LLM-based Multi-Agent Financial Trading Framework 'TradingAgents' v0.4.0 Released

The open-source framework TradingAgents, released by Tauric Research, has launched v0.4.0. This project is a Multi-Agent LLM Financial Trading Framework designed to have LLM-based specialized agents collaborate by mimicking the organizational structure of a real trading firm.

Key Releases and Features

v0.4.0 includes FRED macro data, social sentiment, and a fix for look-ahead bias in decision log memory. Additionally, support for the latest models such as GPT-5.6 and GLM-5.3 has been added, along with improved CLI checkpoint resumption and trader price grounding. Previous versions (v0.2.0~v0.3.1) established LangGraph-based checkpoint resumption, support for various LLM providers (OpenAI, Anthropic, Google, DeepSeek, etc.), and Docker and local Ollama execution environments.

Architecture and Agent Composition

The framework features agents with the following roles that derive optimal strategies through dynamic debate:

• Analyst Team: Responsible for fundamental, sentiment, news, and technical analysis
• Researcher Team: Critically evaluates analysis results from bullish/bearish perspectives
• Trader Agent: Synthesizes analysis and research reports to determine trading timing and size
• Risk Management & Portfolio Manager: Assesses risk and approves/rejects final trades

Usability and Caveats

Users can simulate decision-making by inputting specific tickers and dates via the CLI or Python API. Supported markets include the US, major Asian stock markets, and cryptocurrencies. However, due to the non-deterministic sampling of LLMs and changes in live data, reproducibility is not guaranteed, and it should be treated as a scaffold for multi-agent analysis research rather than a strategy yielding fixed returns.