Open Source
Explore the latest AI open-source projects from GitHub and HuggingFace.
Explore the latest AI open-source projects from GitHub and HuggingFace.
FluidVoice is an open-source voice-to-text dictation app for macOS that runs speech recognition fully on-device. Released under GPLv3 and installable with a single Homebrew command, it has climbed to around 4,400 GitHub stars by offering fast, private dictation that keeps audio and transcripts on the user's Mac rather than streaming them to the cloud. ## Fully Local Speech-to-Text The core promise of FluidVoice is privacy through local processing. Voice is transcribed on-device, so no recordings or text need to leave the machine. This makes it a practical alternative to subscription dictation services for users who handle sensitive material or simply prefer not to depend on a network connection for everyday transcription. ## A Choice of Speech Engines FluidVoice is model-agnostic, supporting a range of speech backends including NVIDIA Nemotron Speech 3.5, Parakeet (Flash, v3, and v2), Cohere, Apple Speech, and OpenAI Whisper. Version 1.6.0 rebuilt the Parakeet implementation for near-instant transcription, narrowing the delay between speaking and seeing words on screen to almost nothing — a meaningful difference for real-time dictation workflows. ## On-Device Enhancement with Fluid Intelligence Beyond raw transcription, FluidVoice offers Fluid Intelligence, a fully local AI runtime that adds smart formatting, context-aware capitalization, and post-processing without API keys or cloud calls. The base app works on its own with any supported speech model and optional cloud AI providers, while Fluid Intelligence layers on private, on-device enhancement for users who want cleaner output without sending data anywhere. ## Native macOS Experience Built in Swift as a native Mac app, FluidVoice includes an adaptive light/dark theme with a compact toolbar switcher and a refreshed, language-first onboarding flow that walks users through voice-engine setup and a real dictation tryout. A Command Mode also lets users trigger actions on their Mac by voice, extending the tool from pure dictation toward hands-free control. ## Considerations FluidVoice is currently macOS-only, with iOS and Windows noted as future goals, so non-Mac users are not yet served. The Fluid Intelligence enhancement layer is kept private rather than open-sourced, which some in the community may see as a caveat to the otherwise GPLv3 project. For Mac users who want fast, private, fully local dictation with a choice of modern speech models, though, FluidVoice is a compelling open-source option.