On-device AI notes on iPhone
Short answer: On-device AI means the work happens on your phone instead of a company's server. In Mmory, transcription, search and answers all run on your iPhone. After setup, it works without a connection for everyday use. The trade-offs are download size, answers that vary by iPhone, and no sync.
What runs where in Mmory
| Step | Where it runs |
|---|---|
| Recording | iPhone |
| Transcription | iPhone, with Apple's on-device speech recognition. A Whisper model on the iPhone may check the transcript |
| Search | iPhone. The index is built and stored there |
| Answers | iPhone |
| Reminders | iPhone |
| Sign-in and purchases | Online, with Clerk, RevenueCat and Apple. No memory content is sent |
| Model downloads | Online, once, after you install. Your content isn't sent |
If on-device speech recognition isn't available, Mmory doesn't fall back to a server. It keeps the recording so you can try again.
Does it work offline?
After setup, yes for everyday use. Recording, transcription, search, answers and reminders run on your iPhone without a connection. Mmory needs the internet after you install it to download its speech and AI models, and to sign in or to buy, check or restore Mmory Pro. It also contacts its sign-in service when it opens, if it can.
The downloads start automatically, including over mobile data, so it's worth opening Mmory on Wi-Fi the first time. They pause in Low Data Mode or when storage is short, and Mmory still works in the meantime.
Which AI model answers?
- iOS 26 or later with Apple Intelligence available: Apple's on-device model.
- iPhones with enough memory: Mmory's private AI model, about 660 MB, once it has downloaded.
- Other iPhones: Mmory quotes the relevant memories instead of writing an answer.
Some questions don't need a model at all. If a memory holds the exact detail you asked for, Mmory looks it up and gives it to you directly.
While the model is downloading, the app says Private AI is still preparing, and answers may be plainer.
The trade-offs
- Size. Models take up storage: about 60 MB for speech, and about 660 MB for the private AI model.
- Capability. A model that fits on a phone is smaller than the largest cloud models. Mmory answers from your saved memories, and it isn't a general chatbot.
- Consistency. Answers depend on your iPhone, as shown above.
- One device. Your memories live on one iPhone. There's no sync, and audio isn't backed up.
Cloud services are the better choice if you need sync across devices, long recordings, or many languages. Choose on-device if you'd rather keep your notes on your own phone and can accept those limits.
Does on-device mean nothing ever leaves the phone?
No, and it's worth being precise. Your memories aren't uploaded to a Mmory server, and audio leaves only if you share it. But transcript text can be included in your own iCloud or computer backup, and the app does make the connections listed above. Our privacy policy lists every provider. You can also read what stays on your iPhone.
Other on-device apps
Aiko and Whisper Notes both describe on-device transcription on their App Store pages, and suit transcribing longer audio better than Mmory does. Mmory's focus is asking questions of short notes.