TradingAgents: Multi-Agents LLM Financial Trading Framework
TradingAgents is an open-source Python framework designed to simulate the operations of a financial trading firm using multiple large language model (LLM)-powered agents. The project aims to provide a modular system where different agents—such as analysts, traders, and portfolio managers—collaborate to make trading decisions based on market data, news, and fundamentals. By structuring the trading process around specialized roles, the framework seeks to mirror how real-world trading teams function, with each agent handling specific tasks like data analysis, strategy formulation, or risk assessment.
The project has gained attention in developer communities due to its recent release history and growing adoption signals. As of the latest updates, TradingAgents has released multiple versions throughout 2026, including v0.7.0 in October, which introduced features such as local execution with ChatGPT Plus or Pro plans, improved backtesting that scores decisions from the first close after each action, startup notifications for newer releases, and packaged wheels for easier installation. Prior releases added capabilities like HTML-formatted reports, support for different LLM providers across agent tiers, parallel analyst processing for faster analysis, and CLI enhancements allowing execution via flags like –ticker and –date without interactive prompts. These iterative updates reflect ongoing efforts to improve usability, reliability, and flexibility.
At a high level, TradingAgents operates by coordinating multiple LLM-driven agents through a structured workflow. Each agent is assigned a specific function—such as researching market conditions, generating trading signals, or managing portfolio exposure—and they interact via defined communication channels. The framework supports integration with various LLM providers, including OpenAI-compatible endpoints, Anthropic, Gemini, Groq, Mistral, NVIDIA, Kimi, Bedrock, and Ollama, allowing users to select models based on performance, cost, or availability. Data inputs are drawn from sources like Yahoo Finance for company news, SEC EDGAR for financial filings, FRED for macroeconomic indicators, and Polymarket for event-driven insights, with safeguards in place to prevent look-ahead bias by ensuring that only information available at the time of a decision is used during backtesting.
The framework is intended for developers, researchers, and technically inclined traders who want to experiment with LLM-based trading strategies in a controlled, reproducible environment. It is particularly suited for those interested in exploring how multi-agent systems can be applied to financial decision-making, testing strategy performance across historical data, or building prototypes that combine natural language reasoning with market analysis. Usage typically involves installing the package via pip, configuring API keys for desired LLM providers, defining a ticker and date range, and running simulations that generate detailed reports of agent interactions and trading outcomes.
While the project shows active development and clear documentation of its features, it remains an early-stage, community-driven initiative. The README and release notes do not indicate corporate backing, formal funding, or guaranteed long-term maintenance. Users should evaluate the software’s suitability for their needs carefully, recognizing that it is provided as open-source code without warranties. Licensing details are not specified in the available source material, so individuals intending to deploy or modify the framework should review the repository’s license file directly. As with any tool involving financial models or automated trading, caution is advised when considering live deployment, and thorough testing in simulated environments is recommended.
Source: https://github.com/TauricResearch/TradingAgents
