I dictate almost everything now. Emails, blog drafts, Slack replies, notes to myself in the car. Over the past few months I have put 44,414 words through Wispr Flow at an average of 161 words per minute, which apparently puts me in the top 5% of its users. I mention that not to brag, but so you know this is not a quick first impression. I lived in that app.
So when I say I cancelled my paid plan last week and moved to a free, open source Wispr Flow alternative, it is worth explaining why.
The tool is FluidVoice, and if you are on a Mac, this is the post I wish someone had written for me a month ago.
Why I paid for Wispr Flow for so long
Dictation only works if it is fast and if it gets out of your way. The reason I stuck with Wispr Flow is that it did both. You hold a shortcut, you talk, and reasonably clean text lands wherever your cursor is. It cleaned up filler words, fixed most of my punctuation, and handled the messy way people actually speak.
That polish is the hard part. Plenty of tools can transcribe. Far fewer can take "um so the thing is basically broken and we should probably look at it" and turn it into a sentence you would actually send. Wispr Flow did that well, which is why I happily paid for it.
The catch is that the clever part, the cleanup, ran in the cloud. Your transcribed text leaves your machine to get processed. For most people that is a fair trade. For me it became the one thing I kept thinking about, particularly when dictating anything I would rather not have sitting on someone else's server.
What a local Wispr Flow alternative looks like
FluidVoice is a free, open source Mac dictation app built by a small team called Altic. It is licensed under GPLv3, the code is public, and it has passed 50,000 downloads. On the surface it does the same job as Wispr Flow: one shortcut, talk, text appears in any app.
The difference is where the intelligence lives.
Everything runs on your Mac
FluidVoice does its transcription with local speech models, so the dictation itself never needs an internet connection. The cleanup, the part that actually makes dictation usable, is handled by a local AI model called Fluid-1. No cloud round trip, no API key, no data leaving your Mac.
That last point is what sold me. I have written before about Apple's own move towards on-device versus cloud AI, and the same logic applies here. If a model can run well on the hardware in front of me, I would rather it did.
Fluid-1 is an optional download of around 3.5GB. The app works without it using the raw speech models, but the polish is the reason you would want it, so I would install it.
The Fluid-1 model cleans up as you go
This is where I expected to be let down, and was not. As I talk, text appears, and Fluid-1 tidies it: capitalisation, punctuation, casing on names, the shape of a sentence. It fixes the kind of mess that makes raw transcription painful to read back.
The part that genuinely surprised me was slang. I do not speak in clean, formal sentences, and I throw in plenty of British shorthand and half-words. I expected a local model to choke on that. It did not. It handled my slang better than the cloud tool I had been paying for, which was not the result I went in expecting.
It changes tone depending on the app
FluidVoice notices which app you are dictating into and adjusts the tone to match. Talk into Slack and you get something casual. Talk into Mail and the same words come out more formal. Talk into a code editor or an issue tracker and it leans structured.
You can write your own prompt for each app, so the behaviour is yours to shape. In practice it means I am not constantly rewriting the same dictated thought to fit where it is going, which is a small thing that adds up across a day.
How it held up across a week of real use
I did not run a controlled test. I used it the way I use any dictation tool, which is constantly and without much patience.
Accuracy was the first thing I watched, and it held. Long sessions did not fall apart, names came through correctly more often than not, and the latency was low enough that I never felt I was waiting for the text to catch up. It felt immediate, which is the bar.
Where it earned the switch was the cleanup matching what I was used to. I was braced for the local model to be a clear step down from the paid cloud version, and to have to accept "good enough for free". That is not what happened. For my day-to-day writing it has been level with what I was paying for.
What you need to run it
FluidVoice needs macOS 15 Sequoia or later. It runs on Apple Silicon, where it is tuned to use CoreML and Metal for low-latency transcription, and it also supports Intel Macs by falling back to Whisper models. You can pick faster or more accurate models depending on whether you care more about speed or precision in a given moment.
That Intel support is worth flagging given Apple is dropping Intel Macs in macOS 27. If your Mac is ageing out of Apple's roadmap, a tool that still runs locally on it is more useful than one that does not.
On first launch it asks for microphone access and accessibility permissions, the latter so it can type into other apps. That is standard for this kind of tool, but worth knowing before you install. The Fluid-1 model is the only large download, and again, it is optional.
It is also a reminder that macOS already ships its own Dictation, which I have covered alongside other Mac apps Apple barely mentions. Built-in Dictation is fine for short bursts. It is the cleanup and tone handling that put something like FluidVoice in a different category.
Should you switch from Wispr Flow?
If you are on a Mac and you dictate regularly, I think it is an easy thing to try, because it costs nothing and your data stays put.
I will be honest about the trade. Wispr Flow is a polished commercial product with a team and a roadmap behind it, and there may be edge cases or platforms where it still suits you better. FluidVoice is younger. But for the core job, turning speech into clean, usable text on a Mac without sending anything to the cloud, it matched my needs closely enough that I cancelled my subscription and have not missed it.
That is the part I keep coming back to. I did not switch to save money, although free is free. I switched because a tool running entirely on my own machine did the job as well as the one I was paying a company to run for me. When that happens, the paid option has to justify itself, and for me it stopped doing that.
If you want to try it, FluidVoice is a free download for Mac.
FAQ
Is FluidVoice really free?
Yes. It is free with no paid tiers, and it is open source under the GPLv3 licence, so the code is public. There is no subscription and no trial that expires.
Does FluidVoice work on Intel Macs?
Yes. Apple Silicon Macs get the most optimised experience, but Intel Macs are supported through Whisper models. You will need macOS 15 Sequoia or later either way.
Does my voice or text leave my Mac?
With the local setup, no. Transcription runs on-device and the Fluid-1 model does its cleanup locally, with no API key required. There are optional cloud providers you can switch on if you want them, but the local-only path keeps everything on your machine.