Dashboard
Try Vito right here, no installing and no account.
Good to know: in the browser Vito uses Whistle, a small speech model — fine for trying, not the best recognition. Better: Soniox with your own key, or the Vito app.
Vito listens to your voice and pastes the finished text right where your cursor is. Here's how to get started:
- Choose how your speech becomes text, under Settings → Speech recognition:
- Cloud (Soniox or AssemblyAI): the most accurate, live transcript while you speak, all languages — needs a free API key; from $0.10 per hour of audio after the starting credit.
- Vito local: free, works offline, nothing leaves your computer — one download of about 1 GB; 25 European languages, the text arrives after you stop. Best for privacy.
- Want automatic AI cleanup? Add a key for Anthropic (Claude) or any OpenAI-compatible endpoint — Groq has a free tier, or point Vito at a local model.
- Set your start/stop hotkey under Settings → Activation and review the other settings.
- Press your hotkey, speak, and Vito pastes the text where your cursor is.
Tip: each API-key field on the settings page has a 'Get a key' button with step-by-step instructions to request that key.
Words per day
Live transcriptidle
Achievements
Settings
General
Activation
Audio
Speech recognition
No live transcript with this provider: the whole recording is sent when you stop, and the text follows once the endpoint is done. Pin the language above rather than 'Automatic' — Whisper guesses it from the first seconds and gets a short dictation wrong.
Runs on this computer, free, and your audio never leaves it: Cactus Compute's Whistle, a speech model of 17 MB. Text appears while you speak, cut at your pauses.
Languages (7): English, Dutch, German, French, Spanish, Italian, Polish. Keyterms from your dictionary are used.
Runs on this computer, free, and your audio never leaves it. Vito installs and runs the engine itself: parakeet.cpp with NVIDIA's Parakeet TDT 0.6B v3.
Languages (25): Dutch, English, German, French, Spanish, Italian, Portuguese, Polish, Czech, Slovak, Hungarian, Romanian, Bulgarian, Croatian, Slovenian, Greek, Swedish, Danish, Finnish, Estonian, Latvian, Lithuanian, Maltese, Russian and Ukrainian — detected automatically. Not suitable for other languages. Keyterms are not used by this engine; your corrections are.
System requirements: about 1 GB of RAM while the engine runs — it stays loaded as long as this card is selected and is freed when you pick another provider — 0.9 GB of disk space, and a 64-bit processor (Windows or Linux x64/ARM64, Apple Silicon or Intel Mac). Speed: about a second per dictation on a recent multi-core CPU, longer on older or low-power machines; the GPU build (Vulkan) is several times faster.
AI cleanup
Cleanup rules
The instructions Vito gives the model. Vito's own sets are always here and can't be changed — pick the one that fits what you dictate, or start your own from it.
A Vito Assist command (“Vito, …”) doesn't use these rules: carrying out an instruction replaces them rather than extending them. The dictionary corrections and the language are sent separately, so you don't need to repeat them here.
Vito Assist
Say “Vito, …” and a short instruction. It isn't typed out — it's applied to your next dictation, or to the text already on your clipboard.
- Say “Vito, …” with your instruction (e.g. “Vito, translate to English”), then stop as usual — with the hotkey, or by pausing so it auto-stops.
- You hear the Vito Assist tone, and the microphone reopens on its own.
- Now speak your text.
- Stop with the hotkey as usual, or just wait for the automatic submit.
The instruction is applied to that dictation, so what gets pasted is the result — not the command. A clipboard command needs no second step: it runs right away.
Vito Assist is text in, text out. It can't generate images, search the internet, or perform actions (open apps, send mail) — so “Vito, draw a dancing horse” won't work.
Give Vito Assist its own, heavier model — answering and transforming ask more of a model than tidy-up does.
Output
Linux — setup & tools
Activation & hotkeys
Backup & restore
A backup holds all your Vito data — settings, dictionary, history, statistics and achievements — in one file. Audio recordings are not included.
Backups are lightly encrypted with a built-in key, so they aren't plainly readable on disk. Because every Vito shares that key it's a deterrent against casual reading — not full security — so still treat a backup as sensitive.
Vito keeps the 4 most recent automatic backups in its data folder, so you always have a way back if something goes wrong.
Sync between computers
Keep your dictations, statistics and costs, achievements, dictionary and cleanup rules the same on every computer, through a folder your own sync app keeps in sync — Dropbox, OneDrive, Nextcloud, iCloud Drive, Google Drive, Syncthing or a network share. Vito talks to no server for this.
Deleting a dictation or giving it a star carries over too. Settings and API keys stay on each computer: microphones, hotkeys and providers often differ between them.
Personal, cross-platform voice dictation: press a hotkey, speak, and get clean text where your cursor is.
Vito uses a "Bring Your Own Key" (BYOK) model: you use your own API keys (AssemblyAI or Soniox for speech recognition, Anthropic or an OpenAI-compatible endpoint for the AI cleanup). Vito charges you nothing itself and sends your audio or text nowhere except to the providers your keys belong to.
Providing, managing and securing those keys — and any costs their use incurs at the providers — are entirely your own responsibility.
Vito does not collect any user data to send anywhere, and there is no tracking or analytics of any kind built into the app. Your settings, dictionary and history stay local on your own computer.
The only data the app sends externally is your speech and the texts that go to the external BYOK providers (speech recognition and AI cleanup). That data is governed by those external providers’ privacy policies.
The browser version on vito.talk counts anonymously how often it is opened, and whether a browser reaches its 1st and 10th dictation — without cookies, and not at all with Do Not Track on. Nothing you say or set is sent. The installed app counts nothing.
Achievement medals and animations use Noto Emoji by Google (static under Apache-2.0, animations under CC BY 4.0), played with lottie-web by Airbnb (MIT). Country flags are flag-icons by Panayiotis Lipiridis (MIT).
Dictionary
Corrections are always applied, regardless of the keyterms setting. Tip: in History, hover over a word and click it to quickly add a correction for it.
| Misheard | Intended |
|---|
Keyterms are sent to the speech recogniser when you enable that in Settings. This is separate from the corrections above — those are always applied to every dictation.
⚠ You're using Soniox for speech recognition. Keyterms are used by AssemblyAI only — they have no effect with Soniox. The corrections above keep working as usual.