A floating-window vocabulary trainer for macOS, with a built-in AI tutor. Type to remember — vocabulary practice and typing drills in one place.
English · 简体中文
- 6 built-in dictionaries (~35,000 words total, all open source)
- Oxford / General English — curated from ECDICT (6,355 words, tagged with Oxford level)
- CET-4 College English Test Band 4 (4,544 words, with phonetics + examples + phrases)
- CET-6 College English Test Band 6 (3,992 words)
- IELTS (5,254 words)
- TOEFL (10,376 words)
- SAT (4,437 words)
- Floating window — a small always-on-top window that follows the current word in the main window
- Pronunciation — Microsoft Edge TTS (free neural voices, needs internet):
en-US-AriaNeuralfor English,zh-CN-XiaoxiaoNeuralfor Chinese - Typing practice — type each word as you see it, with real-time correct/incorrect feedback and per-round stats
- AI tutor (BYOK) — any OpenAI-compatible API, with a customizable base URL (point it at your own relay if you like)
- The conversation is bound to the current word — context follows automatically as you switch words
- Reset conversation — clear the context anytime
- Streaming output (SSE)
- Bilingual UI (EN / 中文) — toggle from the top bar; the dictionary stays the same
- Progress persistence — every word records
seen/correct/status, saved locally
Prerequisites
- Node.js — to run the Electron app
- Python 3.9 +
edge-tts— for pronunciation:python3 -m pip install -r requirements.txt. Pronunciation calls Microsoft Edge TTS and needs an internet connection; ifedge-ttsisn't installed the app still runs, it just stays silent. Install it for the samepython3that's on yourPATH(the app spawnspython3).
# 1. Install dependencies (dictionaries are downloaded automatically)
npm install
# 2. Launch
npm startOn first launch the main window defaults to the Oxford / General English dictionary, starting with 20 words.
| Key | Action |
|---|---|
Space |
Mark as "known" and move to the next word |
Enter (in the input) |
Submit the current input |
Escape (in the input) |
Mark as "unknown" and move to the next word |
S |
Pronounce the current word |
A |
Open the AI chat window |
Open Settings → AI Tutor and fill in four fields:
- Enable AI — turn it on
- API Base URL —
https://api.openai.com/v1, or any OpenAI-compatible endpoint (e.g. a self-hosted relay) - API Key —
sk-... - Model —
gpt-4o-mini, or any OpenAI-compatible model
Open a word card → click the 💬 AI Tutor button. The app automatically sends the current word plus the 5 most recently studied words as context.
Reset the conversation with the 🗑 button in the top-right of the AI window — it clears the context but keeps the linked-word chip.
# Start a mock LLM server + Electron with all windows auto-opened + an auto-sent message
node scripts/mock-llm-server.js & # mock listens on 9876
npm start -- --autoopen --auto-msg # auto-open floating / settings / AI windows + auto-send a messagenpm run package # builds into dist/ as a macOS arm64 .dmgtypingMem_words/
├── main.js # Electron main process (window management + IPC)
├── preload.js # secure contextBridge
├── renderer/
│ ├── i18n.js # EN / 中文 i18n
│ ├── store.js # learning progress / preferences persistence
│ ├── tts.js # pronunciation (Edge TTS only)
│ ├── llm.js # LLM streaming calls + system prompt construction
│ ├── styles.css # global styles (mostly dark)
│ ├── main.js # main study UI logic
│ ├── floating.js # floating-window logic
│ ├── settings.js # settings-window logic
│ └── chat.js # AI chat-window logic
├── windows/
│ ├── main.html
│ ├── floating.html
│ ├── settings.html
│ └── chat.html
├── scripts/
│ ├── download-dicts.js # download / sync dictionaries from GitHub
│ ├── edge_tts_synth.py # Edge TTS → MP3 (pronunciation backend)
│ └── mock-llm-server.js # local mock (for development)
├── data/dicts/ # dictionary JSON (auto-downloaded, git-ignored)
│ ├── oxford.json
│ ├── cet4.json
│ ├── cet6.json
│ ├── ielts.json
│ ├── toefl.json
│ ├── sat.json
│ └── meta.json
└── assets/
- ECDICT — skywind3000/ECDICT (MIT)
- CET / IELTS / TOEFL / SAT — KyleBing/english-vocabulary (MIT)
Dictionaries are downloaded automatically on the first npm install; you can later refresh them incrementally with npm run dicts:download.
- All data (learning progress / preferences / API key) is stored in
~/Library/Application Support/typingmem-words/. - The API key is never uploaded anywhere — it is sent directly to your configured base URL only when calling the model.
- Use Settings → Data → user data directory to view the local files.
- The first cold start of the model can take ~10s (if you use a relay with cold starts).
- Pronunciation requires an internet connection and the
edge-ttsPython package; if it's missing or offline, pronunciation silently does nothing. - Dictionaries load in the order ECDICT → KyleBing during download; any that fail fall back to a minimal built-in set.
MIT
Enjoy 🎉