Runs real models
MLX on Apple silicon, plus Apple's own on-device model. Import from Hugging Face or convert what you already have.
Locaz is a local-first AI app for iPhone, iPad and Mac. Your conversations stay on the hardware you own, and nothing is sent anywhere unless you ask for it.
Download a model, run it on your own silicon, and keep every conversation on the device. Locaz handles the parts that usually go wrong: whether a model fits in memory, which runtime can load it, and what happens when it can't.
MLX on Apple silicon, plus Apple's own on-device model. Import from Hugging Face or convert what you already have.
If a model is too large for the memory you have, Locaz says so before it tries, and offers something that fits.
Memory is off until you turn it on, kept in plain words you can read, edit and delete.
Focused on iPhone, roomy on iPad, a proper document app on Mac. Not one layout stretched three ways.
Every path that could leave the device goes through one gate, and that gate reports what it let through. Turn on Offline Lock and requests are refused rather than queued.
No analytics, no crash uploads, no tracking SDKs, no ads. The app works with the network off.
If a request would leave, Locaz shows exactly what would be sent and asks first, every new kind of content.
Chats, memory and provider settings are encrypted with a key held in your device's keychain.
There is no account, no profile and no server holding your history, because there is nothing collecting it.
The App Store is the easiest way to install and stay updated. Direct downloads are here if you would rather not use it.
Models are not bundled with the app. Locaz downloads the one you choose, verifies it against its published digest, and tells you beforehand how much memory and disk it will take.