A closed MacBook on white linen with a small brass padlock beside it

Offline dictation

Dictate on a Mac with no connection, and no audio leaving it.

Vaitly Voice (VV) recognises speech on the Mac itself, so dictation works on a train, on a plane and behind a firewall. Your audio is not uploaded for standard dictation. The one caveat is the first launch: the speech model downloads once, and that needs internet. After that, hold, speak, paste, offline.

Last updated

What works with the network off

Everything that counts as dictating. Hold your shortcut, speak, let go, and the text is pasted where the cursor was, in whatever app is in front: Mail, Slack, Cursor, Notes, a browser text field. Double-tap the shortcut to latch it if you want both hands free. Recognition runs on the Mac’s own processor, using NVIDIA Parakeet for English by default, so there is no round trip to wait for and nothing to time out.

The features around dictation are local too. Vocabulary teaches VV the spoken form of names and jargon it would otherwise guess at. Snippets turn a spoken trigger into a block of text. History, with the audio, is stored under the app’s data directory on the Mac, and you choose how long it is kept. Transforms, the optional rewrite step, default to Apple Intelligence or a local Ollama model, and both of those run without a connection.

The first run needs internet, once

VV does not ship with the speech model inside the download. On first launch it fetches the default English model, about a gigabyte, and stores it on the Mac. That is the only time standard dictation depends on a connection. If you are setting up a laptop for a trip, launch VV once at home, let the download finish, and dictate a sentence to confirm it pastes. From then on the model is read from disk.

What still needs the network

Three things, none of them dictation. Buying a plan, because checkout and the account that holds the licence live on the web. Cloud transforms, only if you have deliberately chosen a cloud provider for a transform instead of the local default; those send the transcribed text, never the audio, and are labelled as cloud in the app. And moving a licence to another Mac, since a Pro licence covers 2 devices and the swap is recorded against your account. A licence already on the Mac is cached and re-checked periodically when a connection is available, so day-to-day use does not wait on it.

Limits and the privacy boundary

Offline does not mean magic. English is the default model; other languages are optional downloads from inside the app, each needing a connection once, and accuracy in them varies. Accuracy in any language depends on the microphone and on your pace: a laptop mic in a quiet room is fine, a café with the fan on is not, and a steady conversational speed beats rushing. VV transcribes what you say; it does not interpret spoken commands such as “delete the last sentence” or “bold that”, online or off. That is voice control, a different job.

The privacy boundary is the same offline and on. Standard dictation audio is never uploaded. The only data that can leave the Mac is text you send to a cloud transform you configured yourself, plus the licence check, which carries the licence and device identity and nothing you dictated. The privacy page lists exactly what is sent and when.

Compared with cloud dictation

Cloud dictation apps stream your audio to a server, which is why they stop working the moment the connection drops and why they need an account before you can speak. VV vs Wispr Flow walks through the trade-off in detail. If you want the technical version of how recognition runs on the Mac, read on-device recognition. For a broader look at keeping dictation private, see the private dictation guide.

Questions

Does VV work in flight mode?

Yes, once the speech model is on the Mac. Hold the shortcut, speak, and the text is pasted exactly as it would be at your desk. The one thing to do before you board is launch VV once on a connection so the first-run download has finished.

Does the free plan work offline?

Yes. Free dictation needs no sign-in, and the allowance of 1,000 words a week is counted on your Mac and reconciled the next time you are online. The count is kept against an anonymous device identity on the Mac, not an account, and resets every Monday.

Can VV check my licence without a connection?

Yes. A Pro or lifetime licence is cached on the Mac and re-checked periodically when a connection is available. The check confirms the licence and the device; it does not send anything you dictated. You need to be online to buy a plan or move a licence to another Mac.

Do transforms work offline?

The local ones do. Transforms default to Apple Intelligence or a local Ollama model, and both run without a connection. If you have chosen a cloud provider for a transform, that transform is unavailable offline and standard dictation carries on as normal.

What if I lose the connection halfway through the model download?

Reconnect and the download resumes from where it stopped. Until it completes, VV cannot recognise speech, so it will not paste anything. The download is about a gigabyte and happens once per Mac.