Privacy and Local Processing
Super Voice Mode is local-first. Dictation stays on your Mac. Connected Assistant workflows depend on the backend and tools you choose.
Dictation and AI Writing
Section titled “Dictation and AI Writing”The following processing stays on your Mac:
- Microphone capture.
- Voice activity detection.
- Speech-to-text.
- Dictionary replacements.
- Optional AI Correction.
- AI Writing and highlight-and-correct.
- Clipboard preparation and paste.
Normal dictation audio and transcripts are not sent to a cloud service.
When Vocabulary is enabled, Super Voice Mode can read a small amount of nearby text from the active app to help recognize uncommon words in the current dictation. That context is processed on your Mac, is not written to history or settings, and is discarded after the dictation. Sensitive apps and protected text fields are skipped.
Assistant backends
Section titled “Assistant backends”- Local MLX and Ollama: Prompts and replies stay on your Mac. These backends do not have agent tools.
- Hermes: Data goes to the gateway you configured. Hermes tools can contact other services depending on that gateway configuration.
- Claude, Codex, Cursor, and OpenCode: Your request and the context allowed by that agent runtime can be sent to its configured provider, just as when you use that agent directly.
Review the privacy terms of any external backend and enable only the tools you want it to use.
Voice Output
Section titled “Voice Output”Speech synthesis uses Pocket TTS on your Mac. The text is not sent to a cloud text-to-speech service.
Meetings and Notes
Section titled “Meetings and Notes”Meeting transcription and note generation run on your Mac. Meeting audio is not retained after processing. Saved notes, transcripts, and snapshots are ordinary files in your chosen local library.
Treat meeting content as sensitive if you later ask an external Assistant to read or summarize it.
Computer Use
Section titled “Computer Use”Computer Use reads or acts on the window you authorize. It prefers text and accessibility data. A screenshot is used only when the task requires visual information and Screen Recording permission is available.
The selected agent backend may receive the content needed to complete the request. Drafting text never submits it automatically; sending or committing requires an explicit action.
Network activity that is not dictation
Section titled “Network activity that is not dictation”Super Voice Mode can use the network for:
- Downloading the app and on-device models.
- Checking for and downloading updates.
- Validating a Pro license.
- Optional external Assistant backends and their tools.
These cases do not change the local behavior of normal dictation.
Public examples and support
Section titled “Public examples and support”Do not include real names, handles, email addresses, license keys, meeting content, home-directory paths, or private file locations in public examples or support screenshots. Replace them with fictional data before sharing.