The argument in one line.
FluidVoice, a free open-source Mac app, runs both speech recognition and text cleanup entirely on-device, matching or beating paid subscription dictation tools like Wispr Flow without a monthly fee or cloud upload.
Read if. Skip if.
- A Mac developer who dictates code comments, commit messages, or docs and doesn't want a monthly subscription for it.
- Anyone on Apple silicon who's cautious about voice audio or code leaving their machine and hitting a cloud server.
- A Wispr Flow or Superwhisper user curious whether a free local alternative can match the paid cleanup quality.
- You're on Windows or iOS — FluidVoice is Mac-only right now, with those platforms only on a wait list.
- You dictate primarily in a language other than English — you may need extra tuning to get accurate results.
The full version, fast.
FluidVoice is a free, open-source Mac dictation app that runs entirely on-device: a local model called Parakeet transcribes speech, and a second local model, Fluid Intelligence, cleans up capitalization, punctuation, and structure before dropping the result into whatever app has focus. In a live demo it produced correctly formatted, comment-ready text with no lag, and it runs roughly four times faster than cloud competitors like Wispr Flow and Superwhisper because there's no network round-trip. The tradeoffs: it's Mac-only (Windows/iOS are wait-listed), the cleanup model is a ~3.5GB download, and it's slower on Intel Macs. The recommended move is to run it as your everyday Mac tool and keep a cross-platform tool around for when you're not on a Mac.
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
01 · Why Mac dictation isn't good enough
Apple's built-in dictation is fine for texts but fails at capitalization, punctuation, and structure for anything someone will actually read.
02 · FluidVoice: a free, open-source alternative
Introduces FluidVoice — free, open source, and fully on-device, addressing both the privacy and cost objections at once.
03 · How it works: Parakeet + local AI
Parakeet transcribes speech locally; Fluid Intelligence cleans it up; one hotkey drops the result into whatever app has focus.
04 · Live demo in VS Code
One-line install, mic/accessibility permissions, then a live hotkey dictation into a Claude Code terminal that produces clean, comment-formatted text with no visible lag.
05 · FluidVoice vs Apple Dictation, Wispr Flow, Superwhisper
FluidVoice runs about 4x faster than the cloud tools; Wispr Flow and Superwhisper are accurate but paid and cloud-dependent.
06 · Benefits: free, private, fast
Near-instant on Apple silicon; the 'weakest' option on paper turns out to be the only free, open-source, locally-formatted one.
07 · Limitations: Mac-only, model size, Intel support
Windows/iOS are wait-listed, the cleanup model is a ~3.5GB download, Intel runs slower, and non-English dictation may need tuning.
08 · Is it worth it for Mac developers?
A conditional verdict: worth it specifically if avoiding another subscription matters to you.
09 · Best way to use it (dual-tool workflow)
Run FluidVoice as the everyday Mac tool, keep a cross-platform tool around for when you're not on a Mac.
10 · GitHub link and Homebrew install command
Closing CTA: GitHub repo and brew install command in the description, subscribe ask, silent Better Stack end-card bumper.
Lines worth screenshotting.
- FluidVoice runs entirely on-device: a local model called Parakeet transcribes speech, and a second local model, Fluid Intelligence, cleans up punctuation and structure — no audio ever reaches a server.
- The free, open-source dictation tool beats the paid ones on the two things developers actually complain about: privacy and a recurring bill.
- FluidVoice runs roughly four times faster than competing dictation tools because there's no cloud round-trip to wait on.
- The cleanup model alone is about a 3.5 GB download, so setup takes longer than a typical one-line install even though the install command itself is a single line.
- On Apple silicon (M-series chips), the local models run fast enough that dictation feels instantaneous; on Intel Macs it still works, just slower.
- FluidVoice is Mac-only for now — Windows and iOS support exists only as a wait list, so Windows users still need a separate tool.
- Apple's free built-in Mac dictation is fine for a text message but doesn't capitalize, punctuate, or structure text well enough for a commit message or a document someone will actually read.
- A single hotkey captures speech and drops cleaned-up text directly into whatever app has focus — a terminal, Slack, Notes, or an AI coding tool — without picking a target app first.
