Is Your Dictation App Private? On-Device vs Cloud, Explained

When you dictate, something has to turn your speech into text. The big question most people never ask is where that happens. Choosing a private dictation app comes down to understanding one fork in the road: is your audio processed on your own device, or sent to a server in the cloud?

Neither approach is sinister. Both are used by reputable products. But they make different trade-offs, and which one suits you depends on what you dictate and how sensitive it is. This guide explains the difference in plain terms, then shows you exactly what to check before you trust any app with your voice.

Is my dictation private? It depends on where the audio goes

There are two ways an app can transcribe what you say.

On-device (local) transcription runs the speech-to-text model directly on your Mac. Your audio never leaves the machine. There is no upload, so there is nothing on a server to store, leak, or train on. The catch is that it needs a capable chip to run a good model quickly, which is why on-device options tend to shine on Apple Silicon.

Cloud transcription streams your audio to a remote server, transcribes it there, and sends the text back. This is often very fast and very accurate, because servers can run large models that would be heavy on a laptop. The trade-off is simple: your audio leaves your machine to be processed elsewhere.

That is the whole tension. Cloud buys you power and accuracy at the cost of your audio travelling off-device. On-device keeps everything local but leans on your hardware. One is not "good" and the other "evil" — they are different tools for different needs.

What to check in any dictation app

Before you commit, you only need clear answers to three questions. A trustworthy app will answer all of them plainly in its documentation or privacy policy.

1. Is transcription on-device or cloud? This is the foundational question. If it is cloud-based, your audio is being uploaded — which is fine, as long as you know it and the rest of the handling is sound.
2. Is your audio stored or discarded? Processing audio in the cloud and keeping audio recordings are two very different things. Look for explicit wording that audio is discarded after transcription rather than retained.
3. Can you opt out of training? Some services may use data to improve their models. Check whether that happens, and whether you can turn it off — or whether it is off by default.

If an app cannot give you straight answers to these three, that is itself a useful signal.

How the popular apps compare

A quick, fair read of where some well-known options sit. Always confirm against each app's own current documentation, as policies change.

Wispr Flow ($15/mo) is, by its own documentation, a cloud-based dictation tool — your audio is processed in the cloud. That powers its strong accuracy and speed. It also offers a Privacy Mode you can enable for stricter handling. None of this is underhanded; it is a normal cloud design, and the right pick for plenty of people. The key thing to know is simply that it is cloud-first by design. If you want a closer look, see our Dictately vs Wispr Flow comparison.

Superwhisper focuses on on-device, offline transcription, which appeals to people who want their audio to stay local by default.

If you have wondered "is Wispr Flow safe", the honest answer is that it is a legitimate, widely used product — the real question is not safety in some dramatic sense, but whether cloud processing of your audio fits your needs.

Where Dictately sits

Dictately is built around the same trade-off, and we try to be transparent about it.

For English on Apple Silicon, dictation runs fully on-device. Your audio never leaves your Mac — there is no upload and nothing to store on our side. If privacy is your top priority and you mostly dictate in English, this is the mode you want.

For other languages, Dictately uses the cloud, because cloud models currently deliver the accuracy those languages need. In that case your audio is processed in the cloud and then discarded — never stored. You get the quality of a large model without your recordings being kept.

The point is to let you match the tool to the task rather than forcing one compromise on everyone.

How to decide

Think about what you actually dictate.

• Sensitive or confidential work — legal notes, medical details, anything under NDA? Favour on-device transcription so your audio stays on your machine. Dictating in English on Apple Silicon keeps it local.
• Everyday writing, emails, and notes where speed and accuracy matter most? Cloud transcription is a reasonable, convenient choice — just confirm audio is discarded, not stored.
• Multilingual dictation? You will likely rely on the cloud for non-English languages. The thing to verify is retention: processed and discarded is very different from processed and kept.

Privacy is not about fear. It is about knowing where your voice goes and choosing on purpose. Ask the three questions, read the answers, and pick the app that matches the kind of work you do.