TradingAgents

Best for: researchers studying multi-agent LLM systems, using financial trading as a structured problem domain.

TradingAgents, from Tauric Research, is a multi-agent LLM framework for financial trading. It models a trading firm as separate agents — analysts, researchers, traders and risk management — that deliberate and hand off to each other, rather than asking a single model for a verdict. Version 0.3.1 added Alpha Vantage look-ahead filtering, graph-router crash-safety, configurable LLM retry budgets, Bedrock API-key auth, and support for Claude Sonnet 5 and Fable 5.

What makes it useful: the role separation is the research contribution. Assigning distinct responsibilities and letting agents debate surfaces disagreement that a single prompt would average away, and the architecture transfers to any domain where structured deliberation beats a one-shot answer.

Where it falls short: it is a research framework, not a trading system, and running it against real money is your own risk. Look-ahead bias is a persistent hazard in backtesting — the fact that 0.3.1 shipped a fix for it suggests earlier results deserve scepticism. Multi-agent deliberation also multiplies token cost per decision.

Details: Python, Apache 2.0 licensed.