Crypto trading bot that lets an LLM manage the portfolio

What it does This Python framework backtests cryptocurrency strategies against Binance data by piping market prices through a directed graph of technical-analysis nodes. It fetches user-defined timeframes in parallel, runs strategies like MacdStrategy or RSIStrategy, aggregates the resulting signals, and finally asks an LLM to make the portfolio-management call.

The interesting bit The whole pipeline is wired as a LangGraph DAG where both timeframes and strategies are hot-swappable via config.yaml—no core code edits required. The LLM sits at the end not as a chatty assistant but as the actual decision node, weighing ensemble signals and risk limits before it green-lights a trade.

Key highlights

• LangGraph orchestrates data, strategy, risk, and portfolio nodes, parallelizing multi-timeframe analysis.
• Strategies and intervals are dynamically loaded from config.yaml; swap Bollinger bands for RSI or add a 4h window without rewriting code.
• Supports OpenAI, Groq, Anthropic, Gemini, and Ollama through a unified LLM interface.
• Automatically renders workflow graphs and backtest performance charts.
• Backtesting and live signal generation are functional; live order execution is still planned.

Caveats

• Full live trading execution is explicitly marked as planned but not yet implemented, so it currently generates signals rather than placing orders.
• The README truncates mid-sentence, leaving the full scope of the risk-management node and gateway module unclear.

Verdict A solid sandbox for developers who want to experiment with LLM-augmented technical strategies without building a pipeline from scratch. Not for anyone who needs a production live-trading system today.