NeoTrade
A trading cockpit for Indian equities and F&O. A strategy engine watches a universe of
instruments and emits fully-formed trades — size, entry reference, stop, target, the rule
codes that fired, and a composite score — then either executes them (intraday) or routes them
to an approval inbox (long-term). Each user connects their own broker account and chooses per
strategy whether it trades on paper or with real money.
The defining constraint: the AI contribution to any trade's conviction is capped at 30% and
cannot rescue a trade the rules did not already support. That cap is enforced in the type
system, not by convention.
Documentation
Features
- Rule strategy engine — intraday and long-term strategies emitting scored, sized trade
intents; new strategies plug into a registry.
- Composite scoring with a hard AI ceiling — rule conviction decides; the LLM can adjust
within 30% and never below the rule floor.
- Risk-aware sizing — position size derived from account size, per-trade risk, and stop
distance, with exposure limits; never computed inside a strategy.
- Approval inbox — long-term proposals wait for an explicit approve/reject and expire
after three days; intraday proposals execute automatically.
- Backtesting — strategies replay through the same runner and execution simulator used
live, with Indian transaction costs modelled.
- Research and chat — news/sentiment analysis and a tool-using assistant, both through a
self-hosted OmniRoute LLM gateway.
- Live market data over WebSocket, with explicit live/stale/market-closed states.
Tech stack
Backend — FastAPI (Python 3.11+), MongoDB (motor), Redis, pandas/numpy for the indicator
toolkit, yfinance and Zerodha Kite Connect for market data, LangChain/LangGraph for the
research and chat agents, pytest for tests.
Frontend — React 19 + Vite (plain JSX), Tailwind v4, React Router 7, recharts,
framer-motion, axios.
Infrastructure — Docker Compose on an Oracle Cloud VM behind Nginx + Let's Encrypt for the
backend; Vercel for the frontend. Both push-to-deploy from main.
Setup
Docker (recommended)
git clone https://github.com/kushgarg132/NeoTrade.git
cd NeoTrade
# create .env with the values below, then:
docker compose up -d --build backend
MongoDB and Redis are external (managed) services, not compose containers — point
MONGODB_URL and REDIS_URL at them. Also required: GOOGLE_CLIENT_ID, JWT_SECRET,
CORS_ALLOWED_ORIGINS, and OMNIROUTE_API_KEY + OMNIROUTE_BASE_URL for anything using the
LLM. KITE_API_KEY/KITE_API_SECRET are optional — the app treats an unconfigured broker as
a normal state, not an error.
The frontend is served by Vercel; compose runs only backend and ingest.
Manual
# backend (needs Mongo and Redis reachable)
cd backend
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
uvicorn server:app --reload --port 8001
# frontend
cd frontend
npm install
npm run dev
Tests
cd backend && python -m pytest
Project structure
backend/
├── strategies/ # rule strategies (intraday, longterm) + registry
├── scoring/ # composite score, AI cap, rule floor
├── engine/ # runner, sizing, portfolio, execution, backtest, persistence
├── suggestions/ # approval inbox: store, sink, scan, service
├── auth/ # Google sign-in, JWT, refresh tokens, Kite session
├── data/ # market data providers and feeds
├── components/ # analyst/chat agents, shared models, risk rules
├── research/ # LangGraph research agent
├── routers/ # FastAPI endpoints
└── tests/
frontend/src/
├── pages/ # Dashboard, Suggestions, Portfolio, Trading, Watchlist, Scanner, Settings
├── components/ # layout + UI
├── context/ # auth state
└── utils/ # axios instance, interceptors, endpoints
API documentation
With the backend running: Swagger at /api/v1/docs, ReDoc at /api/v1/redoc.
