tauricresearch/tradingagents

TradingAgents is a research framework that simulates a professional trading firm using multiple AI agents powered by large language models (LLMs). The concept behind it is that real-world investment firms do not make decisions with a single analyst, they use specialists in different areas who discuss, debate, and challenge each other before reaching a conclusion. This framework replicates that structure computationally. The system is organized into teams. An Analyst Team consists of four specialized agents: a Fundamentals Analyst that evaluates company financial data, a Sentiment Analyst that processes social media and public opinion signals, a News Analyst that interprets macroeconomic events and global news, and a Technical Analyst that identifies price patterns using indicators like MACD (Moving Average Convergence Divergence) and RSI (Relative Strength Index, common tools in trading that measure momentum and trend). These analysts report to a Researcher Team, which includes both optimistic (bullish) and pessimistic (bearish) researchers who debate the findings to pressure-test conclusions. A Trader Agent then synthesizes all of this into a trade proposal, which a Risk Management team and Portfolio Manager either approve or reject. The framework connects to real or historical market data and supports multiple LLM providers including OpenAI, Anthropic, Google Gemini, and others, so you can swap in different AI models to investigate how they perform. It includes backtesting capabilities, meaning you can run it against historical data to evaluate hypothetical past performance. You would use TradingAgents for academic research into how LLM-based multi-agent systems perform in financial decision-making, or as a starting point for building automated market analysis tools. The README explicitly notes this is intended for research purposes, not as financial or investment advice. It is written in Python and deployable via Docker.