A study app for the Japanese Kanji Aptitude Test (漢検), built for iPhone and iPad. You write each character by hand on the screen and compare it against the answer, rather than reading it and moving on.
- Support — https://cobyapp.github.io/kanji-write/
- Privacy Policy — https://cobyapp.github.io/kanji-write/privacy.html
Every number below comes from the built database, not from an estimate:
| Kanji | 6,053 across all twelve 漢検 levels (10級…1級, including 準2級 / 準1級) |
| Words | 43,138 |
| Four-character idioms (四字熟語) | 398 |
| Example sentences | 6,602 |
| Stroke-order records | 71,836 |
| Practice questions | 18,617 — every 漢検 読み・書き取り and 同音・同訓異字 question is set in an example sentence, as the real paper is; 準1級 95%, 1級 88% reading/writing coverage |
| 漢検 mock papers | Every 級 in its official layout — 大問 order, question counts, points, time limit and pass mark from the 2026年度 papers |
| JLPT vocabulary levels | 6,129 words tagged N5〜N1 |
| Languages | Korean, Japanese, English, Simplified Chinese — including every gloss and sentence translation |
Everything is bundled, so the app runs with no network and no account. Study history stays on the device.
app/ iOS + watchOS + widget app (Tuist, SwiftUI, TCA, GRDB, PencilKit)
data-pipeline/ Python pipeline that builds the shipped SQLite database
docs/ the published support and privacy pages, plus design notes
A Tuist project split into feature modules — AppFeature, Practice,
WritingCanvas, KanjiDetail, Review, TestMode, Worksheet, Reminders,
DictionaryClient, DesignSystem, SharedModels — with an embedded watchOS
app and a home-screen widget. Each feature module has its own test target.
Tuist generates the Xcode project, so it is not checked in. After changing
Project.swift or adding source files:
app/Scripts/generate.shThe script wraps tuist install + tuist generate and lifts the deployment
targets of Tuist's generated package resource bundles, which Xcode 27 would
otherwise reject.
kanjipipe ingests the open dictionary sources, filters them to the 漢検
curriculum, generates practice questions, and writes out/kanji.sqlite. Build
gates in kanjipipe/validate.py refuse to emit a database with starved exam
sections, misaligned glosses, or a level whose kanji count contradicts the
official 漢検 counts.
cd data-pipeline
python -m kanjipipe.build_db # writes out/kanji.sqlite
pytest # pipeline testsEDRDG and Tatoeba publish only a mutable "latest" snapshot, so once they move on the pinned from-scratch build cannot be reproduced. Content this repository owns — the question banks, 四字熟語, 対義語・類義語 and JLPT word levels — can be reloaded into the shipped database in place, with the same content gates:
python scripts/refresh_db.py # updates the app's kanji.sqlite
python scripts/check_question_batch.py FILE # structure of an authored batch
python scripts/check_ambiguity.py FILE # distractors that are also rightThe dictionary content is openly licensed, and the app credits each source on its support page.
| Source | Used for | Licence |
|---|---|---|
| KANJIDIC2 · JMdict | readings, meanings, vocabulary | EDRDG (CC BY-SA 4.0) |
| KanjiVG | stroke order | CC BY-SA 3.0 |
| Tatoeba | example sentences | CC BY 2.0 FR |
| Unihan | character metadata | Unicode |
| GlyphWiki | 旧字 outlines | CC BY-SA 3.0 |
| open-anki-jlpt-decks | JLPT vocabulary levels | MIT |
| Japanese WordNet (NICT) | 類義語 (synonym) pairs | NICT WordNet licence (permissive, BSD-style) |
The source in this repository is not currently released under an open licence. The bundled dictionary data stays under the licences listed above.