Skip to main content
PromptQuorumBuilt for humans. Structured for AI.
Home/Power Local LLM
Voice, Speech & Multimodal

FluidVoice Review: Free Open-Source Mac Dictation App

·6 min readBy Hans Kuepper · Founder of PromptQuorum · Discovery engine for open-weight & open-source AI

FluidVoice (version 1.6.12) is a free, GPL-3.0-licensed Mac app that turns your speech into text with speech models running on your Mac and types the result into the app in front of you. It needs macOS 15 or later, and every statement here comes from the maker's website, privacy policy and GitHub repository.

Key Takeaways

  • Maintenance: the latest release was published on 22 October 2026 and the main branch last changed on 24 October 2026, so no stale-project warning applies.
  • Cost: free; the project asks for GitHub Sponsors support and has no paid tier.
  • Requirements: macOS 15 or later; Apple silicon for every model, Intel Macs only through Whisper.
  • Install: a .dmg on GitHub Releases, or a Homebrew cask named in the README.
  • Closed part: the local Fluid-1 cleanup model is privately maintained, while the app itself is open source.

📍 In One Sentence

FluidVoice is an open-source Mac dictation app that transcribes your voice with an on-device speech model and types the text into the active app.

💬 In Plain Terms

You hold or press a hotkey, say what you want written, and the words appear at the cursor; according to the project, no account is needed and the audio stays on your Mac unless you pick a cloud provider.

📌Note: This review is based on fluidvoice.com, its privacy policy and the GitHub repository, checked on 24 October 2026. PromptQuorum has not installed or timed the app, so accuracy and speed are not assessed.

Read in:
🇺🇸en🇩🇪de🇫🇷fr🇯🇵ja🇨🇳zh🇪🇸es🇧🇷pt🇸🇦ar🇰🇷ko

FluidVoice is a free, open-source Mac app by Altic: you press a hotkey, speak, and the text is typed into the app you are using. According to the project, speech recognition runs on your Mac, and an optional local model cleans up filler words and formatting before the text lands. This review is based on the website, its privacy policy and the GitHub repository, checked on 24 October 2026; PromptQuorum has not tested the app hands-on.

FluidVoice Review: Free Open-Source Mac Dictation App

What Is FluidVoice?

FluidVoice is a macOS dictation app that records while you use a hotkey, transcribes the audio with a speech model you choose and types the result into whatever application has focus. It is made by Altic (altic-dev), and the repository was created in September 2025. The website positions it as a local alternative to cloud dictation services.

Transcription and text cleanup are two separate steps. A speech model turns audio into text; an optional second model, which the project calls Fluid Intelligence or Fluid-1, then rewrites the transcript to remove filler words, fix casing and match the app you are in. The name is easy to confuse: PromptQuorum lists a separate app called Fluid, which is unrelated to this one.

Get It

FluidVoice is a Mac download today; the Windows and iPhone versions are labelled beta on the website, and Linux and Android are only on a waitlist.

Option
How to get it
macOS.dmg and .zip on the GitHub Releases page; the website's "Download for Mac" button; or brew install --cask fluidvoice
WindowsLabelled beta on fluidvoice.com; PromptQuorum did not verify a download
iPhoneBeta through TestFlight; the sign-up form returns an invitation
Websitefluidvoice.com
Privacy policyfluidvoice.com/fluid/privacy-policy
Source codegithub.com/altic-dev/FluidVoice, GPL-3.0

Version reviewed: 1.6.12, published on 22 October 2026 per the GitHub Releases page. The app is free. The licence has been GPL-3.0 since 23 February 2026; earlier versions used Apache 2.0. The project states that fluidvoice.org is not affiliated with it; altic.dev/fluid redirects to fluidvoice.com.

Requirements and Models

The README sets macOS 15 (Sequoia) as the minimum and says every model runs on Apple silicon; Intel Macs work through Whisper models and Apple Speech only.

Nemotron Speech 3.5

What the README says:
Streaming and multilingual variants, about 40 languages, 670 MB and 530 MB downloads, Apple silicon

Parakeet Flash (beta), TDT v2

What the README says:
English only; about 250 MB and 500 MB; lowest-latency live dictation; Apple silicon

Parakeet TDT v3

What the README says:
25 European languages, about 500 MB, Apple silicon

Cohere Transcribe

What the README says:
14 languages including Japanese, Korean, Arabic and Mandarin; about 1.4 GB; Apple silicon

Apple Speech

What the README says:
Built into macOS, no download; languages depend on your system; Apple silicon and Intel

Whisper Tiny to Large

What the README says:
Up to 99 languages; about 75 MB to 2.9 GB; Apple silicon and Intel

Fluid-1 cleanup (optional)

What the README says:
Not a speech model; the README lists about 3.5 GB of disk, the FAQ gives 600 MB for Pico and under 1.5 GB for Mini

Other requirements: microphone access, accessibility permission to type into other apps, and about 1 GB of disk for a voice model. No RAM figure is published.

How to Get Started

The steps follow the project's README; PromptQuorum has not run them.

