VectorBT Pro Agent System

A sophisticated multi-agent system for vectorbtpro that uses specialized AI agents to help with trading strategy development, backtesting, optimization, and analysis.

Status: personal research project. Not investment advice and not production software.

🎨 Web Interface Available!

NEW: Beautiful web interface built with Next.js, shadcn/ui, and Tailwind CSS v3!

# Quick start the web interface
./start_web.sh

# Or manually:
# Terminal 1: cd backend && python main.py
# Terminal 2: cd frontend && npm run dev
# Open http://localhost:3000

See WEB_README.md for complete web interface documentation.

Overview

This system consists of:

• 7 Specialist Agents - Each expert in a specific domain of vectorbtpro
• 1 Master Coordinator - Delegates tasks and synthesizes results
• Web Interface - Beautiful UI for interacting with agents (NEW!)
• CLI Interface - Interactive command-line interface
• Python API - Programmatic access

The Master Agent intelligently routes requests to the appropriate specialist agents, coordinates multi-step workflows, and provides comprehensive summaries.

Architecture

┌─────────────────────────────────────────────┐
│           Master Agent                       │
│  (Coordination, Delegation, Synthesis)       │
└─────────────────┬───────────────────────────┘
                  │
                  │ Delegates to
                  ▼
    ┌─────────────┴─────────────┐
    │                            │
    ▼                            ▼
┌─────────┐  ┌─────────┐  ┌─────────┐
│Portfolio│  │Indicator│  │Backtest │  ...
│ Agent   │  │ Agent   │  │ Agent   │
└─────────┘  └─────────┘  └─────────┘

Specialist Agents

1. Portfolio Agent

Expertise: Portfolio construction, optimization, and rebalancing

Capabilities:

• Portfolio construction
• Asset allocation strategies
• Rebalancing methodologies
• Risk-parity portfolios
• Efficient frontier analysis

Example Tasks:

• "Create an equal-weighted portfolio"
• "Optimize weights for maximum Sharpe ratio"
• "Implement a monthly rebalancing strategy"

2. Indicator Agent

Expertise: Technical indicators and signal generation

Capabilities:

• Technical indicators (RSI, MACD, Bollinger Bands, etc.)
• Signal generation
• Custom indicator development
• Indicator combinations

Example Tasks:

• "Create RSI-based entry signals"
• "Combine MACD and Bollinger Bands"
• "Build a custom momentum indicator"

3. Backtest Agent

Expertise: Strategy backtesting and performance analysis

Capabilities:

• Strategy backtesting
• Performance metrics calculation
• Trade analysis
• Benchmark comparison
• Walk-forward testing

Example Tasks:

• "Backtest a moving average crossover"
• "Calculate Sharpe and Sortino ratios"
• "Compare strategy vs S&P 500"

4. Data Agent

Expertise: Data fetching, cleaning, and preparation

Capabilities:

• Data fetching from various sources
• Data cleaning and validation
• Missing data handling
• Multi-asset alignment
• Time series preparation

Example Tasks:

• "Fetch OHLCV data for tech stocks"
• "Clean and align multi-asset data"
• "Handle missing values in price data"

5. Optimization Agent

Expertise: Parameter optimization and hyperparameter tuning

Capabilities:

• Parameter optimization
• Grid search
• Walk-forward optimization
• Overfitting prevention
• Sensitivity analysis

Example Tasks:

• "Optimize RSI periods"
• "Perform walk-forward analysis"
• "Find optimal moving average lengths"

6. Visualization Agent

Expertise: Data visualization and reporting

Capabilities:

• Interactive charts
• Equity curves
• Drawdown visualization
• Heatmaps
• Performance reports

Example Tasks:

• "Plot equity curve with drawdowns"
• "Create indicator overlay chart"
• "Visualize optimization results"

7. Risk Agent

Expertise: Risk management and analysis

Capabilities:

• Risk metrics (VaR, CVaR, etc.)
• Position sizing
• Drawdown analysis
• Risk-adjusted returns
• Stop-loss strategies

Example Tasks:

• "Calculate Value at Risk"
• "Implement volatility-based position sizing"
• "Analyze maximum drawdown"

Installation

1. Clone the repository:

git clone <repository-url>
cd FernandoGatti

2. Install dependencies:

pip install -r requirements.txt

3. Set up environment variables:

# Create .env file
echo "ANTHROPIC_API_KEY=your_api_key_here" > .env

Usage

Web Interface (Recommended) 🌐

The easiest way to use the system is through the beautiful web interface:

