Instant offline & cloud voice-to-text dictation directly under your cursor.
🌐 Official Website & Live Demo
Press and hold the hotkey, speak, and release — Aura instantly transcribes your speech and types the text right under your cursor in any Windows application. Runs 100% offline via local models (Whisper, NVIDIA Parakeet with CUDA GPU acceleration) or connects to cloud providers (Gemini, Groq, OpenAI, Hugging Face, Custom API).
100% Free & Open Source (AGPL-3.0) — no subscriptions, paywalls, telemetry, or advertisements.
- Global hotkey dictation — hold to talk (
Alt+V), or short-tap to latch recording (toggle mode);Esccancels. - Two recognition modes:
- Local (100% offline & private) — whisper.cpp or NVIDIA Parakeet TDT v3 (sherpa-onnx) on CPU or NVIDIA CUDA GPU. Audio never leaves your computer; download models in one click directly from settings.
- Cloud — Google Gemini, Groq, OpenAI, Hugging Face, or your custom OpenAI-compatible server.
- NVIDIA CUDA GPU acceleration — hardware GPU acceleration for Whisper and Parakeet models with 1-click on-demand runtime downloading and automatic CPU fallback.
- Gaming Mode — automatic detection of 3D games (DirectX / Vulkan) with keyboard hook suspension and model memory unloading for maximum FPS, plus manual tray toggle.
- Idle VRAM Standby (Sleep) — configurable standby timer (15, 30, 60 minutes) to free GPU memory without closing Aura.
- Real-time streaming input — smooth word-level streaming into active fields without flickering or text duplicates (NVIDIA Parakeet).
- Text Replacements & Dynamic Macros — local rules for instant text expansion into snippets, email addresses, or live date/time tokens (
{date},{date_ru},{time},{datetime}). - Focus Guard & Context Editing — ensures transcribed text is never typed into the wrong window if focus shifts (with safe clipboard handoff), and provides 1-click text editing on selected text with a visual AI sparkle indicator.
- Audio device selection, WASAPI Hotplug & Tail Hold — choose physical microphone input devices with dynamic device hotplugging, 450ms tail grace buffer, and non-destructive Silero VAD silence trimming.
- Transcription history with day grouping — up to 300 encrypted dictations (DPAPI) with day grouping (Today / Yesterday / calendar dates), live search, and source filters.
- Custom dictionary — bias recognition towards your names, brands, and technical terms.
- 12 language options — auto-detect, keyboard-layout detection, or fixed selection (ru, en, de, es, fr, it, zh, pt, tr, nl).
- Polished overlay with 6 sound themes — microphone VU meter, recording timer, 6 audio themes (Zen, Rhodes, Sci-Fi, Classic Bell, Bubble, Haptic), and display customization.
- Native High-DPI Tray Menu — modern tray menu with per-monitor DPI scaling, instant gaming mode toggle, and settings access.
- System integration & security — autostart with Windows, Windows DPAPI credential encryption, and one-click diagnostic reports.
Aura combines the speed and privacy of local neural networks with the flexibility of cloud AI providers.
| Feature | Aura | Handy | Wispr Flow |
|---|---|---|---|
| Price | Free (AGPL-3.0) | Free (Open Source) | Paid (subscription) |
| Platforms | Windows (macOS in progress) | Windows / macOS / Linux | Windows / macOS |
| Local Models (Offline) | ✅ Whisper & Parakeet (CUDA GPU / CPU) | ✅ Whisper / Parakeet | ❌ Cloud only |
| Cloud Providers | ✅ Gemini, Groq, OpenAI, HF, Custom | ❌ | ✅ Proprietary cloud |
| Real-time Streaming Input | ✅ | ❌ | ✅ |
| Gaming Mode (D3D/Vulkan) | ✅ Auto + Manual | ❌ | ❌ |
| Text Replacements & Macros | ✅ | ❌ | ➖ |
| Focus Guard Protection | ✅ | ❌ | ➖ |
| Data Encryption (DPAPI) | ✅ | ❌ | ➖ |
| Custom Term Dictionary | ✅ | ❌ | ✅ |
| Transcription History | ✅ Up to 300 items (DPAPI) | ❌ | ✅ |
Official builds target Windows 10 & 11 (macOS port compiles in CI). Handy is a mature cross-platform choice for offline dictation. Wispr Flow offers a polished UI, but requires a paid subscription and proprietary cloud.
