Hold-to-talk speech transcription at your cursor. Transcribe locally on your device with GPU acceleration or through cloud transcription providers. For Windows and Linux.
PortusEchoes lives in your system tray and works across every application on your desktop:
- Hover and speak (Windows): Put your mouse pointer over any text box, editor, or chat app without clicking. Hold your shortcut key (
Ctrl+Alt+Spaceby default) and speak naturally. - Direct typing: On Windows, text is typed directly without touching your clipboard:
- OpenAI Live: BYOK Realtime streaming transcription. Words stream out in real time while you speak, typing into each window as your cursor moves over it.
- Groq: BYOK cloud transcription. Audio is processed after release and the full transcript is typed into the window under your mouse.
- Local Whisper: Fully offline using Whisper. Transcribed locally and typed into the window under your mouse as each segment completes.
- Clipboard delivery on Linux: Linux currently uses clipboard delivery across all providers and models. Your completed transcript is automatically copied to your clipboard. The overlay indicator will let you know when the text is ready to paste (
Ctrl+V) into any target window. - Push-to-Talk Indicator: A small, borderless, click-through overlay pill at the bottom of your screen shows real-time recording and transcription status.
- Sustained Dictation: Dictate uninterrupted for a quick phrase or for up to 30 minutes across all modes and models without artificial time limits or mid-sentence drops. Developers compiling from source can expand this limit even further.
- Full control: Rebind your shortcut to any key combination, switch between local offline models and cloud providers, or control recordings from the system tray menu.
- Hardware Acceleration: Automatic GPU acceleration with multi-threaded CPU fallback.
- Security & Privacy: API keys persist exclusively in OS-native secure storage (Windows Credential Manager / Linux Secret Service). No audio is written to disk.
- Cross-Platform: Windows and Linux support.
Pre-built releases for Windows and Linux are available under GitHub Releases.
- Download
PortusEchoes_<version>_x64-setup.exefrom the latest release. - Run the installer and follow the setup wizard.
- Once installed, PortusEchoes will launch and sit in your system tray.
Choose the package format best suited for your distribution:
- AppImage (Universal standalone):
- Download
portus-echoes_<version>_amd64.AppImage. - Make it executable:
chmod +x portus-echoes_*.AppImage - Run it directly:
./portus-echoes_*.AppImage
- Download
- Debian / Ubuntu Package (
.deb):- Download
portus-echoes_<version>_amd64.deb. - Install it with
dpkg:sudo dpkg -i portus-echoes_*_amd64.deb - Launch PortusEchoes from your application menu or terminal.
- Download
Access the Settings window from the system tray icon or during initial onboarding to configure providers, models, and shortcuts.
- In the Settings tab under Active Provider, select between Local, OpenAI, or Groq. (You can also toggle active providers directly from the system tray menu).
- Under Shortcut, click the input box and press any key combination to rebind your push-to-talk shortcut.
- Standard Models: In the Local tab, click the download icon next to a listed model. The file will download directly into your local application data directory.
- Custom Models: You can use any Whisper GGML model by downloading the
.binfile from the official Hugging Face repository: https://huggingface.co/ggerganov/whisper.cpp/tree/main Select Custom Path in the Local tab and browse to your downloaded.binfile.
- In the OpenAI tab, enter your API key (stored securely in your OS credential store).
- Select a standard model slot (
GPT Live TranscribeorGPT Transcribe), or select Custom Model ID and enter any compatible model (e.g.gpt-4o-transcribe,gpt-4o-mini-transcribe).
- In the Groq tab, enter your Groq API key.
- Select a standard model (
Whisper Large V3 Turbo) or select Custom Model ID and enter a custom Groq model ID.
Run full end-to-end diagnostics at any time from the tray menu or Settings window to verify microphone input streams, global keyboard shortcuts, and provider reachability:
Every provider in PortusEchoes is completely independent. You only need to configure what you want to use:
- Offline Local Only: Transcribe privately on your device using GPU or CPU. Never requires an API key or an internet connection.
- Groq Only: For fast transcriptions without having to manage local models. Just add a Groq API key and transcribe!
- OpenAI Only: Connect your OpenAI API key for real-time live streaming or high accuracy cloud models.
| Capability | Windows | Linux |
|---|---|---|
| Audio Capture | WASAPI | ALSA / PipeWire |
| Local Inference | GPU / CPU | GPU / CPU |
| Voice Activity Detection | Integrated Silero VAD | Integrated Silero VAD |
| Text Delivery | Direct Caret Injection | Clipboard Delivery |
For complete platform details and upcoming Linux caret injection work, see docs/PLATFORM_MATRIX.md.
- Node.js: 18+ and
npm - Rust: 1.88+ and
cargo - CMake: 3.16+
- C/C++ Compiler: MSVC (Windows) or GCC/Clang (Linux)
# Arch Linux
sudo pacman -S alsa-lib pipewire vulkan-loader libx11 libxtst libxkbcommon
# Ubuntu / Debian
sudo apt-get install libasound2-dev libvulkan-dev libx11-dev libxtst-dev libxkbcommon-dev
# Fedora
sudo dnf install alsa-lib-devel vulkan-loader-devel libX11-devel libXtst-devel libxkbcommon-devel-
Clone the repository and configure the environment:
cp .env.example .env
-
Install dependencies:
npm install
-
Launch development mode:
npm run tauri dev
-
Build for production:
npm run tauri build
- Architecture Overview
- Platform Matrix
- Local Inference
- Cloud Providers
- Audio Pipeline
- Shortcuts & Overlay
- Build & Packaging
- Third-Party Licenses & Attributions
PortusEchoes is licensed under the Apache License, Version 2.0.