Most voice apps for the Mac are dictation apps: hold a key, say a sentence, let go, and the words are typed where your cursor is. Psst does that. But its first job is the other one.
Dictation
Short. Aimed at a text field. You already know what you want to say; you just do not want to type it. Hold ⌥ Space, speak, release. The sentence lands in Mail, Notes, Slack, a terminal, whatever is in front. The transcript is not kept, because the text is already where it belongs.
Recording
Long. Aimed at a file. You do not yet know what you think; you are working it out by talking. Tap the bot, walk around, talk for an hour, tap again. The transcript is saved as a text file with the date and time, and it is shown in the panel with a Copy button. This is thinking, not typing.
Why one app
Because they share everything underneath: the same microphone, the same local model, the same Keychain, the same folder. Two apps for this would mean two models downloaded and two things to trust. One small bot on the desktop handles both, and the halo tells you which is happening: copper at rest, red while it records.
The rewrite is a third thing
Select any text, hold ⌃ ⌥ Space, say what to do with it: "shorter", "fix the grammar", "make it a list". The selection is replaced. It needs an OpenAI key, because it is a language task, not a listening task.