Voice & input

Dictate a prompt instead of typing it. Speech is transcribed on your machine by a bundled whisper.cpp speech model — audio never leaves the machine and no account is involved.

Using it

• Press the microphone button in the composer, speak, press again. The text lands in the composer for you to edit before sending.
• Push-to-talk from anywhere is off until you opt in from Settings › Voice & input. Once enabled, hold the global shortcut (default ControlShiftSpace) while Synergoi is focused; press again to stop from another app. A badge stays visible in the window chrome while the shortcut is armed.

The speech model

Three tiers, all running locally. Fast is in the download (a ~57 MB multilingual model) and works out of the box. Balanced (~181 MB) and Best (~547 MB) are one-time downloads from updates.tracewyn.com — bigger model, better transcription, still nothing but your own machine hearing you. Until a tier finishes downloading, Fast keeps working.

Pick the language you speak in the same pane, or leave it on Detect language; choosing English uses the English-only checkpoint, which is more accurate on English. A vocabulary box lets you list names, jargon and project words it keeps mishearing.