The argument in one line.
A free, open-source, on-device dictation app can match paid cloud subscriptions like Wispr Flow and SuperWhisper on speed and real-time accuracy, while removing the privacy risk and the monthly fee.
Read if. Skip if.
READ IF YOU ARE…
- You currently pay $10-15/month for a cloud dictation app like Wispr Flow or SuperWhisper and want a free, private alternative that matches it on speed.
- You're a Mac user comfortable installing an open-source app via Homebrew or a GitHub release and tweaking settings like hotkeys and voice engines.
- You dictate emails, documents, or notes daily and want AI-powered cleanup and a custom dictionary for names and company terms built into your workflow.
SKIP IF…
- You're on Windows or Linux — the app demoed here is macOS-only.
- You need enterprise support, SSO, or team admin controls — this video only covers the individual, self-hosted setup.
The full version, fast.
Two hosts compare FluidVoice, a free open-source Mac dictation app that runs its speech model entirely on-device, against the paid cloud tools Wispr Flow ($15/mo) and SuperWhisper ($10/mo). In a live side-by-side test on an identical sentence, FluidVoice keeps pace with Wispr Flow while showing a real-time word-by-word preview neither paid app offers. The hosts walk through picking a voice-engine model (fast vs. high-accuracy), building a custom dictionary so names and company terms transcribe correctly, and routing dictated text through configurable AI prompts for automatic cleanup. By the end, one host has already uninstalled SuperWhisper. The core conclusion: the underlying speech models are already open or commercially available — the paid apps are selling a polished wrapper, and that wrapper no longer justifies the subscription.
Free for members
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.
00:00 – 00:09
01 · Intro
Andrew previews a free, open-source, on-device dictation app.
00:09 – 00:27
02 · Why Andrew is replacing paid dictation apps
Andrew and Adam compare what they currently pay for Wispr Flow and SuperWhisper.
00:27 – 00:36
03 · Adam's SuperWhisper setup
Adam explains his current paid setup and monthly cost.
00:36 – 01:03
04 · Why Andrew wants a local alternative
Andrew lists his three objections to cloud dictation: privacy/subpoena risk, cloud latency, and weaker UX.
01:03 – 01:21
05 · Introducing FluidVoice
FluidVoice surfaced after the hosts' GitHub 'top repos of the week' show and shot up in stars.
01:21 – 01:39
06 · Adam's first reaction after trying it
Adam describes starring the repo immediately after testing it.
01:39 – 02:15
07 · Installing and customizing FluidVoice
Andrew downloads it for Mac, picks the 'blazing fast' voice engine, and sets hotkeys.
02:15 – 02:42
08 · Wispr Flow speed test
Andrew dictates a test sentence into Wispr Flow and times how long it takes to appear.
02:42 – 03:09
09 · FluidVoice speed test
Andrew dictates the same sentence into FluidVoice, showing the live word-by-word preview.
03:09 – 03:18
10 · Why real-time text preview matters
Andrew highlights being able to see the words land in real time versus just a waveform.
03:18 – 04:03
11 · Open source control and customization
Adam explains that the underlying models are open/available and paid apps just wrap them; FluidVoice lets you keep the source forever.
04:03 – 04:30
12 · Using AI prompts to clean up dictation
Dictated text can be routed through a configurable AI prompt (e.g., format as an email) before pasting.
04:30 – 05:15
13 · Custom dictionaries for names and company terms
Adam shows adding coworker and company names to the custom dictionary so they transcribe correctly.
05:15 – 05:33
14 · Choosing between faster and more accurate models
Adam tested the higher-accuracy Cohere model but stuck with Blazing Fast since he couldn't tell the difference.
05:33 – 06:27
15 · Installation delay to watch for
Both hosts hit a stalled install progress bar that silently resumed after several minutes.
06:27 – 06:54
16 · Why Adam removed SuperWhisper
Adam says he's already uninstalled SuperWhisper after zero issues with FluidVoice.
06:54 – 07:04
17 · Watch the GitHub repo show next
Sign-off pointing to the GitHub repo show where FluidVoice was first surfaced.
Lines worth screenshotting.
