AgentDojo

Training ground simulation environment for backtesting AI agents against historical market regimes.

Overview

AgentDojo is a comprehensive backtesting and simulation platform for Permalator AI trading agents. It allows you to test agent strategies against historical data, simulate different market regimes, and analyze performance metrics before deploying to live markets.

Features

• Historical Backtesting: Test agents against historical market data
• Regime Simulation: Simulate trend, chop, panic, compression, and cascade regimes
• Performance Metrics: Sharpe ratio, max drawdown, win rate, and more
• Agent Comparison: Compare multiple agents side-by-side
• Monte Carlo Simulation: Run multiple simulations for statistical validation
• Visual Analytics: Charts and reports for strategy analysis

Tech Stack

• Backend: Python FastAPI
• Database: PostgreSQL + SQLAlchemy (async)
• Math: NumPy, Pandas, Statsmodels
• Market Data: CCXT integration

Quick Start

# Install dependencies
poetry install

# Set up environment
cp env.example .env
# Edit .env with your configuration

# Run database migrations
alembic upgrade head

# Start server
poetry run uvicorn agent_dojo.main:app --reload

Data Provider

AgentDojo uses CCXT to fetch real historical market data from cryptocurrency exchanges. The default exchange is Binance, but can be configured to use any CCXT-supported exchange (e.g., Bybit, OKX, Kraken).

Data fetched includes:

• Open, High, Low, Close prices
• Volume
• Timestamps

Environment Variables

Variable	Description	Default
DATABASE_URL	PostgreSQL connection string	postgresql://agentdojo:agentdojo123@localhost:5432/agent_dojo
REDIS_URL	Redis connection string (optional)	redis://localhost:6379/0
SECRET_KEY	JWT signing secret	Generate new one!
DEBUG	Debug mode	False
LOG_LEVEL	Logging level	INFO

Project Structure

agent-dojo/
├── agent_dojo/
│   ├── main.py              # FastAPI app
│   ├── core/                 # Config, logging, security
│   ├── models/               # Database models
│   ├── providers/            # Market data providers
│   ├── backtest/             # Backtesting engine
│   ├── simulation/           # Regime simulation
│   └── api/                  # API routes
├── tests/                  # Test suite
├── alembic/               # Database migrations
├── logs/                  # Log files
├── pyproject.toml          # Poetry dependencies
├── Dockerfile             # Docker build
├── docker-compose.yml      # Docker Compose
├── env.example            # Environment template
└── README.md             # This file

API Endpoints

Method	Endpoint	Description
POST	/api/v1/backtest	Run backtest
POST	/api/v1/simulate	Run regime simulation
GET	/api/v1/backtests	List backtests
GET	/api/v1/backtests/{id}	Get backtest results
GET	/api/v1/backtests/{id}/metrics	Get performance metrics
POST	/api/v1/compare	Compare multiple agents
GET	/api/v1/agents/{id}/performance	Get agent performance history

Usage Example

from agent_dojo import BacktestEngine, SimulationConfig

# Configure backtest
config = SimulationConfig(
    agent_id="agent_001",
    symbol="BTC/USDT:USDT",
    start_date="2024-01-01",
    end_date="2024-12-31",
    initial_capital=10000,
    leverage=10
)

# Run backtest
engine = BacktestEngine()
results = await engine.run(config)

# Analyze results
print(f"Sharpe Ratio: {results.metrics.sharpe_ratio}")
print(f"Max Drawdown: {results.metrics.max_drawdown}")
print(f"Total Return: {results.metrics.total_return}%")

License

MIT License - see LICENSE for details.

Links

• Website: https://permalator.com/
• Docs: https://permalator.gitbook.io/documentation/
• X: https://x.com/permalator
• GitHub: https://github.com/permalator