
Your audio never leaves this browser — it is read, transcribed and cut here. Only cloud clip suggestions send anything out, and only the transcript, only if you turn them on.
Starts on its own once the audio has been read.
Runs once a transcript exists, aiming for 15–20 clips.
Work is kept here between visits. The audio is cached for the last two only.
Sends the transcript to your own model and writes the week's collateral from it.
Runs once the transcript is ready, after clip suggestions.
Whisper mishears names and proper nouns the same way every time. Teach it once and every pass through this file is corrected — transcript, captions and exports.