OpenFlow

A macOS-native, GPU-accelerated, streaming speech-to-text menu bar app.

Features • Installation • Usage • Configuration • Tech Stack

---

OpenFlow is an open-source, local-first alternative to Wispr Flow. It provides instant voice dictation into any application using Apple Silicon's Neural Engine (via MLX) for ultra-low latency transcription. All processing happens on-device—no audio ever leaves your Mac.

Features

Feature	Description
Ultra-Low Latency	Optimized for Apple Silicon using MLX (Metal) and faster-whisper. Sub-300ms transcription with the default tiny model.
Private & Local	All processing happens on-device. No audio ever leaves your Mac.
Voice-Reactive UI	Floating overlay that responds to your voice in real-time.
Universal Dictation	Inserts text directly into any active application via Accessibility APIs.
Smart Denoising	Integrated RNNoise for crystal-clear audio capture even in noisy environments.
Easy Installation	Simple setup with make install and make run.

Requirements

• macOS (Apple Silicon recommended)
• Python 3.11 or later
• PortAudio (brew install portaudio)

Installation

git clone https://github.com/sammy4321/OpenFlow.git
cd OpenFlow
make install

The installer creates a virtual environment, compiles RNNoise, downloads Whisper models, and installs dependencies.

Usage

make run

On first run, grant Accessibility permissions when prompted so OpenFlow can insert text into applications. The app runs in your menu bar—click to start dictating.

Configuration

Model Selection

By default, OpenFlow uses the tiny model for maximum speed. You can switch to larger models for better accuracy:

make run MODEL=base
make run MODEL=small
make run MODEL=medium

Supported models: tiny, base, small, medium, large-v3 (availability depends on your machine's memory).

Project Structure

OpenFlow/
├── openflow/
│   ├── main.py          # Entry point, accessibility checks
│   ├── app.py           # Menu bar app, rumps integration
│   ├── whisper_engine.py # MLX / faster-whisper transcription
│   ├── audio.py         # PortAudio capture pipeline
│   ├── streamer.py      # Streaming audio chunks to engine
│   ├── overlay.py       # Voice-reactive floating UI
│   ├── statusbar.py     # Menu bar status icon
│   ├── hotkey.py        # Global shortcut handling
│   ├── accessibility.py # Text insertion via AX APIs
│   ├── config.py        # User preferences, model paths
│   ├── rnnoise.py       # RNNoise denoising wrapper
│   └── updater.py       # Update checks
├── scripts/
│   ├── install.sh       # Main installer
│   ├── bootstrap_python.sh
│   ├── install_rnnoise.sh
│   └── install_models.sh
├── resources/
│   └── icon.png
├── Makefile
├── setup.py
├── requirements.txt
└── README.md

Tech Stack

• MLX — Apple Silicon–optimized ML framework
• faster-whisper — Whisper transcription (CPU fallback)
• mlx-whisper — MLX-native Whisper for lowest latency
• RNNoise — Real-time noise suppression
• rumps — Minimal macOS menu bar app framework
• sounddevice — PortAudio bindings for audio capture

Contributing

Contributions are welcome. Please:

1. Fork the repository
2. Create a feature branch (git checkout -b feature/AmazingFeature)
3. Commit your changes (git commit -m 'Add some AmazingFeature')
4. Push to the branch (git push origin feature/AmazingFeature)
5. Open a Pull Request

License

Licensed under the Apache License 2.0.

Topics

macos speech-to-text dictation whisper mlx apple-silicon menu-bar local-first privacy