A free, installable web app (PWA) that works as a language teacher: AI conversation, role-play scenarios, mini-lessons with exercises, listen & repeat (dictation), spaced-repetition flashcards, progress tracking and a small "student memory". Pure HTML/CSS/JavaScript: no build, no backend, no libraries.
- Interface languages (8): English (default), Português (BR), Español, Français, Deutsch, 中文, 한국어, 日本語
- Practice languages (8): English, Brazilian Portuguese, Spanish, French, German, Mandarin Chinese, Korean, Japanese
- AI: Google Gemini, called straight from your browser with your own API key (get a free one at https://aistudio.google.com/apikey)
- Voice: Chrome's Web Speech API (text-to-speech and speech recognition)
Open the Vercel link of this project (replace with your own deployment URL):
https://YOUR-PROJECT.vercel.app
Why this is the best way to test: the voice system (microphone / speech recognition) only works over HTTPS or localhost, with internet access and microphone permission. The Vercel link is HTTPS, so everything works, including installing the app.
- Open the link in Google Chrome (PC or Android).
- Go to Settings, paste your Gemini API key, choose your interface language, your native language and the language(s) you practice.
- Go to Home and start a conversation. Allow the microphone when Chrome asks.
- Click the green Install app button (top of the screen, on Home or in Settings).
- If the button does not appear: Chrome menu ⋮ → "Install Polyglot AI" (PC: also the install icon at the right end of the address bar; Android: ⋮ → "Add to Home screen" / "Install app").
- Chrome on PC: open the app, click ⋮ in its title bar → Uninstall. (Or open
chrome://apps, right-click the app → Remove.) - Android: long-press the app icon → Uninstall (or Settings → Apps → Polyglot AI → Uninstall).
- Your study data stays in the browser until you clear the site data (Chrome → Settings → Privacy → Site settings → find the site → Clear data). Use Settings → Export data first if you want a backup.
- Download the repository as ZIP from GitHub (Code → Download ZIP) and unzip it, or clone it:
git clone https://github.com/YOUR-USER/YOUR-REPO.git cd YOUR-REPO - Start a local web server in that folder:
python3 -m http.server 8000
- Open http://localhost:8000 in Chrome.
Do not open
index.htmlby double-clicking (file://): speech recognition and the service worker do not work there.localhostis treated as secure, so voice works.
- Put these files in a GitHub repository:
index.html,manifest.json,sw.js,vercel.json,icon-192.png,icon-512.png,icon-maskable-512.png,README.md,README.pt-BR.md. - On https://vercel.com: Add New → Project → Import your repository. Framework preset: Other. No build command, no output directory. Click Deploy.
- Every
git pushredeploys automatically. When you change files, bumpCACHE_VERSIONinsw.jsso installed apps refresh.
- Free conversation with corrections (after each message, at the end of the session, or off).
- Role-play scenarios (restaurant, airport, hotel, job interview, doctor, shopping, taxi, new friend) with an objectives checklist.
- Mini-lessons on any topic: short explanation, 5 audio examples, 3–5 exercises (multiple choice, fill in, translate, word order).
- Listen & repeat / dictation: the app speaks a sentence; you type or say it; differences are highlighted.
- Flashcards with spaced repetition (1, 3, 7, 14, 30 days; a miss restarts the card).
- Level test (A1–C2, with approximate HSK / JLPT / TOPIK equivalents for zh/ja/ko).
- Progress: streak, daily goal, minutes per week, words by stage, most frequent mistakes, lesson history, AI-estimated level (labelled as an estimate).
- Fully separate progress per language: each practice language has its own level, goal, daily goal, streak, minutes, vocabulary, mistakes, lesson history and student memory (max 20 editable items per language, used by the AI; it proposes updates at the end of a session).
- Delete ALL data button (Settings, bottom) to wipe everything and start from zero, with the option to keep your API key.
- Voice: per-language voices, speed 0.5x–2x, "slow" button on each sentence, microphone with target/native language switch, optional hands-free mode.
- Export / import all data as JSON.
- Voices load asynchronously in Chrome; if a language has no voice, the app shows text only and a tip. On Android install voice data in Settings → System → Languages → Text-to-speech.
- The first spoken reply needs a user gesture (a tap) – the app handles this when you press buttons.
- Speech recognition only tells you what the recognizer understood. It is not a pronunciation score.
- Speech recognition in Chrome sends audio to Google's servers; it needs internet.
- Your API key is stored only in your browser's localStorage and sent only to Google's Gemini API. The export file does not include it.
- Study data (sessions, vocabulary, mistakes, memory) is stored locally in IndexedDB. There is no backend and no analytics.
- Conversation text is sent to the Gemini API to generate replies. Do not type information you don't want to share with it.
- Google Chrome (recommended; Edge also works). Android for phone use.
- A Gemini API key and an internet connection.
- HTTPS (Vercel) or
localhostfor voice and installation.
index.html (the whole app) · manifest.json · sw.js (network-first service worker) · vercel.json · icons (icon-192.png, icon-512.png, icon-maskable-512.png).