Convert speech to text locally and control your Mac with voice commands
If you want to type by speaking without sending data to external servers, this app is a fit. FluidVoice processes speech recognition locally on macOS and optionally refines text using local AI. Version 1.6.0 adds the Parakeet engine, which significantly reduces recognition latency, and Fluid Intelligence, a local AI correction feature.
1.6.0 Update: Low-Latency Recognition and Local AI Correction
This update focuses on improving recognition speed and text quality. The Parakeet engine was re-implemented to nearly eliminate the delay between spoken words and the text appearing on screen. The newly added Fluid Intelligence is a local AI model that automatically organizes punctuation and capitalization based on context. This entire process happens inside the Mac, so no data leaves the device. The onboarding process has also been simplified to handle language selection and voice engine configuration in one go.
Two Modes for Controlling Your Mac with Voice
FluidVoice offers features to operate the computer via voice, going beyond simple text input. In Write Mode, you can rewrite selected text or input new content by speaking into any app's text box. In Command Mode, you can launch apps, trigger shortcuts, and execute system actions using voice commands. This allows you to automate workflows without touching the keyboard, which is useful when your hands are busy.
Selecting Voice Models Based on Language and Speed
Users can choose a model that fits their language and speed requirements. Various models are supported, including Nemotron, Parakeet, Cohere, Apple Speech, and Whisper, each with different supported languages and download sizes. For example, Parakeet Flash is English-only and provides the lowest latency, while Nemotron supports approximately 40 languages. Most models require an Apple Silicon Mac, but Whisper models can be used on Intel Macs. Model download sizes range from approximately 75MB to 2.9GB.
Privacy and Data Handling
The app follows a local-first principle. Voice, audio, and converted text do not leave the Mac unless the user explicitly selects a cloud AI provider. You can optionally connect to cloud services like OpenAI or Groq, but the default setting is fully local processing. Analytics data collects only anonymous activity signals and never collects personal information such as voice or input content. This setting can be changed at any time in the settings menu.
Things to Check Before Installing
FluidVoice requires macOS 15.0 (Sequoia) or later and an Apple Silicon Mac. Intel Mac users can only use the Whisper model. Microphone and accessibility permissions are mandatory, and accessibility permission is specifically required to input text into other apps. Disk space requirements are approximately 1GB for voice models, or approximately 3.5GB if including the optional Fluid Intelligence model. It is distributed under the GPLv3 license, while versions prior to February 23, 2026, follow the Apache License 2.0. The official website is altic.dev/fluid; be aware that other domains are not official.
