FluidVoice is a macOS dictation app that performs speech-to-text conversion entirely on-device using local AI models.
The tool addresses the need for fast, private voice dictation by running speech recognition and AI enhancement locally on your Mac without sending audio to cloud services or requiring API keys. It uses on-device models including Parakeet, Nemotron Speech, Whisper, and others, combined with a proprietary local AI enhancement layer called Fluid Intelligence that handles smart formatting, context-aware capitalization, and post-processing entirely offline.
FluidVoice suits users who prioritize privacy and speed in dictation workflows and want to avoid subscription services for voice-to-text features. The tool emphasizes zero latency between speaking and text appearing on screen, making it practical for real-time dictation. It works as a standalone dictation app with any supported speech model, with optional Fluid Intelligence enhancement available for users wanting additional on-device post-processing. The README positions it as a local alternative to Whisper Flow, though it does not provide detailed comparisons to other dictation solutions.
The project maintains active development with recent releases introducing performance improvements to the Parakeet speech model implementation, a new fully local AI enhancement system, and refined user onboarding. Platform expansion is underway, with iOS and Windows versions on a waitlist. The codebase is open source under GPLv3, though the Fluid Intelligence enhancement layer remains privately maintained to support sustainable free-tier access to core dictation features.
