Fuck You FlowOpen model. Public code. Somebody charged you for it anyway.
Free dictation that lives on your PC.
Wispr Flow raised 361 million dollars and hands you 2,000 words a week before the meter starts at 15 dollars a month. We put the same thing behind one key and left the meter off. Greek and English, everything on your PC, nothing in the cloud.
Windows 10 and 11. Version 0.9.0 beta. 1.4 GB with every model inside. No internet connection after setup. MIT license. The SHA256 checksum is on the release page.
Why this is free
The speech model was already open.
OpenAI published Whisper on 21 September 2022. Its own repository says it in one line: Whisper's code and model weights are released under the MIT License.
Public weights, public code, running on a graphics card you already paid for. That has been true for four years.
What the companies added on top was a login, a monthly fee and a counter that stops at 2,000 words a week. Wispr Flow raised 361 million dollars doing it and reached a valuation of 2 billion. The recognition underneath is the same open model, and it was already yours.
So we wrapped it in one key and handed it back. It costs zero because it cost us close to zero. It runs on your machine because that is where it could always run. It asks for no account because we want to know nothing about you.
Take it. Read the code. Fork it. The MIT licence even lets you sell it.
What is inside
Six things the app does on its own, on your machine, without asking anyone.
Greek that holds up
Question marks come from grammar rules or from the rise of your voice. Spoken self-corrections are applied before the text lands.
Snippets
Say a phrase, get a whole block of text. Signatures, addresses, the replies you send every day.
Nothing leaves
No server, no analytics, no account. Password fields are detected and stay empty.
Correct it once. It learns.
Every dictation is kept in History, one file on your disk. Fix a word there and the fix becomes a dictionary rule. The next time you say it, it lands the way you wrote it.
Your words, your rules.
Names, brands, the terms of your trade. Add them to the dictionary and the engine spells them your way, every time. Rules made in History show up here too, ready to edit.
How much you said, how much you saved.
Words per day, median wait, minutes saved against typing. What you said yesterday is one click away. All of it stays in one file on your disk and goes nowhere else.
Every graphics card. It picks the model itself.
NVIDIA, AMD and Intel all run through Vulkan. The first start reads your machine and picks the model that fits it. Change the key, the microphone or the model here whenever you like.
One key.
Press right Alt.
A small window with a waveform appears. It listens.
Speak.
Greek, English or both. Each finished phrase is transcribed while you keep talking.
Press right Alt again.
The text lands where your cursor was. With periods, commas and question marks.
Works in any program: Word, Chrome, Slack, Outlook, even a terminal.
Speed
Measured on an RTX 3070 with the large model. Wait is the time from key release to text on screen.
The wait depends only on the last phrase, because earlier phrases were transcribed while you spoke. AMD and Intel cards run through Vulkan at the same speed. Without a graphics card it runs on the processor, with a wait of a few seconds.
| Words | Speech | Wait |
|---|---|---|
| 32 | 17.5 s | 0.59 s |
| 108 | 82 s | 0.17 s |
| 179 | 77 s | 0.53 s |
| 312 | 152 s | 0.88 s |
The bill
Every row verified. The valuation and the funding are from TechCrunch, 17 August 2026. The prices and the weekly word count are from the Wispr Flow pricing page, read on 6 September 2026.
| Wispr Flow | Fuck You Flow | |
|---|---|---|
| Price | Free to 2,000 words a week, then 12 to 15 dollars a month | 0, forever |
| Words per week | 2,000 on the free plan | As many as you can say |
| Company valuation | 2 billion dollars (August 2026) | One domain name |
| Investor money | 361 million dollars | One weekend |
| Where your voice goes | Their servers | Into the text. Then nowhere. |
| Account | Required | None |
| Internet | Always | Only to download it |
| Greek | One of 100+ languages | First language, mixed with English in the same sentence |
| Source code | Secret | Open, MIT, on GitHub |
The whole comparison, row by row, with the sources: the free Wispr Flow alternative for Windows.
The models that do Greek well have been open and free for four years. Someone had to put them behind a key. We did.
Questions
Is it a free alternative to Wispr Flow?
Yes. Same job, one-key dictation into any Windows program, no subscription, no account, and no weekly word limit. Open source under MIT on GitHub.
Where does my voice go?
Nowhere. Recognition runs on your own machine with the Whisper model. After setup no internet connection is needed. Password fields are detected and left empty.
Does it need an NVIDIA card?
No. NVIDIA, AMD and Intel through Vulkan. Without a graphics card it runs on the processor, more slowly.
Why does Windows show a blue warning during install?
The installer carries no purchased code-signing certificate. Click More info, then Run anyway. The SHA256 checksum is on GitHub.
What does beta mean?
It runs daily on our machines and is now going out to yours. Some programs may refuse the paste, some cards are untested. Whatever breaks, we want it written down.
Why the name?
A dictation company was valued at 2 billion dollars for something open models have done for free for two years. The name is the reply. On social media we write FU Flow.
Tell us
Three doors. Whatever broke, whatever you want, whoever you are.
Something broke
Tell us the program, the graphics card and what you saw. The Diagnostics tab has Recent problems ready to copy.
Report a bug on GitHubSomething you want
A feature, a language quirk, a program it should paste into. Say how you would use it and we read it.
Open an idea on GitHub