LLM Trading Lab is a framework for running AI-managed trading experiments where large language models make portfolio decisions under predefined rules.
The project addresses the question of whether large language models can generate alpha or make sound trading decisions using real-time market data. It does this by implementing a live trading engine that accepts LLM-generated trade selections, enforces hard constraints like automated stop-loss rules, and logs all decisions with full transparency. The framework maintains a complete forward-only record of trades, preserving historical data and research artifacts without revision, enabling auditability and comparison against benchmarks like the S&P 500 and Russell 2000.
The tool suits researchers and developers interested in empirically testing LLM decision-making in measurable domains where outcomes are concrete and irreversible. It works best for those who want to run their own AI-driven trading experiments using the same structure and methodology, rather than for production trading systems. The project explicitly points users toward a separate framework called LIBB for those seeking a more general experimental platform for LLM research.
The project maintains a complete archive of historical artifacts from its initial micro-cap experiment, including decision logs, research evaluations, and performance analytics. The codebase includes a production-grade trading engine alongside research tooling, with support for multiple data sources and standard financial metrics like Sharpe ratio, Sortino ratio, and drawdown analysis. The maintainer actively develops extensions to the framework for new experiments and welcomes contributions including bug reports, refactors, and collaboration from high-quality contributors.
