Intelligence that stays where it belongs.
Most AI works by sending what you type to someone else’s computer. Ours runs on yours, wherever we can. Here is what we promise, how it works, and where the limits are.
Five principles
- 1
Your device is the boundary.
For our on-device apps, the work happens on the chip in your hand. There is no Minimo server that sees your document or the answer.
- 2
No account by default.
Our on-device apps work without an email address, a phone number or a login. We can’t lose data we never had.
- 3
Nothing to train on.
Where an app never sends us your files, we can’t train on them. That is a property of the architecture, not a line in a policy.
- 4
Offline is the normal state.
If an app needs the network, for example for exchange rates, it says so on its card and in its privacy label.
- 5
The right model, not the biggest.
We build on proven on-device models for language, speech and translation, and add our own detection rules, so each app does one job well.
What happens when you use a Minimo on-device app.
You give it something
A document, a sentence, a voice note or a photo. It is read from your phone’s storage into memory.
The app works on it
A compact model on your phone, plus our own rules, does the work on the phone’s own chip. No request is made.
You get a result
A summary, a redacted copy, a transcript or a translation. You decide whether to save or share it.
Memory is cleared
When you close the task, the working copy is released. Only what you chose to save remains.
Cloud AI and on-device AI, side by side.
| Question | Typical cloud assistant | Minimo on-device apps |
|---|---|---|
| Where your text goes | To a data centre, often abroad | Nowhere. It stays on your phone |
| Who could read it | The provider, its staff, sub-processors | You |
| Works without signal | No | Yes |
| Account required | Usually | No |
| Used to train models | Depends on settings and plan | Not for content we never receive |
Some Minimo apps do use the network or our servers. The Products page says which.
Three capabilities, each doing one job.
Language
On-device language models read your documents, summarise them and answer questions about them.
Recognition
Our detection rules find emails, phone numbers, addresses and card-like numbers so they can be removed before anything is shared.
Speech
On-device speech models turn speech into text, and text into another language.
What on-device can’t do. Yet.
Privacy has a cost. We would rather tell you than have you find out.
- On-device models know less about the world than the largest cloud models. They are best at the job they were built for.
- The language features need a recent iPhone or iPad with enough on-device AI capability. Older devices can’t run them.
- Very long documents take longer on older phones. A 200-page PDF may need a minute rather than seconds.
- On-device language coverage is still thin for Malay, Khmer, Lao and Burmese. See the Multicultural page.
See the principles at work.
A live check of this website, where your data goes, our audit status and our limits are on the Trust and proof page.