# Quick start (runs both backend and frontend)
./start_web.sh

# Or start manually:
# Terminal 1 - Backend:
cd backend && python main.py

# Terminal 2 - Frontend:
cd frontend && npm install && npm run dev

# Open browser to http://localhost:3000

Features:

• Modern dashboard with real-time stats
• Visual agent cards showing capabilities
• Task submission with examples
• Beautiful results display
• Task history tracking
• Dark mode by default

See WEB_README.md for detailed web interface documentation.

Interactive CLI

Command-line interface for terminal users:

python examples/interactive_cli.py

Commands:

• Type any question or request in natural language
• agents - Show all available agents
• history - Show task history
• info <agent> - Get details about a specific agent
• help - Show help
• exit - Exit the system

Example Interactions:

You: Create a simple moving average crossover strategy and backtest it

[Master Agent processes and delegates to indicator and backtest agents]

You: Optimize the parameters for best Sharpe ratio

[Master Agent delegates to optimization agent]

You: Show me the risk metrics and equity curve

[Master Agent delegates to risk and visualization agents]

Programmatic Usage

from agents import MasterAgent

# Initialize
master = MasterAgent()

# Display available agents
master.display_agents()

# Process a request
result = master.process_request(
    "Create a momentum strategy using RSI and backtest it"
)

# Access results
print(result["summary"])
print(result["task_plan"])

# Get agent info
info = master.get_agent_info("portfolio")
print(info)

# View history
history = master.get_task_history()

Basic Examples

python examples/basic_usage.py

This runs several predefined examples demonstrating different capabilities.

How It Works

1. User Request → You ask a question or make a request
2. Analysis → Master Agent analyzes the request
3. Planning → Creates a task plan identifying which agents are needed
4. Delegation → Delegates tasks to specialist agents
5. Execution → Each agent processes their assigned task
6. Synthesis → Master Agent combines results into a coherent summary
7. Response → You receive a comprehensive answer with code examples

Example Workflows

Simple Workflow

Request: "Create RSI indicator signals"
→ Indicator Agent generates signals
→ Master Agent returns code and explanation

Complex Multi-Agent Workflow

Request: "Build and optimize a complete trading system"
→ Data Agent: Fetches historical data
→ Indicator Agent: Creates technical indicators
→ Backtest Agent: Backtests the strategy
→ Optimization Agent: Optimizes parameters
→ Risk Agent: Analyzes risk metrics
→ Visualization Agent: Creates charts
→ Master Agent: Synthesizes complete report

Configuration

Environment Variables

Create a .env file:

# Required
ANTHROPIC_API_KEY=your_api_key_here

# Optional
DEFAULT_MODEL=claude-3-5-sonnet-20241022

Agent Configuration

You can customize agents by modifying their system prompts in agents/<agent_name>_agent.py.

Advanced Usage

Using Individual Agents

from agents import PortfolioAgent, BacktestAgent

# Use a specific agent directly
portfolio_agent = PortfolioAgent()

# Get capabilities
capabilities = portfolio_agent.get_capabilities()

# Process a task
from utils.message_types import TaskRequest
import uuid

task = TaskRequest(
    task_id=str(uuid.uuid4()),
    requester="user",
    agent_type="portfolio",
    description="Create an equal-weighted portfolio",
    parameters={"symbols": ["AAPL", "GOOGL", "MSFT"]}
)

response = portfolio_agent.process_task(task)
print(response.result)

Custom Agents

Create your own specialist agent:

from agents.base_agent import BaseAgent
from utils.message_types import AgentCapability

class MyCustomAgent(BaseAgent):
    def __init__(self):
        super().__init__(agent_type="custom")

    def get_capabilities(self) -> AgentCapability:
        return AgentCapability(
            agent_type=self.agent_type,
            description="My custom agent description",
            capabilities=["capability1", "capability2"],
            example_tasks=["Example task 1", "Example task 2"]
        )

    def get_system_prompt(self) -> str:
        return "Your custom system prompt here..."