  1. 1
    Install FluidVoice
    Why it matters: Run `brew install --cask fluidvoice` or open the .dmg from the Releases page.
  2. 2
    Grant permissions
    Why it matters: Allow microphone and accessibility access; both are required for dictation and for typing into other apps.
  3. 3
    Set a hotkey and a model
    Why it matters: Pick a global hotkey, then choose a voice model during onboarding; Apple Speech needs no download, the others download once.
  4. 4
    Dictate
    Why it matters: Click into any text field, use the hotkey and speak; enable Fluid-1 or add a cloud provider key only if you want the text cleaned up.

Features Confirmed by the Sources

Every item below comes from the website, the privacy policy or the README; none has been independently tested.

  • Three modes. Dictation types text; Write Mode writes or rewrites text in a field you select; Command Mode controls the Mac by voice.
  • Per-app tone. You can assign different prompts to different apps, and the app detects which one is focused.
  • Live preview. A transcription overlay, with a notch-aware layout, shows words as you speak.
  • Dictionary and history. A custom dictionary learns spellings; transcription history and audio history are optional and stored locally.
  • Hold Space to Mute. An opt-in setting pauses a hands-free recording while you hold Space.
  • FluidMeet. Version 1.6.12 includes meeting transcription with speaker names, Markdown, text or JSON export, and summaries that run on the Mac or with your own provider.

Data Flow and Privacy

The privacy policy, last updated on 5 July 2026, says FluidVoice needs no account and does not collect or send raw audio, transcription text, selected text, prompts, window titles or file paths to FluidVoice-operated servers, and does not use them to train AI models. Fluid-1 inference runs locally; its model file is downloaded from a host such as Hugging Face, which retrieves weights only. Preferences stay on the Mac, provider API keys go into the macOS Keychain, transcription history is a setting you can turn off, and audio history is off by default.

The app does use the network in three cases: downloading models, checking for updates and sending anonymous or pseudonymous analytics, which Settings can disable. The analytics fields listed are app and macOS version, chip family, a random install identifier, feature flags, word-count ranges and latency figures. If you enable a cloud provider, the text needed for that feature goes to the provider you configured, under its own privacy policy.

📌Note: These are the maintainers' declarations, not audit results. The app's code is public, but PromptQuorum has not read it or inspected the app's network traffic, and the Fluid-1 runtime is not open source.

What We Could Not Verify

  • Speed and accuracy. The website's figures (under 100 ms perceived latency, speaking about 3.7 times faster than typing, 350,000+ downloads) and the user quotes are the project's own; PromptQuorum measured none of them.
  • Fluid-1 quality. How well the local cleanup model preserves meaning, and how it compares with cloud providers, was not tested.
  • Intel performance. Intel Macs are limited to Whisper and Apple Speech, and their speed was not checked.
  • Windows and iPhone. Both are beta; their feature set, models and data handling were not reviewed.
  • Installer signing. Whether the .dmg is notarized by Apple was not checked.

Competitors and Alternatives

Handy

Platforms:
macOS, Windows, Linux
Price / license:
Free / MIT
Key difference:
Cross-platform push-to-talk dictation with local Whisper and Parakeet; no built-in local cleanup model
Articles about Handy (2)

Also mentioned in:

+5 more not shown

MacWhisper

Platforms:
macOS
Price / license:
Freemium / Proprietary
Key difference:
Mac app with a one-time purchase for local Whisper and Parakeet transcription; closed source
Articles about MacWhisper (2)

Also mentioned in:

AquaVoice

Platforms:
macOS, Windows, iOS, Android
Price / license:
Freemium / Proprietary
Key difference:
Cloud dictation service that types into the active app, so it is not an offline option
Articles about AquaVoice (2)

These are same-category apps from the PromptQuorum directory; confirm each app's current price, license and platforms on its own page.

Compare in directory

Frequently Asked Questions

Can I use FluidVoice on an Intel Mac?

Yes, but only with Whisper models or Apple Speech; the README says Intel support arrived in version 1.5.1 and that every other model needs Apple silicon.

Does the local cleanup need an account or an API key?

No: the README says Fluid-1 needs no cloud and no API keys. Keys are only for the optional OpenAI, Groq or custom providers and are stored in the macOS Keychain.

What is FluidMeet?

A meeting feature inside the app: it transcribes calls, labels speakers, exports Markdown, text or JSON, and lets Claude Code, Codex and other coding agents read saved meetings through an Integrations setting.

How do I avoid look-alike download sites?

The project says fluidvoice.org is not official and points to its GitHub Releases page for official downloads, which is also where the version number in this review comes from.

How do I try beta builds?

Switch on Settings, Automatic Updates, Beta Releases; the Releases page shows such prereleases, for example 1.6.12-beta.1 on 13 October 2026, before a stable version follows.

Verdict

FluidVoice is a free, open-source and actively updated way to dictate into any Mac app, with speech models that by the project's account run on the device and an optional local cleanup step. Against that, it requires macOS 15, its local cleanup model is not open source, the Windows and iPhone versions are still beta, and none of its speed or accuracy claims could be checked here. It suits Mac users who want hotkey dictation without a subscription and can test it on their own hardware; anyone who needs proven accuracy, Intel-Mac speed or a stable Windows version should try it first.

Sources

← Back to Power Local LLM

MCP