How dictation works

When you hold the hotkey, Dettivo opens an audio buffer immediately. It does not wait for the speech-to-text model to warm up — the audio is captured, then handed to the warmed model when it becomes ready. First dictation feels slightly slower than the second; the model prewarm hides that gap after one use.

After capture, audio runs through your selected STT engine — usually a bundled WhisperKit model. The raw transcript is what you get in Raw mode. Polish then applies deterministic dictionary substitutions, replacement rules, snippet expansions, punctuation normalization, and code/path/flag post-processing. Enhanced adds an optional model rewrite on top.

Finally, the result is inserted into the focused app via Accessibility. If insertion is blocked — secure fields, sandbox restrictions — Dettivo falls back to clipboard with a five-second undo window.