Kudu v0.1.2

Private, fast voice dictation. Hold a key, speak, release: clean text appears wherever your cursor is. Speech recognition runs on your own computer. No account, no cloud, no telemetry.

Install

The apps aren't signed with paid Apple / Microsoft certificates yet, so your system will warn that they come from an unidentified developer. The commands below install Kudu and clear that warning. They're also in each download's install script, which you can read before running.

macOS (Apple Silicon, macOS 14+)

curl -fsSL https://github.com/manasdutta04/kudu/releases/latest/download/install-macos.sh | bash

Or by hand: download Kudu-macOS.dmg (or the .zip), drag Kudu into Applications, then run this once in Terminal so macOS lets the unsigned app open:

xattr -cr /Applications/Kudu.app

On first launch, allow Microphone and Accessibility when Kudu asks. Hold fn to dictate.

Windows 10 / 11 (64-bit) · beta

In PowerShell:

irm https://github.com/manasdutta04/kudu/releases/latest/download/install.ps1 | iex

Or download the installer, Kudu-Windows-x64-Setup.exe (when SmartScreen appears, click More info → Run anyway), or the portable Kudu-Windows-x64.zip (in PowerShell in its folder run Unblock-File .\kudu.exe). Hold Ctrl + Win to dictate.

Linux (x86_64, X11 and Wayland) · beta

curl -fsSL https://github.com/manasdutta04/kudu/releases/latest/download/install.sh | sh

The script installs Kudu to ~/.local/bin and the app menu, and asks for sudo once to let Kudu read the keyboard (input group) and type through a virtual keyboard (/dev/uinput). Log out and back in after the first install. Hold Ctrl + Super to dictate.

Or install a package: sudo apt install ./Kudu-Linux-x86_64.deb (Debian, Ubuntu, Mint) or sudo dnf install ./Kudu-Linux-x86_64.rpm (Fedora, openSUSE), then add yourself to the input group once: sudo usermod -aG input $USER and log out and in. Any other distribution: extract Kudu-Linux-x86_64.tar.gz and run ./kudu/kudu.

First launch

Kudu downloads its speech model once (Mac ~450 MB; Windows/Linux: Accurate 482 MB or Compact 251 MB), then works offline. On Windows and Linux, smart formatting and Command mode are an optional extra download in Settings (491 MB or 1.1 GB).

What's in this release

New ways to install: a Windows installer, and .deb and .rpm packages for Linux. Also included from 0.1.1: Windows and Linux no longer close when downloading the speech model; problems show an error and are written to kudu.log (Windows: %LOCALAPPDATA%\Kudu\data\kudu.log, Linux: ~/.local/share/kudu/kudu.log).

New to Kudu? This is still the first public version:

• macOS: hold fn (or any shortcut you set) to dictate; formatting while you speak with Apple's on-device model (lists, numbers, units, math, @mentions, email layout, code and terminal commands); Command mode to rewrite or draft text; styles per app; a personal dictionary that learns from your corrections, with an Undo notice; names and terms spelled the way they appear on screen; snippets; history and Insights with a streak heat map. Apple Silicon only.
• Windows and Linux (beta): the same text cleanup, styles, dictionary, snippets, learning and shortcuts as the Mac app, with on-device speech (Parakeet or Moonshine) and an optional small language model (Qwen2.5). Please report anything that doesn't work on your setup.

Files are listed with SHA-256 checksums in SHA256SUMS.txt.

Kudu v0.1.1

Private, fast voice dictation. Hold a key, speak, release: clean text appears wherever your cursor is. Speech recognition runs on your own computer. No account, no cloud, no telemetry.

Install

The apps aren't signed with paid Apple / Microsoft certificates yet, so your system will warn that they come from an unidentified developer. The commands below install Kudu and clear that warning. They're also in each download's install script, which you can read before running.

macOS (Apple Silicon, macOS 14+)

curl -fsSL https://github.com/manasdutta04/kudu/releases/latest/download/install-macos.sh | bash

Or by hand: download Kudu-macOS.dmg (or the .zip), drag Kudu into Applications, then run this once in Terminal so macOS lets the unsigned app open:

xattr -cr /Applications/Kudu.app

On first launch, allow Microphone and Accessibility when Kudu asks. Hold fn to dictate.

Windows 10 / 11 (64-bit) · beta

In PowerShell:

irm https://github.com/manasdutta04/kudu/releases/latest/download/install.ps1 | iex

Or by hand: download Kudu-Windows-x64.zip, extract it, then in PowerShell in that folder run Unblock-File .\kudu.exe (or, when SmartScreen appears, click More info → Run anyway). Hold Ctrl + Win to dictate.

