Skip to docs content
Capture

Voice notes

Dictate when typing is slow. Audio goes to OpenAI Whisper, comes back as text, and lands either in the document you're writing or as a new project note.

Two places to start dictating

There are two voice entry points in the workspace:

  • Mic button in the chat composerTap the mic icon next to the send button to dictate a message to the research chat. Tap once to start, once again to stop. The transcript drops into the chat input where you can edit it before sending.
  • Voice dictation commandOpen the command palette with Cmd+K and pick "Voice dictation". A modal opens with a single round record button. Stop, review the transcript, then either insert it into the current note at the cursor or save it as its own note in the project file tree.

What happens to the audio

The recording is streamed to OpenAI Whisper for transcription and then discarded; only the transcribed text is stored in your project. If you save the transcript as a note, that note is a normal Tiptap note the research chat can read alongside your papers.

Transcripts are editable like any other note. Fix mis-heard terms, add formatting, split into sections, or hand the raw text to the research agent and ask it to clean up the dictation.

Browser support

Chrome, Edge, and Firefox record in WebM or Ogg. Safari and iOS record in MP4. Whisper accepts all three. If your browser blocks microphone access, the dictation button shows a specific error (denied, no device, or in use by another app) so you know what to fix.

Long recordings
Whisper accepts up to about 24MB per upload. For a continuous one-hour dictation, record in shorter segments and stitch the transcripts together in the note.
Keep going