Offline AI Reminder for iPhone and iPad · last updated 23 September 2026
Offline AI Reminder collects nothing, sends nothing and has no accounts. Your reminders, the times you finished them and your score exist only on your device. There is no server to send them to, because the app has no networking code in it at all.
Every privacy policy says something like the above. This one can be tested in about ten seconds, and you should test it:
The language model that answers you is a file inside the app bundle. It is the reason the download is large, and it is the reason nothing has to leave.
| What | Where | Leaves the device? |
|---|---|---|
| Your reminders — titles, notes, dates, repeat rules, lists | A file in the app's own container | No |
| What you have ticked off, and when | The same file, filed by the day it was due | No |
| Your points, streak and level | Computed from the above; nothing extra is stored | No |
| Your settings — theme, week start, notification preferences | The app's own preferences | No |
| What you type to the assistant | In memory, for the length of the conversation | No |
A second copy of your reminder file is kept beside the first, so that a damaged file does not cost you your history. Both are inside the app's own container and both go when the app does.
If a file ever turns out to be unreadable, it is moved aside and kept rather than deleted, under a dated name in that same container — on the principle that data you cannot read is still yours, and throwing it away is not the app's decision to make. It goes when the app is deleted, like everything else.
Nothing is written to a log that leaves the device. The app has no crash reporter and no diagnostics upload; on the rare failure path it prints a line to the device's own console, which never leaves the phone and which nobody but you can read.
All of it is deleted when you delete the app. There is nothing left behind and nothing held anywhere else.
The assistant runs a language model on your device. What you type goes to that model and nowhere else.
It is also worth knowing how little of the work the model actually does. Dates, times, repeat rules, counts and points are worked out by the app itself — deterministically, from your own data. The model's only job is to add a sentence of its own underneath, and a sentence containing any figure is discarded before you see it. That is a correctness decision rather than a privacy one, but it has the same effect: your reminders are never handed to a language model to be re-described.
You can switch the model's contributions off entirely in Settings → Assistant → Written notes. The assistant keeps working, because the understanding was never the model's job.
The app asks for exactly one thing.
| Permission | Asked when | Used for |
|---|---|---|
| Notifications | The first time you have a reminder, not at launch | Telling you a reminder is due. Nothing else. |
If you say no, the app keeps working — your reminders are all still there and the calendar and score are unaffected. Nothing will pop up to tell you about them, which is most of the point of a reminder app, so it is worth saying yes. You can change your mind in iOS Settings at any time.
Reminders you mark high priority are delivered at the Time Sensitive level, which is what lets them through a Focus. Normal and low ones are not, deliberately: an app that marks everything urgent has marked nothing urgent.
The app is not directed at children and collects nothing from anybody. It does include an on-device assistant that answers free text, so it is not rated 4+, and that is the honest answer rather than a convenient one.
Data-protection law gives you rights of access, correction, deletion, portability and objection. They are satisfied here by construction rather than by process: the only copy of your data is on your device, in your hands. Deleting the app deletes it. There is nothing to request from anybody, because nobody else has anything.
The app is built on llama.cpp and ggml, MIT licensed, Copyright © 2023-2024 The ggml authors. Permission is granted, free of charge, to use, copy, modify, merge, publish, distribute, sublicense and sell copies of that software, subject to the notice being included; it is provided "as is", without warranty of any kind.
The bundled language model is Llama 3.2 1B Instruct, used under the Llama 3.2 Community License Agreement, Copyright © Meta Platforms, Inc. All Rights Reserved. Built with Llama.
Neither component is a service. Both are code and data inside the app, and neither one talks to anybody.
If a future version of the app changes what it does with data, this page will be updated before that version is released, and the date at the top will change. If a future version ever gained a network feature, it would be described here, in plain words, before you could install it.
Questions about this policy, or about anything the app does with what you type into it: sivavimelrajhen@gmail.com
There is no account to close, no mailing list to leave and no data to request — so an email here is a question, never a form to fill in.