AI Trading Agent

Autonomous AI-powered trading system for Indian stock markets (NSE/BSE)
Chat with an AI agent that analyzes markets, executes trades, and manages risk — all from a single dashboard.

DISCLAIMER: This software is for educational and research purposes only. Not financial advice. Trading involves substantial risk of loss. Paper trading results are simulated. Always consult a SEBI-registered investment advisor. See full disclaimer.

AI Trading Agent is an open-source, self-hosted trading toolkit that combines:

• AI Chat Interface — Talk to Claude Code in natural language: "analyze RELIANCE", "buy NIFTY PE at 23900", "show me oversold stocks"
• Real-time Dashboard — Dark-themed web UI with live market data, portfolio tracking, and agent monitoring
• Autonomous Scanner — Scans your watchlist every N minutes, scores stocks 0-10 on technical confluence
• Paper + Live Trading — Switch between paper trading and real broker execution with one click
• Multi-Broker Support — Groww, AngelOne, Zerodha, Upstox — add accounts from the dashboard
• Background Agents — Persistent trade monitors that track SL/target and auto-exit positions
• Trading Memory — AI remembers patterns, rules, lessons, and past trades across sessions
• Skills Cache — Common queries execute instantly without AI having to regenerate commands
• Self-Modifying Dashboard — Ask the AI to change the UI itself: "add a watchlist panel", "make font bigger", "add BANKNIFTY to market pulse"

You type → AI analyzes → Executes trade → Monitors position → Auto-exits at SL/Target
          ↓                                    ↓
    Saves to memory                  Live P&L in dashboard

"Add a sector heatmap panel" → AI edits the dashboard code → Refresh browser → Done

• Node.js 22+
• Python 3.12+ (for dashboard)
• Claude Code CLI (npm install -g @anthropic-ai/claude-code)

# Clone
git clone https://github.com/Manjussha/AI-trader.git
cd AI-trader

# Install Node dependencies
npm install

# Install Python dependencies (for dashboard)
pip install starlette uvicorn httpx

# Option 1: Start everything with one command
npm run agent-ui

# Option 2: Start separately
node node-bridge.mjs          # Terminal 1: Start the data bridge (port 3001)
python dashboard/run.py        # Terminal 2: Start the dashboard (port 8000)

Open http://localhost:8000 — start chatting with the AI agent.

npm run cli                    # Interactive terminal REPL
npm run bot                    # Autonomous scanner + dashboard
npm run bot:paper              # Scanner + auto paper trade on strong signals

No API key needed for market data — NSE India public API + Yahoo Finance are used for all price data.

• Chat with Claude Code in natural language
• See tool executions (curl calls, analysis) in collapsible blocks
• Stop button + Escape key to interrupt
• Type a new message while AI is responding — auto-cancels previous
• Chat history persists across refreshes

• Accounts — Switch between paper trading and real broker accounts
• Active Agents — Background trade monitors with live P&L
• Portfolio — Cash, holdings, unrealized/realized P&L with live prices
• Market Pulse — NIFTY live price, day range, auto-refreshes
• Trading Memory — Patterns, rules, lessons, trades, notes

The AI chatbot has full access to read and edit the dashboard code. Ask it to customize the UI:

> "Add BANKNIFTY to the market pulse panel"
> "Make the chat font size 14px"
> "Add a new panel showing top 5 gainers"
> "Change the accent color to green"
> "Add a keyboard shortcut for quick buy"

The bot uses Read to inspect current code, Edit to make surgical changes, and tells you to refresh the browser. It has access to all project files:

Browser (localhost:8000)
    │ WebSocket
    ▼
Python Starlette + Uvicorn          ← Dashboard server
    │ Spawns claude -p subprocess   ← Claude Code CLI (AI backend)
    │ httpx async HTTP              ← Bridge client
    ▼
Node.js Bridge (localhost:3001)     ← REST API wrapping all JS modules
    │
    ├── GrowwClient (NSE/Yahoo)     ← Free market data (no auth)
    ├── Analytics (RSI, MACD...)    ← Technical indicators
    ├── Patterns (22 candlestick)   ← Pattern recognition
    ├── Greeks (Black-Scholes)      ← Options calculator
    ├── Paper Trade Engine          ← Virtual portfolio
    ├── Trade Journal               ← Performance tracking
    ├── History Analyzer            ← Similarity matching
    ├── Broker Adapters             ← Groww/AngelOne/Zerodha/Upstox
    └── Skills + Memory             ← Persistent caches

Why this architecture?

• Node bridge keeps NSE connections warm (cookies, TLS keep-alive) — sub-second market data
• Python dashboard uses Starlette + native WebSocket for real-time streaming
• Claude Code CLI as AI backend — no API key needed, full tool execution capabilities
• Skills cache — common queries execute instantly without AI regenerating commands

Every stock gets scored across 10 independent technical factors:

• Score >= 6 — Alert fired (terminal + Telegram)
• Score >= 7 — Auto paper trade with ATR-based sizing

Manage accounts directly from the dashboard — add API credentials, test connection, switch active account with one click.

Market data (prices, indices, historical OHLCV) uses free NSE India + Yahoo Finance APIs. Broker credentials are only needed for real order execution.