- A $10-15/month cloud dictation subscription carries three costs at once: the fee, sending every spoken word to a third-party server, and the latency of a round trip to that server.
- The underlying speech-to-text models used by paid dictation apps are open or commercially available to anyone — what you're actually paying for is the polished Mac wrapper around them.
- Open source means you keep the exact version you have forever, even if the company behind a paid alternative disappears, pivots, or starts charging more.
- A live real-time text preview — seeing your own words appear as you speak instead of a moving waveform — proves a dictation tool heard you correctly before you even stop talking.
- Building a custom dictionary of the specific names and company terms you say every day is the single highest-leverage setup step for any dictation tool, paid or free.
- A model that scores a couple of points higher on accuracy but runs slower rarely wins in daily use, because most people can't perceive that small a difference.
- A progress bar stuck at a low percentage during a local AI model download isn't necessarily a failed install — some downloads silently resume and finish minutes later.
- Once a free, open-source alternative matches a paid tool's reliability, the paid tool's remaining case shrinks down to convenience features alone.
Open source dictation now matches the paid apps
LOCAL VS CLOUD
A free, open-source, on-device dictation app running on this Mac matched paid cloud tools like Wispr Flow and SuperWhisper on speed and real-time accuracy, while removing the privacy risk and the monthly fee.
02Why Andrew is replacing paid dictation apps
- A $10-15/month cloud dictation subscription carries three costs at once: the fee, sending every spoken word to a third-party server, and the latency of a round trip to that server.
04Why Andrew wants a local alternative
- Local, on-device speech models remove the subpoena and data-retention risk of cloud dictation entirely, since audio never leaves the machine.
05Introducing FluidVoice
- Actively surfacing under-starred open-source repos, not just the already-popular ones, is how you find genuinely useful tools before they blow up.
07Installing and customizing FluidVoice
- A free, open-source tool can match a paid app's setup polish — choice of voice engine, hotkey, and continuation key — so 'open source' doesn't have to mean 'less configurable.'
08Wispr Flow speed test
- The only fair way to compare two dictation tools is a live side-by-side test on the identical sentence, not marketing claims or isolated demos.
09FluidVoice speed test
- Seeing your own words appear on screen as you speak, not just a moving waveform, is what proves a dictation tool actually heard you correctly.
10Why real-time text preview matters
- A visible live transcript removes the 'is this even working' anxiety, which is a bigger everyday usability problem than any small accuracy gap between models.
11Open source control and customization
- The underlying speech-to-text models are open or commercially available to anyone; what paid dictation apps actually sell is the polished wrapper around them.
- Open source means you keep the version you have forever, even if the company behind a paid alternative disappears, pivots, or starts charging more.
12Using AI prompts to clean up dictation
- Routing raw dictated text through a configurable AI prompt before it lands in the target app can auto-strip filler words and apply formatting rules you set once.
13Custom dictionaries for names and company terms
- Building a custom dictionary of the specific names and company terms you say every day is the single highest-leverage setup step for any dictation tool, paid or free.
14Choosing between faster and more accurate models
- A model that scores a couple of points higher on accuracy but runs slower rarely wins in daily use, because most people can't perceive that small a difference.
15Installation delay to watch for
- A progress bar stuck at a low percentage during a local model download isn't necessarily a failed install — some downloads silently resume and finish minutes later.
16Why Adam removed SuperWhisper
- Once a free, open-source alternative matches a paid tool's reliability, the paid tool's remaining case shrinks down to convenience features alone.
Terms worth knowing.
- On-device (local) speech model
- A speech-to-text model that runs entirely on your own computer instead of sending audio to a cloud server, so no recording ever leaves the machine.
- Custom dictionary (dictation)
- A per-app word list you add to a dictation tool so it correctly transcribes specific names, products, or company terms it would otherwise misspell.
- AI prompt cleanup
- An automatic step where raw dictated text is passed through a configurable AI instruction (e.g., 'remove filler words, format as an email') before it's pasted into the target app.
- Voice engine (model tier)
- The specific speech-recognition model a dictation app uses to convert audio to text; different tiers trade transcription speed against accuracy.
Things they pointed at.
Lines you could clip.
00:44
“I don't like it for a few reasons. Number one, I'm sending it up into the cloud... who knows who's gonna subpoena what that I said.”
