FluidVoice
FluidVoice is an open-source dictation app for macOS under GPL-3.0. It types what you say into any app through the accessibility APIs, with the speech model running on the Mac: Nemotron Speech 3.5, Parakeet, Cohere Transcribe, Apple Speech or Whisper. It adds a command mode, a write mode that rewrites selected text, and per-app prompts. Its optional Fluid Intelligence enhancement model is local but not open source.
What does FluidVoice do?
FluidVoice describes itself as an "open source voice-to-text dictation app for macOS with on-device AI enhancement." It is a menu-bar app with a global hotkey and a live transcription overlay that supports the notch. Beyond plain dictation it has a Command Mode that launches apps and runs shortcuts, and a Write Mode that rewrites selected text.
The README lists the models it downloads with their sizes: Nemotron Speech 3.5 (about 40 languages, about 670 MB), Nemotron 3.5 Multilingual (about 530 MB), Parakeet Flash beta (English, about 250 MB), Parakeet TDT v3 (25 languages, about 500 MB), Parakeet TDT v2 (English), Cohere Transcribe (14 languages, about 1.4 GB), Apple Speech (built in), and Whisper from Tiny to Large (99 languages, about 75 MB to 2.9 GB). Enhancement of the text can run through the local Fluid Intelligence model, about 3.5 GB, or optionally through OpenAI, Groq or a custom provider with keys held in the macOS Keychain.
Key facts
- Licence: GPL-3.0 from 2026-02-23 onward; versions published before that date were Apache-2.0, per the README.
- Fluid Intelligence, the local enhancement model, is "a separate, privately maintained local AI runtime" and is not open source; it needs about 3.5 GB of disk and no cloud or API keys.
- Speech models and sizes as the README lists them: Nemotron Speech 3.5 about 670 MB, Nemotron 3.5 Multilingual about 530 MB, Parakeet Flash about 250 MB, Parakeet TDT v3 about 500 MB, Cohere Transcribe about 1.4 GB, Whisper 75 MB to 2.9 GB.
- Requirements: macOS 15.0 Sequoia or later; an Apple Silicon Mac for all models; Intel Macs supported through Whisper models from version 1.5.1.
- Install with brew install --cask fluidvoice or from GitHub Releases; build from source in Xcode with Swift Package Manager.
- Audio history with ZIP export, per-app prompt sets, auto-updates with a beta channel.
- Anonymous analytics are enabled by default and can be disabled, per the README.
- Latest tagged release when read: v1.6.9 on 2026-08-18; 11,613 stars on 2026-09-17.
What does it replace, and where does it fall short?
FluidVoice replaces Wispr Flow and Superwhisper on a Mac with the same shape: a hotkey, on-device speech recognition, and a rewrite step. Its model list is the widest of the dictation apps on this site, and the write and command modes cover the polish-my-text and do-something-for-me jobs the paid apps sell. It sits beside VoiceInk on the open-source alternatives to Wispr Flow page.
Where it falls short: macOS only, 15.0 or later, with Apple Silicon needed for every model except Whisper. The enhancement model that makes the text read well, Fluid Intelligence, is closed; the README says it is kept private "so we can sustainably offer the core dictation experience for free." Anonymous analytics are on by default until you turn them off.
How does FluidVoice run?
brew install --cask fluidvoice, or download the release from GitHub. The README asks for macOS 15.0 or later, an Apple Silicon Mac for all models, and about 1 GB of disk per voice model, plus about 3.5 GB for Fluid Intelligence if you install it. Intel Macs run the Whisper models only. Building from source uses Xcode and Swift Package Manager. There is no hosted or paid tier; the project takes GitHub Sponsors.
Who is FluidVoice for?
A Mac user on Apple Silicon who wants dictation with a choice of the newest local models and a rewrite step without a subscription. Someone on Windows or Linux should look at Handy, OpenWhispr or SpeakoFlow; someone who needs every component open should note that the enhancement model here is closed.
What limits does the README state?
"FluidVoice is on macOS today." "Fluid Intelligence is a separate, privately maintained local AI runtime." Apple Silicon is required for all models, with Intel limited to Whisper. The README says iOS and Windows are on the way; neither exists in the README as of the read date.
