Karukan is an open-source Japanese input method system for Linux and macOS, powered by a neural kana-kanji conversion engine. It turns romanized or kana keystrokes into natural Japanese text using modern NLP rather than static dictionaries alone. The project has about 740 stars on GitHub and was actively updated in October 2026.
Converting kana to the right mix of kanji, hiragana and katakana is the hardest part of typing Japanese, because the same reading maps to many possible words. Karukan attacks the problem with a neural conversion engine written in Rust, giving it both speed and smarter context-aware predictions. It plugs into the desktop through the Fcitx5 input method framework, so it works across applications on Linux, with macOS support as well. The result is an IME that learns toward natural, modern Japanese instead of the stiff suggestions of older dictionary-only systems.
For users, the neural approach pays off most on long sentences and compound words, exactly the cases where traditional dictionary-driven IMEs guess wrong and force tedious reconversion. And because conversion happens entirely on the device, everything you type stays on your own machine.
Karukan is completely free and open source under the Apache-2.0 license. There are no subscriptions, accounts or paid tiers; you install it from the GitHub repository and it runs locally on your machine.
Karukan is for Japanese speakers and learners who type on Linux or macOS and are dissatisfied with their default input method. Developers interested in NLP-driven text input, or in Rust implementations of language tooling, will also find the codebase worth studying.