docs
Post-processing is a second model that rewrites the recognised text to your instruction — remove filler words, make it a business letter, turn it into a list. The instruction is written in ordinary words and is called a prompt. It is off by default, because it is the one step that noticeably slows a dictation down.
The main switch. While it is off the rest of the tab does nothing and the text is inserted as recognised.
An editor that runs on your own computer. Nothing leaves the machine, but such models weigh a gigabyte and up. The catalogue next to it downloads one in a click.
The text goes to a service you name — OpenAI, Groq, your own server. Faster and cleverer, but the text leaves the computer. The key is stored encrypted, and a test button checks the address before you rely on it.
Your instructions to the editor. The ticked ones run in order, top to bottom. Each has a name, the instruction itself, an optional shortcut of its own, and a field to try a phrase without dictating anything.