Download VibeVox
Windows 10 / 11 · 64-bit
Version 0.2.16 • ~648 MB
Unzip → run VibeVox.exe. If Windows SmartScreen appears: More info → Run anyway (the app is unsigned).
Requires Windows 10 build 19041 or higher
Changelog
v0.2.16
September 2026
- The first dictation after starting the app no longer occasionally goes missing — on some Windows 11 computers the first pasted text of a session could be lost
- Pasted dictations no longer show up in the Windows clipboard history (Win+V), and cloud clipboard no longer carries them to your other devices
- A dictation is no longer lost silently when the cursor is somewhere text cannot go — the app tells you, and the text waits on the clipboard for Ctrl+V
- On Windows 10 the recording pill no longer takes the focus away from the window you are dictating into
- Proofreading no longer pastes a model's refusal, or your text unchanged, over the selection — the selection stays as it was and the app says what happened
- “Start with Windows” shows the real state: if VibeVox is switched off in the Windows startup list, Settings sees it, and switching it on works again
- Your hotkey pressed together with Ctrl or Shift no longer starts a dictation, and the arrow keys can now be chosen as the hotkey
- With noise suppression on, the first dictation no longer waits three seconds for it to load
- Many smaller reliability fixes — quitting during a recording, switching the speech model in Settings, saving settings
v0.2.15
September 2026
- The first dictation after you turn the computer on no longer comes back empty — the microphone takes about three seconds to wake up, and that wait used to land on your first take
- Proofreading keeps the text you selected when the dictation starts from a mouse side button or from the tray icon — the click itself used to clear the selection before the app could read it
- The recording pill shows how far the recognition has got, and it no longer jumps, changes size or flickers while it moves between recording and recognising; the indicator itself is noticeably bigger, and the brief pill that used to flash before recording starts is gone
- The pill can no longer sit on screen showing nothing, and the mouse no longer falls straight through it
- Settings keep what you chose: a switch no longer flips back on its own a moment later, and a new API key typed right after removing the old one is no longer lost
- The app reaches the tray about two seconds sooner after it starts
- Uninstalling asks you to close VibeVox first instead of going ahead while it is still running
- Translation and Proofreading work with the slower, cleverer models too — if one runs out of time the next attempt gets longer, and the app now says the model was too slow instead of just failing
v0.2.14
September 2026
- Long dictations: the limit is now five minutes, and a long take is transcribed whole — the Norwegian model used to hear only its first thirty seconds and say nothing about the rest
- Pauses inside a dictation no longer cost you waiting time: the silence is skipped instead of being sent to the recognition
- A quiet microphone no longer swallows the ends of phrases, and when the input level in Windows is set low the app now says so
- The first short word of a phrase is dropped far less often
- Translation and Proofreading now work on long texts — past about five thousand characters they used to give up with nothing but an error
- The tray menu shows the language you translate into, so you no longer have to open the settings to remember it
v0.2.13
August 2026
- Windows lets every screen have its own scale, and everything the app draws itself — the Settings window, the tray menu, the recording pill — now comes out the right size on each of them, including when you drag it from one screen to another
- The Settings window remembers the screen you moved it to and opens there next time; leave it on your main screen and it keeps appearing wherever the tray is
- Dragging the Settings window by its title bar: the window now stays under the cursor instead of drifting away from it
- Live typing finishes at once when you stop after a pause — the app no longer spends time listening to a silence you have already finished
- The recording pill steps out from under the Windows search window again
- And the small things: no empty strip under “Transcribing…” and no empty column beside the tray menu on a screen with a different scale
v0.2.12
August 2026
- The interface, the tray menu and the recording pill are now read aloud by a screen reader — checked with Windows Narrator and with NVDA. Every switch says its name, whether it is on, and what it does
- The recording pill no longer hides behind the Windows search window, and it holds one place for the whole dictation instead of moving as it goes from recording to transcribing
- The Settings window opens already drawn — no white or black flash first — and it can no longer come up showing placeholder values instead of your own settings after a cold start
- When a setting needs a restart, the app now waits until you close the Settings window, and the tray icon comes back every time
- Messages about the microphone fit inside the notification again, together with the path to the setting you need
- If animation effects are turned off in Windows, the recording pill no longer pulses and the spinner no longer spins
- And the small things: Windows’ tray-icon list names the app “VibeVox” instead of its filename; a new dictation started while the previous one is still being transcribed is politely declined instead of piling up; and a problem report now says which machine and which screen it came from
v0.2.11
August 2026
- The interface now speaks 14 languages — Norwegian, Italian, Dutch, Polish, Swedish and Ukrainian join the eight that were already there. The language lists inside Settings are written in your own language too, and sorted the way you read them
- Each speech model remembers the language you last used it with: switching to the Norwegian pack and back no longer leaves a foreign language behind. And “Auto-detect” is finally saved when you pick it
- The recording pill names the work it is doing — “Translating…”, “Rewriting…”, “Proofreading…” instead of “Transcribing…” for the whole length of an AI request. And it appears the moment you press the key, while the microphone is still opening
