AI Multi-Agent Trading Copilot

A modular SaaS platform combining 80+ quantitative strategies from 151 Trading Strategies (Kakushadze & Serur, 2018) with a 6-agent LLM architecture powered by Claude.

---

Architecture

┌─────────────────────────────────────────────────────────────────┐
│                        Signal Pipeline (DAG)                     │
│                                                                   │
│  ┌──────────────────────────────────────────┐                    │
│  │           Analyst Team (parallel)         │                    │
│  │  [Fundamental] [Technical] [Sentiment]    │                    │
│  │               [Macro]                     │                    │
│  └──────────────────────┬───────────────────┘                    │
│                         ↓                                        │
│              ┌──────────────────────┐                            │
│              │   Bull / Bear Debate  │                            │
│              └──────────┬───────────┘                            │
│                         ↓                                        │
│              ┌──────────────────────┐                            │
│              │     Risk Manager      │                            │
│              └──────────┬───────────┘                            │
│                         ↓                                        │
│              ┌──────────────────────┐                            │
│              │    Trader Agent       │ ← claude-opus-4-6          │
│              └──────────┬───────────┘                            │
│                         ↓                                        │
│                  Final Signal + TP/SL                             │
└─────────────────────────────────────────────────────────────────┘

Tech Stack

Layer	Technology
LLM Orchestration	LangGraph + LangChain
Primary LLM	Claude Opus 4.6 (Trader) + Sonnet 4.6 (Analysts)
Backend	Python 3.11 + FastAPI
Frontend	Next.js 14 + TradingView Lightweight Charts
Database	TimescaleDB (PostgreSQL 15)
Cache / Bus	Redis 7
Containerization	Docker Compose

Quick Start

Prerequisites

• Docker Desktop
• An Anthropic API key (get one at console.anthropic.com)
• (Optional) Polygon.io and Alpha Vantage API keys for live data

1. Clone and configure

git clone <repo>
cd ai-trading-copilot
cp .env.example .env

Edit .env and set:

ANTHROPIC_API_KEY=sk-ant-...

2. Start the full stack

docker compose up --build

This starts:

• PostgreSQL/TimescaleDB on port 5432
• Redis on port 6379
• FastAPI backend on port 8000
• Next.js frontend on port 3000

3. Open the app

Navigate to http://localhost:3000

• Demo account: [email protected] / demo1234

---

Local Development (without Docker)

Backend

cd backend
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate
pip install -r requirements.txt

# Start with local postgres/redis running
uvicorn app.main:app --reload --port 8000

Frontend

cd frontend
npm install
npm run dev

---

API Reference

All endpoints under /api/v1. JWT required via Authorization: Bearer <token>.

Method	Endpoint	Description
POST	/api/v1/signals/generate	Trigger multi-agent analysis
GET	/api/v1/signals/{id}	Retrieve signal details
GET	/api/v1/portfolio/positions	Current positions + P&L
POST	/api/v1/portfolio/execute	Execute a trade signal
GET	/api/v1/agents/status	Health of all 6 agents
GET	/api/v1/backtest/{strategy}	Run historical backtest
POST	/api/v1/debate/trigger	Force bull/bear debate
WS	/ws/v1/signals/stream	Real-time signal stream

Example: Generate a Signal

curl -X POST http://localhost:8000/api/v1/signals/generate \
  -H "Content-Type: application/json" \
  -d '{"ticker": "AAPL", "timeframe": "1D", "asset_class": "stocks"}'

Sample Signal Response

{
  "signal_id": "550e8400-e29b-41d4-a716-446655440000",
  "ticker": "AAPL",
  "direction": "LONG",
  "entry_price": 227.50,
  "stop_loss": 220.10,
  "take_profit_1": 238.60,
  "take_profit_2": 245.90,
  "take_profit_3": 257.20,
  "confidence_score": 78.4,
  "agent_votes": [
    {"agent": "FundamentalAnalyst", "direction": "LONG", "confidence": 82.0, "reasoning": "..."},
    {"agent": "TechnicalAnalyst", "direction": "LONG", "confidence": 75.0, "reasoning": "..."},
    {"agent": "SentimentAnalyst", "direction": "LONG", "confidence": 68.0, "reasoning": "..."},
    {"agent": "MacroAnalyst", "direction": "NEUTRAL", "confidence": 55.0, "reasoning": "..."}
  ],
  "reasoning_chain": ["EMA crossover bullish", "RSI not overbought", "Earnings beat last quarter"],
  "strategy_sources": ["price_momentum_3.1", "ema_crossover_3.11", "earnings_momentum_3.2"]
}

