Seven specialized LLM agents working in concert — analyzing fundamentals, gauging sentiment, scanning news, and executing technical analysis — to produce explainable, research-grade trading decisions.
TradingAgents is an open-source, research-driven framework built by Tauric Research with LangGraph. It simulates a professional trading institution with clearly specialized AI agents that collaborate through structured communication and debate.
Seven specialized roles — fundamental, sentiment, news, and technical analysts, researchers, a trader, and a risk manager — work together to evaluate markets and form consensus-driven decisions.
From strategy construction and data integration to backtesting, simulated execution, and risk management — all within a unified, modular framework.
Every decision is explained, evaluated, and traced. Preserves reasoning chains, transaction logs, and performance metrics for rigorous analysis.
Pure LLM-powered. Runs on standard hardware with API-based access to OpenAI, DeepSeek, Claude, and other leading models — no specialized compute needed.
TradingAgents simulates a real trading institution. Each role has distinct tools and constraints, enabling rigorous debate, cross-validation, and risk-aware execution.
Built with FastAPI + Vue 3, MongoDB, and Redis — delivering a modern, high-performance platform for both research and hands-on learning.
Authentication, role-based access control, and comprehensive operation audit trails for enterprise security.
Centrally manage model parameters, data sources, and trading strategies through an intuitive interface.
MongoDB for flexible data storage and Redis for multi-level caching — performance at every layer.
Analyze multiple tickers simultaneously with intelligent filtering and screening capabilities.
Create custom watchlists with comprehensive company profiles, financials, and real-time agent analysis.
Practice with realistic order execution in a simulated exchange environment with full transaction logging.
Manage LLM providers dynamically — switch between OpenAI, DeepSeek, Claude, and others on the fly.
Generate and export detailed analysis reports, backtest results, and performance summaries.
Receive alerts on trade executions, risk warnings, and important market events through built-in notification channels.
We combine structured multi-agent communication with a modular architecture to deliver capabilities that traditional quant frameworks cannot match.
Agents don't just report — they debate. Analysts present evidence, researchers challenge assumptions, and the trader synthesizes a consensus. This structured debate produces more robust decisions than any single model can achieve.
Every trade comes with a complete reasoning chain. You can inspect exactly why each agent recommended a course of action, what evidence they considered, and how risk constraints shaped the final decision.
Swap in new models, data sources, or entire agent roles without rewriting the framework. The plug-and-play architecture makes TradingAgents a living platform for experimentation.
Test strategies against historical data with a full simulated exchange. Review transaction logs, cumulative returns, and risk metrics to validate every approach before deployment.
Supports OpenAI GPT-4, DeepSeek, Anthropic Claude, and more — with the ability to assign different models to different agent roles based on their specific strengths.
Integrated with Tushare, AKShare, BaoStock, Finnhub, Yahoo Finance, Alpha Vantage, Wind, Eastmoney Choice, and more — covering global markets.
Grounded in academic research with published papers and experimental results. TradingAgents is built for researchers, by researchers — not a black box.
The TradingAgents-AI hosted platform provides a guided, hands-on entry point for students and practitioners to learn multi-agent trading concepts without infrastructure setup.
From open-source exploration to hosted platform and private deployment — find the right fit for your goals.
Connect to 10+ financial data providers and all major LLM platforms — all managed through a unified configuration interface.
