What Voxor sends to the cloud, and what Local Mode does not

Before dictating something confidential, ask one question: does my voice leave this computer?

Voxor gives you two answers because it has two ways to process a recording. In cloud mode, audio is sent away for transcription. In Local Mode, recognition and cleanup happen on your machine. The switch changes the route taken by the content of the session; it is not a privacy label placed on top of the same route.

Here is what that means in practice.

A dictation in cloud mode

Hold the Voxor key and start speaking. The app streams the recording to the services that transcribe it. Uploading while you talk reduces the wait at the end, although Voxor waits for the completed result before putting text into the active field.

Voxor uses the recording to produce the transcript and then discards it. It does not keep a library of raw voice recordings.

The text is different. Cloud transcripts can appear in your account history so you can find and reuse them. They stay there until you delete them. This is the distinction worth remembering: the recording is temporary, while the resulting text can be stored as part of the product.

Cloud transcription and cleanup involve service providers. Their names and roles can change as the product changes, so we keep the current list in the privacy policy rather than copying a list here that may go stale.

What changes in Local Mode

Turn on Local Mode and the audio takes a shorter trip: from the microphone to a speech model running on the same computer. The cleanup step also runs there. The voice and transcript from that session are not sent to Voxor’s cloud for processing.

There is no quiet fallback. If the local model cannot run, Voxor shows an error instead of sending the recording to a server. We consider that behavior part of the feature. A local mode that becomes cloud mode when something goes wrong is not a boundary anyone can rely on.

Local dictation also works without a connection. The model has to fit on your machine, so the choice of model and its speed depend on the hardware you have. Server models have more compute available. Local models keep the session on the device and continue working offline. Which one fits depends on the work and the computer in front of you.

Local Mode does not make the whole app offline

You still have a Voxor account. The app still needs to know whether the account is active, can check for updates, and may occasionally connect for services unrelated to the content you dictated.

That does not send the words from a Local Mode session to the cloud. It simply means “local” describes the speech and cleanup path, not every network request the application will ever make.

We spell this out because broad claims create the wrong expectation. A useful privacy promise should have edges. Ours is that, in Local Mode, speech recognition and cleanup run on your computer without a silent cloud fallback.

Images and agents only go where you send them

Voxor can use an image as extra context, but it does not attach a screenshot to ordinary dictation. You have to choose the image-context feature. When you do, the image is sent for that operation, held for the session, and then discarded rather than added to transcript history.

Voice agents are another deliberate route out of the standard dictation path. An agent may send text to a webhook you configured. If you trigger it, that endpoint receives the information needed for the request. From there, the rules of the service behind the webhook apply.

Neither feature is hidden traffic attached to every recording. Both begin with an action you take, which is why the interface needs to make that action clear.

What you can delete

Cloud transcript history and raw audio do not share a retention rule. The audio is not kept as a permanent recording, but transcript history remains available until you remove it.

You can delete transcripts in the app. For an account export or account deletion, contact support@voxor.ai. Some billing records may have to remain for accounting or legal reasons even after an account is closed; the privacy policy describes those exceptions.

For sensitive work, the practical routine is short. Turn on Local Mode before you begin, check that you are using ordinary dictation rather than an image or agent action, and stop if Voxor says the local model is unavailable. The app will not make the cloud decision on your behalf.

We will keep updating this article when the product’s data path changes. For the complete terms, current service-provider list, retention details, and your privacy rights, read the Voxor privacy policy.

Related reading: Local Mode, dictation for Windows, and dictation for Mac.