Video Summary
TradingAgents is a Python multi-agent LLM framework that simulates a Wall Street trading firm with role-based agents.
Architecture: four parallel analysts (fundamentals, sentiment, news, technical), a bull/bear researcher debate, a trader, risk team, and a portfolio manager.
LangGraph orchestrates agents as nodes with checkpointing, resumability, and a persistent decision log that learns from past trades.
Setup is straightforward (clone, pip install, set LLM API key) and supports many model providers, but runs can be costly and it's research-only (no live brokerage).
Repo is Apache 2.0, backed by an arXiv paper from UCLA, and reached 53k+ stars rapidly; v0.2.4 added structured-output decision agents and Docker support.
TradingAgents is an open-source Python framework that simulates a full trading firm using separate LLM agents for different roles; its uniqueness lies in role-based, debate-driven decision making and full auditability of each step.
Four parallel analysts (fundamentals, sentiment, news, technical) produce reports; two researchers (bull and bear) debate those reports; a trader proposes trades; risk management evaluates proposals; the portfolio manager gives final approval.
LangGraph models each agent as a node in a directed graph, checkpointing transitions so runs can be resumed after crashes and maintaining a persistent decision log for later reflection.
No — the project is a research framework and runs a simulated exchange/backtest; it doesn't include live brokerage integration, and authors warn it's not financial advice.
Setup is simple (clone, pip install, set LLM API key) and supports many providers, but costs can be significant because each run makes many LLM calls across parallel agents.
"Somebody just open-sourced a hedge fund."
A Python project called Trading Agents has gained significant traction on GitHub, surpassing 53,000 stars and trending this weekend.
This project is a multi-agent LLM framework that simulates the operations of a real Wall Street trading firm, allowing users to run a complete simulated firm by pointing it at a stock ticker.
"Real hedge funds don't have one person picking stocks; they have entire teams that argue with each other to produce a defensible position."
The framework consists of various specialized roles, each represented by an LLM agent.
Roles include a fundamentals analyst, sentiment analyst, news analyst, technical analyst, bullish and bearish researchers, a trader agent, a risk management team, and a portfolio manager.
Each agent operates using LLMs that communicate with one another, simulating the collaborative yet competitive environment of an actual trading firm.
"Every completed run appends its decision and its reasoning to a markdown file in your home directory."
The analyst team consists of four parallel agents that provide insights based on different analyses: fundamentals, sentiment, news, and technical.
The researcher team comprises a bull and a bear agent who debate and analyze the reports from the analysts before passing their conclusions to the trader agent.
The trader proposes trades based on the debate, while the risk management team assesses each proposal, and the portfolio manager gives final approval.
"Every agent is a node in a directed graph, and every transition between nodes is checkpointed."
The framework utilizes LangGraph, which allows every transition and decision made by the agents to be logged and resumable in case of a crash.
A persistent decision log records the decisions made during evaluations, learning from previous trades to enhance future performance.
"Setup is honestly painless."
Users can clone the repository, set up a Python environment, and run the CLI to begin using the framework. It is compatible with major language model providers, including OpenAI GPT, Google Gemini, and others.
This project is suited for quant researchers, hobbyist traders, fintech founders, and indie hackers interested in the practical applications of multi-agent LLM systems.
"The authors explicitly say this is a research framework, not financial advice."
Running this framework can incur significant costs due to the number of calls made to LLMs, as each analysis cycle requires multiple parallel calls and evaluations.
The simulated exchange operates only as a backtest and does not offer live brokerage integration, so users will need to implement their own trading solutions if they wish to trade in real-time.
