Whisper knows English — it doesn't know your colleague's name, your repo, or the product in the window in front of you. Context spelling closes that gap: when you start dictating, Tota glances at the visible text of the window you're dictating into and uses the names and labels it finds there to spell them correctly in your transcription.
How It Works
- At the moment you start recording, Tota reads the visible text of the frontmost window through macOS Accessibility — the same permission it already uses to insert text.
- Names and interface labels found there bias the speech engine while it transcribes, and a sound-alike pass fixes any that were still misheard.
- Windows and tabs you looked at in the minutes before dictating get a say too, so a name you just read somewhere else is still spelled right after you switch apps.
Everything happens on your Mac and lasts only for that dictation: harvested terms are never stored, never added to your Dictionary, and never sent anywhere.
Seeing and Undoing Corrections
When a context correction changes your transcription, it appears in the corrections card with the original wording, a one-click Revert, and a don't correct again option for a term Tota keeps getting wrong.
What Tota Never Reads
- Password fields and other secure input
- Terminal windows
- The text field you're dictating into
- Tota's own windows
Turning It Off
Context spelling is on by default. Toggle Spell what's on your screen in Settings > Corrections to turn it off — with it off, Tota does no screen reading at all.
