TradingAgents: Multi-Agent LLM Framework That Simulates a Trading Firm
TradingAgents is trending on GitHub with 371 stars per day, reaching 33.6K total stars. The open-source framework simulates a professional trading firm using specialized LLM-powered agents working in coordinated roles — fundamental analysts, sentiment experts, technical analysts, traders, and risk managers.
The framework operates through a structured pipeline: the analyst team concurrently gathers market data, the research team discusses and evaluates findings, the trader makes decisions based on the analysis, and the risk management team assesses decisions against current market conditions. This mirrors how actual trading desks operate, with each agent bringing specialized expertise.
Built on LangGraph for modularity, TradingAgents supports multiple LLM providers including OpenAI, Google, Anthropic, xAI, OpenRouter, and Ollama. It uses quick-thinking models for data retrieval and deep-thinking models for analysis — requiring no GPUs to run. Users interact through CLI or programmatically via Python's .propagate() function.
Experiments show significant improvements over baseline models in cumulative returns, Sharpe ratio, and maximum drawdown. The project originated from a December 2024 research paper (arxiv.org/abs/2412.20138) by Tauric Research and has since grown into a production-ready framework.
GitHub: https://github.com/TauricResearch/TradingAgents
License: Apache 2.0
← 返回所有文章
The framework operates through a structured pipeline: the analyst team concurrently gathers market data, the research team discusses and evaluates findings, the trader makes decisions based on the analysis, and the risk management team assesses decisions against current market conditions. This mirrors how actual trading desks operate, with each agent bringing specialized expertise.
Built on LangGraph for modularity, TradingAgents supports multiple LLM providers including OpenAI, Google, Anthropic, xAI, OpenRouter, and Ollama. It uses quick-thinking models for data retrieval and deep-thinking models for analysis — requiring no GPUs to run. Users interact through CLI or programmatically via Python's .propagate() function.
Experiments show significant improvements over baseline models in cumulative returns, Sharpe ratio, and maximum drawdown. The project originated from a December 2024 research paper (arxiv.org/abs/2412.20138) by Tauric Research and has since grown into a production-ready framework.
GitHub: https://github.com/TauricResearch/TradingAgents
License: Apache 2.0
评论