Two small local models can replace one paid cloud subscription
FluidVoice shows that chaining two small on-device models — one for transcription, one for cleanup — can match paid cloud dictation tools on quality while removing the monthly fee and the privacy risk.
- Apple's built-in Mac dictation is fine for a text message, but it doesn't capitalize, punctuate, or structure a sentence well enough for a commit message or a document someone will actually read.
- The two real objections developers have to dictation tools are privacy (recordings leaving the machine) and cost (another monthly subscription) — not accuracy.
- FluidVoice is free and open source, and every part of the speech-to-text process happens on the Mac itself — nothing is uploaded to a server.
- Running entirely on-device removes both developer objections at once: nothing leaves the machine, and there's no recurring bill.
- A local model called Parakeet transcribes raw speech, then a second local model, Fluid Intelligence, acts as an editor — fixing capitalization, punctuation, and structure before the text lands.
- One hotkey drops the cleaned-up text into whatever app has focus — a terminal, Slack, Mail, or Notes — so there's no per-app integration to set up.
- Setup is a one-line install followed by granting mic and accessibility permissions — the same ritual every Mac app requires, not extra friction specific to this tool.
- In a live demo, holding the hotkey and speaking a full sentence into a code editor produced correctly capitalized, punctuated text formatted like an actual code comment with no visible lag.
- FluidVoice runs about four times faster than the cloud competitors because there's no network round-trip to wait on.
- Wispr Flow and Superwhisper are accurate paid tools, but they still cost a subscription and still send audio to the cloud — for many developers that second point is the dealbreaker before price even comes up.
- On Apple silicon, the local models run fast enough that the dictation is effectively instant.
- The tool expected to be the weakest option (free, open source) turned out to be the only one doing smart local formatting without a subscription or a cloud trip.
- FluidVoice is Mac-only right now — Windows and iOS exist only as a wait list — and the cleanup model alone is about a 3.5 GB download, so setup takes longer than the one-line install suggests.
- It works on Intel Macs, just slower than Apple silicon, and dictating in a language other than English may need extra tuning to get accurate results.
- The honest verdict is conditional, not a blanket yes: it's worth it specifically if avoiding another monthly subscription matters to you.
- The recommended setup is to run FluidVoice as the everyday tool on Mac while keeping a cross-platform tool around for when you're not on a Mac — combining free local performance with platform coverage.
Terms worth knowing.
- Parakeet
- The local speech-recognition model FluidVoice uses to convert spoken audio into raw text entirely on-device.
- Fluid Intelligence
- FluidVoice's second local model, which edits Parakeet's raw transcript — fixing capitalization, punctuation, and structure before it's inserted.
- Wispr Flow
- A paid, subscription-based cloud dictation tool that FluidVoice is positioned against in this video.
- Superwhisper
- Another paid dictation competitor mentioned as accurate but still cloud-dependent and subscription-priced.
Things they pointed at.
Lines you could clip.
“You'd think the free thing we already have on Mac would be good enough, but it's not.”
“This is actually four times faster than using other ones already out there.”
“On Apple silicon, it's fast enough that you can forget it's even thinking.”
“The free one that was supposed to be the weakest is the only one that's free open source and does the smart formatting right there on our machine.”
Word for word.
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
The bait, then the rug-pull.
Every Mac already has a dictation button — and it's not good enough for real work. This video tests the free, open-source app built to replace it, and the subscription tools built on top of it.
Named ideas worth stealing.
FluidVoice's two-model pipeline
- Parakeet — local speech-to-text
- Fluid Intelligence — local cleanup & formatting
Speech runs through two on-device models in sequence: one converts audio to raw text, the second acts as an editor fixing capitalization, punctuation, and structure before the hotkey drops it into the active app.
How they asked for the click.
“The GitHub link and the brew command are both in the description below. If you enjoy coding tips and tricks like this, be sure to subscribe to the BetterStack channel.”
single closing line pairs the tool's install link with a subscribe ask — no hard sell, no separate spoken sponsor read despite the channel's own Better Stack product appearing as a silent end-card bumper
Visual structure at a glance.
