beyond Copilot+ devices. Use Foundry Local to run open-source models locally. With new tooling in Foundry Toolkit, optimize and prep your models for local AI deployments. Run custom AI workloads locally across GPU, NPU, or CPU with Windows ML, now with support for web apps through WebNN. Learn how the platform enables on-device inference across all Windows PCs to help you ship performant, scalable, and secure AI-powered apps on Windows.

Seating for this session is first-come, first-served. Add it to your schedule to plan your day and arrive early to secure a spot.

To learn more, please check out these resources:
 * https://aka.ms/build26/BRK260


𝗦𝗽𝗲𝗮𝗸𝗲𝗿𝘀:
 * Anastasiya Tarnouskaya
 * Aditi Narvekar
 * Jordi Janer


𝗦𝗲𝘀𝘀𝗶𝗼𝗻 𝗜𝗻𝗳𝗼𝗿𝗺𝗮𝘁𝗶𝗼𝗻:
This is one of many sessions from the Microsoft Build 2026 event. View even more sessions on-demand and learn about Microsoft Build at https://build.microsoft.com

BRK260 | English (US) | Windows


Breakout | (300) Advanced


#MSBuild

Chapters:
0:00 - Session Introduction and Overview of Local AI for Windows
00:00:40 - Speakers Introduction: Anastasia and Aditi from Microsoft
00:03:11 - Benefits of Running AI Models Locally: Privacy, Latency, Network Independence, and Cost Savings
00:21:28 - Announcement of Windows MLCLI Preview
00:25:04 - Analyzing and Optimizing the Model for Hardware Support
00:27:24 - Running Optimized Models with Windows ML
00:33:19 - Switching to NPU for Faster Inference
00:35:25 - Overview of Voicemod and its Real-Time AI Voice Transformation
00:36:06 - Voicemod’s Use of Windows ML for Low Latency AI Processing

Microsoft Developer

Microsoft Build session covering how to build local AI-powered Windows apps using the Foundry on Windows stack, which includes Windows AI APIs, Foundry Local, and Windows ML. Key announcements include Windows AI APIs expanding to CPU and GPU support (previously NPU-only), Foundry Local reaching general availability with new Qwen vision-language and speech models, the new Windows ML CLI (in preview) for model conversion/optimization/benchmarking, Windows ML 2.0 with ONNX runtime improvements, and WebNN support for bringing hardware-accelerated AI to web apps in Chromium browsers. Live demos cover speech recognition, structured JSON extraction with Phi Silica on GPU, video super resolution in ClipChamp, inventory management with Qwen 3.5 VLM, and real-time voice transformation with VoiceMod — all running locally across NPU, GPU, and CPU without cloud dependencies.

Build Apps w/ Local AI for Unmetered Intelligence on every Windows PC | BRK260