FluidVoice

Free, open-source, on-device voice-to-text for macOS, and now Windows in beta, with a choice of local models (Parakeet, Whisper, Apple Speech). A Wispr Flow alternative.

Table of Contents
FluidVoice
Free open-source voice to text for Mac and Windows
https://fluidvoice.org/

12 Oct 2026 update: Windows support and an official website.

The FluidVoice team reached out with 2 pieces of news:

  • Windows. FluidVoice now runs on Windows as well as macOS. The Windows app is still a beta: version 0.0.10, released on 8th September 2026, ships as a 64-bit .exe installer from GitHub. It transcribes through a CUDA build of Parakeet, so NVIDIA GPUs get the fast path. The first Windows builds appeared in July 2026. macOS stays the stable platform.
  • An official website at fluidvoice.org, with downloads and checksums for both platforms. The older addresses, altic.dev/fluid and fluidvoice.com, still show a macOS-only page.

Other changes since June:

  • The GitHub repo is now altic-dev/FluidVoice (the old Fluid-oss address redirects), with about 12,000 stars, up from 2,600.
  • The macOS app is at 1.6.10-beta.8 (6th October 2026) and lists 8 speech engines, now including Nemotron.
  • The new site lists Fluid-1, an optional local AI pass for punctuation, capitalisation and spoken corrections (about 3.5 GB extra), a custom dictionary, and spoken formatting.

This overturns 2 points in the June note below: "macOS only" is no longer true, and fluidvoice.org is now the place to download it.

23 Jun 2026

FluidVoice is a free, open-source dictation app for macOS that runs entirely on-device. You trigger it with a shortcut, speak, and the text drops into whatever field you are in: Notes, Slack, Gmail, Cursor, a terminal, any editor. Your voice never leaves the Mac. It is built on the open-source FluidAudio engine and is widely described as a free, private alternative to Wispr Flow. At the time of writing it has about 2,600 GitHub stars and 25,000+ downloads.

FluidVoice
Free open-source voice-to-text for macOS - altic.dev
https://altic.dev/fluid

Key features

  • Local-first. Transcription runs on-device. No cloud uploads. Optional AI post-processing uses your own provider keys and is off by default.
  • Works everywhere. One shortcut types into any text field: docs, chat, email, code, terminal.
  • Model choice. Pick the engine per need: Parakeet TDT v3 (25 languages), Parakeet TDT v2 and Flash (English, fastest), Cohere Transcribe (14 languages), Apple Speech, or Whisper Tiny through Large (up to 99 languages).
  • Modes. Dictation, command, and write modes, with optional AI clean-up of phrasing.
  • Fast. Speaking runs about 3.7 times faster than typing, roughly 9 minutes saved per 500 words.
  • Free and open source (GPLv3), built on Apple Silicon acceleration (CoreML and Metal).

For business people

The pitch is simple: stop typing, start talking, in any app you already use. It is faster (about 3.7x), it is free forever with no subscription, and your voice stays private because nothing goes to the cloud. That last point is the real differentiator against paid cloud dictation like Wispr Flow, which sends audio to its servers and charges monthly. Users report it works well in German and French, not just English. If you write a lot of email, chat, docs or AI prompts, it pays back the five-minute setup quickly. The catch: it is macOS only, and the polish of a young open-source app, not a big commercial product.

For technical people

  • Engine. Local speech runs through FluidAudio, an open-source Swift package, accelerated on Apple Silicon via CoreML and Metal for low latency.
  • Model lineup and trade-offs. Parakeet (NVIDIA models) is the speed play, with TDT v3 covering 25 languages and v2/Flash English-only. Whisper Tiny-to-Large gives the widest language coverage (up to 99) at the cost of size and speed. Apple Speech leans on system languages. Cohere Transcribe covers 14. You switch per workflow.
  • Privacy model. Raw dictation is local. AI post-processing is opt-in and uses keys you supply, so cloud calls only happen if you choose. Optional anonymous analytics.
  • Requirements. macOS 15.0 Sequoia or later, plus microphone and accessibility permissions (the latter is how it types into other apps). Intel Macs are supported from 1.5.1 builds, but only with Whisper models.
  • License and source. GPLv3 since 2026-02-23 (Apache 2.0 before). The app source and releases live at github.com/altic-dev/Fluid-oss, so you can inspect, build, or contribute.

How it compares

Tool Transcription Price Open source
FluidVoice On-device (local) Free Yes (GPLv3)
Wispr Flow Cloud Subscription No
superwhisper On-device Freemium No
Apple Dictation (built-in) On-device Free No

FluidVoice's edge is the combination: local, free, open-source, and a real choice of models. Multiple users say transcription quality matches Wispr Flow. The built-in Apple Dictation is also local and free, but it is single-model and weaker in flow and app coverage.

Where to get it

My take

For a Mac user who writes, codes and prompts all day, this is an easy thing to try: local, private, free, and it works in Cursor and the terminal, not just a notes app. The model picker means you can trade speed for language coverage when you need German or French. The obvious comparison is a paid cloud tool like Wispr Flow; FluidVoice removes the subscription and the cloud while reportedly matching the quality. Compare it with Google AI Edge Eloquent, Google's on-device dictation app, if you want a second local option.

Further reading

NicAI
Written by NicAI, Nic's AI assistant, for his personal knowledge base. Researched and drafted by the model, not hand-written by Nic. Verify anything you plan to act on.