Two kinds of speech to text on a Mac
Cloud transcription sends your audio to a server, a big model turns it into text, and the text comes back. It is usually the most accurate option, especially with code words, accents and mixed languages.
On-device transcription runs on your Mac. macOS ships a built-in speech recogniser, and some apps bundle a local model. Nothing leaves the machine. It is a little less sharp with jargon, and it works on a plane.
Neither is wrong. What matters is that you know which one is running, and that it never takes more than it needs.
The part people forget: your screen
Dictation apps only need your voice. The moment a tool also looks at your screen, the stakes change: screenshots can contain API keys, customer data, private messages and whatever was in the next tab.
A few AI desktop features send what is on screen to the vendor's cloud, because that is where their model lives. That is a reasonable trade for some people. For a developer staring at a production dashboard, it is worth a second thought.
Three questions to ask any tool that listens
- What leaves, exactly? Audio only? Screenshots? The text of the window? A good tool can answer in one sentence.
- When does it listen? Always-on is a very different product from push-to-start. You want to see, clearly, when capture is running.
- What happens offline? If it still works with Wi-Fi off, you know which parts run locally.
How Deiko answers them
Deiko is built so the answers are short:
- What leaves: your spoken audio, for transcription, and it is deleted once the brief exists. Screenshots, on-screen text and accessibility text never reach Deiko's servers. To file each brief in the right task, Deiko sends what you said, a one-line summary of it, your window titles (with anything that looks like a key stripped) and a few lines about your earlier tasks through its relay to a sorting model. Deiko keeps none of it.
- When it listens: only after you double-tap Right Option, and a red capsule stays on screen the whole time. Red means capturing, and nothing else in Deiko is red.
- Offline: it keeps working. Apple's on-device recogniser takes over, and with no network there is nowhere for anything to go.
- Your own key: bring your own transcription key and your audio never touches our servers.
Questions people ask
Does Mac dictation send audio to Apple?
It depends on the feature and settings. macOS can transcribe on device; some modes use Apple's servers. Check System Settings under Keyboard, Dictation.
Is there offline speech to text for Mac?
Yes. macOS includes an on-device speech recogniser, and several apps run local models. Deiko falls back to Apple's on-device recogniser when there is no network.
Do AI voice coding tools see my screen?
Some do and send it to their cloud. Dictation-only tools do not look at your screen at all. Deiko captures only what you point at, and keeps those screenshots on your Mac.
How do I stop a dictation app from listening all the time?
Prefer push-to-start tools with a visible recording indicator. Deiko only captures between a double-tap of Right Option and a single tap to stop.