Press a hotkey. Speak. Clean text lands at your cursor — in any app.
A free, open-source, cross-platform voice-dictation tool. You bring your own API key; there's no subscription.
Vito is short for Voice In, Text Out. It's a small background app that turns your voice into polished, ready-to-use text — anywhere you can type. It comes with a friendly, built-in web interface to manage everything — settings, history, dictionary and more — shown just below.
You press a hotkey, speak naturally, and Vito:
- Transcribes your speech with a best-in-class speech-to-text engine, then
- Cleans it up with AI — stripping the "uhm"s and false starts, fixing punctuation, building proper sentences, and
- Pastes the finished text straight at your cursor — in your editor, your browser, Slack, an email, anywhere.
No window to switch to, no copy-paste dance. Talk, and the text appears where you're already working.
Dictation apps like Wispr Flow, Superwhisper, Aqua Voice and Willow are excellent — but most of them charge a monthly subscription, typically €12–€30 a month, whether you use them a little or a lot. (Side-by-side comparisons on vito.talk.)
Vito flips that model:
| Subscription apps | Vito | |
|---|---|---|
| Cost | Fixed €12–€30 / month | Only the API you actually use |
| Idle months | You still pay | You pay nothing |
| Your key | Locked to their backend | Bring Your Own Key — your accounts |
| Source | Closed | Open source (AGPL-3.0) |
| Tracking | Varies | None. Zero telemetry. |
| Languages | Varies | 60, UI translated into all 60 |
The app itself is free. You only pay the speech-to-text and AI providers directly, for exactly what you dictate. Stop using it for a month and that month costs you €0 — no recurring charge, ever.
Vito uses a Bring Your Own Key (BYOK) model.
New to "Bring Your Own Key"? Here's what it means.
Instead of paying Vito, you create your own free account with the AI providers Vito uses, generate a personal API key (a secret password that lets software use the service on your behalf), and paste it into Vito once. From then on, Vito talks to those services directly, using your account, and you're billed by them — pay-as-you-go, usually fractions of a cent per use.
Nothing runs through a middle-man server. It's your key, your account, your data, your bill.
Speech-to-text is billed per hour of audio. In practice that's about 15 cents (€0.15) for a solid hour of continuous talking — and almost nobody dictates non-stop.
| Your usage | Roughly costs |
|---|---|
| A busy hour of solid dictation | ~€0.15 |
| Normal daily use (email, notes, chat) | a few euros / month |
| 100 hours of non-stop speech in one month | ~€15 |
To spend €15 in a month — a typical monthly subscription price elsewhere — you'd have to speak one hundred hours without pausing. Even heavy, everyday users usually land at just a few euros a month.
Start free — no credit card. Pair AssemblyAI (speech-to-text, free starting credit) with Groq (AI cleanup, free tier) and you can set Vito up and start dictating without paying anything or entering a card.
Raw speech-to-text gives you a literal transcript. Vito adds an AI cleanup pass that turns it into text you'd actually send:
- Removes fillers — "uh", "uhm", "you know", "I mean…", repeated words and false starts.
- Fixes punctuation and capitalization, and builds well-structured sentences.
- Understands spoken commands — say "new line" / "new paragraph" and it becomes real formatting.
- Turns named emoji into emoji — "thumbs up emoji" → 👍 — but only when you're clearly naming one.
- Applies your dictionary — names and jargon come out spelled the way you want.
- Runs on the model you choose — Claude, any OpenAI-compatible endpoint like Groq (with a free tier), or a fully local model (Ollama, LM Studio) so the text never leaves your machine.
If cleanup ever errors or times out, Vito pastes the raw transcript instead — a dictation is never lost. Both the raw and cleaned versions are kept in your history.
Beyond cleaning up, Vito can take a spoken command. Start a dictation with "Vito, …" and a short instruction — it isn't typed out, it's carried out on your next dictation, or on the text already on your clipboard:
- Reshape — "Vito, translate to English", "Vito, summarise this", "Vito, turn this into bullet points".
- Answer — "Vito, question" then ask it, or "Vito, reply to this".
- Clipboard — "Vito, summarise the text on the clipboard" — the result goes back to your clipboard, ready to paste.
It runs through the AI-cleanup pass, so it needs cleanup on — and you can point it at its own, heavier model when you want. Usage gets its own statistics and achievements.
Vito dictates in 60 languages, and its entire interface is translated into all 60 as well.
