Autonomous cryptocurrency quantitative trading strategy research framework. Evolved from the autoresearch paradigm — instead of optimizing LLMs, this framework lets AI agents autonomously discover, evolve, and execute crypto trading strategies.
Disclaimer: This project is for research and educational purposes. Trading cryptocurrencies carries significant risk. Past performance does not guarantee future results. Use at your own risk.
autoresearch-crypto is a modular quantitative trading framework designed for:
- Strategy Research: Automated search and optimization of trading strategies via evolutionary algorithms
- Backtesting: High-fidelity backtesting with realistic fee/slippage modeling
- Live Trading: Production-ready integration with Nado DEX and OKX
- Agent-Driven Evolution: ATLAS multi-strategy evolution and GEPA reflective evolution engines
| Feature | Description |
|---|---|
| 11 Built-in Strategies | Trend, mean-reversion, scalping, grid, hybrid momentum, adaptive, and more |
| ATLAS Evolution | Multi-agent competitive evolution across strategy archetypes |
| GEPA Reflection | Reflective learning with edge-case detection and strategy-death recovery |
| Market Regime Detection | Automatic regime classification (trending / ranging / volatile) |
| Multi-Timeframe Analysis | Ensemble signals across 1m, 5m, 15m, 1h, 4h timeframes |
| Maker/Taker Fee Optimization | Hybrid execution: POST_ONLY (Maker) for entry/take-profit, IOC (Taker) for stop-loss |
| Walk-Forward Validation | Out-of-sample testing with multiple windows |
- Python 3.10+
- uv package manager (recommended)
- A single NVIDIA GPU (optional, for strategy search — 8GB+ VRAM recommended)
# 1. Clone the repository
git clone <repo-url> autoresearch-crypto
cd autoresearch-crypto
# 2. Install dependencies
uv sync
# 3. Prepare your environment
cp .env.example .env
# Edit .env and add your API keys / private keys# Download ETH 5-minute data (60 days)
uv run python prepare_crypto.py --symbol ETHUSDT --interval 5m --days 60# Quick scan (5 minutes)
uv run python search_eth_optimal.py --quick
# Full search (recommended, ~20 minutes)
uv run python search_eth_optimal.py# Backtest with default strategy
uv run python backtest_quant.py --symbol ETHUSDT --interval 5m --days 60
# Backtest with a specific checkpoint
uv run python backtest_quant.py \
--symbol ETHUSDT --interval 5m --days 60 \
--checkpoint checkpoints/hybrid_mm_eth60d.ptuv run python live_nado_quant.py \
--ticker ETH \
--interval 5m \
--capital 100 \
--checkpoint checkpoints/hybrid_mm_eth60d.pt# Testnet paper trading (recommended for testing)
uv run python live_binance_quant.py \
--symbol BTCUSDT --interval 5m --demo --capital 100
# Live trading (real money — only after strategy is stable)
uv run python live_binance_quant.py \
--symbol BTCUSDT --interval 5m --live --capital 500 --leverage 2autoresearch-crypto/
├── live_nado_quant.py # Nado DEX live trading
├── live_okx_quant.py # OKX live trading
├── live_binance_quant.py # Binance live trading
├── backtest_quant.py # Backtesting engine
├── train_quant.py # Strategy training / search
├── search_eth_optimal.py # ETH optimal strategy search
├── search_deep.py # Deep parameter search
├── prepare_crypto.py # Market data downloader
├── dex/ # Core framework
│ ├── strategies/ # Strategy implementations (11 classes)
│ ├── indicators.py # Technical indicators
│ ├── evolution.py # ATLAS evolution engine
│ ├── reflection.py # GEPA reflective evolution
│ ├── scoring.py # Risk-adjusted scoring
│ ├── market_regime.py # Regime detection
│ ├── config.py # Centralized constants
│ └── live/common.py # Shared live trading utilities
├── checkpoints/ # Strategy checkpoints (kept: quant_model.pt)
├── docs/ # Documentation
│ ├── STRATEGIES.md # Strategy catalog
│ └── 策略.md # Original Chinese strategy notes
└── scripts/ # Utility scripts
├── evolve.py # Run ATLAS evolution
├── evolve_gepa.py # Run GEPA evolution
└── train_all.py # Train all strategy types
| Strategy | Type | Best For | Frequency |
|---|---|---|---|
| HybridMM | Mean-reversion + Momentum | Range + Trend | Low (~2/day) |
| TrendStrategy | Bollinger Band Mean-reversion | Range-bound | Very Low |
| ScalpStrategy | High-frequency Scalping | High Volatility | Very High |
| TrendFollowStrategy | Trend Following | Strong Trends | Low |
| AdaptiveHybridStrategy | Self-adapting | Mixed Regimes | Medium |
| GridStrategy | Grid Trading | Sideways | Medium |
| PureActionStrategy | Extremes Reversal | Overbought/Oversold | Medium |
See docs/STRATEGIES.md for full details.
Competitive evolution where multiple agents (Alpha, Beta, Gamma, Delta) evolve different strategy archetypes simultaneously:
uv run python scripts/evolve.py --generations 30Reflective learning with forced hypothesis logging every 5 rounds:
uv run python scripts/evolve_gepa.py --cycles 30The framework uses a risk-penalized score instead of raw Sharpe ratio to prevent agents from favoring high-leverage / high-drawdown strategies:
Score = Sharpe * (1 - |MaxDD|)^-1 * TradePenalty * EdgeGuard
All trading defaults are centralized in dex/config.py:
| Parameter | Default | Description |
|---|---|---|
INITIAL_CAPITAL |
10000.0 | Backtest initial capital |
COMMISSION |
0.02% | DEX Maker fee |
SLIPPAGE |
0.02% | Execution slippage |
EVAL_MAX_DRAWDOWN |
30% | Hard disable threshold |
| Exchange | Status | File |
|---|---|---|
| Nado DEX | Supported | live_nado_quant.py |
| OKX | Supported | live_okx_quant.py |
| Binance | Supported | live_binance_quant.py |
Contributions are welcome. Please open an issue or pull request.
MIT — see LICENSE.