TauricResearch/TradingAgents

What it solves

TradingAgents addresses the complexity of financial trading by simulating the structure of a professional trading firm. Instead of relying on a single LLM, it uses a team of specialized agents to analyze market data, debate strategies, and manage risk, reducing the likelihood of oversight and improving the quality of trading decisions.

How it works

The framework uses LangGraph to coordinate a multi-agent workflow divided into four primary stages:

1. Analyst Team: Specialized agents gather data on fundamentals (financials), sentiment (social media/news), global macro trends, and technical indicators (MACD/RSI).
2. Researcher Team: Bullish and bearish researchers engage in structured debates to challenge the analysts' findings and balance potential gains against risks.
3. Trader Agent: Synthesizes the reports from analysts and researchers to determine the timing and size of a trade.
4. Risk & Portfolio Management: A risk management team evaluates volatility and liquidity, and a Portfolio Manager makes the final decision to approve or reject the transaction.

The system includes a decision log that records past trades and reflections to improve future performance and supports checkpointing to resume interrupted runs.

Who it’s for

It is designed for researchers studying multi-agent AI systems and financial analysis, as well as developers looking to build LLM-powered trading scaffolds.

Highlights

• Multi-Provider Support: Compatible with OpenAI, Google, Anthropic, xAI, DeepSeek, Qwen, GLM, MiniMax, and local models via Ollama.
• Global Market Coverage: Works with any market covered by Yahoo Finance, including US, HK, Tokyo, London, India, Canada, Australia, China, and Crypto.
• Memory System: Maintains a persistent decision log to inject lessons from previous trades into current analyses.
• Fault Tolerance: Implements LangGraph checkpointing to resume analysis from the last successful step after a crash.