Browse and download AI models from Hugging Face — directly inside Whisper AI
Running an open-weight model on your phone used to require a developer setup. We've changed that.
Whisper AI is a private AI assistant that doesn't collect your data. Everything runs on your device — your conversations, your context, your models. Nothing leaves.
With this update, we've made it significantly easier to choose which model powers your assistant. Whisper AI now includes a built-in Hugging Face browser, giving you direct access to thousands of open-weight models without leaving the app.
What's new
From within Whisper AI, you can now:
- Search and browse models from Hugging Face directly in the app
- See a compatibility rating for each model before you download — very well, okay, poorly, or bad — based on your specific device
- Filter by all models, text-only, or vision-capable models
- Download and run models locally, with no external accounts or sign-ins required
The compatibility ratings matter. On-device AI performance varies significantly depending on your hardware. Rather than showing you a generic catalogue and letting you discover what works through trial and error, Whisper AI tells you upfront how well each model will run on your device. We'd rather be honest about that than have anyone feel misled.
Why Hugging Face
Hugging Face hosts one of the largest libraries of open-weight AI models in the world. These are models that no single company owns or controls — they're publicly available, free to download, and free to use.
That matters for a private AI assistant that doesn't collect your data. Open-weight models are inspectable. Their behaviour isn't dictated by a platform whose incentives may not align with yours. You can verify what you're running.
Until now, accessing this library on a mobile device meant either using a developer-focused tool or navigating a process that assumed a certain level of technical comfort. The Hugging Face integration in Whisper AI removes that barrier. If you can tap a button, you can browse and download an open-weight model.
What this means in practice
Running your own model used to mean knowing what a GGUF file is, where to find it, how to get it onto your device, and whether it would actually run. That's a reasonable ask for a developer. It's an unreasonable ask for most people.
Whisper AI handles all of that. You browse, you check compatibility, you download. The model runs locally. Your conversations stay on your device.
This is what we mean when we say that private AI should be accessible, not just possible.
A note on device compatibility
Newer devices — particularly those with more recent chips — will see a wider range of models rated as compatible and will generally achieve better performance. Older devices still run a fully private, on-device AI assistant. They work best with smaller, more efficient models, which the compatibility filter will surface for you automatically.
Some features within Whisper AI — including image support and Apple Intelligence integration — also require newer hardware. We're upfront about this because we think you deserve to know what your device can do before you invest time in a setup.
Part of a broader picture
The Hugging Face integration sits alongside a growing set of options inside Whisper AI: Apple Intelligence, GitHub Copilot, Open Router, and custom API support. You can switch between providers within a conversation and choose your point on the privacy spectrum.
On-device models — particularly open-weight ones from Hugging Face — represent the highest point on that spectrum. Nothing leaves your device. No company sees your prompts. No account is required.
That's the option we want to make as easy to reach as any other.