English · Deutsch

RudariFlow

Local speech-to-text dictation app for Windows, powered by whisper.cpp. Global hotkey, push-to-talk or toggle mode, automatic paste of the transcribed text.

v0.4.0 — Windows. In-process whisper-rs backend with persistent model, hotkey-press warmup, and streaming partial transcripts. NVIDIA GPU recommended for speed; CPU fallback included for AMD / Intel / no-GPU systems. macOS and Linux planned for later releases.

See CHANGELOG.md for the full history.

Made by oggi.

Features

• Local transcription via in-process whisper-rs — no cloud required, no subprocess per dictation
• Persistent model: loaded once on first use and reused across dictations
• Hotkey-press warmup: pressing PTT preloads the model in parallel so it's hot by the time you finish speaking
• Streaming partial transcripts: text appears in the overlay as Whisper emits each segment
• Auto backend detection: uses NVIDIA CUDA when available, falls back to CPU otherwise
• Custom Vocabulary: inject domain terms (names, jargon, acronyms) to bias recognition
• No-speech detection: silent recordings show an overlay notice instead of pasting nothing
• Clipboard-safe paste: your previous clipboard contents are saved and restored around auto-paste
• Multiple Whisper models selectable: tiny → large-v3-turbo, auto-downloaded on selection
• Languages: auto-detect or pick from 14 fixed languages (EN, DE, FR, IT, ES, …)
• Push-to-talk and toggle modes
• Configurable global hotkey (capture any chord from the settings UI)
• Floating recording pill with live waveform and cancel button
• Auto-paste via simulated typing (works with any application)
• System tray icon — closing the window minimises to tray instead of quitting
• Optional: start with Windows login
• UI available in English and German (auto-detected from OS locale)

System Requirements

• OS: Windows 10/11 x64
• GPU (recommended): NVIDIA with a CUDA-capable driver for full speed
• CPU fallback: Works without a GPU or on AMD/Intel — significantly slower (~10-30×). For CPU-only users we recommend the small or medium model.
• Note: The first run of each model on a new GPU JIT-compiles CUDA kernels (~30-60s, one-time)
• RAM: the selected whisper model stays resident from first dictation onward. large-v3-turbo ≈ 1.6 GB, small ≈ 500 MB, tiny ≈ 80 MB.

Installation (for end users)

Download the latest RudariFlow_x.y.z_x64-setup.exe from the Releases page and run it.

The installer is unsigned, so Windows SmartScreen will show an "Unknown publisher" warning — click More info → Run anyway to proceed. Code signing may be added in a later release.

Development

Prerequisites

• Rust (MSVC toolchain on Windows)
• Node.js ≥ 20
• Visual Studio Build Tools with the C++ workload (for cargo build)
• CUDA Toolkit 12.x (required to compile whisper-rs with the cuda feature)

Setup

# 1. Clone the repo
git clone https://github.com/oggii/RudariFlow.git
cd RudariFlow

# 2. Frontend dependencies
npm install

# 3. Fetch CUDA runtime DLLs (~80 MB)
powershell -ExecutionPolicy Bypass -File scripts/setup-whisper.ps1

# 4. Run in dev mode
npm run tauri dev

Production build

npm run tauri build

Produces:

• src-tauri/target/release/rudariflow.exe (portable)
• src-tauri/target/release/bundle/nsis/RudariFlow_x.y.z_x64-setup.exe (NSIS installer)
• src-tauri/target/release/bundle/msi/RudariFlow_x.y.z_x64_en-US.msi (MSI installer)

Architecture

• Tauri 2 (Rust backend + Webview frontend)
• Frontend: Vanilla TypeScript + Vite
• Audio capture: cpal (cross-platform low-level audio I/O)
• Transcription: in-process whisper-rs (whisper.cpp Rust bindings) with the cuda feature; runtime fallback to CPU
• Auto-paste: enigo (keyboard simulation)
• Hotkey: tauri-plugin-global-shortcut
• Autostart: tauri-plugin-autostart

Licence / Credits

RudariFlow is released under the MIT License — free to use, modify, redistribute, and incorporate into closed-source projects, with attribution.

Initial Tauri scaffolding based on albertshiney/typr. Uses whisper.cpp (MIT) for transcription.

© 2026 oggi.