The fastest way to install Aura on Windows 10/11:
winget install Malashka.AuraDownload the installer from GitHub Releases or via SourceForge Mirror and run it. You can explore the interactive settings mockup and live demo on our Official Website.
For local mode, download a model from the "Speech" tab in Settings (Base model is a great starting point for fast CPU inference, Large v3 Turbo (Q5) for NVIDIA CUDA). For cloud mode, provide an API key for your chosen provider (Groq and Google Gemini offer free tiers).
First launch — "Windows protected your PC"? The installer is not yet signed with an Authenticode certificate, so Windows SmartScreen displays a standard warning for new open-source binaries. Click More info → Run anyway. The entire codebase is open-source for independent audit and self-compilation.
| Action | Default |
|---|---|
| Start recording | hold Alt + V |
| Finish and type text | release the hotkey |
| Latch recording (toggle mode) | short tap Alt + V |
| Cancel recording | Esc |
The hotkey, language, recognition engine, and overlay settings are configurable via Settings (tray icon → "Open Settings").
Prerequisites: Rust (stable), Node.js 18+, WebView2 (preinstalled on Windows 10/11).
git clone https://github.com/malashkadev/aura.git
cd aura
npm install
npm run dev # development mode
npm run build # NSIS/MSI installer in src-tauri/target/release/bundle/Whisper.cpp sidecar binaries live in src-tauri/binaries/. To update them to a newer release, run python install_whisper.py.
Run test suite:
cd src-tauri
cargo test- Local mode runs entirely offline on your machine (CPU / GPU) and never transmits audio over the network.
- Cloud mode sends audio recordings, generated transcripts, selected text (only when selection editing is active), and custom dictionary terms directly to your chosen provider over TLS with zero telemetry.
- Secure local storage — API keys and transcription history are encrypted using Windows DPAPI (
CryptProtectData) with file access restricted to the current user and SYSTEM via ACL. - Update checks run only on demand or when explicitly enabled in settings.
- Gaming Mode — automatic detection of 3D games and graphics APIs (DirectX / Vulkan) with safe suspension of the low-level keyboard hook and model memory unloading to maintain 100% FPS in games, plus manual tray toggle.
- Idle VRAM Standby (Sleep) — automatic model unloading from dedicated GPU memory when idle for 15, 30, or 60 minutes, with instant reload on next activation.
- Local Text Replacements & Dynamic Macros — custom expansion rules to replace spoken triggers with text snippets, email addresses, or live date/time tokens (
{date},{date_ru},{time},{datetime}). - Native High-DPI Tray Menu — custom WebView2 tray window with per-monitor DPI awareness and multi-monitor cursor placement, featuring instant gaming mode toggle.
- Dynamic Microphone Hotplug (WASAPI) — real-time detection and graceful reconnection when headsets or USB microphones are plugged or unplugged.
- 300-Item Encrypted History — day grouping (Today / Yesterday), live search, source filtering, and DPAPI encryption.
- 450ms Grace Buffer (Tail Hold) — expanded post-release tail hold buffer and non-destructive Silero VAD to eliminate clipped word and sentence endings.
- New Sound Themes — added "Bubble" (water droplet) and "Haptic" (mechanical tactile switch) sound themes alongside Zen, Rhodes, Sci-Fi, and Classic Bell.
- Windows Package Manager Support — install and update via one command:
winget install Malashka.Aura.
- macOS support — native port (global hotkeys via
CGEventTap, CoreAudio capture) is in the codebase and compiles in CI. Remaining items: macOS whisper binary,.appbundle, Accessibility-permission flow, and hardware testing.