Privacy
How we handle your data
Exactly which parts of Infina run on your computer and which talk to a server.
Plenty of apps say “private” and leave it there. This page is the specific version: what runs on your own computer, what talks to a server, and how to check for yourself.
The short version: Infina hears you and types for you on your own computer. Your voice and your words stay there. That is not a setting you have to find — it is what the app does. One optional add-on changes part of it, and the rest of this page is precise about which part.
On your computer: nothing you say leaves
This is Infina as it comes — the $99/year plan on its own, and the free trial. Every step runs on your own computer:
- Waiting for the trigger word. A tiny on-device listener, listening for one thing. It cannot record, store, or send audio.
- Hearing you. Your speech is transcribed on your own computer — on a Mac by Apple's Neural Engine, on Windows by the laptop's own processor.
- Tidying the text. Punctuation and formatting are applied by rules running on your own computer.
- Voice commands. Heard on your computer and understood on your computer. Which commands it understands differs by platform — see Windows and Mac side by side.
Your audio never leaves. Your dictated text never leaves.
One exception, on your first few minutes
The on-device speech model is a few hundred megabytes, and it downloads after you install. Until it lands, your computer cannot hear you at all — so during that window only, and only if you speak before it finishes, your speech is transcribed on our servers so you are not left talking to an app that is still unpacking.
It ends by itself: the moment the on-device model answers, that path stops being used. There is no setting and nothing expires. The text is still tidied by the rules on your computer, and if you would rather wait, just don't speak until Settings → Models shows the download finished.
Why you don't have to trust a switch for this
It is enforced twice, in two places that don't depend on each other. The app resolves every voice path to on-device whenever the add-on isn't active, ignoring whatever settings happen to be saved. And the servers independently refuse cloud work for an account that isn't entitled to it. A stale or edited setting on your computer cannot talk them into it.
On-device is also a floor, never a ceiling. No individual setting can quietly move one part of the app back to the cloud.
If you add Infina Cloud: what changes
Infina Cloud is an optional add-on. It exists because polishing text well needs models bigger than your own computer can run. With it active:
- Your dictation goes encrypted to the cloud for both push-to-talk and hands-free — that is the part you paid for. It passes through our servers on the way, comes back as text, and is never stored — not by us, not by the provider. Either lane can be moved back on-device in Settings → Models.
- Voice commands are still heard on your computer. Always, subscriber or not — commands are short and latency matters more than polish. What gets sent for a subscriber is the command's text, to work out what you meant, never the audio.
One switch undoes all of it: setting Settings → Models to On-device puts every path back on your computer, even with an active subscription.
Storing is separate, and it is your choice: Privacy mode (on by default) saves nothing. Turn it off and your dictations are saved to your account as history.
Wi-Fi drops, dictation doesn't
If the network goes away, Infina falls back to the on-device pipeline on its own rather than failing — and switches back when the connection returns. Nothing to notice, nothing to do.
What talks to a server no matter what
Being straight about the rest, because “nothing leaves your computer” is a claim about your dictation, not about the whole app:
- Signing in, and checking your plan.
- App updates.
- Product analytics — counts and timings, never your dictated text or your audio.
- Anything you deliberately sync, like your snippets and dictation history.
The exception worth knowing about
When a voice command runs entirely on your computer, the command's text is sent to us to be counted.
The reason is dull: a command that never touches a server also never gets counted, and the free trial is measured in commands and words. So the text is sent, checked to see whether it counts at all (things like “send it” are free), counted, and discarded. It is never written to a database.
Dictation does not do this. When you dictate on-device, only the number of words is sent, never the words.
Checking it yourself
Don't take our word for it — turn off Wi-Fi and use it. Without the add-on, dictation and voice commands keep working exactly as before, because none of it needed a server. That is the whole claim, and it takes thirty seconds to test. Give the on-device model time to finish downloading first, or you will be testing the one thing above that does need a connection.