What is TradingAgents?
TradingAgents models a trading research workflow through specialized LLM agents. Analysts examine company fundamentals, market news, social sentiment and technical indicators in parallel. Bullish and bearish researchers debate their reports before trader, risk and portfolio management agents produce a decision. The documented execution workflow uses a simulated exchange.
The framework runs through an interactive or unattended command-line interface and can be embedded in Python applications. Installation is available from source or with Docker. Model providers can be configured separately for analysts and managers, including local inference through Ollama or OpenAI-compatible servers. Saved reports include run settings and a self-contained HTML page.
Historical analysis restricts data according to its availability on the analysis date. SEC EDGAR supplies statements as filed, while historical statements from vendors without publication dates are withheld. A persistent memory log records decisions and later outcomes, and optional checkpoints allow interrupted workflows to resume. TradingAgents is intended for research, not financial, investment or trading advice.
Key capabilities
- Specialized analysis Runs fundamentals, news, sentiment and technical analysts in parallel before research debates begin.
- Decision review Routes analyst findings through bullish and bearish researchers, a trader, risk analysts and a portfolio manager.
- Portfolio context Accepts cash and position information so trading and risk assessments can account for existing holdings.
- Historical evaluation Runs analyses across tickers and dates, restricts dated data access and scores eligible decisions against subsequent market closes.
- Persistence and recovery Stores decisions and outcome reflections, with optional checkpoints for resuming interrupted runs.
- Configurable execution Supports unattended CLI runs, Python usage, Docker deployment and separate model providers for reasoning tiers.
Where TradingAgents fits
TradingAgents fits financial AI research workflows that compare agent roles, model choices and decision outcomes across instruments and analysis dates. Developers can embed its analysis graph in Python experiments or schedule unattended CLI runs, then inspect reports and the decision memory log.
Before deployment, users should verify model access, provider usage limits, market-data availability and the historical coverage of each vendor. Historical fundamentals are restricted when publication dates cannot be established. Existing holdings must be supplied explicitly for portfolio-aware analysis, and checkpoint compatibility should be checked when upgrading.
Install
pip install .Reviews of TradingAgents
Be the first to share your experience with TradingAgents.
Write a review
Loading review form…