New · Meet Fluid-1, our custom-trained local model

Your voice never leaves your Mac.

Q: Is FluidVoice free?

Yes. FluidVoice is completely free and open source.

Q: What macOS version does FluidVoice require?

FluidVoice requires macOS 15.0 Sequoia or later, plus microphone access and accessibility permissions for typing.

Q: Does FluidVoice support Intel Macs?

Yes. Current version supports Intel Macs by using Whisper models.

Q: Which speech models does FluidVoice support?

FluidVoice supports Nemotron Speech 3.5 (Ultra Fast Low Latency and Multilingual), Parakeet Flash (Beta), Parakeet TDT v3, Parakeet TDT v2, Cohere Transcribe, Apple Speech, and Whisper Tiny/Base/Small/Medium/Large.

Q: What is Fluid Intelligence and the Fluid-1 model?

Fluid Intelligence is a local AI runtime which runs the Fluid-1 model. It delivers advanced on-device dictation enhancement — smart formatting, context-aware capitalization, adaptive tone based on the active app, and post-processing — all running locally on your Mac. It is fully optional (~3.5 GB download), requires no cloud or API keys, and no data leaves your machine. FluidVoice works great on its own without it.

Q: How many languages does FluidVoice support?

Language support is model-specific: Nemotron Speech 3.5 supports ~40 languages, Parakeet Flash and Parakeet TDT v2 are English-only, Parakeet TDT v3 supports 25 languages, Cohere Transcribe supports 14 languages, Apple Speech depends on macOS system languages, and Whisper supports up to 99 languages depending on model size.

Q: Does FluidVoice work offline?

FluidVoice is local-first and uses local speech models for dictation. AI enhancement can use optional providers such as OpenAI, Groq, or custom providers, so cloud use depends on the features and providers a user enables.

Q: What permissions does FluidVoice need?

FluidVoice needs microphone access for voice capture and accessibility permissions for typing into other apps.

Q: What license does FluidVoice use?

From 2026-02-23 onward, FluidVoice is licensed under the GNU General Public License v3.0 (GPLv3). Versions before that date used Apache License 2.0.

Q: What does FluidVoice analytics collect?

Optional anonymous analytics, FluidVoice does not collect voice, raw audio, transcribed text, prompts etc. No sensitive info leaves your mac

FluidVoice turns rough, rambling speech into polished, ready-to-send text in any app. Free forever, open source, and 100% on-device.

$brew install --cask fluidvoice

—

Zero data leaves your MacFree forever · open sourceCustom-trained Fluid-1 model

100,000+ downloads · macOS 15.0+ · Apple Silicon & Intel

Scroll to see more

Watch FluidVoice and Fluid Intelligence work

Raw FluidVoice dictation — speech hits the terminal as fast as you can talk.

Try it yourself

Take FluidVoice for a spin

You talk. Fluid-1 — our custom-trained local model — turns it into a polished prompt before it ever hits an app. Watch it happen.

Claude CodeFileEditView

Fri 9:41 AM

Claude Code v2.1.12

▐▛███▜▌ ▝▜█████▛▘ ▘▘ ▝▝

Welcome back!

Opus 4.5 · Claude Team · FluidVoice

~/dev/contact-form

❯ ▍

Read(ContactForm.tsx)

Read 156 lines

I'll fix both issues — email validation and

a loading spinner on submit.

Update(ContactForm.tsx)

──────────────────────────────────────────────────────────────────────

Edit file ContactForm.tsx

╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌

22 const handleSubmit = async (e) => {

23 e.preventDefault();

24 + if (!validateEmail(email)) {

+ setError('Please enter a valid email');

+ return;

+ }

25 }

Bash(npm test)

4 passed

The demo plays automatically — press ⌥ anytime to replay

Why it's different

Trained for how you actually talk. Built to never leave your Mac.

Most dictation apps stream your microphone to the cloud and hand you a raw transcript. FluidVoice does everything on-device — then goes one step further with Fluid-1, a custom-trained model that turns messy speech into text that's ready to send.

You speak

Hold one hotkey and talk normally — ums, stumbles, and mid-sentence changes of mind included.

Transcribed on-device

A local speech model converts your audio to raw text. Your voice never touches a server.

03 · Fluid-1

Fluid-1 rewrites it

Our custom-trained post-processing model strips filler, fixes casing, formats structure, and matches the tone of the app you're in.

Fluid-1 is trained on real dictation — so it knows what you meant, not just what you said. And because it runs locally, that intelligence costs you nothing and exposes nothing.

