FluidVoice is a voice-to-text dictation app for macOS that types what you say into any text field, from email and chat to terminals and code editors such as Claude Code. You hold a global hotkey and speak, a local speech model converts the audio to text, and an optional enhancement step strips filler words, fixes capitalization, and formats the result before it lands in the active app. Speech recognition runs on-device, so audio stays on the Mac unless you opt in to a cloud provider. iOS and Windows versions are in preparation.
Beyond plain dictation, Write Mode writes or rewrites text inline in any app (select text and rewrite it, or dictate new content), and Command Mode controls the Mac by voice: launching apps, running shortcuts, and triggering system actions. Per-app prompt sets adapt the output to whatever app you are in, and a live overlay shows words as you speak and fits around the MacBook notch.
Enhancement runs either through Fluid Intelligence (a separately maintained, fully local model, also called Fluid-1) or through an OpenAI, Groq, or custom provider key stored in the macOS Keychain. Every enhancement layer, audio history, and analytics are optional, and core dictation works with only microphone and accessibility permissions and a hotkey.
| Speech model | Best for | Languages | Hardware |
|---|---|---|---|
| Nemotron Speech 3.5 | Streaming multilingual dictation | About 40 | Apple Silicon |
| Parakeet Flash | Lowest-latency live English | English | Apple Silicon |
| Parakeet TDT v3 | Fast multilingual default | 25 | Apple Silicon |
| Parakeet TDT v2 | Fastest English-only | English | Apple Silicon |
| Cohere Transcribe | High-accuracy multilingual | 14 | Apple Silicon |
| Apple Speech | No download | System languages | Apple Silicon and Intel |
| Whisper (Tiny to Large) | Broad compatibility | Up to 99 | Apple Silicon and Intel |
Features
- On-device speech recognition: Nemotron, Parakeet, Cohere Transcribe, Apple Speech, and Whisper models run locally
- AI enhancement: optional cleanup of filler, casing, and structure through the local Fluid Intelligence model or an OpenAI, Groq, or custom provider
- Write Mode: write or rewrite text in any text field, inline
- Command Mode: launch apps, run shortcuts, and trigger system actions by voice
- Global hotkey: start voice capture from anywhere without switching apps
- Smart typing: inserts text into any app through macOS accessibility APIs
- Live preview: a real-time, notch-aware transcription overlay in several sizes
- Per-app configuration: different prompt sets for different apps
- Audio history: optional local recording history with storage budgets and ZIP export
- Usage stats: daily usage tracking in the toolbar
- Menu bar integration: quick access to status and settings, with an optional beta update channel