- When a request to the AI fails you are told, instead of quietly receiving your raw dictation. Translation and the update check also work again on machines whose Windows certificate store had gone stale
- Rewriting a selection is more reliable: the copy no longer reads the empty middle of another program’s copy, and if the AI declines the request, its refusal never lands in your document
- A switch that cannot work no longer sits there looking switched on: it now names the setting standing in the way, and the tab it is on
- Help rewritten from scratch in all 14 languages: the questions and the tips describe the program as it is today
- And the small things: a sharp icon at every size, a tray icon that no longer stays grey after a reboot nor leaves a dead copy behind after a restart, a downloaded language pack that survives deleting the program folder, and detailed logs by default — so a problem report names its cause
v0.2.10
August 2026
- Beautiful writing: turn Rewrite on, select nothing, and your dictation arrives as if you had typed it — punctuated, with the hesitations and false starts gone. Numbers, names and dates are left alone, and if the AI answers with something else you get your original dictation instead
- The AI model list now comes from the provider itself: a “Refresh list” button next to the model field, you can type a name by hand, and when a model has been retired VibeVox says so instead of showing a generic failure
- OpenRouter added — one key, hundreds of models. Perplexity removed from the list: it is a search service, and our models are gone from its catalogue
- The tray and the Settings window no longer disagree: switch Translate on in the tray and the switch in Settings follows
- The recording pill no longer shows black corners on Windows 10
- The download now carries README.txt, with the English text first
v0.2.9
August 2026
- The default shortcut is now a double tap on Ctrl — on many laptops the top row sends media keys, so F2 never reached VibeVox at all. Any key can still be chosen in Settings, and there is a built-in test that tells you what is wrong with yours
- Live typing keeps up much better: a finished phrase now appears as soon as you pause, instead of waiting for the next one
- Speaker verification cuts out only the words that were not yours, instead of blocking the whole dictation; History keeps the complete text either way
- Silence is no longer transcribed — the model can no longer invent a short phrase out of room noise
- The start and stop sounds no longer stutter while the computer is busy transcribing
- The app now says when your microphone is muted in Windows or held by another program, instead of showing a generic error
- Copied images and files survive dictation — rewriting a selection or restoring the clipboard no longer turns them into plain text
- New installs keep no history by default, there is a one-click diagnostics bundle under Support, and installing over an older version no longer mixes old and new program files
v0.2.8
July 2026
- If a model download fails you now see why — and you can install the model by hand from a folder
- Your own word list: names and terms the model keeps mishearing are now written correctly
- Text now arrives via the clipboard, so modern apps (the new Notepad and friends) no longer receive it truncated
- VibeVox can mute the system sound while you dictate
- The recording pill shows a live waveform of your voice; the microphone meter is recalibrated and warns you when the Windows input level is set too low
- Translate and Rewrite are two switches now, exactly like in the tray menu
- Recording limit is up to 4 minutes — the speech model cannot transcribe longer takes
- Many reliability fixes: the pill no longer gets stuck on screen, the clipboard survives two quick dictations in a row, and settings changed in one window are no longer reverted by another
v0.2.7
July 2026
- Settings, the recording indicator and the tray menu now open reliably — after unpacking the download with Windows Explorer they could previously stay blank
- If Windows is missing the system component the interface needs, VibeVox now offers to install it in one click
- If no microphone is connected, VibeVox says so instead of showing a generic error
- Dictating into a window that runs as administrator now puts your text on the clipboard with an explanation, instead of losing it silently
- Checking for updates no longer downloads the full package unnecessarily
v0.2.6
July 2026
- Hands-free: if your camera disconnects mid-session, it now switches off cleanly — no more stuck camera light
- Live typing stays in the right window even if you switch apps while dictating
- Changing a setting while you're recording no longer disturbs your next dictation
v0.2.5
July 2026
- Live typing (beta): words appear on screen as you speak, not only at the end
- New Audio panel: microphone picker, sensitivity slider and a live input level meter
- Push-to-talk (Hold) recording mode and a draggable floating recording indicator with live "Transcribing…" status
- AI translation & rewrite with your own API key (OpenAI, Claude, Gemini and more) — including a one-click key test
- Norwegian dictation fully offline — the model downloads right from Settings
- Transcription history you can review and clean up in one click
- Faster and more resilient: quick first dictation, self-healing settings, clear error messages, lower memory use, and a tray icon that pins itself to the taskbar
- Beta polish: dictation no longer lands in the wrong window when you switch apps, a disconnected Bluetooth mic no longer sticks recording, and Settings closes instantly
- Lighter download (~700 MB) — the Norwegian model installs from within the app; the portable build offers a desktop shortcut on first launch
v0.2.4
June 2026
- Fixed: the Settings window no longer opens by itself after a Windows restart
- Autostart is now a single, reliable entry
v0.2.3
May 2026
- Norwegian dictation (Bokmål & Nynorsk) via an optional on-demand NB-Whisper pack
- Tray menu AI toggle: turn AI on/off and switch between Translate and Rewrite in one click
- Scrollable language picker with a recognized-languages info popup
- Lighter INT8 quantized models for slower PCs
v0.2.0
February 2026
- Bring-your-own-key AI with Translate and Rewrite modes across 9 providers
- Parallel AI translation with progressive injection