This release is mostly about meetings. There is a floating panel you can pull up mid-call to ask questions, jot context for the summary and read the live transcript without leaving the app you are in. Meeting notes can be published as a public link, summaries can cite the transcript so you can jump straight to the moment something was said, and there is a library of ready-made summary formats so you are not starting from a blank prompt. On macOS, a freeze that could lock the app seconds after launch is gone, along with a whole class of stuck-shortcut bugs. The app is also a lot smaller.

New

• The Meeting Panel — Press Cmd+\ (Ctrl+\ on Windows) during a meeting and a floating panel opens over whatever app you are in. Jot context for the summary on any plan, and on Pro ask questions and follow the live transcript as it comes in. It never appears in your own screen share.

• Public share links (Pro) — Publish a meeting note as a link anyone can open. The summary and transcript go; the audio, screenshots, your private notes and your Ask AI conversations stay behind. Revoke it whenever you like.

• A library of ready-made templates (Pro) — Forty-four summary formats: standups, one-on-ones, discovery calls, user interviews, post-mortems, lecture notes and more. Add one and edit it however you like.

• Citations in your summaries (Pro) — Switch on summary citations and every bullet carries the timestamp it came from. Click one and the note jumps to that moment in the transcript and cues the audio there.

• Ask AI gets its own tab (Pro) — Notes now have a dedicated Ask AI tab with saved conversations in the sidebar, grouped by recency, renameable, and still there after a restart. The chat box sits at the bottom of every tab, so you can ask from the summary without losing your place.

• Rank your microphones — Microphone choice is now an ordered list rather than a single pick. Drag your headset above your webcam above the built-in mic, and Vowen opens the highest-ranked one that is actually plugged in when recording starts.

• A separate mic order for meetings — Meetings can follow their own microphone ranking, so calls prefer the desk mic while dictation keeps preferring your headset.

• More than one custom server (Pro) — Add as many OpenAI-compatible speech endpoints as you want, each with its own name, URL, model and optional key. They show up as ordinary model cards when you pick a model, and Vowen checks it can reach the server before saving.

• Teach the dictionary how you get misheard — Any dictionary word can now carry the ways it gets misheard. Add "cuber netties" and "kubernets" to Kubernetes and they are corrected in your dictations automatically.

• Echo cancellation for meeting notes (macOS) — On speakers, the other side of the call used to get recorded twice, once cleanly and once bleeding back through your mic. Vowen now cancels it against the actual output mix, and switches itself off when you move to headphones. Needs macOS 14 or later on Apple silicon.

• Tones and new notes from the idle pill — The idle pill picks up a tone switcher and a New Note button with a template picker, so you can pin a tone or start the right kind of note without opening anything. Using the Notch indicator style, the same options are drawn natively in the notch.

• Tags that come with the template — Give a template a set of tags and every note it produces is tagged from the moment it starts recording.

• Mic-only templates — A new "Capture system audio" switch per template. Turn it off for in-person meetings, one-on-ones and voice journals, where recording what is coming out of your speakers is just noise.

• Choose the browser and profile — A workflow that opens URLs can now say which browser, and which profile within it, to use, so work links land in your work profile instead of wherever your system default points.

• Find, back and forward — Cmd+F (Ctrl+F on Windows) or / focuses the search box on the notes, transcribe, history, dictionary, threads and expansions pages. Cmd+Left and Cmd+Right (Alt on Windows) step back and forward through where you have been, browser-style.

Improved

• Vocabulary that reaches the provider — Your dictionary now goes through each provider's own biasing parameter, Deepgram and ElevenLabs keyterms, AssemblyAI, Speechmatics and the rest, instead of one generic prompt, and it reaches meeting notes and file transcription where several models silently ignored it before. It also used to be truncated alphabetically, so a long dictionary quietly stopped working somewhere around "C"; it now keeps what you added most recently. A few engines still have no biasing hook at all, and those now say so in the logs instead of pretending.

• A much smaller download — Vowen ships a lot lighter: bundled SDKs, duplicated dependencies and unused Chromium locales cleared out. On Apple silicon the app is now about 373 MB. One model did go with it: the on-device GigaAM Russian engine has been retired, and Russian still transcribes on every cloud model and on multilingual Whisper.

• Music that stops the instant you record (Pro, macOS) — Muting background audio moved off AppleScript onto the native controller, so it lands in milliseconds instead of half a second and the first words of a video no longer leak into your recording. Plugging in headphones mid-recording no longer un-silences your Mac, and a crash can no longer leave it muted.

• Pausing falls back to muting (Pro) — When the app you are playing from refuses a pause, or reports nothing is playing when something clearly is, Vowen now checks the output device itself and mutes instead of quietly doing nothing.

• The meeting indicator, rebuilt — A responsive level meter, an audio-reactive glow, a live status label and a red stop button that reads as stop. The meeting-detected banner now carries the meeting app's logo and a split Start button for picking a template, and the ending banner counts down instead of just appearing.

