Say it once.
It's already written.
Hold fn, speak, let go. Kudu writes it in SlackGmailCursorNotionWhatsAppVS Code
Slack, clean and punctuated, and your voice never leaves your computer.
Works wherever you can type
Slack
Gmail
Notion
VS Code
Cursor
WhatsApp
Outlook
Linear
Figma
GitHubSlackGmailNotionVS CodeCursorWhatsAppOutlookLinearFigmaGitHub
Discord
Google Docs
Obsidian
Teams
Telegram
Word
Jira
ChatGPT
Chrome
MessagesDiscordGoogle DocsObsidianTeamsTelegramWordJiraChatGPTChromeMessages…and every other app with a text field. No plugins, no setup.
Try it
Hold the key.Say it messy.
Let go, and Kudu writes what you meant, formatted for the app you're in.
Press and hold fn, or the space bar
Slack# launch
Ravi 10:42
Are we still on for Friday?
You said…
Press and hold fn, or the space bar
Speed
Three times faster than typing.
Hey team, quick update on the launch. The landing page needs a final review, pricing is updated with the new plans, and the announcement goes out Thursday at 3:30.
Hey team, quick update on the launch. The landing page needs a final review, pricing is updated with the new plans, and the announcement goes out Thursday at 3:30.
Both start together…
0 s
typical wait from letting go to text, on an M-series Mac
0 bytes
of your voice sent to any server. Ever.
0%
open source, MIT licensed, free to use and change
Every app, its own voice
Fast. Accurate. Everywhere.
How it works
Talk like a person. Read like a writer.
Speak the way you think
Ramble, pause, change your mind halfway through. Kudu drops the ums and the false starts, and when you say “no wait”, it keeps what you meant.
Formatted while you speak
It learns your words
You said
um let's meet at five no wait at six and uh bring the the the slides
Kudu wrote
Let's meet at 6 and bring the slides.
Made for how you write
Everything it needs, nothing it doesn't.
hey can't make it today, tomorrow work?
Command mode
Select text, hold the Command shortcut and say what to do: make it formal, shorter, fix the grammar, or draft a reply from scratch.
Hey, are you free for lunch tomorrow? Let's do 12 if that works for you.
A style for every app
Formal in email, relaxed in chat, all lowercase with friends. Kudu knows which app you're in.
cal.com/manas/30min
Snippets
Say a trigger, get your link, address or sign-off.
Your shortcut
fn, Ctrl + Win, Right Alt, anything you like.
Works offline
The speech model runs on your computer. Planes, trains, no Wi-Fi: all fine.
148
wpm average
12 days
current streak
41,208
words this year
Insights
Words per minute, words cleaned up, your streak, and where you dictate the most. Kept on your computer, like everything else.
Privacy
Your voice stays home.
Most dictation apps send your voice to a server to be transcribed. Kudu never does. What you say, and what it becomes, stays on your computer.
Read the privacy policy →Transcribed on your computer
Speech recognition runs locally: the Apple Neural Engine on a Mac, your CPU on Windows and Linux.
No account, no cloud
Nothing to sign up for. The only network use is downloading the speech model once.
No analytics, no telemetry
Kudu doesn't track you, phone home, or report crashes anywhere.
Open source, MIT
Every line is on GitHub. Read it, build it yourself, change it.
Download
On the computer you already use.
A native Mac app, built in Swift. Speech runs on the Apple Neural Engine.
- Apple Silicon (M1 or newer)
- macOS 14 Sonoma or later
- Smart formatting with Apple Intelligence on macOS 26
A light background app with a tray icon. Speech runs on your CPU, no GPU needed.
- Windows 10 or 11, 64-bit
- Any 4-core CPU from the last ten years
- No GPU, no Visual C++ install needed
Works on X11 and Wayland: GNOME, KDE Plasma, Hyprland, Sway and more.
- .deb, .rpm and .tar.gz packages
- X11 and Wayland (GNOME, KDE, Hyprland, Sway)
- Ubuntu 22.04+, Debian 12+, Fedora 38+, Arch
Or install from a terminal (Terminal):
curl -fsSL https://github.com/manasdutta04/kudu/releases/latest/download/install-macos.sh | bashQuestions
Good questions.
Yes. Kudu is open source under the MIT license: no subscription, no account, no word limits. The speech models it downloads are open models too.
Stop typing. Start saying.
Free for macOS, Windows and Linux. Set up in a minute, works offline after that.