TradingAgents

In TradingAgents, a team of LLM agents plays out the roles of a real trading firm. Fundamentals, sentiment, news, and technical analysts each read the market from their own angle, bullish and bearish researchers argue those readings out in structured debate, and a trader agent turns the reports into a decision. Nothing reaches the simulated exchange until a risk management team and a portfolio manager sign off. Tauric Research documented the design in an arXiv paper and ships it under Apache-2.0, where it is closing in on 97,000 GitHub stars.

Recent releases added a verified data-access contract along with FRED and Polymarket as data vendors. A provider registry covers OpenAI-compatible endpoints, Bedrock, Groq, and local Ollama. The project says outright that it exists for research and is not financial advice; outcomes shift with the backbone model, the temperature setting, and the quality of the data you feed it.