• Long recordings on Gemini — Recordings past about twenty minutes used to die with an opaque "socket hang up", because the size check measured the audio before encoding inflated it by a third. They are now compressed and, past roughly half an hour, uploaded through Gemini's Files API instead of being crammed into the request, and the uploaded copy is deleted as soon as the transcript comes back.

• A live transcript that reads like one — Words in flight now settle in place with a colour change instead of vanishing and re-flowing into a different block, and speech split across two emissions no longer arrives as "ch at". Long uninterrupted turns on AssemblyAI now close within seconds in the live preview instead of running on unfinalized.

• Waveforms with actual shape — Meeting recordings mix your mic with system audio far below it, and the old waveform flattened the whole remote side into a one-pixel line. Existing recordings get recomputed the first time you open them.

• Search your notes settings — The meeting-notes settings modal gets the search box the main settings already had. Type "obsidian" or "citations" and the matching rows surface across all five tabs at once.

• Shift-click to select a range — Select one note, shift-click another, and everything between them comes along. Works on the notes and transcribe lists, and follows whatever filter and sort you have applied.

• Escape means escape — Esc now closes the top-most thing and only that, so a picker over a modal closes the picker. In the tone, workflow and expansion editors it asks before throwing away unsaved changes, and on a list it clears your selection.

• One scroller per page — Dictionary, Threads, Expansions, Models and Workflows now keep their banner and search bar pinned while just the content scrolls, matching the rest of the app.

• Quieter chrome — Dropped the shadows from every recording surface and the mistinted focus rectangles from every clickable one.

• Tones inside installed web apps (macOS) — A tone assigned to a site used to go dead the moment you opened that site as an installed app. Website rules now match inside PWAs, and installed web apps show up in the app picker.

Fixed

• The freeze seconds after launch (macOS) — Vowen could lock up about two seconds after starting, with the tray icon still there, settings dead, quit dead, and force-quit the only way out. The input hook that caused it now runs off the main thread.

• Every shortcut going dead at once (macOS) — Under load macOS switches off our key tap, and if it happened mid hold-to-record the recording could never be stopped, and while it was stuck every other shortcut stopped working too. It now recovers on its own within seconds instead of needing a restart.

• A bare "D" starting dictation (macOS) — Stuck modifier flags could turn an ordinary keystroke into a full chord, so a plain "D" started dictation and a plain "P" started command mode. On-screen and accessibility keyboards still fire shortcuts through the path that was always safe.

• Text expansion quietly switching off (macOS) — The shared input hook could be stopped out from under the text expander when something unrelated let go of it, and snippets would just silently stop firing mid-session.

• The last word going missing — Trailing-silence trimming was being applied to every cloud model by accident, clipping the final word on Deepgram, ElevenLabs, Mistral and others, while Retry transcribed it correctly. Streaming models also lost the last buffer when you stopped.

• Undoing a cancellation — Undo left the indicator stuck on "connecting" for the rest of the take, brought the recording back in the wrong mode so the shortcut that started it could not stop it, and dropped the tone and language you were recording with. It is also no longer offered after a hold-to-record take, where there was nothing to resume into.

• The idle pill's own shortcut — Starting a recording from the idle pill made your keyboard shortcut a no-op, so the pill advertised "Hold to dictate" and then only the mouse could stop it.

• Meetings recording your mic only (macOS) — On FaceTime and Phone calls, system-audio capture could attach and deliver nothing but silence, so you would find out afterwards that the other side was never recorded. Vowen now notices within seconds and switches to a different capture path mid-recording.

• Meeting notes ignoring your settings — An auto-started meeting could resolve its settings before they had finished loading, so "Identify Speakers" silently stopped applying and the model, language and template fell back to defaults.

• Summaries using the wrong provider — Picking a summary provider that was not connected silently fell back to a different one, so you kept seeing an error from a provider you had not chosen. It now tells you which one is not connected and where to fix it.

• The indicator opening expanded — The recording indicator appears exactly where your cursor already is after clicking Start, so it read that as a hover and opened its card at the beginning of every meeting. It now waits to be hovered on purpose.

• The indicator cut off mid-dictation — Dismissing a banner while still recording could tear the indicator down or shrink it until the live preview was clipped.

• Speaker setup you never asked for — Starting a meeting with a cloud model and Identify Speakers on downloaded a local diarization bundle and showed a preparing banner for an engine the recording never touches.

• Vowen detecting itself as a meeting (Pro) — Dictating or using command mode could raise a meeting-detected banner off Vowen's own microphone use, sometimes announcing a Teams meeting purely because Teams was open in the background.

• Everything slightly too big — In the installed app the recording indicator, its card and the meeting panel all rendered about 11% larger than intended, because the zoom the layout was designed against only applied in development.

Update from within Vowen (Settings → General → Check for updates), or download the latest build at vowen.ai.