Questions people ask
Is FluidVoice open source?
The app is: GPL-3.0 since 2026-02-23, and Apache-2.0 for versions published before that date. One part is not: Fluid Intelligence, the optional local model that enhances the transcript, which the README calls a separate, privately maintained runtime. Dictation works without it.
Does FluidVoice send my voice anywhere?
Not unless you choose to. The README says your voice, audio and transcribed text never leave your machine unless you explicitly opt in to a cloud AI provider such as OpenAI or Groq for enhancement. Speech recognition runs on the Mac. Anonymous analytics are on by default and can be turned off.
Which Macs does FluidVoice run on?
macOS 15.0 Sequoia or later. All models run on Apple Silicon; Intel Macs are supported through the Whisper models from version 1.5.1. Budget about 1 GB of disk per speech model and about 3.5 GB for the optional Fluid Intelligence model.
How does FluidVoice compare with Wispr Flow and Superwhisper?
FluidVoice covers the same jobs, dictate, rewrite selected text and run commands by voice, with the speech model on the device and no subscription. Wispr Flow and Superwhisper add phone apps and vendor support. The open-source alternatives to Superwhisper page compares it with VoiceInk, Handy, OpenWhispr and the others.
Which speech models does FluidVoice support?
Nemotron Speech 3.5 and Nemotron 3.5 Multilingual (about 40 languages), Parakeet Flash, Parakeet TDT v3 (25 languages) and v2, Cohere Transcribe (14 languages), Apple's built-in Speech, and Whisper from Tiny to Large (99 languages), with sizes from about 75 MB to about 2.9 GB, per the README.
Sources
- FluidVoice README and GPL-3.0 LICENSE: github.com/altic-dev/FluidVoice, read 2026-09-17.
Compared with the others
On the open-source alternatives to Wispr Flow page, FluidVoice is the pick for a mac with the newest local models. GPL-3.0, on-device Nemotron, Parakeet, Cohere Transcribe, Apple Speech or Whisper, with a write mode that rewrites selected text and a command mode.
Also on that page: Handy for all three desktops, nothing else running, VoiceInk for a mac, with modes per application, SayIt for windows, with an editable cleanup step.
On the open-source alternatives to Superwhisper page, FluidVoice is the pick for a mac with the newest local models. GPL-3.0, on-device Nemotron, Parakeet, Cohere Transcribe, Apple Speech or Whisper, with a write mode that rewrites selected text and a command mode.
Also on that page: VoiceInk for a mac, with modes per application, Handy for all three desktops, nothing else running, SayIt for windows, with an editable cleanup step.
More voice dictation apps
A free, offline push-to-talk dictation app for macOS, Windows and Linux that runs Whisper or NVIDIA Parakeet on your own machine and pastes the text at your cursor, under MIT.
A dictation and meeting-notes app under MIT that runs Whisper or Parakeet locally or cloud models with your own key, detects Zoom, Teams and FaceTime calls, and labels speakers on device.
A macOS dictation app under GPL-3.0 whose README names Superwhisper and Wispr Flow as the tools it replaces, with local models, per-app modes, a personal dictionary and an AI assistant mode.
A speech-to-text app inside the Epicenter monorepo under AGPL-3.0 that records, transcribes with a provider you choose, cloud or local, optionally polishes the text, and pastes it at the cursor.
An AGPL-3.0 voice typing app for Windows that ships seven local speech models including Nemotron 3.5 ASR and three Qwen3-ASR sizes, with an editable AI cleanup step through Ollama or any OpenAI-compatible endpoint.
An MIT desktop voice assistant for Windows, macOS and Linux that extends Handy's dictation core with a trigger phrase, a local cleanup model, screen-aware answers and Kokoro speech, with no account and no telemetry.
Added September 17, 2026. Every claim above comes from the project's README, LICENSE or model card, read on September 17, 2026, or from the GitHub API on the date shown in the panel. Found an error? Write to reveneau@licheo.com and it is fixed in the next weekly pass. Repository: github.com/altic-dev/FluidVoice.
Running one of these inside your own environment, with your own data and your own security rules, is the kind of work Reveneau does. Read how a forward deployed engagement works.