FluidVoice
Fluid-1
Optional local AI pass that fixes capitalisation, punctuation and spoken corrections.
FluidVoice
FluidVoice is a free open source voice to text app for Mac and Windows. macOS v1.6.10-beta.7 and Windows 0.0.10 are available as official pre-release builds.
The official website of FluidVoice, developed by Altic (altic-dev).
- Free to use
- Open source
- Local-first
- No required subscription
- 10k+ GitHub stars
- 8 speech engines
- GPLv3 open source
- Local audio stays on device
Current release
FluidVoice v1.6.10-beta.7 at a glance
A compact, verifiable snapshot for anyone checking the download before installing it.
- Version
- 1.6.10-beta.7
- DMG size
- 71.8 MB
- SHA-256
433ffc1513cd07b4a93c4a545da614bd590d7bf4db0cea36ed7c5cddd4ea4c23
- Spoken Formatting in Custom DictionaryNew lines, paragraphs, tabs, spaces, punctuation and symbols can be inserted by voice using a start word you configure.
- More resilient audio switchingBluetooth route and audio-format changes are handled more reliably in this release.
- External-mic fixesUpdates cover 3.5 mm microphones and wired-headset capture while a MacBook is used closed.
The basics
What is FluidVoice?
FluidVoice is a free, open-source dictation app developed by Altic (altic-dev). It can place recognised speech directly into editable fields across the apps you already use.
Local speech engines include Parakeet, Nemotron, Whisper, Cohere Transcribe and Apple Speech. Hardware, macOS and language availability vary by engine, so the model picker in your installed version is the final compatibility check.
For cleaner output, you can also enable Fluid Intelligence, an optional local AI layer that improves punctuation, capitalisation, sentence structure, and spoken corrections before the text is inserted.
- AI tools ChatGPT, Claude, Gemini
- Code editors VS Code, Cursor, Windsurf, Xcode
- Terminals Terminal, iTerm, Warp, Ghostty
- Team chat Slack, Discord, Teams
- Notes & docs Notes, Obsidian, Notion, Google Docs
- Other common fields Email, browsers, web forms
Direct insertion depends on the active app exposing an editable field through macOS Accessibility. Compatibility can vary in protected or unusual input controls.
Start to finish
From speech to text at your cursor
Choose your preferred hotkey, start dictation, speak naturally, then use the same shortcut to stop. In local mode, the steps below run on your Mac before the text reaches the active cursor.
FluidVoice
Listening
ListeningSpeak naturally.
Capturing speechYour words become text in real time.
send the updated report today
Optional AI refiningLocal cleanup runs here when enabled.
WritingPolished text appears at your cursor.
Send the updated report today.
ReadyYour dictation is complete.
The same real macOS dictation, shown during capture and after the finished text is inserted.
Beyond dictation
Advanced Features
Custom Dictionary
Teach it names and jargon by voice, or import a list you already have.
Write Mode
Rewrite, shorten or reformat selected text without leaving the app.
Command Mode
Speak an instruction instead of text, with app-aware actions in supported workflows.
Also included: spoken punctuation with no AI needed, local dictation history, per-app prompts, live preview, usage stats, and light and dark themes.
Public community feedback
What FluidVoice users are saying
Short excerpts from public GitHub discussions. Each card links to the original post and uses the author’s public profile avatar.
10k+ developers have starred FluidVoice on GitHub. These quotes are unrated comments, not scored reviews — GitHub Discussions has no rating system.
Verified GitHub discussion
Hands down the best dictation tool I have tried, and the one that I actually keep using!
Verified GitHub discussion
I just downloaded the app and tested the speed and accuracy in German and I am blown away.
Verified GitHub discussion
I’m new to this app and thanks for building it, it’s fantastic.
Local by default
Built for Privacy
Your voice and local transcripts stay on your device during local dictation.
Local speech recognition and the optional Fluid-1 cleanup layer run on your Mac. External AI providers are separate, optional choices and receive text only when you configure and use them.
- 1You speakFluidVoice captures audio during the active dictation session.
- 2Your Mac processes itThe selected local speech model converts audio to text on-device.
- 3Text reaches your cursorThe result is inserted into the active editable field.
Performance
Release-backed responsiveness
Current builds report first-word capture below 100 ms for the revised Parakeet path on supported hardware, a figure first published with v1.6.7. Your result still depends on the Mac, speech model, audio device and workload.
First-word capture latency
<100ms
This is a release-supported result for the revised capture path, not a universal benchmark for every model and Mac.
- <100ms Reported first-word capture latency for the revised Parakeet path, published with v1.6.7
- Local Voice audio and local transcripts stay on your Mac during local dictation
- 8engines Choose a speech engine for your language, hardware and preferred trade-off
The whole path stays on one machine
- 01Microphone
- 02On-device speech model
- 03Selected local runtime
- 04Text in a supported editable field
The local path is straightforward: FluidVoice captures the active dictation, sends it to your selected on-device speech model, and inserts the result into the current editable field. If you enable Fluid-1, cleanup also runs locally before insertion.
Cloud AI providers are optional and separate from this local path. When one is configured, the text sent through that provider is handled under that provider’s terms.
Eight engines, one app
Pick the speech model that fits you
FluidVoice lets you choose among speech engines with different language, hardware and responsiveness trade-offs, then change the selection in the app.
A starting speech model for four common priorities.
| If you want | Start with | Languages |
|---|---|---|
| Fastest English | Parakeet Flash | English |
| Everyday multilingual | Parakeet TDT v3 | 25 |
| Strong multilingual accuracy | Cohere Transcribe | 14 |
| Widest coverage | Whisper | Up to 99 |
Why people switch
Free, and not a trial
A qualified comparison of common approaches. Individual products and operating-system features vary.
| FluidVoice | Typical cloud app | Built-in dictation | |
|---|---|---|---|
| Price | Free | Often subscription-based | Included with the OS |
| Speech processing | Local with local models | Often remote | Varies by OS and language |
| Text cleanup | Local Fluid-1 or an optional provider | Often cloud-assisted | Usually limited |
| Model choice | Multiple speech engines | Varies by product | Usually one system engine |
| Source availability | GPLv3 app code | Often proprietary | Proprietary OS component |
Before you install
Common questions
What is FluidVoice?
FluidVoice is a voice-to-text app that turns speech into editable text on your Mac. It can use downloaded speech models locally, then insert the result into supported text fields across the apps you already use.
Does FluidVoice work offline?
Local dictation works offline after the speech model and any local cleanup assets you use are installed. Internet access is still needed for updates, new model downloads and optional cloud providers.
Do I need Fluid-1?
No, it is optional. Basic dictation works without it. Install it if you want local AI cleanup and formatting on top of speech recognition.
Stop typing what you could just say.
Download the app, grant microphone and Accessibility permissions, choose a speech model and set the shortcut that fits your workflow.