Linux (x86_64, X11 and Wayland) · beta

curl -fsSL https://github.com/manasdutta04/kudu/releases/latest/download/install.sh | sh

The script installs Kudu to ~/.local/bin and the app menu, and asks for sudo once to let Kudu read the keyboard (input group) and type through a virtual keyboard (/dev/uinput). Log out and back in after the first install. Hold Ctrl + Super to dictate.

Or by hand: extract Kudu-Linux-x86_64.tar.gz and run ./kudu; the settings window shows the same permission commands.

First launch

Kudu downloads its speech model once (Mac ~450 MB; Windows/Linux: Accurate 482 MB or Compact 251 MB), then works offline. On Windows and Linux, smart formatting and Command mode are an optional extra download in Settings (491 MB or 1.1 GB).

What's in this release

Fixed on Windows and Linux: the app closed when you clicked Download for the speech model. Downloads work now. If anything else goes wrong, Kudu shows an error instead of closing, and writes details to kudu.log (Windows: %LOCALAPPDATA%\Kudu\data\kudu.log, Linux: ~/.local/share/kudu/kudu.log). Please attach that file when you report a problem.

New to Kudu? This is still the first public version:

• macOS: hold fn (or any shortcut you set) to dictate; formatting while you speak with Apple's on-device model (lists, numbers, units, math, @mentions, email layout, code and terminal commands); Command mode to rewrite or draft text; styles per app; a personal dictionary that learns from your corrections, with an Undo notice; names and terms spelled the way they appear on screen; snippets; history and Insights with a streak heat map. Apple Silicon only.
• Windows and Linux (beta): the same text cleanup, styles, dictionary, snippets, learning and shortcuts as the Mac app, with on-device speech (Parakeet or Moonshine) and an optional small language model (Qwen2.5). Please report anything that doesn't work on your setup.

Files are listed with SHA-256 checksums in SHA256SUMS.txt.

Kudu v0.1.0

Private, fast voice dictation. Hold a key, speak, release: clean text appears wherever your cursor is. Speech recognition runs on your own computer. No account, no cloud, no telemetry.

Install

The apps aren't signed with paid Apple / Microsoft certificates yet, so your system will warn that they come from an unidentified developer. The commands below install Kudu and clear that warning. They're also in each download's install script, which you can read before running.

macOS (Apple Silicon, macOS 14+)

curl -fsSL https://github.com/manasdutta04/kudu/releases/latest/download/install-macos.sh | bash

Or by hand: download Kudu-macOS.dmg (or the .zip), drag Kudu into Applications, then run this once in Terminal so macOS lets the unsigned app open:

xattr -cr /Applications/Kudu.app

On first launch, allow Microphone and Accessibility when Kudu asks. Hold fn to dictate.

Windows 10 / 11 (64-bit) · beta

In PowerShell:

irm https://github.com/manasdutta04/kudu/releases/latest/download/install.ps1 | iex

Or by hand: download Kudu-Windows-x64.zip, extract it, then in PowerShell in that folder run Unblock-File .\kudu.exe (or, when SmartScreen appears, click More info → Run anyway). Hold Ctrl + Win to dictate.

Linux (x86_64, X11 and Wayland) · beta

curl -fsSL https://github.com/manasdutta04/kudu/releases/latest/download/install.sh | sh

The script installs Kudu to ~/.local/bin and the app menu, and asks for sudo once to let Kudu read the keyboard (input group) and type through a virtual keyboard (/dev/uinput). Log out and back in after the first install. Hold Ctrl + Super to dictate.

Or by hand: extract Kudu-Linux-x86_64.tar.gz and run ./kudu; the settings window shows the same permission commands.

First launch

Kudu downloads its speech model once (Mac ~450 MB; Windows/Linux: Accurate 482 MB or Compact 251 MB), then works offline. On Windows and Linux, smart formatting and Command mode are an optional extra download in Settings (491 MB or 1.1 GB).

What's in this release

This is Kudu's first public release.

• macOS: hold fn (or any shortcut you set) to dictate; formatting while you speak with Apple's on-device model (lists, numbers, units, math, @mentions, email layout, code and terminal commands); Command mode to rewrite or draft text; styles per app; a personal dictionary that learns from your corrections, with an Undo notice; names and terms spelled the way they appear on screen; snippets; history and Insights with a streak heat map. Apple Silicon only.
• Windows and Linux (first beta): the same text cleanup, styles, dictionary, snippets, learning and shortcuts as the Mac app, with on-device speech (Parakeet or Moonshine) and an optional small language model (Qwen2.5). Please report anything that doesn't work on your setup.

Files are listed with SHA-256 checksums in SHA256SUMS.txt.