AI Trading Bot — All-Time Record

Fully automated intraday trading bot for NIFTY options on Angel One SmartAPI, with Claude AI for trade narrative + post-trade review.

Backend: FastAPI on a DigitalOcean droplet (Docker) | Frontend: Next.js on Vercel | Data: Angel One SmartAPI

---

Roadmap

Phase	Budget	Style	Goal	Status
A	₹1.25L	Intraday NIFTY	Compound to ₹1.5L	✅ Active
B	₹1.5L	Intraday + Swing	Compound to ₹15L	🔒 Locked
C	₹15L+	Options Selling (Straddle / Iron Condor)	Monthly income	🔒 Locked

---

Strategy

The bot runs a single live strategy: ATR Intraday.

ATR Intraday

• Signal: SMA50 + SMA20 + EMA9 + RSI + MACD + Bollinger + Volume + ATR-vol filter
  • VWAP + ORB + 15m trend/RSI + PDH/PDL + 12 candlestick patterns
  • PCR + OI walls + herd-gate + S/R zones (RBD/DBR + structure)
• Timeframe: 5-minute bars
• Score range: −10 to +10. Live entry threshold: ≥ ±8 (config default ±6, hard-floor of 8 enforced in strategies/trend_strategy.py to skip 6–7 traps).
• R:R: 1:3.0 — SL = ½ × min(5m ATR, 50)pts (≥ 5pts floor); TP = SL × 3.0
• Entry window: 09:30–15:20 IST. Mandatory square-off at 15:20.
• Strike selection: Walk ATM ± 10 strikes in ₹50 steps until premium falls in ₹155–₹165 range. Closest match used as fallback.
• Zone gating: Pre-market watch zones (PDH/PDL/ORB/weekly) computed at 09:00. Trade rejected if signal fights an active zone within 30pts.
• Zone-reversal early entry: Independent of score — at most one entry/day. When price taps a watch zone and the last 5m candle confirms rejection (bearish at resistance / bullish at support), force-entry CE/PE.

C-ICT was active previously and will be rebuilt from scratch later. Musashi / Raijin / SMC / Expiry-Day Gap research are not in the live path.

---

Data Pipeline

Angel One SmartAPI   ←  primary: real NSE bars + option chain + VIX
        ↓ if unavailable
Cycle skipped        ←  never trade on bad data

5-minute option chain snapshots are persisted to build a real historical premium dataset over time (mitigates the BS-vs-live premium mismatch that hurts naive backtests).

---

Architecture

DigitalOcean droplet (Docker)
└── api  (FastAPI + APScheduler in one process)
    ├── api/server.py           # REST + WebSocket, snapshot every 5s
    ├── core/bot_runner.py      # AsyncIOScheduler — schedules every cycle
    │
    ├── strategies/
    │   ├── trend_strategy.py   # Order execution, SL/TP, zone gating
    │   ├── signal_scorer.py    # ATR Intraday scoring engine (atr_only)
    │   ├── indicators.py       # EMA, RSI, ATR, VWAP, swing structure
    │   ├── patterns.py         # 12 candlestick pattern detectors
    │   └── vix_filter.py       # India VIX regime helper
    │
    ├── data/
    │   ├── angel_fetcher.py    # Angel One singleton — TOTP, bars, option LTP, VIX
    │   ├── market.py           # RealMarketData — daily + intraday indicators
    │   ├── option_chain.py     # OI walls, PCR, max-pain, herd-gate
    │   └── oi_data.py          # OI helpers
    │
    ├── core/
    │   ├── brain.py            # Claude AI trade remarks
    │   ├── broker.py           # MockBroker (paper) + AngelOneBroker (live)
    │   ├── memory.py           # SQLite — trades, signal_log, daily_summaries
    │   ├── records.py          # All-time records tracker
    │   ├── journal.py          # Daily journal JSON + Claude review (haiku)
    │   ├── ipc.py              # Flag-file IPC: pause, force-trade, day-bias, settings
    │   ├── paper_seller.py     # Buyer-vs-seller paper P&L comparison
    │   ├── sr_levels.py        # RBD/DBR institutional zones, structure detection
    │   ├── zone_briefing.py    # Pre-market PDH/PDL/ORB/weekly watch zones (9 AM)
    │   ├── greeks.py           # Black-Scholes delta/gamma/theta/vega
    │   ├── trade_analyst.py    # Post-trade Claude analysis
    │   └── angel_error_log.py  # Append-only error log
    │
    ├── backtesting/
    │   ├── engine.py           # ATR Intraday bar-by-bar replay
    │   ├── metrics.py          # Win rate, Sharpe, drawdown, profit factor
    │   └── charges.py          # NSE brokerage + STT + exchange + GST
    │
    └── db/
        └── trading.db          # SQLite

