Vowen 0.5.6 Release notes
New
Reminders you set by voice. Ask command mode to remind you about something at five, or in twenty minutes, and Vowen brings it back in its own banner. It survives sleep, so one that comes due while the lid is shut still arrives when you open it.
Gemini 3.5 Transcribe. Google's dedicated speech model is now a provider. Dictation streams word by word, files and meetings come back with native speaker labels, and your dictionary reaches it as a real biasing field instead of a prompt.
Choose which Gemini model transcribes. The Gemini card carries a choice between
gemini-3.5-transcribeandgemini-3.5-flash, so the general-purpose model stays available for anyone who prefers it.Drop a file anywhere on the Transcribe page and the dialog opens with it attached, instead of you having to open the dialog first.
The macOS Share menu, Open With, and dock drops all hand files to the same place. Vowen selects them in the Transcribe dialog rather than transcribing on its own, so the model, language and diarization choices stay yours.
Drags from Voice Memos, Photos and Mail. Those apps hand over a file promise rather than a file, which Chromium cannot read, so the drag used to do nothing at all. Vowen now resolves them natively while the Transcribe dialog is open. (macOS)
Timers say when they are done. A timer set through command mode raises a banner the moment it runs out, instead of finishing quietly.
Mute either side of a meeting, mid-meeting. Hover the meeting indicator for separate mutes for your microphone and for system audio, live on the running recording. Switching system audio on stands the capture up without restarting.
Play a recording from the voice log. Press and hold any entry on the home page and its audio opens in place, with the same player the notes and transcription pages use.
Screenshots that reach Claude Code and Codex. Pointing at things while dictating into a terminal agent now works. A terminal cannot carry an image, so Vowen sends the chord those agents listen for and they read the screenshot off the clipboard themselves. (macOS)
Improved
Meetings that run for hours. Audio is compressed before it is uploaded and request timeouts scale with the size of what is being sent, so long recordings stop failing on the clock rather than on anything the provider did.
A live preview that reconnects. Every streaming provider hangs up eventually: Gemini at ten minutes, Deepgram and OpenAI at an hour, AssemblyAI at three. The preview notices and re-opens instead of going quiet for the rest of the meeting.
One card on the meeting indicator, not two. The hover card lost the notes panel that duplicated the meeting panel, and teaches the panel shortcut as real keycaps instead. It also stops opening revealed just because the cursor was already where it appeared.
Integrations settings, rebuilt. Calendar, Obsidian, webhooks and export in meeting-notes settings now share one layout, with what each one saves shown as cards rather than a column of switches.
Watch folders that read as a list, with what happens to each folder stated on its own row.
Half a second to cancel. Holding the stop button to throw a take away takes about half a second now, down from over a second, and the ring around the button fills in step with it.
Space plays, everywhere. Space toggles playback in every player in the app, and the angle brackets step the speed. Both stand down while you are typing or while a button has focus.
A transcript editor that keeps up. Typing in a long transcript re-rendered every row on every keystroke, which is exactly the lag it felt like. Editing is smooth now, whatever the length.
Cmd+F in settings (Ctrl+F on Windows) focuses the settings search box, the way it already did on the notes, dictionary and transcribe pages.
A quieter sidebar. Dropped the section heading over the AI pages.
Fixed
A stray space and a missing capital in Electron apps. Chromium apps take seconds to build the accessibility tree Vowen reads the caret from, and VS Code measured over four. Vowen now warms that tree when you switch apps, and when it still cannot read the caret it leaves your text exactly as dictated. (macOS)
A button read as an empty text field. A focused button answers the caret questions anyway, and answers them as though it were an empty field at the start of a sentence. Smart spacing now only acts on something that really is a text field. (macOS)
Instructions that pinned a language and kept one. Custom instructions asking for output in one language while also asking to keep the language you spoke translated every dictation into the pinned one. The language you spoke now wins that contradiction.
The connectors catalog failing to search. Composio started appending a note about custom servers to its search results, which made every catalog search come back unreadable.
A console window flashing when speaker detection runs. Speaker detection and the local speech engines spawned a visible console window on every run. Those are hidden now. (Windows)
Speaker detection timing out on long meetings. It was given a flat ten minutes regardless of length, so an hour-long meeting was killed while it was still working. The budget scales with the recording now, and a run that does blow it no longer leaves your retry queued behind it. (Windows)
Pausing apps that publish no metadata. An app can own the media session without publishing a title, Stremio among them, and those could not be paused at all. Vowen reads the playing state directly now and pauses them instead of muting your Mac. (macOS)
Settings jumping to the front when notes start. Starting meeting notes from the idle pill, a shortcut or a meeting banner brought the settings window forward over whatever you were in. It stays where it was, and only comes up when something genuinely needs you.
A failed note opening on the wrong tab. A note whose summary failed opened on Transcript, the one tab that cannot show why. It opens on Summary now, with the reason rendered read only so a keystroke cannot save it over the note.