NeoGPT is a multilingual AI chatbot built for India's linguistic diversity. It understands how people actually speak — Hinglish, Tanglish, Banglish, code-mixed conversations — without needing translation. Built as a college project by Rahul Chowdhury.
User: "Yaar, aaj market mein kya hua? TCS ke baare mein bata."
Neo: "TCS aaj ₹3,842 pe close hua — 1.2% ki badhot ke saath 🚀"
neogpt/
├── frontend/ # React + Vite web app → Vercel
├── backend/neogpt/ # Express + Node.js API → Render
└── extension/ # Chrome MV3 extension → Chrome Web Store
| Feature | Status |
|---|---|
| 35+ Indian languages + code-mixed dialects | ✅ Live |
| Real-time web search via Tavily | ✅ Live |
| Persistent threads with shareable URLs | ✅ Live |
| Redis thread cache — <50ms resume | ✅ Live |
| Dual-LLM fallback with exponential backoff | ✅ Live |
| Emotion-aware, tone-matching responses | ✅ Live |
| Token usage tracking + plan limits | ✅ Live |
| JWT auth with per-user API keys | ✅ Live |
| Chrome extension — summarize any webpage | ✅ Live |
| Context-aware chat inside extension | ✅ Live |
| Voice AI (regional accents) | 🔜 Roadmap |
| RAG document Q&A | 🔜 Roadmap |
| WhatsApp / Telegram integration | 🔜 Roadmap |
Frontend — React 18, Vite, Tailwind CSS, Framer Motion, React Router
Backend — Node.js, Express, MongoDB Atlas, Redis (ioredis), Groq SDK, Tavily
AI / Inference — Groq (openai/gpt-oss-20b primary, llama-3.3-70b-versatile fallback)
Auth — JWT, bcrypt, UUID-based API keys
Infra — Vercel (frontend), Render (backend), MongoDB Atlas, Upstash Redis
- Node.js ≥ 18
- MongoDB Atlas account (free tier works)
- Redis instance — Upstash free tier recommended
- Groq API key
- Tavily API key
cd frontend
npm installCreate frontend/.env:
VITE_API_BASE_URL=http://localhost:5000npm run dev # dev server at localhost:5173
npm run build # production build → dist/Deploy to Vercel:
vercel --prod
# set VITE_API_BASE_URL=https://your-render-url.onrender.comfrontend/src/
├── pages/
│ ├── LandingPage.jsx # marketing page
│ ├── AuthPage.jsx # login / signup
│ └── ChatPage.jsx # main chat UI
├── pages/landing/
│ ├── constants.js # all data, pricing, CHAT_DEMO
│ ├── ChatPreview.jsx # looping live demo widget
│ ├── HeroSection.jsx # navbar + hero + particles
│ └── ProductSections.jsx # features, business, pricing
├── components/
│ ├── Navbar.jsx # top nav with token usage bar
│ ├── ChatSidebar.jsx # thread list, mobile drawer
│ ├── ChatContainer.jsx # message bubbles
│ ├── InputBox.jsx # textarea with send button
│ └── LanguageSelect.jsx # 35 language dropdown
└── api/
├── apiClient.js # axios + 401 interceptor
└── authAPI.js # login, register, chat history
cd backend/neogpt
npm installCreate backend/neogpt/.env:
GROQ_API_KEY=your_groq_key
TAVILY_API_KEY=your_tavily_key
MONGO_URI=mongodb+srv://user:pass@cluster.mongodb.net/neogpt
REDIS_URL=rediss://your-upstash-url:6379
JWT_SECRET=your_jwt_secretnode server.js # starts on port 5000Deploy to Render:
- Build command:
npm install - Start command:
node server.js - Add all env vars in Render dashboard
backend/neogpt/
├── server.js # Express app, CORS, route registration
├── chatbot.js # generate() — LLM + tool call loop
├── models/
│ ├── chat.js # Chat schema (username, threadId, messages[])
│ └── ...
