Skip to content

THE FIELD JOURNAL

Foundry Local on an AI PC: What It Is and First Steps

By Local AI Scout · Beginner guide

Paper-cut illustration of a slim laptop with abstract lines on its screen, three small chip cards beside it with one glowing softly, and a shelf of paper model boxes

By Local AI Scout · Local · Computers · Guide · Docs-only · Checked November 18, 2026

Paper-cut illustration of a slim laptop with abstract lines on its screen, three small chip cards beside it with one glowing softly, and a shelf of paper model boxes

Short answer: Foundry Local is Microsoft's toolkit for running AI models on the user's own device. Microsoft describes it as "an end-to-end local AI solution for shipping applications," built around an SDK that developers put inside their apps, with a curated model catalog and automatic use of the GPU or NPU when available. There is also a command-line tool, which Microsoft labels a preview, that lets you list models and chat with one in a terminal. It runs on Windows, macOS on Apple silicon, and Linux, and needs no Azure subscription. It is not a chat app like LM Studio. If you own a Copilot+ PC and want to see its NPU do something, the preview CLI is a short experiment. If you just want a private chatbot, start with Ollama or LM Studio. We haven't installed it.

This guide follows one reader: someone who bought a laptop sold as an "AI PC," read our AI PC explainer, and wants to know whether Foundry Local is the app that finally uses the NPU.

What Foundry Local is, in Microsoft's words

From Microsoft Learn's "What is Foundry Local?" page (last updated October 22, 2026):

  • An SDK first. It provides an SDK for C#, C++, JavaScript, Rust and Python, a curated catalog of optimized models, and automatic hardware acceleration. Microsoft says the runtime adds "approximately 20 MB" to an app.
  • Curated, not everything. The catalog covers chat models (Microsoft's examples include GPT OSS, Qwen, DeepSeek, Mistral and Phi) and speech-to-text (Whisper). Microsoft says it is "designed for shipping production applications, not for general-purpose model experimentation."
  • Hardware picked for you. It "detects the hardware available on your device and selects the best execution provider," using GPUs and NPUs when available and falling back to the CPU.
  • OpenAI-compatible. Apps that already use the OpenAI SDK can point at a Foundry Local endpoint "with minimal code changes." An optional local server is available.
  • Single user. Microsoft says it is optimized for "a single user" on hardware-constrained devices, and points to dedicated server frameworks for serving many people.

Does it send my prompts to Microsoft?

Microsoft's FAQ says prompts and outputs "are processed locally." It lists two network uses: downloading model files and hardware plug-ins the first time a model runs, and "optional diagnostics" if a user chooses to share logs. What Microsoft may collect is governed by the product terms. That matches the pattern in our does local AI phone home? guide: the chat stays on the device, the downloads don't.

Try it: the preview CLI in five commands

Microsoft's CLI page, last updated August 14, 2026, says the CLI "is available in preview" and calls it "a companion tool for exploration and management." These are the documented commands. We haven't run them, and preview commands can change.

  1. Install. On Windows: winget install Microsoft.FoundryLocal. On a Mac with Apple silicon: brew tap microsoft/foundrylocal, then brew install foundrylocal.
  2. Check it installed: foundry --version.
  3. See the catalog: foundry model list. The first run downloads the execution providers for your hardware. To see versions for a particular chip, add --device (for example --device gpu); our AI PC explainer uses --device npu to look for NPU builds.
  4. Chat: foundry chat phi-4-mini. Microsoft says using an alias like phi-4-mini lets Foundry Local "automatically select the best variant for your hardware." Type /exit to leave.
  5. Tidy up: foundry cache list shows downloaded models; foundry cache remove phi-4-mini deletes one.

If you see "Request to local service failed," Microsoft's tip is foundry server restart.

Older tutorials may show foundry model run ‡. The current CLI page uses foundry chat.

Foundry Local, Ollama or LM Studio?

You want…Start withWhy
A private chat window, no terminalLM StudioA desktop app built for chatting
Any open model, and other apps that talk to itOllamaThe biggest catalog and the most app integrations
To see your Copilot+ PC's NPU run a modelFoundry Local (preview CLI)Microsoft says it uses NPUs when available
To ship an app that runs AI on customers' devicesFoundry Local SDKThat's what Microsoft built it for

Our AI PC explainer explains why Ollama and LM Studio mostly don't use the NPU yet, and our Windows guide covers the usual setup.

Choose Foundry Local if…

  • You have a Copilot+ PC and want to try the NPU without building anything.
  • You're a developer planning an app that runs AI on the user's device, and you're happy with a curated catalog.

Choose something else if…

  • You want a chat app you can hand to a non-technical colleague.
  • You need a specific model that isn't in Microsoft's catalog.
  • You need to serve one model to several people at once; Microsoft points to server frameworks for that. See Ollama vs llama.cpp vs vLLM.

Frequently asked questions

What is Foundry Local?

It's Microsoft's toolkit for running AI models on the user's own device. Microsoft describes it as an end-to-end local AI solution for shipping applications, with an SDK, a curated model catalog and automatic hardware acceleration. A command-line tool, labeled as a preview, lets you browse models and chat with one.

Does Foundry Local use the NPU?

Microsoft says it detects the hardware on your device and accelerates inference on GPUs and NPUs when available, falling back to the CPU. Use the model list to see which versions exist for your hardware.

Is Foundry Local free, and do I need Azure?

Microsoft's FAQ says no Azure subscription is required and that there are no per-token costs because models run on your device. Use is governed by the product terms and the licenses of the models you run.

Does Foundry Local work on a Mac?

Yes. Microsoft lists Windows, macOS on Apple silicon and Linux as supported, and documents a Homebrew install for the preview CLI on Apple silicon Macs.

Sources

  • Microsoft Learn, "What is Foundry Local?" (last updated October 22, 2026), opened November 18, 2026: https://learn.microsoft.com/en-us/azure/foundry-local/what-is-foundry-local
  • Microsoft Learn, "Use the Foundry Local CLI (preview)" (last updated August 14, 2026), opened November 18, 2026: https://learn.microsoft.com/en-us/azure/foundry-local/how-to/how-to-use-foundry-local-cli
  • Microsoft Learn, "Foundry Local CLI reference," opened November 18, 2026: https://learn.microsoft.com/en-us/azure/foundry-local/reference/reference-cli

How we checked: a documentation-based guide, not a hands-on test. We haven't installed Foundry Local for this page. Sources opened November 18, 2026 unless stated. ‡ marks something we could not confirm on an official page, or our own reading. See our editorial policy.

Keep exploring: Apps · Models · All guides

THE SCOUT LETTER

Good finds.
Less noise.

Useful local AI apps, practical guides and the updates worth your time. A thoughtful note in your inbox.

Free to join. Unsubscribe with one click.