Changelog
What changed in FnFlow, newest first.
0.14.0 17 October 2026
- Model library. Settings → Recognition now lists 18 Whisper models, from tiny (32 MB) to large-v3 (3.1 GB), including English-only and smaller compressed versions. Download any of them, switch with one click, delete the ones you don't need. Every model runs on your Mac and is checked before use.
- A paused download now resumes only for the same model, so switching models mid-download can't mix files.
0.13.1 12 October 2026
- Dictation now pastes into text fields that macOS doesn't expose to accessibility, such as some fields in Electron apps and web pages. When the app can't confirm the paste, it says “Paste sent” and keeps the text in history.
0.13.0 9 October 2026
- A friendlier first minute. The setup window shows the gesture (hold the key, talk, let go), a progress bar and a clear checkmark next to each permission, with a line on what each one is for.
- Your first dictation happens right in setup. Say the suggested phrase and see “It works!” before you go anywhere else.
- If your 🌐 key is already busy, the app picks right ⌘ for you instead of sending you to Keyboard settings. If a permission doesn't register within 10 seconds, setup tells you what to do.
- When an update has downloaded, the menu shows Restart to Update, so you don't have to quit the app to get it.
0.12.3 8 October 2026
- Updates now download in the background and install the next time the app quits, so you stay on the latest version without a prompt.
- Anonymous usage counts now cover the setup steps, “opened today”, a rough time to the first dictation and whether a dictation failed or couldn't be pasted – so we can see where people get stuck. Still no ID and no text, and still off with one switch in Settings → General. The Privacy page lists every event.
0.12.2 8 October 2026
- The speech model starts getting ready the moment the app opens, while you grant permissions – the first dictation waits less.
- Quitting the app always stops its background recognition process, even mid-task, so no memory is left in use.
- Anonymous usage counts are no longer sent twice when permissions are granted.
0.12.1 8 October 2026
- Running straight from the installer? The app offers to move itself to Applications, where it updates itself and starts faster.
- The very first transcription after install says “Preparing recognition – first time only” instead of looking stuck while the Mac readies the model.
0.12.0 8 October 2026
- Right ⌘ works out of the box. If your 🌐 key is busy, the app uses right ⌘ – no trip to Keyboard settings. Switch between fn and right ⌘ in Settings → General. Shortcuts like ⌘C no longer start a dictation.
- Try it right in the setup window. Hold the key, say a few words and see them appear before you go anywhere else.
- Half the download for the accurate model. The background model is now large-v3-turbo in 8-bit (874 MB instead of 1.6 GB) – the same text in our test on 32 real dictations, and a little faster.
0.11.0 8 October 2026
- Setup in one button. Grant Access asks for the microphone and Accessibility in turn and notices on its own when you've allowed them – no restart. Input Monitoring is no longer needed on most Macs.
- The setup window checks the 🌐/fn key setting and tells you when it still starts Apple's dictation or switches the input language.
0.10.0 7 October 2026
- Anonymous usage counts: installed, set up, first dictation and dictated today – with the app version only, no ID and no text. Turn them off in Settings → General. The Privacy page lists exactly what is sent.
0.9.9 7 October 2026
- The setup window now shows that recognition works right away with the built-in model, and that a more accurate model downloads in the background on Wi-Fi. One click turns that download off.
0.9.8 6 October 2026
- Automatic updates. The app checks for new versions once a day and installs them in a click. You can also use Check for Updates… in the menu.
- Report a Problem… in the menu opens an email to us with the app and macOS version filled in.
0.9.7 6 October 2026
- Your own vocabulary. Add names and jargon in Settings → Vocabulary; Whisper uses them as spelling hints. Replacements fix finished text with any engine.
0.9.6 5 October 2026
- Whisper is built in. No Homebrew, no terminal. A light model ships with the app, so it works right after install.
- The larger model downloads in the background on Wi-Fi, with pause, resume and checksum verification, and the app switches to it on its own.
- Runs on macOS 15 Sequoia and later (was macOS 26). Apple's speech engine remains available on macOS 26.
0.9.5 5 October 2026
- Accidental taps of fn shorter than 0.6 s no longer start a transcription.
- Silence now shows “No speech heard” instead of an error, and Whisper's phantom phrases like “Thank you.” are removed.
- Recognizing while recording is on by default.
0.9.4 5 October 2026
- Pick the engine, language and model from the menu bar and the history window.
0.9.2 4 October 2026
- History opens about 20× faster on large histories. Handles sleep and wake.
0.9.1 4 October 2026
- Idle CPU use down from 13–16% to under 0.5%.
- Whisper is kept ready while you talk: about 1.2 s from releasing fn to text.
0.9.0 3 October 2026
- New native interface: menu bar popover, glass recording pill, three-column history, settings and a setup assistant.
- New interface in English.