Say what the step does and let Whisper type it into the description for you.
Voice input has its own provider and its own key, separate from the one the writing actions use. Both live in the same dialog: open the menu next to the ✨ button in a step description and choose AI API Settings, then scroll to AI Voice Input Settings.
Test Connection checks it before you rely on it. Settings save as you type.
Your operating system will ask for microphone permission the first time. On macOS that prompt appears once, and if you dismiss it you will need to allow Folge under Privacy and Security.
Transcription hints is the setting that turns this from a novelty into something usable. Whisper is good at ordinary speech and bad at names it has never seen, so a list of the words you actually say fixes most of what it gets wrong.
Put your product names, systems, colleagues and acronyms in there, separated by commas: Northwind Ops, OAuth, Bergamo Foods, SKU, Kalmar. Without it, a product name comes back spelled the way it sounds.
The audio you record, and nothing else. No screenshot is involved, so this is the one AI feature in Folge that never sends a picture of your screen.
It goes to the provider you configured, on your key and your account. If recordings of internal discussion cannot leave the building, point the custom provider at a Whisper server you run: see Set Up AI in Folge.
Folge captures a screenshot on every click, so a guide writes itself as you work.