---

Strategies Implemented

Strategy	Ref	Asset Class
Price Momentum	3.1	Stocks
Earnings Momentum	3.2	Stocks
Value Factor	3.3	Stocks
Pairs Trading	3.8	Stocks
Mean Reversion (Clusters)	3.9	Stocks
EMA Crossover Systems	3.11-3.13	Stocks
Stat Arb (Markowitz)	3.18	Stocks
Sector Rotation	4.1	ETFs
VIX Basis	7.2	Volatility
FX Carry	8.2	Forex
ANN Crypto	18.2	Crypto
Sentiment (Naive Bayes)	18.3	Crypto
Macro Momentum	19.2	Global Macro
Announcement Trading	19.5	Global Macro

---

Pricing Tiers

Tier	Price	Features
Free	$0	Paper trading, 3 strategies, delayed data
Retail	$49/mo	All strategies, real-time data, 100 signals/day
Pro	$199/mo	API access, custom agent tuning, webhooks
Enterprise	$499/mo	Dedicated infra, white-label, SLA

---

Project Structure

ai-trading-copilot/
├── backend/
│   ├── app/
│   │   ├── main.py              # FastAPI app entry point
│   │   ├── config.py            # Settings (pydantic-settings)
│   │   ├── agents/              # 6 LangGraph agents
│   │   │   ├── base.py
│   │   │   ├── fundamental.py
│   │   │   ├── technical.py
│   │   │   ├── sentiment.py
│   │   │   ├── macro.py
│   │   │   ├── risk_manager.py
│   │   │   └── trader.py
│   │   ├── pipeline/
│   │   │   ├── graph.py         # LangGraph DAG
│   │   │   └── state.py         # Shared state TypedDict
│   │   ├── api/routes/          # REST endpoints
│   │   ├── strategies/          # Pure quant implementations
│   │   ├── models/              # Pydantic response models
│   │   ├── db/                  # SQLAlchemy + init SQL
│   │   ├── data/                # Market data service
│   │   └── auth/                # JWT + middleware
│   ├── requirements.txt
│   └── Dockerfile
├── frontend/
│   ├── app/                     # Next.js App Router pages
│   │   ├── page.tsx             # Landing
│   │   ├── dashboard/           # Main trading terminal
│   │   ├── signals/             # Signal history
│   │   ├── portfolio/           # Portfolio tracker
│   │   ├── pricing/             # Subscription tiers
│   │   └── login/               # Auth
│   ├── components/              # React components
│   │   ├── TradingChart.tsx     # TradingView chart
│   │   ├── SignalCard.tsx       # Signal display
│   │   ├── AgentStatus.tsx      # Agent health indicators
│   │   └── Sidebar.tsx          # Navigation
│   ├── lib/
│   │   ├── api.ts               # API client + mock data
│   │   └── websocket.ts         # WS client hook
│   └── Dockerfile
├── docker-compose.yml
├── .env.example
└── README.md

---

Risk Management Rules

The Risk Manager agent enforces portfolio-level constraints from the spec:

• Max drawdown: 15% (circuit breaker stops new signals)
• Position sizing: Kelly criterion, half-Kelly default, max 5% per position
• Sector concentration: Max 30% per sector
• Correlation limit: New positions must have ρ < 0.7 with existing
• Vol targeting: 10% annualized portfolio volatility
• Dollar neutrality: Enforced for stat-arb strategies (3.18)

---

Development Notes

• Anti-repainting: All signals generated on confirmed (closed) bars only, enforced at framework level
• Mock mode: If no API keys are set, the system runs with realistic mock data — fully functional for development and demos
• Agent fallbacks: If any analyst agent fails, the pipeline continues with remaining agents' outputs
• Audit trail: Every agent decision and debate transcript is logged to agent_logs table

---

Built with Claude API · LangGraph · FastAPI · Next.js