See all 60 languages
- Western Europe (11): English, Dutch, German, French, Spanish, Portuguese, Italian, Catalan, Galician, Basque, Welsh
- Northern Europe / Baltic (7): Danish, Swedish, Norwegian, Finnish, Estonian, Latvian, Lithuanian
- Slavic (12): Russian, Ukrainian, Belarusian, Polish, Czech, Slovak, Slovenian, Croatian, Serbian, Bosnian, Bulgarian, Macedonian
- Rest of Europe / Turkic (7): Greek, Romanian, Hungarian, Albanian, Turkish, Azerbaijani, Kazakh
- Middle East (4): Arabic, Hebrew, Persian, Urdu
- South Asia (9): Hindi, Bengali, Punjabi, Marathi, Gujarati, Tamil, Telugu, Kannada, Malayalam
- East & Southeast Asia (8): Chinese (Mandarin), Japanese, Korean, Vietnamese, Thai, Indonesian, Malay, Tagalog (Filipino)
- Africa (2): Swahili, Afrikaans
Together these cover roughly 75–80% of the world as a native language — nearly every language with more than 50 million speakers — and, counting people who use one of them as a second or lingua-franca language (English, French, Arabic, Russian, Hindi, Spanish, Indonesian, Swahili…), around 90%+ of humanity can use Vito in a language they understand.
- No tracking. No telemetry. No analytics. Vito never phones home.
- Your history, statistics and settings stay on your own device.
- The only thing that leaves your computer is the audio you dictate, sent directly to the speech-to-text provider, and — if cleanup is on — the transcript, sent directly to the AI provider. Nothing more, and only when you're actively dictating. Run cleanup on a local model and even the transcript never leaves your machine.
- Because it's Bring Your Own Key, that data flows through your own accounts, not ours. There is no Vito server in the middle.
| Purpose | Providers |
|---|---|
| Speech-to-text | Soniox · AssemblyAI |
| AI cleanup | Anthropic (Claude), or any OpenAI-compatible endpoint — Groq (free tier), OpenAI, or a local model (Ollama, LM Studio) that keeps cleanup fully on-device |
You only need a key for the speech-to-text provider to get started; the AI-cleanup key is optional but recommended.
Step-by-step setup guides, one per provider: Soniox · AssemblyAI · Anthropic · Groq (a free-tier option for AI cleanup) — or browse all setup guides on vito.talk.
Vito is simple to use, but powerful under the hood.
- 🎙️ Press-to-talk or push-to-talk — quick tap to toggle, or hold to talk and release to stop.
- ✨ AI cleanup of every dictation (fillers, punctuation, new lines, spoken emoji).
- 🪄 Vito Assist — say "Vito, …" with a command to translate, summarise, answer or reshape your dictation (or your clipboard), instead of typing it out.
- ⏸️ Auto-stop on pause — Vito can stop by itself when you stop speaking.
- 📋 Paste, type or clipboard-only injection modes for maximum app compatibility.
- 📁 Upload an audio file to have it transcribed to text.
- 🌗 Light & dark mode, switching automatically with your OS preference.
- 🌍 UI translated into 60 languages.
- 🪟 Mini always-on-top overlay you can place anywhere on screen.
- 🧪 Built-in demo mode with sample content, so a fresh install has something to show.
- 🔎 Searchable history of everything you've dictated.
- 📖 Dictionary for names, jargon and common mishearings — corrected automatically.
- 🔊 Play back and save your recordings.
- 🏆 Achievements & gamification to track milestones.
- 📊 Detailed statistics on usage and cost.
- 🚀 Start with your computer, so Vito is always ready.
- 🔁 Backup & restore with automatic rolling backups.
- ⬆️ Built-in update checker — tells you when a new version is out, and on Windows installs it for you.
- 🎧 Configurable media behavior while dictating — do nothing, duck the audio, or auto pause/play.
- 🐧 Linux-friendly — detects missing utilities/libraries and explains how to bind hotkeys in each desktop environment.
Vito is cross-platform: Windows, Linux and macOS.
-
Windows — grab
Vito-Setup-<version>.exefrom the Releases page. Per-user install, no admin needed. -
Linux — two builds, and which one suits you depends on your desktop:
Flatpak AppImage Best for GNOME, KDE everything else — wlroots compositors (niri, Sway, Hyprland), X11, minimal setups Typing into other apps needs the RemoteDesktop portal, which GNOME and KDE provide always works — wtypeon Wayland,ydotoolon X11Updates flatpak update, like any other appdownload the new file yourself Setup none none on Wayland; ydotool+ a udev rule only on X11The AppImage always works. The Flatpak is the nicer way to live with Vito — it installs and updates itself — but the sandbox has one hard requirement: it can only type where the RemoteDesktop portal exists. On a compositor without one, a sandboxed Vito can still hear you and clean up your text, but it can only put the result on the clipboard for you to paste. It says so on the settings page rather than leaving you guessing.
