TradingAgents: open multi‑agent trading framework
TradingAgents, a new open‑source Python framework, enables multi‑agent LLM systems for experimenting with trading strategies—providing a playground for research into agent coordination, market signals, and automated decision logic. It packages multi‑agent orchestration specifically for quant and trading workflows. (x.com/quantscience_/status/2038591142854291522)
The official repository shows 44.5k stars and 8.1k forks and is released under an Apache‑2.0 license, with a v0.2.2 release noted in March 2026. The framework’s technical report on arXiv is listed as arXiv:2412.20138 and names Yijia Xiao, Edward Sun, Di Luo, and Wei Wang as authors, with the record first submitted on Dec 28, 2024 and last revised on Jun 3, 2025. Tauric Research and the paper report backtest results of up to 30.5% annualized returns and improvements in cumulative returns, Sharpe ratio, and maximum drawdown versus baseline models in their experimental section. The codebase explicitly implements seven specialized agent roles—Fundamentals, Sentiment, News, Technical analysts, Bull/Bear Researchers, Trader, and Risk Manager—and uses structured outputs plus the ReAct prompting pattern to orchestrate debates and decisions. Recent release notes document multi‑provider LLM support added across 2026 releases (v0.2.0 and v0.2.2), naming compatibility with GPT‑5.x, Gemini 3.x, Claude 4.x and explicit model coverage for GPT‑5.4, Gemini 3.1, and Claude 4.6. Product and community pages highlight LangGraph integration for modular agent wiring, a demo/backtesting terminal, and a localized deployment built on FastAPI + Vue 3 with MongoDB and Redis plus connectors for data sources like Alpha Vantage, Tushare and AkShare. (tradingagents-ai.com tradingagents-ai.github.io )
