FluidVoice
Turn speech into polished text instantly with on-device AI. Free, open-source, and 100% private with no data leaving your Mac.
FluidVoice is an open-source alternative to:
FluidVoice is a free, open-source voice-to-text tool for macOS that transforms natural speech into ready-to-use text with on-device AI. Unlike cloud-based dictation apps, FluidVoice processes everything locally, ensuring your voice never leaves your device.
Key Features
- On-device processing: Local speech models convert audio to text without sending data to servers
- Fluid-1 AI: Custom-trained model that refines raw transcripts into polished, context-aware text
- System-wide input: Works in any app with a single hotkey
- 40+ languages: Supports multilingual dictation with models like Nemotron, Whisper, and Apple Speech
- Adaptive tone: Custom prompts for different apps (Slack casual, Mail formal, GitHub structured)
- Real-time speed: ~3.7x faster than typing with <100ms latency
- Offline functionality: Works without Wi-Fi or internet connection
- Apple Silicon optimized: CoreML and Metal for low-latency, battery-efficient performance
- Multiple modes: Write Mode, Command Mode, and Direct Dictation
- AI post-processing: Choose between local Fluid Intelligence, OpenAI, Groq, or custom providers
Tech / Self-Hosting
FluidVoice is a native macOS application that runs entirely on-device. The open-source project includes:
- Local speech models (Nemotron Speech 3.5, Parakeet Flash, Apple Speech)
- CoreML and Metal acceleration for Apple Silicon
- Custom Fluid-1 model for post-processing
- Brew installation for easy setup (
brew install --cask fluidvoice)
No self-hosting is required as the application is designed to run locally on your Mac. The source code is available on GitHub under GPLv3 license.