├── routes/
│ ├── ChatRoutes.js # POST /chat · GET /history · GET /:threadId
│ └── summarizeRoutes.js # POST /summarize (for extension)
├── Auth/
│ ├── config.js # PLAN_LIMITS — free/pro/premium/premium+
│ ├── models/User.js # User schema with plan + token tracking
│ ├── middlewares/
│ │ ├── User.js # JWT verify middleware
│ │ └── TokenLimit.js # monthly quota check + auto-reset
│ └── routes/UserRoutes.js # register, login, profile, plan update
└── utils/
└── multerConfig.js # file upload config
| Method | Endpoint | Auth | Description |
|---|---|---|---|
POST |
/user/register |
— | Create account |
POST |
/user/login |
— | Returns JWT |
GET |
/user/ |
JWT | Get current user |
PUT |
/user/language |
JWT | Update language preference |
POST |
/chat |
JWT | Send message, get AI reply |
GET |
/chat/history |
JWT | All threads for user |
GET |
/chat/:threadId |
JWT | Messages in a thread |
POST |
/summarize |
JWT | Summarize webpage (extension) |
| Plan | Monthly tokens | Price |
|---|---|---|
| Free | 50,000 | ₹0 |
| Pro | 500,000 | ₹250/mo |
| Premium | 2,000,000 | ₹700/mo |
| Premium+ | Unlimited | ₹1,200/mo |
Tokens reset on the 1st of each month. Counter increments from completion.usage.total_tokens after every Groq call.
| Variable | Required | Description |
|---|---|---|
GROQ_API_KEY |
✅ | Groq inference API |
TAVILY_API_KEY |
✅ | Real-time web search |
MONGO_URI |
✅ | MongoDB Atlas connection string |
REDIS_URL |
✅ | Redis for 24h thread cache |
JWT_SECRET |
✅ | Token signing secret |
PORT |
— | Default: 5000 |
The NeoGPT extension lets you summarize any webpage, ask questions about its content, and get AI-powered explanations of selected text — all powered by the same backend.
cd extension
# no build step required — pure MV3- Open
chrome://extensions/ - Enable Developer mode (top right)
- Click Load unpacked
- Select the
extension/folder - Open the extension popup → Settings → paste your NeoGPT JWT token
extension/
├── manifest.json # MV3 manifest
├── background.js # service worker — message hub, API calls, 1h cache
├── content.js # DOM extraction, scroll-to-section, highlight, toolbar
└── popup/
├── popup.html # entry point
├── App.jsx # tab shell — Summary / Chat / Settings
├── views/
│ ├── SummaryView.jsx # TLDR + key points + section cards
│ ├── ChatView.jsx # context-aware chat with selection actions
│ └── SettingsView.jsx # token, language, cache controls
└── components/
├── SectionCard.jsx # clickable section with scroll-on-click
└── ChatBubble.jsx # message bubble with markdown
- Summarize page — extracts headings + content, sends to
/summarize, returns TLDR + bullet points + section summaries - Section navigation — click any section card → smoothly scrolls to + highlights the DOM element (yellow fade, 2s)
- Context-aware chat — ask questions about the full page or a highlighted selection
- Floating toolbar — select any text on the page → mini toolbar appears with Summarize / Explain / Translate
- 1h cache — summaries cached by URL in
chrome.storage.localto avoid repeat API calls - Language toggle — English, Hindi, Hinglish, Tamil, Bengali and more
After logging in to neogpt-blue.vercel.app:
- Open DevTools → Application → Local Storage
- Copy the value of
authToken - Paste it into the extension Settings tab
The extension uses the same
/chatand/summarizeendpoints as the web app, so your token quota is shared.
Browser / Extension
│
▼
Express API (Render)
│
┌────┼────┐
│ │
JWT Rate limit
verify (20 req/min)
│ │
└────┬────┘
│
Token quota
(plan check)
│
generate() — chatbot.js
│
┌────┼────────────┐
│ │
Redis cache MongoDB fallback
(24h TTL) (thread restore)
│
Groq LLM call
│
tool_calls? ──yes──▶ Tavily search ──▶ second LLM pass
│no
▼
save to Redis + MongoDB
increment tokensUsed
return { reply, tokensUsed }
Model fallback order (on rate limit / error):
openai/gpt-oss-20b— primaryllama-3.1-8b-instant— fast fallbackopenai/gpt-oss-120b— heavy fallbackllama-3.3-70b-versatile— final fallback
Each model gets 3 attempts with exponential backoff (2s → 5s → 10s) on 429s before switching.
# terminal 1 — backend
cd backend/neogpt
cp .env.example .env # fill in your keys
node server.js
# terminal 2 — frontend
cd frontend
echo "VITE_API_BASE_URL=http://localhost:5000" > .env
npm run dev
# extension — load unpacked from extension/ in chrome://extensions# fork → clone → branch
git checkout -b feature/your-feature
# make changes, then
git commit -m "feat: your feature description"
git push origin feature/your-feature
# open a PR- Voice AI — regional accent synthesis (Tamil, Bengali, Marathi)
- RAG — upload PDFs, get answers in your language
- Thread forking — clone and branch conversations
- WhatsApp / Telegram integration
- Adaptive learning — remember your slang over time
- Mobile app (React Native)
MIT — see LICENSE
Built with ❤️ by Rahul Chowdhury · UEM Jaipur