Recently, the Tauric Research team open-sourced an intriguing project called TradingAgents. This Python framework simulates the complete workflow of a hedge fund using a multi-agent approach and has already garnered 20.1k stars on GitHub.


The core idea is to decompose the trading process into specialized roles: fundamental analysts, sentiment analysts, news analysts, and technical analysts handle domain-specific data analysis; long and short research teams debate perspectives; finally, traders consolidate reports, which are evaluated by the risk management team before execution.


Notably, the team emphasizes this is purely a research framework—trading performance depends on data quality, model selection, and other factors. The project requires FinnHub and OpenAI API keys and recommends starting with lightweight models like gpt-4.1-mini due to the framework's high API call volume.

Installation is straightforward:

```bash

git clone https://github.com/TauricResearch/TradingAgents.git

cd TradingAgents

pip install -r requirements.txt

```

The CLI interface allows interactive selection of stock symbols, dates, etc.:


Code integration is intuitive:

```python

from tradingagents.graph.trading_graph import TradingAgentsGraph

ta = TradingAgentsGraph(debug=True)

_, decision = ta.propagate("NVDA", "2024-05-10")

```

The team published a 38-page technical paper on arXiv, and the framework remains under active development—the latest commit on July 3rd fixed an infinite loop issue. For those exploring LLM applications in quantitative trading, this project offers a solid practical starting point.

Paper: https://arxiv.org/pdf/2412.20138

GitHub Repo: https://github.com/TauricResearch/TradingAgents