Flatpak (GNOME, KDE):
flatpak remote-add --user --if-not-exists vito https://vinceecniv.github.io/Vito/vito.flatpakrepo flatpak install --user vito io.github.vinceecniv.vito
AppImage (anything): download
Vito-<version>.AppImagefrom Releases, make it executable, and run it.Either way there is nothing to configure first: where the portals are available Vito asks the desktop for its hotkey and for permission to type, and you confirm once. See
docs/flatpak.mdanddocs/linux-portals.md. -
macOS — download
Vito-<version>.dmgfrom Releases and drag Vito to Applications. Universal (Apple Silicon and Intel), macOS 11 or newer. Vito lives in the menu bar, not the Dock.The build is not signed with an Apple Developer certificate. Double-clicking a freshly downloaded copy shows "Vito cannot be opened": use right-click → Open, and if macOS still refuses (it does on recent versions), go to System Settings → Privacy & Security and press Open Anyway under the message about Vito. You only do this once.
Finder also shows a small ⌛ next to Vito's name. That is macOS reporting that the app has not been through notarisation — not that something is still copying, and not something approving the app clears. It is cosmetic: the ⌛ stays until the app is notarised, while Vito itself runs perfectly well. Then grant two permissions in System Settings → Privacy & Security:
Permission Needed for Without it Microphone recording your voice Vito records silence Accessibility the global hotkey, and pasting into other apps no hotkey; injection falls back to an error, use clipboard only mode and paste yourself You grant these once; updating Vito keeps them. (If you ever build Vito yourself without the signing certificate, macOS ties the grant to that exact binary instead, and every rebuild resets it — with the confusing symptom that the switch still shows as on while Vito reports the permission as missing. Select Vito in the list, remove it with −, and let Vito ask again.)
After installing, open Vito, paste in a speech-to-text API key, pick your hotkey, and start talking.
Vito is one Go binary with an embedded web UI — no GUI toolkit, no build step for the front end. It uses cgo (miniaudio), so you build natively per platform with a C compiler available.
go build -o vito ./cmd/vitoWindows build details
cgo needs a C compiler (mingw-w64):
winget install --id BrechtSanders.WinLibs.POSIX.UCRT # mingw-w64, no admin
# open a fresh shell so gcc is on PATH, then from the repo root:
pwsh -File packaging/build-windows.ps1 # -> vito.exeTo build the installer (calendar-versioned year.month):
pwsh -File packaging/build-installer.ps1 # next free number this month
pwsh -File packaging/build-installer.ps1 -Version 2026.7.2 -TagLinux dependencies & setup
Text delivery picks the first route that works, so what you need depends on your desktop. On GNOME and KDE
the RemoteDesktop portal handles it and there is nothing to install. Elsewhere Vito uses wtype, and
ydotool remains selectable for X11 and older setups — that is the only route that still wants a daemon
and a udev rule for /dev/uinput. Clipboard-only mode always works.
Runtime helpers (miniaudio dlopens ALSA/PulseAudio, so there's no link-time audio dependency):
- PipeWire / PulseAudio — audio capture/playback.
- wl-clipboard (
wl-copy/wl-paste) — clipboard. - wtype — the paste keystroke where the portal is unavailable (niri, wlroots).
- ydotool + ydotoold (optional) — the older injection route, for X11 and as a last resort.
- libnotify (
notify-send) — state notifications. - pactl (optional) — duck media while dictating; pausing goes over MPRIS and needs nothing installed.
sudo dnf install wl-clipboard wtype libnotify pulseaudio-utils # FedoraBuild the AppImage:
bash packaging/build-appimage.sh 2026.7 # from Linux
pwsh -File packaging/build-appimage.ps1 -Version 2026.7 # from Windows, via DockerBuild the Flatpak (manifest in packaging/flatpak/, published to a self-hosted repository by
.github/workflows/flatpak.yml on every release tag):
flatpak-builder --user --install build packaging/flatpak/io.github.vinceecniv.vito.ymlThe settings page checks for the helpers above, reports which delivery route is actually in use, and gives
per-desktop instructions for binding the hotkey. Background on the portals and the sandbox:
docs/linux-portals.md and docs/flatpak.md.
