# Suzume > Japanese tokenizer that runs in the browser. Its WebAssembly binary is under 269KiB gzipped, with its dictionaries bundled in — no server, no external dictionary download, no install step for the reader. Suzume segments Japanese text into words with part-of-speech tags and base forms. The same core ships as WebAssembly for browsers and JavaScript runtimes, and natively for Python, Go, C/C++, and a CLI. Unknown words are handled with pattern-based candidate generation rather than a large dictionary. The links below point to the canonical HTML documentation. ## Key pages - [Suzume home](https://suzume.libraz.net/) - [Docs](https://suzume.libraz.net/docs/getting-started): Split Japanese text into words with part-of-speech tags and base forms from JavaScript, Python, Go, C++ or the CLI. Try the live demo, then install and run it. ## Guide - [Getting Started](https://suzume.libraz.net/docs/getting-started): Split Japanese text into words with part-of-speech tags and base forms from JavaScript, Python, Go, C++ or the CLI. Try the live demo, then install and run it. - [Installation](https://suzume.libraz.net/docs/installation): Install the Suzume Japanese tokenizer: @libraz/suzume from npm or a CDN for browsers, Node.js, Deno and Bun, suzume from PyPI, or Go and C/C++ from source. - [User Dictionary](https://suzume.libraz.net/docs/user-dictionary): Keep domain terms as one token or set their part of speech and base form: write a TSV user dictionary, load it at runtime, or compile it to binary. - [Migrating to 0.10](https://suzume.libraz.net/docs/migrating-to-0.10): Every breaking change in Suzume 0.10 by surface: npm, Python, C++, C ABI, both CLIs and dictionary files, each as the old form, the new form and the reason. ## Concepts - [How It Works](https://suzume.libraz.net/docs/how-it-works): How Suzume tokenizes Japanese with a small built-in dictionary, pattern-based unknown-word candidates and connection scoring, and when to use which tool. - [Modes, Offsets, and Normalization](https://suzume.libraz.net/docs/analysis-modes): Choose normal, search or split mode, merge compounds, map token offsets back to the original string in each binding, and control character normalization. - [Glossary](https://suzume.libraz.net/docs/glossary): Plain-language definitions of POS, base form, conjugation forms, copula, formal nouns, unknown words and other Japanese tokenization terms, with real output. - [Speed and Accuracy](https://suzume.libraz.net/docs/performance): Reproducible WASM speed measurements under Node and word-boundary agreement on Suzume's own test data, with what each figure covers and what it leaves out. ## API Reference - [JavaScript / WASM](https://suzume.libraz.net/docs/api): Reference for @libraz/suzume in browsers, Node.js, Deno and Bun: Suzume.create options, analyze, generateTags, user dictionaries, Morpheme fields and errors. - [Python](https://suzume.libraz.net/docs/python): Tokenize Japanese in Python with the suzume package from PyPI: constructor options, analysis results, Morpheme fields, tags, user dictionaries and errors. - [Python CLI](https://suzume.libraz.net/docs/python-cli): The suzume command from the Python wheel: tokenize Japanese from arguments or standard input, pick an output format, load dictionaries, print tags or JSON. - [Go](https://suzume.libraz.net/docs/go): Tokenize Japanese in Go with github.com/libraz/go-suzume, a cgo binding with embedded dictionaries: setup, analysis modes, offsets, tags, user dictionaries. - [C / C++](https://suzume.libraz.net/docs/cpp): Link Suzume's C++17 wrapper or C ABI into native code: install, CMake and pkg-config, analysis results, modes, tags, errors, ownership and ABI compatibility. - [Part-of-Speech Labels](https://suzume.libraz.net/docs/pos-labels): Every coarse POS and extended POS label Suzume returns, with examples, and the field that carries them in JavaScript, Python, Go, C++ and the C ABI. - [Native Developer CLI](https://suzume.libraz.net/docs/cli): Reference for suzume-cli, the native developer command: analyze text in several output formats, compile and validate dictionaries, run tests and benchmarks. ## MeCab Comparison - [Differences from MeCab](https://suzume.libraz.net/docs/mecab-comparison): Where Suzume and MeCab split the same Japanese text differently, grouped into merging, splitting and notation topics, with the shared comparison baseline. - [Compounds, Quantities, and Names](https://suzume.libraz.net/docs/mecab-compounds): How Suzume keeps kanji compounds, quantities, dates and names together and which suffixes it leaves visible, compared with MeCab output in tested examples. - [Verbs and Grammar](https://suzume.libraz.net/docs/mecab-grammar): Token boundaries for compound verbs, auxiliaries, particles, and colloquial and classical forms: what Suzume keeps whole and what it splits, versus MeCab. - [Text and Notation](https://suzume.libraz.net/docs/mecab-text): How Suzume tokenizes everyday kana words, expressive spellings, technical notation such as identifiers and URLs, and symbols, compared with MeCab output. - [POS Classification](https://suzume.libraz.net/docs/pos-differences): Why the same Japanese token gets different part-of-speech labels in MeCab and Suzume, grouped by lexical category, predicate and context, and how they map. ## Build & Contribute - [Native Build](https://suzume.libraz.net/docs/native-build): Build the Suzume C++17 static library and native CLI from source with CMake, plus the optional shared C library, embedded dictionaries, examples and WASM. - [Testing Guide](https://suzume.libraz.net/docs/testing): How the Suzume core repository tests the C++ core, WASM and Python bindings, native CLI and cross-binding parity, and how to add tests, benchmarks and CI. ## Japanese (日本語) - [日本語版インデックス](https://suzume.libraz.net/ja/llms.txt): The same index in Japanese, covering the /ja/ documentation.