TradingAgents is a multi-agent financial trading framework built on large language models that simulates the structure and decision-making processes of real-world trading firms. The framework decomposes complex trading tasks into specialized agent roles, each powered by LLMs, that collaborate to evaluate market conditions and make informed trading decisions. The repository is written in Python and serves as both a research platform and a practical trading system, with its technical foundations documented in an arXiv paper at arxiv.org/pdf/2412.20138.
The framework organizes agents into distinct teams that mirror professional trading operations. The Analyst Team includes a Fundamentals Analyst that evaluates company financials and performance metrics, a Sentiment Analyst that aggregates news headlines and social media sentiment, a News Analyst that monitors macroeconomic indicators and global events, and a Technical Analyst that applies technical indicators like MACD and RSI to detect trading patterns. A separate Researcher Team comprises bullish and bearish researchers who critically assess the analysts' insights through structured debates to balance potential gains against risks. The Trader Agent synthesizes reports from analysts and researchers to make actual trading decisions, while the Risk Management and Portfolio Manager team continuously evaluates portfolio risk through volatility and liquidity assessments, with the Portfolio Manager providing final approval or rejection of transaction proposals before execution on a simulated exchange.
The project supports multiple LLM providers including OpenAI, Anthropic, Google, Groq, Mistral, NVIDIA, Kimi, DeepSeek, and any OpenAI-compatible endpoint. Users can configure the framework through environment variables with automatic API-key detection, and the system supports local models via Ollama as well as AWS Bedrock. The CLI interface allows users to select tickers, analysis dates, LLM providers, and research depth, with real-time progress tracking as agents execute their analyses. The framework works with any market covered by Yahoo Finance, supporting US equities, international markets including Hong Kong, Tokyo, London, India, Canada, and Australia, Chinese A-shares, and cryptocurrencies.
According to GitGenius activity tracking, the repository has processed 328 issues and pull requests with a median response latency of 37.4 hours and a mean of 423.2 hours. The most active contributor is Yijia-Xiao with 323 tracked events, followed by Tauric-Research with 52 events and slc03 with 11 events. Enhancement requests represent the most common issue label with 4 instances. The repository shares overlapping contributors with openclaw/clawhub, berriai/litellm, and openai/codex, indicating active engagement within the broader AI and trading research communities.
Recent releases show rapid development velocity. Version 0.3.0 introduced a verified data-access contract, expanded provider registry, FRED and Polymarket data vendors, and CI gates. Version 0.2.5 added a grounded Sentiment Analyst, GPT-5.5 model coverage, multi-region support for Qwen, GLM, and MiniMax, and remote Ollama support. Version 0.2.4 introduced structured-output agents, LangGraph checkpoint resume, persistent decision logs, and Docker support. The framework explicitly disclaims that it is designed for research purposes and should not be considered financial, investment, or trading advice, as trading performance varies based on model selection, temperature settings, trading periods, data quality, and other non-deterministic factors.
