An open-source financial trading framework utilizing multi-agent LLM systems. It simulates the dynamics of a real-world trading firm by deploying specialized AI agents—ranging from fundamental and sentiment analysts to traders and risk managers—who collaborate to analyze market data and formulate trading strategies, primarily designed for research purposes.
One Minute Overview#
TradingAgents is a research framework simulating a real-world trading firm. It uses LangGraph to build a multi-agent system comprising analysts, researchers, traders, and risk managers who collaborate and debate to evaluate markets and make decisions.
Core Value: Decomposes complex trading tasks into specialized roles, providing a scalable and modular platform for researching LLM applications in financial decision-making.
Quick Start#
Installation Difficulty: Medium - Requires Python environment setup and third-party API keys.
# Clone project and install dependencies
git clone https://github.com/TauricResearch/TradingAgents.git
cd TradingAgents
conda create -n tradingagents python=3.13
conda activate tradingagents
pip install -r requirements.txt
Is this suitable for me?
- ✅ AI Finance Researchers: Looking to explore multi-agent collaboration in financial analysis.
- ✅ Quant Developers: Need to build modular trading strategy prototypes based on LLMs.
- ❌ Seeking Stable Profits: This project is for research; performance varies due to model non-determinism and API costs, and it is not investment advice.
Core Capabilities#
1. Agent Collaboration - Simulating Real Workflows#
The system splits trading tasks into specialized roles: fundamental, sentiment, news, and technical analysis, mimicking a human team workflow. Actual Value: Increases the comprehensiveness and depth of market analysis through specialized division of labor.
2. Bull/Bear Debate - Optimizing Decisions#
Introduces bullish and bearish researchers to critically assess analyst conclusions and engage in structured debates to balance gains against risks. Actual Value: Effectively reduces single-perspective bias and provides more robust decision grounds.
3. Dynamic Risk Control#
Built-in Risk Manager and Portfolio Manager agents evaluate volatility and liquidity in real-time, holding the final veto power. Actual Value: Emphasizes drawdown and extreme risk prevention while pursuing returns.
Tech Stack & Integration#
Languages: Python (3.13+) Core Framework: LangGraph Key Dependencies: OpenAI API (LLM), Alpha Vantage / yfinance (Data) Integration: Python SDK / CLI Tool
Maintenance Status#
- Activity: Recently officially released and open-sourced with high community interest.
- Updates: Codebase and documentation are actively maintained.
- Community: Active Discord community and supported by an arXiv paper.
Commercial & Licensing#
License: Open Source (Specific type not explicitly defined, code is public).
- ⚠️ Disclaimer: The framework is designed for research purposes. Trading performance varies based on model selection and data quality; it is not intended as financial advice.
Docs & Learning#
- Quality: Comprehensive
- Official Docs: GitHub README
- Examples: Includes CLI demos and Python usage snippets
- Paper: arXiv:2412.20138