• Picks
  • Discover
  • Alternatives
  • About

© 2026 Willet

  • Newsletter
  • About
  • FAQ
  • Terms
  • Privacy & Cookies
Join the Discord (opens in a new tab)
Hitoshi Togasaki avatar

Karukan

Japanese input method with neural kana-kanji conversion

  • System utilities
  • AI models and runtimes
View on GitHub

Featured by Willet

What it is

A Japanese input method for Linux (as an fcitx5 add-on) and macOS that converts kana to kanji using small GPT-2 and Qwen3 based models run locally through llama.cpp. It offers live conversion as you type, context-aware suggestions, learning from chosen candidates, a SudachiDict-based system dictionary, user dictionaries in Mozc and Google IME formats, and emoji lookup. People who write Japanese on Linux or Mac desktops can use it in place of dictionary-only IMEs, with the model downloaded once and run offline afterward.

Images

Facts

Stars
740

Forks
61
Commits
89
Contributors
5
Language
Rust
License
Apache-2.0
Last commit
Oct 10, 2026
On GitHub since
August 2024
First release
February 2026
Platforms
macOS, Linux, CLI
Self-hostable
No
Alternative to
Google Japanese Input, Microsoft IME, ATOK
Author
Hitoshi Togasaki
Tags
  • linux-desktop
  • llm-inference
  • drivers
  • translation
  • local-ai

Last synced from GitHub on Oct 20, 2026

Share and more

Comments

No comments yet. Start the conversation.

The 10 best overlooked open-source projects, every Monday.