Skip to content

How-ToGPU

How to Use Microsoft's AI Dev Gallery to Try Local AI Models on Your PC

AI Dev Gallery is Microsoft's free, open-source Windows app with more than 25 local AI samples that can run on your CPU, GPU or NPU. Install it from the Microsoft Store, pick a sample, read the model card and download the model. After that it works offline. Microsoft recommends 16GB RAM, 20GB free space and 8GB VRAM for GPU samples.

By , Editorial team

4 min read

How to Use Microsoft's AI Dev Gallery to Try Local AI Models on Your PC

Want to see what your PC can do with local AI, without learning Python or setting up a command-line tool? Microsoft's AI Dev Gallery is a free Windows app that bundles more than 25 interactive samples powered by AI models that run on your own hardware. It's aimed at developers, but it's also one of the easiest ways to check whether your graphics card or Copilot+ PC NPU can handle a given task. This guide is based on Microsoft's documentation and the project's GitHub page. Forge PC hasn't benchmarked the samples, and performance depends on your hardware and the model you pick.

According to Microsoft Learn, AI Dev Gallery is an open-source app for Windows developers who want to add AI features to their apps. It includes:

  • More than 25 interactive samples powered by local AI models, including samples for all of the Windows AI APIs.

  • A model browser for finding, downloading and running models from Hugging Face and GitHub. Microsoft says these can use your PC's NPU, CPU or GPU, depending on what your device supports.

  • Source code for every sample, written in C#, which you can export as a standalone Visual Studio project.

The GitHub repository lists topics including Phi-3, Mistral, Whisper and Stable Diffusion, and its listed topics also include ONNX Runtime and DirectML. The app is MIT-licensed and, according to the repository, still in public preview.

What you need

Microsoft's GitHub page lists these device requirements:

Requirement

Microsoft's guidance

Windows version

Windows 10 version 1809 (build 17763) or later

Architecture

x64 or ARM64

Memory

At least 16GB recommended

Disk space

At least 20GB free recommended

GPU

8GB of VRAM recommended for running samples on the GPU

Microsoft Learn adds that some models may need a minimum amount of dedicated GPU memory to work properly. If you're not sure how much your card has, see how to check VRAM usage in Windows. For more on what NPUs do, read what is an NPU and a Copilot+ PC.

Step 1: Install the app

The simplest route is the Microsoft Store: search for AI Dev Gallery and install it. You don't need a Microsoft account to use it, according to Microsoft's FAQ.

Developers can also build it from source. Clone the repository, open AIDevGallery.sln in Visual Studio 2022 or later (with the Windows application development workload), set AIDevGallery as the startup project and run it. Microsoft notes that on ARM64 Copilot+ PCs you should build and run the solution as ARM64 rather than x64.

Step 2: Pick a sample

Open the app and browse the samples. When you run a sample, you choose which model to use. Microsoft says the app includes popular open-source models and also models and APIs from Microsoft Foundry on Windows.

Before you download a model, read its model card. Microsoft "strongly" recommends this, because model cards explain what a model is for, its limitations and its licence.

Step 3: Download models and go offline

Models download to your PC the first time you use them. After that, Microsoft says the app works offline. You only need an internet connection to download more models from Hugging Face or GitHub. AI models can be large, so keep an eye on your drive space. Our guide to freeing up disk space in Windows 11 can help.

Step 4: Check which chip is doing the work

To see whether a sample is running on your GPU or NPU, open Task Manager's Performance tab while it runs. Our guide to checking NPU usage in Windows 11 shows where to look. If a sample is slow on the CPU, try a smaller model or one that supports your GPU.

Step 5 (for developers): Export the code

Every sample has a code view. Microsoft says you can export a sample into a new Visual Studio solution that contains only the code and model files it needs. This is the app's main purpose: a working starting point for adding local AI to a Windows app.

