Lexantir

A small notebook for words.

Capture English words anywhere you read — then re-read them, woven back into a story of your own.

Lexantir is a macOS vocabulary app built on Content and Language Integrated Learning (CLIL) — the idea that you learn a language fastest by using it to learn something else.

It is not a flashcard app. The loop is different:

You read in the wild → capture words with a double-click → the app infers the domain of today's words → an LLM writes you a short passage that re-uses every word you caught today, in context that's actually about your interests. You read it back. The words stick because they came home to a story.

flowchart LR
    A["📖 Read anywhere<br/>(Safari · Notes · PDF)"] --> B["✦ Double-click<br/>a word"]
    B --> C["🪄 Floating card<br/>translation · meanings · audio"]
    C --> D["📥 Add to today"]
    D --> E["🧠 AI infers the<br/>domain of today's words"]
    E --> F["✍️ Generates a passage<br/>re-using every word"]
    F --> A

While reading anywhere on macOS, double-click an English word. A small floating card appears right by your cursor — the word, its part of speech, a Chinese translation, multiple meanings — and it's spoken aloud automatically. The card floats over every other app without stealing focus, so your text selection in Safari survives.

Two buttons: Add daily or + Example. That's the whole interaction. Capture cost ≈ zero.

A restrained, Morandi-inspired palette — desaturated, dusty, warm. Warm-white "notebook paper" with a faint dot grid, cream cards, soft shadows, restrained flat sand buttons, Georgia serif throughout. No vivid greens, reds, or blues anywhere.

Reference points: Apple Notes · the Claude app on Mac · the Duolingo spell screen · a paper dot-grid notebook.

• JDK 23 · JavaFX 23.0.1 · Maven — imperative JavaFX, no FXML
• SQLite (sqlite-jdbc) — local store at ~/Library/Application Support/WordBook/wordbook.db
• JNativeHook — global mouse / keyboard capture
• JNA → Cocoa — floating-panel window level, foreground yielding
• Gson · DeepSeek (deepseek-v4-flash) for article generation
• Translation via Google's free keyless endpoint

Requires JDK 23 (the project is built/tested on Zulu 23, arm64).

cd wordbook-java
export JAVA_HOME=~/.sdkman/candidates/java/current

mvn -q javafx:run        # dev run
./package-mac.sh         # build → sign → install /Applications/WordBook.app

package-mac.sh builds a shaded fat jar, wraps it with a bundled JRE via jpackage, sets LSUIElement (menu-bar agent), code-signs, and installs to /Applications.

⚠️ First launch needs Accessibility + Input Monitoring permission (System Settings → Privacy & Security) for the global capture to work.

wordbook-java/
├── src/main/java/com/wordbook/
│   ├── ui/            MainWindow (the controller), Theme, I18n, components, modals
│   ├── service/       Translator · TtsService · DeepSeekService
│   ├── capture/       GlobalCapture (JNativeHook)
│   ├── platform/      MacNative (JNA → Cocoa)
│   └── db/            Database (SQLite)
├── src/main/resources/  icons · styles/app.css
├── packaging/         WordBook.icns · appicon_src.png
└── package-mac.sh

• Notarized distribution (Developer ID) so it opens cleanly on other Macs
• Natural offline TTS (premium say voices) — replace the basic system voice
• Article library browsing UI (articles are already persisted)
• Unified bundled illustration set for the achievement card (31 domains + fallback)
• Mobile (the capture gesture is desktop-only by nature — needs a rethink on iOS/Android)

• Privacy: the DeepSeek API key lives only in your local SQLite DB — never in source.
• Naming: the repo is Lexantir; the app is still internally named WordBook (bundle id com.wordbook.app). A full rebrand is on the to-do list.

Changelog · Contributing · Security policy · Code of conduct · License