Dolores - AI Crypto Trading Agent

Dolores is an autonomous AI trading agent powered by DeepSeek-V3, Qlib (Microsoft AI), and LightGBM. She analyzes multi-coin market structures, sentiment data (Funding Rate, OI), and news to make risk-managed trading decisions.

🚀 Features

• Multi-Coin Prediction: Uses Qlib to predict relative strength (Ranking) of BTC, ETH, SOL, BNB, DOGE.
• Sentiment Analysis: Integrates Funding Rate Z-Scores, Open Interest changes, and Fear & Greed Index.
• AI Reasoning: DeepSeek LLM analyzes quantitative data + news context to generate trading plans.
• Risk Management:
  • Max 3 open positions.
  • Max 50% NAV exposure per decision.
  • Hard stop-loss and take-profit logic.
• Automated Cycle: run_daily_cycle.py orchestrates the full pipeline.

🛠️ Setup

1. Install Dependencies:

   pip install -r requirements.txt

   (Note: Qlib requires specific installation steps, refer to Qlib Docs)

2. Environment Variables: Create a .env file:

   DEEPSEEK_API_KEY=your_key_here
   OKX_API_KEY=your_key (optional for public data)
   OKX_SECRET_KEY=your_key
   OKX_PASSPHRASE=your_key

3. Run the Cycle:

   python3 run_daily_cycle.py

📂 Architecture

• fetch_okx_data.py: Fetches OHLCV & Sentiment data.
• generate_multi_coin_signals.py: Computes technical indicators.
• prepare_multi_coin_qlib.py: Formats data for Qlib.
• inference_qlib_model.py: Runs LightGBM model inference.
• fetch_onchain_and_news.py: Aggregates news & on-chain metrics.
• DeepSeek_Agent.py: The "Brain" - generates decisions.
• run_daily_cycle.py: The "Heartbeat" - runs the loop.

📊 Data Storage

• Market Data: Local CSVs in data_features/ and qlib_data/.
• State: portfolio_state.json (Track positions).
• Logs: agent_decision_log.json (Latest decision).

⚠️ Disclaimer

This is experimental software. Use at your own risk.