BROKER=groww        # or angelone | zerodha | upstox

# Groww
GROWW_API_KEY=
TOTP_SECRET=

# AngelOne
ANGELONE_API_KEY=
ANGELONE_CLIENT_ID=
ANGELONE_PASSWORD=
ANGELONE_TOTP_SECRET=

# Zerodha
ZERODHA_API_KEY=
ZERODHA_API_SECRET=
ZERODHA_ACCESS_TOKEN=

# Upstox
UPSTOX_API_KEY=
UPSTOX_API_SECRET=
UPSTOX_ACCESS_TOKEN=

# Telegram (optional)
TELEGRAM_BOT_TOKEN=
TELEGRAM_CHAT_ID=

Extend BaseBroker — implement 8 methods:

// src/brokers/my-broker.js
import { BaseBroker } from './base.js';

export class MyBroker extends BaseBroker {
  constructor(config) { super(config); this.name = 'MyBroker'; }

  async authenticate()       { /* return access token */ }
  async placeOrder(params)   { /* return { orderId, status } */ }
  async cancelOrder(id)      { /* cancel pending order */ }
  async getHoldings()        { /* return holdings array */ }
  async getPositions()       { /* return positions array */ }
  async getFunds()           { /* return { available, used, total } */ }
  async getOrderList()       { /* return orders array */ }
  async getOrderDetail(id)   { /* return single order */ }
}

Register in src/brokers/index.js and set BROKER=my-broker in .env.

npm run cli                    # Interactive REPL
npm run bot                    # Autonomous scanner
npm run bot:paper              # Auto paper trade on score >= 7
npm run bridge                 # Start Node.js data bridge
npm run agent-ui               # Start full dashboard (bridge + Python)
npm run q RELIANCE             # Quick scan single stock
npm run advisor                # Standalone AI advisor
npm run monitor                # Real-time market monitor
npm run portfolio              # Portfolio viewer
npm run stock                  # Stock detail viewer

node trading-bot.mjs [options]

  --mode        watch | paper | live     Default: watch
  --watchlist   RELIANCE,TCS,INFY       Comma-separated NSE symbols
  --interval    5                        Minutes between scans
  --capital     100000                   Paper trading capital (INR)
  --risk        1                        % of capital to risk per trade
  --min-score   6                        Alert threshold (0-10)
  --index       "NIFTY 50"              Index to screen

Get alerts on your phone and control the bot remotely.

Setup: Set TELEGRAM_BOT_TOKEN and TELEGRAM_CHAT_ID in .env, then run npm run bot.

Use all tools via natural language inside Claude Desktop:

{
  "mcpServers": {
    "ai-trader": {
      "command": "node",
      "args": ["C:/path/to/AI-trader/src/server.js"]
    }
  }
}

AI-trader/
├── dashboard/                    # Python web dashboard
│   ├── run.py                   # Entry point (starts bridge + uvicorn)
│   ├── app.py                   # Starlette routes + WebSocket
│   ├── claude_chat.py           # Claude Code CLI streaming engine
│   ├── agents.py                # Background agent registry
│   ├── bridge.py                # Async HTTP client to Node bridge
│   └── static/index.html        # Single-page dark-theme UI
├── node-bridge.mjs              # REST API wrapping all JS modules (port 3001)
├── trading-bot.mjs              # Autonomous scanner bot
├── cli.mjs                      # Interactive terminal REPL
├── src/
│   ├── server.js                # MCP server (30+ tools)
│   ├── groww-client.js          # NSE India + Yahoo Finance + Groww API
│   ├── analytics.js             # RSI, MACD, BB, ATR, VWAP, SuperTrend...
│   ├── patterns.js              # 22 candlestick patterns
│   ├── greeks.js                # Black-Scholes options calculator
│   ├── fo-skill.js              # F&O strategy builder + regime detection
│   ├── paper-trade.js           # Virtual trading engine
│   ├── trade-journal.js         # Performance tracker
│   ├── history-analyzer.js      # Historical similarity matching
│   ├── telegram.js              # Telegram bot (alerts + control)
│   └── brokers/
│       ├── base.js              # Abstract broker interface
│       ├── groww.js             # Groww adapter
│       ├── angelone.js          # AngelOne Smart API
│       ├── zerodha.js           # Zerodha Kite Connect
│       ├── upstox.js            # Upstox v2
│       └── index.js             # Broker factory
├── tools/                       # Utility scripts
├── dashboards/                  # Terminal-based dashboards
├── data/                        # Runtime data (gitignored)
│   ├── paper-portfolio.json     # Paper trading state
│   ├── skills.json              # AI skills cache
│   ├── trading-memory.json      # Persistent trading memory
│   └── accounts.json            # Broker account configs
├── CLAUDE.md                    # AI context file
├── package.json
└── .env                         # Credentials (gitignored)

• .env and data/accounts.json are in .gitignore — credentials never committed
• All API keys stay local on your machine
• Paper trading is fully isolated — zero real money
• Live mode requires explicit account activation in the dashboard
• Real broker orders require AI to double-confirm with user

1. Fork the repo
2. Create a feature branch (git checkout -b feature/your-feature)
3. Commit changes (git commit -m "Add your feature")
4. Push to branch (git push origin feature/your-feature)
5. Open a Pull Request

MIT — free to use, modify, and share. See LICENSE.

This project is an open-source educational tool. By using this software, you agree:

1. Not financial advice. Nothing in this codebase constitutes financial, investment, or trading advice. All signals and trade plans are algorithm-generated for educational purposes.

2. Risk of loss. Trading equities, futures, and options (F&O) carries substantial risk. You may lose your entire capital. F&O losses can exceed initial investment.

3. No liability. Authors and contributors shall not be held liable for any financial losses arising from use of this software.

4. Backtests are not predictions. Historical results do not predict future performance.

5. Regulatory compliance. Automated trading may be subject to SEBI regulations and broker terms of service. Ensure compliance.

6. Live trading. Real broker mode places real orders with real money. Use only after thorough paper testing.

7. Data accuracy. No guarantees about accuracy or timeliness of market data.

Always consult a SEBI-registered investment advisor before making real trading decisions.

Built for Indian retail traders who want institutional-grade tools without institutional-grade cost.

Get Started • Dashboard • Brokers • Features