Project Structure

FernandoGatti/
├── agents/                     # Agent system
│   ├── __init__.py
│   ├── base_agent.py           # Base agent class
│   ├── master_agent.py         # Master coordinator
│   ├── portfolio_agent.py      # Portfolio specialist
│   ├── indicator_agent.py      # Indicator specialist
│   ├── backtest_agent.py       # Backtest specialist
│   ├── data_agent.py           # Data specialist
│   ├── optimization_agent.py   # Optimization specialist
│   ├── visualization_agent.py  # Visualization specialist
│   └── risk_agent.py           # Risk specialist
├── utils/                      # Utilities
│   ├── __init__.py
│   └── message_types.py        # Message types for communication
├── examples/                   # CLI examples
│   ├── basic_usage.py          # Basic examples
│   ├── interactive_cli.py      # Interactive CLI
│   └── test_setup.py           # Setup verification
├── backend/                    # FastAPI backend (NEW!)
│   ├── main.py                 # API server
│   └── requirements.txt        # Backend dependencies
├── frontend/                   # Next.js frontend (NEW!)
│   ├── app/                    # Next.js app directory
│   │   ├── globals.css         # Tailwind + theme
│   │   ├── layout.tsx          # Root layout
│   │   └── page.tsx            # Main dashboard
│   ├── components/             # React components
│   │   ├── ui/                 # shadcn/ui components
│   │   ├── agent-card.tsx      # Agent display
│   │   ├── task-form.tsx       # Task submission
│   │   ├── result-display.tsx  # Results view
│   │   └── stats-card.tsx      # Statistics
│   ├── lib/                    # Utilities
│   ├── package.json            # Frontend dependencies
│   └── tailwind.config.ts      # Tailwind config
├── requirements.txt            # Core dependencies
├── start_web.sh                # Web interface launcher
├── README.md                   # This file
└── WEB_README.md               # Web interface docs

Requirements

Core System

• Python 3.8+
• vectorbtpro >= 1.0.0
• anthropic >= 0.18.0
• rich >= 13.0.0
• pydantic >= 2.0.0
• python-dotenv >= 1.0.0

Web Interface (Optional)

• Node.js 18+ (for frontend)
• FastAPI (for backend API)
• Next.js 14 (frontend framework)
• See backend/requirements.txt and frontend/package.json for details

Tips for Best Results

1. Be specific - The more specific your request, the better the results
2. Use multiple steps - Break complex tasks into steps for the agents
3. Review code - Always review generated code before using in production
4. Iterate - Refine your requests based on the results
5. Combine agents - Complex workflows often benefit from multiple agents

Example Requests

Data & Preparation:

• "Fetch OHLCV data for AAPL, GOOGL, MSFT from 2020 to 2024"
• "Clean the data and handle missing values"

Indicators & Signals:

• "Create RSI indicator with 14-period"
• "Generate buy signals when RSI crosses above 30"
• "Combine MACD and Bollinger Bands for entry signals"

Backtesting:

• "Backtest a simple moving average crossover strategy"
• "Calculate all performance metrics including Sharpe ratio"
• "Compare against buy-and-hold benchmark"

Optimization:

• "Optimize RSI periods from 10 to 30"
• "Perform walk-forward optimization to prevent overfitting"
• "Find parameters that maximize risk-adjusted returns"

Portfolio:

• "Create a risk-parity portfolio with 5 assets"
• "Optimize portfolio weights for maximum Sharpe ratio"
• "Implement monthly rebalancing"

Risk Management:

• "Calculate Value at Risk at 95% confidence"
• "Implement volatility-based position sizing"
• "Analyze maximum drawdown and recovery periods"

Visualization:

• "Plot equity curve with drawdown overlay"
• "Create heatmap of optimization results"
• "Visualize rolling Sharpe ratio"

Troubleshooting

Issue: API Key Error

• Make sure ANTHROPIC_API_KEY is set in .env file
• Verify the API key is valid

Issue: Module Not Found

• Run pip install -r requirements.txt
• Make sure you're in the correct directory

Issue: Agent Not Responding

• Check your internet connection
• Verify API rate limits haven't been exceeded

Contributing

Contributions are welcome! Areas for improvement:

• Additional specialist agents
• Enhanced task planning algorithms
• Better result synthesis
• More example workflows
• Performance optimizations

License

No license has been set for this repository yet.

Support

For issues, questions, or contributions, please open an issue on GitHub.

---

Built with ❤️ using Claude 3.5 Sonnet and vectorbtpro