Truly on-device
Models run locally on Apple silicon. Your prompts and replies never touch a network — because there is no backend to send them to.
Mura runs powerful open language models entirely on your phone. No sign-up, no subscription, no servers — your conversations are yours alone, and they work even in airplane mode.
Everything happens on your device. Nothing is uploaded, logged, or trained on.
Models run locally on Apple silicon. Your prompts and replies never touch a network — because there is no backend to send them to.
Download a model once, then chat on a plane, a train, or anywhere with no signal.
No sign-up, no email, no subscription. Open the app and start typing.
On supported iPhones, Mura taps Apple's on-device Foundation model — instant, with nothing to download.
Accelerated by Apple's MLX framework, with answers that stream in token-by-token.
Multi-turn memory and saved chats — all stored privately on-device, ready to pick back up.
Pick from a curated, memory-aware catalog of leading open models — Mura downloads them straight from Hugging Face and tells you which ones fit your iPhone.
New models added regularly · all trademarks belong to their respective owners.
Your conversations never touch a server.
There is no server.
Mura has no analytics on your chats, no cloud sync, and no telemetry on what you say. Privacy isn't a setting you toggle — it's the way the app is built.
Everything you need to know about running AI privately on your iPhone.
Mura supports leading open models including Meta Llama 3.2, Google Gemma 3, Qwen3, Microsoft Phi-4 mini, Mistral 7B and SmolLM3 — and Apple Intelligence on supported iPhones. Every model you download runs completely offline on your device, and the catalog keeps growing.
Yes. Once you've downloaded a model, Mura works completely offline — on a plane, underground, or anywhere with no signal. The only time a connection is needed is the one-time model download. All chatting happens locally on your device.
Your conversations are processed entirely on your device — there's no account, no cloud sync, and nothing is ever used to train a model. The only data we collect is anonymous, aggregated usage and crash diagnostics to improve the app, never the content of your chats. See our Privacy Policy for the full detail.
Mura is built for iPhone and optimized for Apple silicon. It runs on iOS 26 and later; larger models need a more recent iPhone with enough memory, so Mura's catalog tells you exactly which models fit your device. The built-in Apple Intelligence option requires an Apple Intelligence-capable iPhone.
Download Mura from the App Store, pick a model to download (or start instantly with Apple Intelligence), and begin chatting. There's no account creation and no login — just download and use.
Yes. You can adjust generation settings such as temperature and maximum response length to shape how the model replies, and switch between models at any time to change the assistant's style, speed and capabilities.