CLI-based multi-agent trading system using LLMs (github.com)

TradingAgents, an open-source multi-agent trading system powered by large language models (LLMs), has been released to the research community. The framework simulates a realistic trading firm by deploying specialized agents—such as fundamental analysts, sentiment experts, news interpreters, technical analysts, and risk managers—that collaboratively analyze market data and debate strategies to inform trading decisions. This multi-agent architecture modularly decomposes complex financial tasks, enabling scalable and robust analysis through structured inter-agent dialogue and dynamic risk assessment. Designed primarily for research, TradingAgents integrates various LLMs for different cognitive roles, using models like o1-preview and gpt-4o for deep and fast thinking respectively, while supporting cost-effective alternatives for experimentation. It leverages real-time data via the FinnHub API and OpenAI’s APIs, with flexibility for users to customize configurations, debate rounds, and model selections. The framework also features a CLI interface for easy interaction, allowing users to track agent decisions live. Built with LangGraph for modularity, TradingAgents is set to advance AI-driven financial trading research through transparent, collaborative agent-based modeling, inviting contributions from the community and promising soon-to-be-released curated datasets for backtesting. This project highlights how multi-agent LLM setups can enhance financial decision-making complexity beyond single-agent AI systems.