Privacy and telemetry

Because the models run locally, your prompts and files are processed on your PC. Microsoft's README notes that the app itself logs basic telemetry, covered by the Microsoft privacy statement. Microsoft also says it can't guarantee that third-party models from sources such as Hugging Face meet its Responsible AI standards, so check each model's licence before using it in your own projects.

How it compares with other local AI apps

AI Dev Gallery is a showcase and developer tool, not a daily chatbot. For everyday local chat, an app such as LM Studio or Ollama is a better fit; see our best free local AI apps. If you want Microsoft's command-line way to run models locally, see how to run AI models with Foundry Local.

Our verdict

AI Dev Gallery is a free, open-source Microsoft app with more than 25 local AI samples that can run on your CPU, GPU or NPU. Install it from the Microsoft Store, pick a sample, read the model card, then download the model. After that, it works offline. Microsoft recommends 16GB of RAM, 20GB of free space and 8GB of VRAM for GPU samples. It's a good way to test what your hardware can do, and developers can export any sample as a Visual Studio project.

The bottom line

AI Dev Gallery is a free, open-source Microsoft app with more than 25 local AI samples that can run on your CPU, GPU or NPU. Install it from the Microsoft Store, pick a sample, read the model card, then download the model. After that, it works offline. Microsoft recommends 16GB of RAM, 20GB of free space and 8GB of VRAM for GPU samples. It's a good way to test what your hardware can do, and developers can export any sample as a Visual Studio project.

Sources

  1. 01Microsoft Learn, What is the AI Dev Gallery?
  2. 02Microsoft on GitHub, ai-dev-gallery repository README

About the author

Forge PC Staff

Editorial team

The shared byline for articles researched, written and edited by the Forge PC editorial team. How we source, check and correct our work is set out in our editorial policy.

More from Forge PC Staff

Read next

Vertical vs Horizontal GPU Mount: Is Showing Off Your Graphics Card Worth It?

Versus

Vertical vs Horizontal GPU Mount: Is Showing Off Your Graphics Card Worth It?

Horizontal mounting gives a direct PCIe connection with no extra parts. Vertical mounting shows off the card and stops sag, but needs a bracket and riser. Lian Li's and Cooler Master's kits include PCIe 4.0 risers, which limit a PCIe 5.0 card to PCIe 4.0 link speeds, and the fans end up facing the side panel.

Forge PC Staff4 min

How to Turn Off Mouse Acceleration in Windows 11 (Enhance Pointer Precision)

How-To

How to Turn Off Mouse Acceleration in Windows 11 (Enhance Pointer Precision)

To turn off mouse acceleration in Windows 11, go to Settings > Bluetooth & devices > Mouse and switch off Enhance pointer precision (or untick it under Additional mouse settings > Pointer Options). Then set your DPI or pointer speed, and check your mouse software for any extra acceleration setting.

Forge PC Staff3 min

How to Dual Boot Windows 11 and Linux (Ubuntu) Without Losing Your Files

How-To

How to Dual Boot Windows 11 and Linux (Ubuntu) Without Losing Your Files

To dual boot Windows 11 and Ubuntu, back up first. Installing Ubuntu on its own SSD is the safest option and avoids turning off BitLocker. On a single drive, turn BitLocker off and shrink C: in Disk Management. Then create a USB installer with balenaEtcher and choose 'Install Ubuntu alongside Windows'. Ubuntu needs at least 25GB.

Forge PC Staff4 min

Windows Shows Less Usable RAM Than Installed? How to Find the Cause and Fix It

How-To

Windows Shows Less Usable RAM Than Installed? How to Find the Cause and Fix It

A small gap between installed and usable RAM is normal hardware reservation. A big gap usually comes from integrated graphics reserving memory, a Maximum memory limit set in msconfig, or the removememory and truncatememory boot options. If installed RAM itself is too low, reseat the sticks, check the slots and BIOS, then test each stick.

Forge PC Staff4 min