TradingAgents: A Team of AI Agents That Debates Before Making a Trade

TradingAgents is an open-source framework that models an entire trading firm with specialized LLM agents, from analysts to a bull-versus-bear debate to a portfolio manager, and it is explicitly a research tool, not investment advice.

What it is

TradingAgents is an open-source multi-agent LLM framework from Tauric Research. Instead of asking one AI to analyze a stock, it simulates a miniature trading firm where every role is an AI agent. There are four analysts covering fundamentals, sentiment, news, and technical indicators. Two researchers, one bullish and one bearish, debate the evidence. A trader agent proposes a specific trade. A risk team evaluates volatility and liquidity. Finally, a portfolio manager approves or rejects the simulated transaction.

The framework supports cloud providers like OpenAI, Claude, and Gemini, and can also run fully local through Ollama. Recent releases add persistent decision logs and LangGraph checkpoint resume, so a workflow interrupted mid-run can pick up where it left off.

Why it matters

• It shows how a complex problem can be decomposed across specialized agents rather than dumped on one model.
• The bull-versus-bear debate forces cross-checking instead of letting a single line of reasoning win too easily.
• The pipeline is complete: analysis, debate, trading proposal, risk review, and final approval.
• It supports both cloud and local models, so you can run it without external API dependencies.
• It is a practical case study for learning LangGraph and multi-agent system design.

How it works

You clone the repo, install the package, and run tradingagents in the terminal. It asks which stock to analyze, on what date, with which model, and at what research depth. The four analysts produce separate reports. The bullish and bearish researchers argue from opposite sides. The trader agent turns the debate into a concrete proposal. The risk team checks volatility and liquidity. The portfolio manager makes the final call.

Recent versions persist decision logs as a journal that later sessions can learn from, and LangGraph checkpoints let you resume from a saved node if a run is interrupted. Docker support is available for containerized runs.

Caveats

The repo states clearly that this is a research tool, not investment advice. Running the same ticker on the same date twice can produce different results depending on the model, temperature, data, and timing. Do not use the output as a signal to trade real money. The framework is valuable as a multi-agent design case study, but its outputs are probabilistic and non-deterministic.

Who it's for

This is for developers and researchers interested in multi-agent architecture, LangGraph, and financial workflow simulation, not for anyone looking for an automated trading signal.