macOS build details
Nothing to install beyond the Xcode Command Line Tools — audio goes through CoreAudio and text injection through CoreGraphics, both part of macOS, so there are no external helpers like Linux needs:
xcode-select --install
bash packaging/make-signing-cert.sh # once per machine, see below
bash packaging/build-macos.sh 2026.8 # -> dist/Vito-2026.8.dmgThat produces a universal Vito.app (arm64 + x86_64, merged with lipo) inside a DMG, with a .sha256
beside it. Must run on macOS: cgo needs the real SDK, and the icon, signature and disk image are made with
Apple's own tools (sips, iconutil, codesign, hdiutil).
Two things the bundle carries that the binary alone cannot:
NSMicrophoneUsageDescriptioninInfo.plist. Without it macOS denies the microphone outright rather than asking, and dictation silently records nothing.LSUIElement, which keeps Vito out of the Dock and the app switcher — it lives in the menu bar, like the tray icon elsewhere.
make-signing-cert.sh creates a self-signed code-signing certificate in your login keychain, which the
build uses when it is there and falls back to ad-hoc when it isn't. It is worth the one command: macOS
stores a permission grant as a code signing requirement, and signed ad-hoc that requirement is the exact
binary — so every rebuild quietly invalidates the user's Accessibility approval. Signed with a certificate
that stays put, the requirement is identifier … and certificate root … and the grant survives updates.
Sign every release with the same certificate, and never commit it. The script also asks once for your
login keychain password, to add codesign to the key's partition list — without that macOS interrupts every
build with a password prompt, twice over, once per architecture of the universal binary.
It is still not a Developer ID and still not notarised, so the right-click-to-open step and the ⌛ remain.
Only an Apple Developer ID plus notarytool removes those.
Configuration & hotkeys
First run creates ~/.config/vito/config.json (chmod 600) with defaults and an auth token. Configure
everything from the web UI at http://127.0.0.1:4573 while the daemon runs — changes apply to the next
dictation without a restart.
On Windows and macOS you set the hotkey in the app itself, under Settings → Activation. On macOS it needs the Accessibility permission; until you grant it the settings page shows the hotkey as denied and says so.
On Wayland the binding belongs to the desktop, not to Vito — an application cannot grab a key for itself, so it has to ask. Vito asks through the GlobalShortcuts portal: you confirm the shortcut once, it then appears in your desktop's own keyboard settings, and Vito can send you straight there to change it. Push-to-talk works too, because the portal reports the release as well as the press. This needs GNOME or KDE; other compositors do not implement the portal yet.
Where the portal is unavailable, bind a signal to a key in your compositor instead. The signals are not a lesser route — they cost nothing, launch no process, need no binary path (identical for a Flatpak, an AppImage or a plain binary) and never wait for a squashfs mount, so they stay supported permanently:
- GNOME / KDE: Custom Shortcuts → command
pkill -USR2 -f 'vito serve'(start/stop), andpkill -USR1 -f 'vito serve'(cancel). - niri (
config.kdl):binds { Mod+D { spawn "pkill" "-USR2" "-f" "vito serve"; } Mod+Shift+D { spawn "pkill" "-USR1" "-f" "vito serve"; } }
SIGUSR2 toggles a dictation, SIGUSR1 cancels it (cleanup is the settings toggle, not a hotkey). On
Windows and macOS the daemon registers a global hotkey itself. Which route is actually in use, and the
exact commands per desktop, are shown in the app under Settings → Activation.
Vito's full source is here for anyone to read, audit, and adapt.
It's released under the GNU Affero General Public License v3.0. Use it, change it, build on it, and yes — charge for it if you like. The AGPL asks one thing in return: if you distribute a changed version, or run one as a service for other people, those people get the source too, under the same licence. Improvements stay where everyone can reach them.
Not open contribution. Bug reports, ideas and translation corrections are very welcome; code contributions are not accepted, so the copyright stays in one pair of hands and the licence can be changed later. Forking is welcome and the licence guarantees it — see CONTRIBUTING.md.
The licence covers the code, not the name Vito or the waveform logo — a fork is welcome under a name and icon of its own.
Vito is free and always will be. If it saves you a subscription — or just some typing — you can say thanks:
Vito — Voice In, Text Out · Speak anywhere, get clean text · Made with ❤️ and voice.