FreeFlowDownload
Changelog · 1.5.1

⌥ shortcuts leave the microphone alone.

By Yiftach FreemanUpdated

  • A ⌥ press during “Writing it down…” is no longer lostPress ⌥ again while the previous take was still being transcribed and nothing happened: no recording, no indicator, and whatever you said went nowhere. Cohere takes a second or two to write a take down, so this was easy to hit. The hold is now remembered and the next take starts the moment the previous one is pasted — keep holding, and watch for the indicator.
  • Silence no longer comes out as “Thank you.”Hold ⌥ without speaking and the model was handed a second of room noise, which it confidently transcribed as “Thank you.” The silence check now looks for a moment of actual speech-level sound instead of averaging the whole take, so an empty hold shows “Didn’t catch that” and pastes nothing.
  • ⌥ chords no longer start a takePressing ⌥ as part of a shortcut — ⌘⌥← to switch tabs, ⌥⌫, ⌥e for an accent — opened the microphone, ducked your audio and flashed the indicator for a few milliseconds before the chord cancelled it. On AirPods every one of those flipped the headset into its mono call profile: sound in one ear, crackle, and a volume that sometimes never came back. Holding ⌥ now waits 150 ms before it touches anything, so a shortcut never reaches the microphone at all.
  • Your volume always comes backDucking put the volume back only if it was still exactly where FreeFlow had left it — a Bluetooth headset renegotiating its level was enough to strand it at 25%, and the next take ducked from there. It now comes back unless you turned it up yourself mid-take, and, outside Incognito, every duck and restore is written to the log.
  • The language menu knows your modelCohere Transcribe speaks 14 languages and nothing else. Picking Hebrew for it quietly typed English, with one grey caption as the only warning. A mode’s language menu now lists only what its model can transcribe, and a mode already set to something its model can’t do says so in red. Hebrew, and every language outside those 14, needs a Whisper model.
  • Speed and accuracy on every modelThe meters only fit beside a trash icon, so installed models showed them and the ones you could still download didn’t — and “Accuracy” wrapped onto two lines. They now sit on the same line as each model’s description, on every row, at any window size. Hover a description for the model’s story, and click the ⓘ next to Voice models for what “avg WER” and the meters actually measure.
  • Modes give the settings the roomThe list of modes opened wide and the settings squeezed in beside it. The list now starts narrow and the settings take the rest — drag the divider if you want it otherwise.

Newer: 1.5.2. Older: 1.5.0. Every release.

Download the current versionUpdates install themselves from inside the app.
Download now