azooKey/azooKey-Desktop

azooKey‑Desktop – Japanese Input Method for macOS

What it is

• An open‑source Japanese IME (input method editor) for macOS that replaces the system’s Kana‑Kanji conversion with a neural model called Zenzai. The project also bundles optional LLM‑based “smart conversion” and integrates with the Tuner personal‑optimization service.

Why it matters

• Traditional Japanese IMEs rely on rule‑based dictionaries. azooKey‑Desktop swaps those for a modern neural network (gguf‑format) that can learn from your typing history, custom dictionaries, and a profile prompt, delivering higher‑accuracy conversions.
• The repo ships the model weights via Git LFS, so the heavy AI component is part of the source tree rather than a separate download.

Key features (as listed in the README)

• Zenzai neural conversion – high‑precision Kana‑Kanji conversion.
  • Profile‑prompt support.
  • History‑learning (adapts to your usage).
  • User‑dictionary.
  • Integration with the personal‑optimization system Tuner.
• LLM‑assisted “nice‑feeling” conversion – an optional large‑language‑model step that can suggest more natural phrasing.
• Live conversion – results appear as you type.
• Native AZIK support – compatibility with the AZIK keyboard layout.

How to get it (end‑user)

1. Release package – download the .pkg from the Releases page and install.
2. Log out of macOS and log back in.
3. Open System Settings → Keyboard → Input Sources, click +, choose Japanese → azooKey, and confirm.
4. Select azooKey from the menu‑bar icon.

Alternative: brew install azooKey (Homebrew). After installation you still need to log out/in and add the input source.

Building from source (for developers)

• Prerequisites
  • macOS 15+ (tested on 15, works on 14/26 but unverified)
  • Xcode 26.1+ (personal team signing works if you’re not in the Apple Developer Program)
  • Git LFS (required to pull the model files)
  • SwiftLint (code‑style checks)
• Clone with submodules & LFS

  git lfs install
  git clone https://github.com/azooKey/azooKey-Desktop --recursive
  cd azooKey-Desktop
  # If you already cloned without LFS/submodules:
  git submodule update --init
  git -C azooKeyMac/Resources/zenz-v3.1-small-gguf lfs pull
  git -C azooKeyMac/Resources/base_n5_lm lfs pull
  

• Configure signing – open azooKeyMac.xcodeproj, change the Team to your Personal Team, and replace the bundle identifier (dev.ensan…) with one you own.
• Build & install

  ./install.sh   # produces the same result as the .pkg installer
  

  After the script finishes, repeat the log‑out / input‑source steps above.
• Troubleshooting – the README lists common signing errors, Xcode version issues, and how to verify that the model files are actually present (they should be tens of megabytes, not the 134‑byte LFS pointer).

Community & extensions

• Discord server for support and contribution: https://discord.gg/dY9gHuyZN5
• Forks for other platforms:
  • fcitx5‑hazkey – Linux client.
  • azooKey‑Windows – Windows client.
  • azoo‑key‑skkserv – SKK server implementation.

Funding

• The project accepts GitHub Sponsors.
• It is supported by Japan’s IPA Mito‑IT talent‑development program (2024).

Bottom line azooKey‑Desktop is a genuine AI‑enhanced software project: it bundles a neural language model for Japanese input, offers optional LLM‑based refinements, and provides full build instructions for macOS developers who want to contribute or customize the IME.