All help topics

AI Polish and Cloud Data

Which AI Polish options keep your text on your Mac, and what is sent when one does not.

Whether any of your text leaves your Mac depends entirely on which AI Polish option you picked. Go to Settings > AI Polish to check your current selection.

Nothing leaves your Mac

These options keep your transcription and the polish request entirely on your device.

  • None. No AI step runs at all.
  • Apple Intelligence. Apple’s model, running directly on your Mac.
  • EG-1. The model built by Envious Labs, running directly on your Mac.
  • Ollama, when you pick a model you downloaded. Ollama also offers hosted models that run on its own servers, and those do send your transcribed text. EnviousWispr shows the hosted models under a separate heading so you can tell which kind you are choosing.

Your text is sent

OpenAI, Gemini, Claude, and Ollama’s hosted models all run on their own servers. Your account is with that company, on their terms. Envious Labs is not in the middle of it. Everything goes straight from your Mac to the provider, so Envious Labs never sees your text either.

What is sent

  • The text that needs polishing.
  • The instructions for cleaning it up, plus your custom words.
  • The name of the app you are dictating into.
  • Your API key, so the provider can confirm the request belongs to you. Ollama’s hosted models use your Ollama sign-in instead of an API key.

What is never sent

  • Your audio. That holds on every single option without exception.
  • Your other transcripts, or your History.

EnviousWispr adds nothing to the request that identifies you or your Mac. The provider still sees the ordinary details of any internet connection, such as your IP address.

With OpenAI and Gemini, EnviousWispr also asks them not to keep a copy of that request. That is a request to the provider, not something EnviousWispr can enforce, and your text still has to reach their servers to be polished.

If polish fails

You still receive the tidied-up version of your dictation from immediately before the AI step. A network problem, a slow provider, or a reply that arrives cut short costs you the polish, never your words.