← Back to Discover

TradingAgents

Open-source Python project in which LLM-powered agents act as analysts, researchers, a trader and risk managers to debate a stock and reach a simulated trading decision. Built for research, not advice.

FinanceAgent frameworks
Visit website ↗

What it does

TradingAgents splits a trading decision across specialised LLM agents. An analyst team covers fundamentals, sentiment, news and technical indicators, then bullish and bearish researchers debate the findings before a trader drafts a decision. A risk management team reviews the proposal and a portfolio manager approves or rejects it. If approved, the order goes to a simulated exchange. The project states that it is a research framework and should not be taken as financial or investment guidance. It installs as a Python 3.11+ package with an interactive CLI, flag-driven runs for scripts, or a Docker setup. It works with many LLM providers through API keys, with a ChatGPT Plus or Pro subscription, or with local models through Ollama and other OpenAI-compatible servers. Runs can include backtesting over tickers and dates, and reports can be saved, including as a single HTML page.

Capabilities

  • Specialised analyst, researcher, trader, risk and portfolio manager agents
  • Structured bullish versus bearish debates before a decision
  • Interactive CLI plus flag-driven runs for scheduled jobs
  • Works with many LLM providers, ChatGPT subscriptions and local models
  • Backtesting over a grid of tickers and dates
  • Saved reports, including a single-page HTML version

When to use it

  • You want to study how a team of LLM agents reasons about a stock
  • You are researching multi-agent workflows in a finance setting
  • You want to backtest agent-made decisions against historical dates