03:31
“The models are either open or available, and these companies have packaged it into a nice Mac app for you. Fluid Voice is open source.”
05:57
“I've already removed Super Whisper from my computer.”
04:49
“It's never gonna spell your name wrong when I speak to it because it knows exactly how Andrew Warner is spelled.”
Word for word.
Read-along
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.
00:01This little dictation app is free, open source, and uses AI on your computer, and I'm going to show you how you can install it and use it right now. Presented by Zapier, the AI automation company.
00:12Adam, everyone's using audio dictation apps on their computers. They're paying $15 a month for it. I'm using WhisperFlow,
00:18and we've got a better solution, one that's completely free and works way better. What do you use? What do you pay?
00:24I use this thing called Super Whisper. It's using all the same models that all these other tools use. And I'm paying it's about $10 a month, uh, but I don't like to think about it too much.
00:3510 is fine. This is I pay $15
00:37a month because I pay, um, monthly for this. I don't like it for a few reasons. Number one, I'm sending it up into the cloud, which I don't have any worries about any private information, but who knows who's gonna subpoena what that I said, numb number one.
00:48Number two, if it's going up into the cloud, it's taking longer. And number three, I happen to believe that there's a better user experience, and WhisperFlow is not giving it to us, and neither are these other apps. So you and I did the GitHub show, and we talked about the top, uh, repos of the week.
01:03And then I said, I've gotta attach something to it that people are not, like, starring enough, and it's this. It is something called Fluid Voice.
01:11Look. It shot up after we did our GitHub show, which is incredibly popular. If you're not watching it, folks, you gotta watch it.
01:17We reveal these types of things all the time. I'm I'm one of those stars, Andrew. I'm that star right at the kinda top third there.
01:23Was like, oh, man. After I tried it and I started using it, immediately came back to the to the repository,
01:28gave it a star because I I think these guys are gonna be really successful. It's an awesome product. It's fantastic because
01:35you get to customize it. You you get to download this or get the whole, uh, the whole, um, what is it called?
01:42The code base, and you could change it to however you want. Or you could just go to their website, and you can get it. I got it and downloaded it for Mac.
01:49The setup is super duper easy. I then once I had it, I got to pick which voice engine I wanted. And I said, I want the blazing fast one, but you can go even, um, you can go for whatever model you like.
02:00And then you can customize it here by telling it what you want to use as your hotkey, what you want to use as your continuation key. It will even do some edits on your text if you want it.
02:11Let me show you what it looks like compared to WhisperFlow. On WhisperFlow, I'm gonna hit the f n key, I'm gonna say, my name is Andrew, and I'm just gonna be talking directly into the screen, and I'm gonna talk super fast because I'm a New Yorker at heart, and I will always talk super fast.
02:25And you can see on the very bottom of my screen how all my WAV files are coming in. Let's see how long it takes for it to come through. Let go.
02:31It's thinking. Not not bad. Right?
02:34Yeah. That's good. I think it basically got it.
02:36Now I'm going to switch over to, um, Fluid, and watch this. My name is Andrew, and now I'm gonna be using this Fluid Voice thing. And notice on the bottom of my screen, you cannot just see that it's typing with the WAV thing, but it's also showing the words, which lets me see that what I'm saying is actually going through.
02:54And it's letting me see what kind of features I've got on and where it's planning to to paste it, in this case, into, um, Chrome. Let's let go of the button, the hotkey, and let's see how fast it pasted in.
03:07That's amazing. Right. And the fact that I could see it in real time is fantastic.
03:11Really good piece of software. I'm so glad that that we got to share this with people. What do you have to add on this?
03:17You've been using this a lot longer than I have. I mean, there's there's so much. So so number one, I think you glossed over something I think it's really important here.
03:24You know, Whisper Flow, Super Whisper, all these other ones, they're all companies Mhmm. That are charging us to use this software. The models are are either open or available, and these companies have packaged it into a nice Mac app for you.
03:36Fluid Voice is open source. You could download the source. You could save it forever.
03:41You don't ever have to pay these guys money. Other people are gonna come in here and and contribute to it. They're gonna change it.
