The open-source Wispr Flow alternative
You talk. It types.
Hold a key, say the thing, let go. Punctuated text appears wherever your cursor already is, in any app. Free and open source for macOS and Linux, and every word is transcribed on your own machine.
$0 forever No account No cloud, ever MIT licensed
Everyone else: $12 to $25 a month This one: $0
Real time, no cuts. The key goes down, the words show up, the message gets sent.
Pricing
Don't rent your own voice.
Dictation apps want $12 to $25 a month to send your audio to a server you don't own. Your laptop can already do this. It does it while the Wi-Fi is off. So the price here is nothing, and it stays nothing.
-
No subscription, no trial, no card
There's no paid tier to upgrade to. The whole app is the free tier.
-
No account to create
Install it, hold the key, talk. Nothing to sign up for, nothing to log into.
-
No usage limits
No weekly word caps. Dictate a novel on a plane if you like.
Wispr Flow Pro is listed at $12/mo billed annually and $15/mo month-to-month (wisprflow.ai/pricing, checked August 2026). Other cloud dictation apps land somewhere between $8 and $25 a month.
How it works
Three seconds to learn.
Nothing to memorise. You never say "period" or "new paragraph". Talk the way you'd talk to a person and the punctuation shows up on its own.
Hold the key
Right ⌘ on macOS, Right Alt on Linux, or whichever key you'd rather use. A small pill shows up to say it's listening.
Say the thing
Ramble, pause, restart halfway through. The mic keeps a 300 ms pre-roll, so your first syllable never gets eaten.
Let go
Punctuated, capitalised text lands wherever your cursor already was. Words show up live while you're still talking.
Every app you already use
If you can type there, you can talk there.
WhisprCatch types at the operating-system level. No plugin, no extension, no per-app integration. It just goes into whatever window has focus.
Privacy
Your voice never leaves the room.
The speech model runs as int8 ONNX on your own CPU. Not "encrypted in transit". Not "deleted after 30 days". Your audio is decoded in memory, never written to disk, and never sent anywhere at all.
-
Works with the Wi-Fi off
Once the model is downloaded, dictation is fully offline. Try it on a plane.
-
No telemetry, not even the "anonymous" kind
The app phones home for nothing. There's no analytics SDK in the binary to argue about.
-
Don't trust us, check
Every line is MIT-licensed. Read the audio pipeline, or just watch the network tab stay empty.
Run whisper-catch doctor
$ whisper-catch doctor
microphone granted
input monitoring granted
accessibility granted
model parakeet-0.6b · ready
inference cpu · int8 · onnx runtime
network not used
hold right cmd and speak.Two models to pick from: NVIDIA Parakeet 0.6B (~660 MB, the accurate one) or Moonshine base (~64 MB, the tiny one). Swap in settings whenever.
Honest comparison
WhisprCatch vs. cloud dictation.
Paid tools like Wispr Flow are genuinely good, and they do a few things we can't. Here's the whole table, including the rows where we lose.
| WhisprCatch | Typical cloud dictation app | |
|---|---|---|
| Price | $0 forever | $12 to $25 a month |
| Cost over 3 years | $0 | $430 to $900 |
| Where your audio is transcribed | Your CPU | Their servers |
| Works offline / on a plane | Yes | No |
| Account required | None | Yes |
| Source code | MIT, on GitHub | Closed |
| Linux | Yes, .deb | Rarely |
| macOS | Yes, Homebrew | Yes |
| Windows, iOS, Android | Not yet | Usually |
| AI tone & style rewriting | No, it types what you said | Yes |
| Learned personal dictionary | Not yet | Yes |
| Usage limits | None | Weekly word caps on free tiers |
Want the long version? Read the full Wispr Flow alternative comparison →
Install
One command. One minute.
A first-run wizard handles permissions and downloads the model. Then you hold a key and talk.
macOS
Apple Silicon, macOS 11+. Installs and updates through Homebrew.
Grant Accessibility, Input Monitoring and Microphone, then quit and reopen once. macOS only re-reads permissions at launch. After that, hold Right ⌘ and talk.
Linux
Ubuntu & Debian, x86-64. Download the .deb, or install it from the terminal.
Download .debOne polkit prompt for keyboard access, then hold Right Alt and talk. Works on X11 and Wayland.
On Fedora, Arch, or an older Mac? It builds from source with
cargo build --release -p whisper-catch.
Build instructions →
FAQ
Fair questions.
Is there a free, open-source alternative to Wispr Flow?
Yes. WhisprCatch works the same way: hold a key, talk, let go, and the punctuated text is typed wherever your cursor is. The difference is where it runs. Everything happens on your own machine, it costs nothing, there's no account, and the whole thing is MIT-licensed on GitHub. It also has a real Linux build, which most paid dictation apps don't.
How much does WhisprCatch cost?
Nothing, and there's no upgrade to sell you. No paid tier, no trial, no card. Cloud dictation apps generally charge $12 to $25 a month, which works out to somewhere around $430 to $900 over three years.
Does my voice leave my computer?
No. The speech model runs on your CPU as int8 ONNX. Nothing is uploaded while you dictate, audio is never written to disk, and once the model is downloaded the app works with the Wi-Fi off.
Which platforms does it support?
macOS 11 and up on Apple Silicon through Homebrew, plus Ubuntu and Debian on x86-64 through a .deb. Everything else can build from source. No Windows, iOS or Android yet, and Intel Macs are still on the list.
Do I need a GPU or an internet connection?
Neither. The models run int8 on the CPU at roughly 25 times realtime, so a laptop handles it fine. The one time WhisprCatch touches the network is the model download during setup.
Which speech recognition model does it use?
Either NVIDIA Parakeet 0.6B (about 660 MB, the more accurate one) or Moonshine base (about 64 MB, tiny and quick), both as int8 ONNX via transcribe-rs. macOS starts on Moonshine, Linux starts on Parakeet, and you can switch in settings.
Which apps does it type into?
Whichever one has focus. Keystrokes are injected at the operating-system level, so there's no plugin or integration to install. Editors, browsers, chat apps, terminals, note apps, AI chat windows: if you can type there, this works there.
What do I need to do on macOS after installing?
Grant three permissions (Accessibility, Input Monitoring, Microphone), then quit and reopen the app. macOS only re-reads permissions when an app starts, which trips up a lot of people. The first-run wizard opens each pane for you, and whisper-catch doctor tells you which grants are live.
What can Wispr Flow do that WhisprCatch can't?
Quite a bit. It has Windows, iOS and Android apps, AI rewriting that adjusts your tone, a dictionary that learns your jargon, snippets, and team billing. WhisprCatch is macOS and Linux only, and it won't ship your text off to a model to be reworded. If you want cross-device sync and AI rewriting, pay for the cloud tool. If you want free, private, offline dictation, take this one.
Is it really open source?
Yes, MIT-licensed, all of it, at github.com/AviroopPaul/whisper-catch. Read the audio pipeline, check for yourself that nothing phones home, build it, fork it.
Go on then
Stop paying to talk.
Free, open source, and running on your own machine in about a minute.