Hold
Use a single key or key combination in a browser, IDE, chat, document, or game.
Local-first dictation · 100 languages
Speak naturally.
Type anywhere.
Hold one key, say what you mean, and release. VoicePrompt recognizes any of Whisper's 100 languages, formats the result for the active app, and puts it where you are already working.
Free preview. No account, subscription, or payment.
English and Slovenian share a tuned automatic mode. Every other language is one searchable selection away. Latest GitLab release.
Actual hotkey overlay
Local-firstPrivate audio pipeline
Model memory controlRelease after 15 min or keep loaded
100 languagesEN + SL Auto by default
Release gatedVerified before publishing
Designed for focus
Graphite, Evergreen, or Ember. Always dark, never loud.
The shortest path from thought to text
Three moments. Zero ceremony.
No recording window, transcript editor, or copy-and-paste routine. VoicePrompt works around your current app.
Speak
A compact wave, equalizer, or microphone orb confirms the app is hearing you. Speak in any supported language.
Release
Your complete text lands once by paste, or stays on the clipboard when an app blocks input.
Built for daily use
Fast enough to disappear.
Thoughtful enough to trust.
100-language recognition
Your language is already in the model.
Search and pin any language supported by Whisper large-v3. English and Slovenian get evidence-preserving automatic routing by default, including a dedicated Slovenian slang profile.
Your microphone is not a cloud service.
Speech recognition runs on your PC using the local processor you select. Audio is never stored. Optional text cleanup and self-hosted server modes are explicit choices.
Keep talking. The wait stays short.
Complete speech blocks are prepared while you speak. The full recording remains available for automatic fallback, then VoicePrompt pastes once.
Useful controls, without hidden behavior.
01Smart local formattingGet sentence boundaries, punctuation, and paragraphs without enabling an AI provider.
02Bounded app contextMatch casing, spacing, and writing style while excluding password fields.
03Corrections that learnTeach a recurring mishearing once from local Recovery.
04Selected-text commandsSay “Command …” or “Ukaz …” for an explicit AI transformation.
05Four writing modesVerbatim, Clean, Polish, or structured Prompt.
06Recovery comparisonCopy the delivered result or untouched original.
07Model memory controlRelease weights after 15 idle minutes or keep the same model ready.
The application users actually download
Current interface.
Current capabilities.
These views come from the current Windows release. Six focused workspaces guide setup, dictation, audio, Engine & AI, recovery, and system tools.
Readiness at a glance
Confirm the runtime, hotkey, microphone, language, and recognition engine from one screen.
- 100 language profilesOne multilingual model
- English + Slovenian AutoTuned bilingual default
- Single keys or combosSelective global capture
- Hold or toggleTwo activation styles
- Paste or Copy onlyReliable output choices
- Buffered long promptsOne ordered result
- Model memory policyBalanced or keep loaded
- Safe recording fallbackComplete audio retained if streaming setup fails
- Exact voice commandsRouted before prose formatting
- Live input testNo second audio capture
- Smart formatting and contextLocal, bounded, configurable
- Correction learningImmediate deterministic fixes
- Four writing modesVerbatim through Prompt
- Result and original recoveryBounded local history
- Per-application profilesExact executable rules
- Portable backupsPrivate fields excluded
- Compatible server modeExplicit self-hosted option
- Verified in-app updatesStable or Preview
A privacy model you can explain
Local audio.
Visible choices.
The default path is microphone to local engine to focused app. There is no account, analytics SDK, hidden upload, or saved recording.
- Optional AI cleanup sends completed text only; selected text is sent only after an explicit Command or Ukaz action.
- Compatible server mode sends audio only to the server you configure.
- Manual update checks and confirmed downloads contact only the official GitLab release endpoints.
Microphone16 kHz mono
on device
VoicePromptLocal engineone paste
Your appFocused window
Start in a few minutes
One download.
One install command.
VoicePrompt installs its private Python environment, pinned speech runtime, tray interface, and Windows integration fixes. The multilingual model downloads once on first start.
Windows 11 x64Python 3.11+NVIDIA GPU recommended · CPU supported
powershell -ExecutionPolicy Bypass -File .\install.ps1- 1
Download and extract the latest release.
- 2
Open PowerShell in the extracted folder.
- 3
Run the command, then update future versions from inside VoicePrompt.
Less typing. More thinking.
Your next sentence is faster out loud.
Download VoicePrompt, choose from 100 languages, and keep your hands on the work that matters.