TradingAgents — Multi-Agent LLM Financial Trading Framework

Introduction

TradingAgents is an open-source framework that models a trading firm as a team of specialized LLM agents. Each agent plays a distinct role — analyst, researcher, risk manager, trader — collaborating through structured workflows to make informed trading decisions.

What TradingAgents Does

• Simulates a multi-agent trading firm with role-based LLM agents
• Provides fundamental, technical, and sentiment analysis through specialized analyst agents
• Includes a risk management agent that evaluates portfolio exposure before trades
• Aggregates signals from multiple agents into a unified trading decision
• Supports backtesting against historical market data for strategy evaluation

Architecture Overview

TradingAgents uses a LangGraph-based orchestration layer to coordinate agents. A market data pipeline feeds real-time and historical data to analyst agents, whose outputs flow to a portfolio manager agent for aggregation. A risk management agent applies constraints before the final trade execution decision. All agent interactions are logged for auditability.

Self-Hosting & Configuration

• Requires Python 3.9+ with LangChain and LangGraph dependencies
• Configure API keys for market data providers and LLM services via environment variables
• Customize agent prompts and risk parameters through configuration files
• Supports multiple LLM backends including OpenAI, Anthropic, and local models
• Historical market data can be sourced from free APIs or local CSV files

Key Features

• Role-based agent architecture mirrors real trading firm workflows
• Built-in backtesting engine for evaluating strategies on historical data
• Multi-source analysis combining fundamental, technical, and news sentiment signals
• Risk management layer with configurable position limits and drawdown thresholds
• Transparent decision logs showing each agent's contribution to the final trade

Comparison with Similar Tools

• QuantConnect / Backtrader — traditional algorithmic trading; TradingAgents uses LLM reasoning
• CrewAI — general multi-agent framework; TradingAgents is purpose-built for financial markets
• AutoGen — multi-agent conversations; TradingAgents provides domain-specific trading roles
• ai-hedge-fund — similar concept; TradingAgents offers more structured role separation and risk controls

FAQ

Q: Can TradingAgents execute real trades? A: The default setup is for analysis and backtesting only. Broker integration requires additional configuration.

Q: What market data sources are supported? A: Yahoo Finance, Alpha Vantage, and custom CSV data sources out of the box.

Q: Does it work with local LLMs? A: Yes. Any LangChain-compatible model can be used, including Ollama-hosted models.

Q: Is this financial advice? A: No. TradingAgents is a research and educational tool, not a financial advisory service.

Sources

• https://github.com/TauricResearch/TradingAgents
• https://arxiv.org/abs/2412.20138