Loved by the community

From developers to multilingual teams

100,000+

Downloads

Fluid-1

On-device AI model

Free

Forever, no tiers

GPLv3

Open source

40+

Languages

What's inside

Powerful features, seamlessly integrated

Capture, intelligence, and speed in a single local pipeline — one hotkey away in any app.

Real-time speed

Speaking is ~3.7x faster than typing. Text lands instantly so ideas stay in motion.

Speaking

~3.7x

faster than typing

System-wide input

One shortcut, any text field. Email, docs, chat, terminal, and code editors all work the same.

+ any app

Works offline

Local speech models process everything on-device. No Wi-Fi, no problem.

0 bytes leave this Mac

Fluid Intelligence

On-device AI polishes formatting, capitalization, and post-processing. Zero data leaves your Mac.

Predefined modes

Write Mode, Command Mode, and Direct Dictation - switch context instantly.

WriteCommandDictate

Multilanguage support

40+ languages across models. From Nemotron to Whisper, speak naturally in your language.

Hello+ 40 languages

Apple Silicon optimized

CoreML and Metal deliver low-latency transcription while preserving battery efficiency.

engine · CoreML + Metal

target · Apple Silicon M-series

AI post-processing

Local or cloud — Fluid Intelligence, OpenAI, Groq, or custom providers. You choose.

Fluid IntelligenceOpenAIGroqCustom

Adaptive tone

Different apps, different tone

Customize prompts for each app so your dictation fits the context.

You said

“hey so um the deploy broke because of the config change from yesterday, I fixed it but we should probably review those before pushing to prod going forward”

#engineering

Tone: Casual

Rohit 9:42 AM

hey team — deploy broke from yesterday's config change. fixed it now, but we should probably start reviewing configs before pushing to prod 🙏

Customize per app

Write custom prompts for any app. Slack casual, Mail formal, GitHub structured — you define it.

Detects the app

FluidVoice recognizes which app is focused and selects the matching tone profile automatically.

Same voice, right tone

Speak naturally. Fluid-1 rewrites your words to fit the context without changing your meaning.

Performance

Built for speed you can feel

FluidVoice keeps inference local and latency low so your writing flow stays uninterrupted.

Perceived latency

<100ms

review the config before prod86ms

Fast enough to feel immediate while moving through thought-heavy tasks.

Real-time factor

Up to

Peak throughput varies by hardware, model, and audio length.

Local-first dictation

Local speech models keep dictation on-device; optional AI providers are user-controlled.

Multi-hour long-form windows

Handles extended transcription sessions without manual chunking.

Architecture highlights

mic→on-device model→CoreML · Metal→text in any app

Model lineup spans Nemotron Speech 3.5, Parakeet Flash, Parakeet TDT v3/v2, Cohere Transcribe, Apple Speech, and Whisper, with optimized routing to keep transcription responsive across short bursts and long dictation sessions.

Nemotron Speech 3.5Nemotron 3.5 MultilingualParakeet Flash (Beta)Parakeet TDT v3Parakeet TDT v2Cohere TranscribeApple SpeechWhisper Tiny / Base / Small / Medium / Large

What the community says

Loved by developers, built for how you work

Free forever

“Thank you for making it free, forever! Hope to see it support more languages soon!”

GitHub discussion

Community-first growth

“Very nice! I imagine monetization in future might be easier cause you're being distributed for free. Smart plan.”

Early adopter

Open source trust

“Thanks for making it open source”

Developer comment

Multilingual reliability

“It works fine in German by the way (Tahoe, Mac Mini M2 Pro, external mic). This opens a host of new possibilities for interacting with my Mac.”

German workflow

Intent recovery

“Wow, impressive. I stumbled while dictating and the transcriber caught it and output the word I had intended to say!”

Daily dictation

Head-to-head speed

“This is a really impressive app, speedy and accurate. It's just as good as Wispr Flow in my testing.”

Comparison test

Product polish

“This is unbelievably slick. Congrats, and well done.”

Launch feedback

Open source goodwill

“Solid product! and super cool of you to make it open source :)”

Community comment

Language + media

“I just did a quick test in French using dictation and also tried transcribing a video — both worked very well.”

French test

Developer adoption

“Love the project. Thanks for building this!”

GitHub reaction

Retention

“Having been one of the early users, I want to say thank you and amazing job!”

Early user

Momentum

“All the best of success with this. Congrats :)”

Community support