TradingAgents: Six LLM agents collaborate to make trading decisions

This is a multi-agent framework that faithfully replicates the organizational structure of a real trading firm. Specialized LLM agents, ranging from fundamental, sentiment, news, and technical analysts to traders, risk managers, and portfolio managers, each perform their designated roles. The core concept goes beyond individual model judgments, deriving final trading strategies through dynamic debates among the agents.

It supports global stock markets and cryptocurrencies covered by Yahoo Finance. It handles tickers from various regions, including the US, Hong Kong, Tokyo, London, India, China A-shares, Canada, and Australia, automatically identifying benchmarks and company information appropriate for each market. It offers broad support for major LLM providers such as OpenAI, Google, Anthropic, xAI, DeepSeek, Qwen, GLM, and MiniMax, and also allows running local models via Ollama.

Designed with a modular architecture based on LangGraph, it offers high flexibility. Through the CLI interface, you can set the ticker, analysis date, LLM provider, and research depth, and track the agents' progress in real time. It also supports importing it directly as a Python module to initialize the graph and retrieve decisions within your code. It is designed for research purposes and explicitly states that it does not provide actual investment advice.