TradingAgents is an open-source multi-agent framework designed for financial research, simulating a professional trading firm through a collaborative ecosystem of LLM-powered agents. The architecture decomposes complex market analysis into specialized roles: Fundamental, Sentiment, News, and Technical Analysts provide data-driven insights, which are then critiqued by bullish and bearish researchers. These insights undergo structured debate before being synthesized by a Trader and a Risk Management team, with final execution authority resting with a Portfolio Manager. The framework supports a diverse array of LLM providers, including OpenAI (defaulting to GPT-6 Sol and Luna), Google Gemini, Anthropic Claude, and local models via Ollama. Technical features include point-in-time data integrity using SEC EDGAR for fundamental filings, preventing look-ahead bias, and a decision-log memory system that enables agents to reflect on realized returns and past successes. Users can perform large-scale backtesting across ticker and date grids, utilizing LangGraph for stateful checkpoint resumes. The system also offers optional social sentiment screening via TypeSafe's Jev to filter irrelevant StockTwits and Reddit posts, ensuring high-quality input for the Sentiment Analyst.