Voice notes
Record onto the board. Transcribe when you want it, into a note beside it.
Drag Voice out of the toolbar and press record.
Recording happens natively, in Umbra's own process rather than through the webview, and writes straight to disk as you speak. The waveform is drawn from the audio itself, so a quiet recording still shows something rather than a flat line.
The audio is an asset — a file in your data folder — never text on the element. A recording embedded in the board's data would be re-synced in full every time you renamed it.
Transcription
Transcription is on demand. Press Transcribe and the transcript arrives as a fresh note placed beside the recording — searchable, editable, and readable by the AI like any other writing. It is not hidden on the voice block, because text on a voice block is none of those things.
What we cannot promise yet
Transcription asks the AI CLI you have configured. Not every CLI can read an audio file — most handle text, images and PDFs only. If yours cannot, the note says so plainly and your recording is untouched. Everything else about a voice note works with no network at all.
Permission
macOS asks for the microphone the first time you record. Umbra records only while you hold the button down and stores the file locally; nothing is uploaded anywhere.