03:47There's gonna be versions of this. The fact that this is open means that this is free, and it's going to be free. And you get the control.
03:54Now you don't have to customize it if you don't want to, but you could. But what you've already pointed out is when you go into the settings, there's a lot of custom available there, including, you know, running all of your transcriptions through AI prompts before Right.
04:11They get pasted into the application. And so you could have set up, I wanna I wanna prompt for writing emails. Well, now it's gonna remove your ticks.
04:20It's gonna add, you know, a new line after dear Andrew. I love the show that we do together. It'll add the the new line there.
04:27It'll add the comma, and you can customize those prompts. You can also customize the dictionary.
04:32I've already when I get one of these tools, the first thing I do is I drop into the custom words, and I add in all the names of the people that I work with that I say out loud all the time. I add the names of the companies that I work with because they're all spelled weird. And immediately, once you put them in this dictionary, again, with Fluid Voice is gonna use AI to plug in those words.
04:54And so it's never gonna spell your name wrong when I speak to it because it knows exactly how Andrew Warner is spelled. It knows how my name is spelled. My partner, Jesse Pucci, it knows all of that.
05:04Is it j e s s I, j e s I e, j e s s e? Like, fluid voice knows. So I've put all of that in in place, and agree.
05:13I'm on the blazing fast model too, but I did actually test out a couple of the other models. Mhmm. I tested one of the even faster ones.
05:20I also tested one that was I don't remember which it was. Yeah. The cohere high accuracy.
05:25It's it got, like, two more accuracy points, but it's a little slower. And I decided, no.
05:31I can't tell a difference. And so I'm I'm on the same one that you're on. The the only weird thing, you know, when I first downloaded this, I installed it.
05:39I clicked blazing fast. It started to install. And then nothing happened.
05:43And nothing happened. And I actually wrote you a Slack message. It was like, oh, man, Andrew.
05:48Like, I was wrong. I in the video, I said this is gonna be amazing, and it's not.
05:53I'm sorry. But you know what? I waited a couple more minutes.
05:56The thing installed,
05:57and it totally works. And I've I've already removed Super Whisper from my computer. That's a good note because the same thing happened to me.
06:04And because it happened to you, I knew to watch out for it. What happens is it will say 3%, uh, installed, and then it just stops, and it won't do anything.
06:13And then a few minutes later, maybe ten, fifteen minutes later, it shifts from 3% to 98 to a 100, and then it's done. So Yep. That that part wasn't working well.
06:23Everything is Little inconveniences
06:24to me like that are gonna be worth the $10 a month. Plus, like, since it's installed and since it's running, I've had zero issues. Everything has worked great.
06:33The little pop up works great. I love that you can see the text as you're speaking instead of just the little thing that's moving. Because half the time, like, wait.
06:42Can it actually hear me? Is it connected to my Bluetooth headphones that are in the other room? Like like, is this working?
06:48Seeing the the text on the screen is is amazing too. And it's free. Did we mention that it's free?
06:53I love that it's free. I love that it's open source. Fluid Voice is one of many apps that we showed in the GitHub repo show.
06:59Click the link here, and we'll show you the others. See you there.
The Hook
The bait, then the rug-pull.
Andrew Warner opens with a flat claim: the free app beats the $15-a-month one. What follows is a live, unscripted speed test — same sentence, same moment, two apps racing to paste it into the screen — and by the end his co-host has already deleted his paid subscription's app from his computer.
Named ideas worth stealing.
05:15list
Voice engine tiers (speed vs. accuracy)
- Blazing Fast (Parakeet) — English only, fastest transcription
- Cohere High Accuracy — multilingual, a couple points more accurate, noticeably slower
FluidVoice ships multiple downloadable voice-engine models; the two hosts both landed on the fast tier after testing the high-accuracy one and finding the gap imperceptible in daily use.
Steal forpicking a default speech model for any personal dictation or transcription tool
How they asked for the click.
VERBAL ASK
06:54next-video
“Fluid Voice is one of many apps that we showed in the GitHub repo show. Click the link here, and we'll show you the others.”
Soft cross-promotion to their own GitHub-repos-of-the-week show, framed as 'here's where we found this' rather than a hard sales pitch.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