Vercel (Next.js frontend)
└── frontend/                   # Dashboard, signal radar, P&L, backtest, journal

---

Setup

1. Install dependencies

pip install -r requirements.txt

2. Configure .env

# Angel One SmartAPI
ANGEL_API_KEY=...
ANGEL_CLIENT_ID=...
ANGEL_PASSWORD=...
ANGEL_TOTP_TOKEN=...        # base32 secret from Angel One 2FA setup

# Claude AI
ANTHROPIC_API_KEY=sk-ant-...

# Mode
TRADING_MODE=paper          # paper = simulation | live = real money

All financial parameters (capital, lot size, SL/TP, premium target) live in config.py — not env vars. See feedback_deploy_habits.md for the rationale.

3. Daily token routine — automatic on droplet

Nothing to do daily. Angel One TOTP login is automatic — data/angel_fetcher.py reads ANGEL_API_KEY / ANGEL_CLIENT_ID / ANGEL_PASSWORD / ANGEL_TOTP_TOKEN from .env and generates a fresh session on first authenticated call. The bot runs 24/7 on the DigitalOcean droplet so this happens automatically every container start and as needed during the session.

scripts/get_token.py remains as a manual sanity check only:

# Run inside the api container if you ever need to verify creds work
docker compose exec api python scripts/get_token.py

It does not need to be scheduled on a laptop or as a cron — the droplet handles its own auth.

---

Run Locally

# Start FastAPI backend (Bot runs inside the same process via APScheduler lifespan)
python -m uvicorn api.server:app --host 0.0.0.0 --port 8000 --reload

API at http://localhost:8000, WebSocket at ws://localhost:8000/ws.

---

API Endpoints

Endpoint	Method	Description
/api/auth/token	POST	Login (returns JWT)
/api/snapshot	GET	Full dashboard snapshot
/api/bot/debug	GET	Live ATR signal score
/api/bot/pause	POST	Pause trading
/api/bot/resume	POST	Resume trading
/api/bot/bias	POST	Set day bias (BULLISH / NEUTRAL / BEARISH)
/api/trade/force	POST	Queue a manual trade
/api/live/preflight	POST	Validate token, contract, LTP, margin
/api/backtest	POST	Run ATR Intraday backtest
/api/journals	GET	List daily journals
/api/journals/{date}	GET	Get a specific day's journal
/api/signal-log	GET	Every 5-min evaluation (trade or no-trade)
/api/paper-comparison	GET	Buyer-vs-seller paper P&L per signal
/api/chart-data	GET	NIFTY 5m candles + S/R + EMA + POC
/api/event-blocks	GET/POST	Event-blocked dates (Budget, RBI MPC)
/api/market-holidays	GET/POST	NSE market holidays
/api/angel/session	POST	Force fresh Angel One TOTP login
/ws	WebSocket	Live snapshot every 5s

---

Backtesting

POST /api/backtest
{
  "strategy": "ATR Intraday",
  "symbol": "NIFTY",
  "period": "60d",
  "interval": "5m",
  "capital": 125000,
  "risk_pct": 2.0,
  "rr_ratio": 3.0,
  "daily_loss_limit_pct": 5.0
}

Backtest replays real Angel One historical bars. For realistic results (bar-level SL + slippage modeling), use scripts/backtest_live_atr.py.

---

Paper vs Live

	Paper Mode	Live Mode
TRADING_MODE	paper	live
Real money	No	Yes
Orders placed	Simulated (MockBroker)	Real Angel One orders
Market data	Angel One	Same

Always run paper for at least 1–2 weeks after material strategy changes. Verify signal scores in the Signal Radar, then flip TRADING_MODE=live.

---

Deploy

git pull
docker compose build --no-cache
docker compose up -d --force-recreate

Or use deploy.sh. The GitHub Actions workflow (.github/workflows/deploy.yml) SSHs into the droplet on push to main and runs the same sequence.

---

Security

• All credentials in .env — never committed (.gitignore)
• Logs automatically mask API keys, passwords, tokens (core/security.py)
• Never push .env to git