Verified project record

TauricResearch/TradingAgents

A multi-agent LLM framework that mirrors a real-world trading firm, deploying specialized analyst, researcher, trader, and risk-management agents to collaboratively evaluate market conditions and inform trading decisions. It is structured for research and simulation purposes and supports multiple LLM providers and financial data integrations.

Project overview

The framework assigns specialized agents to debate findings across fundamentals, sentiment, news, technicals, and risk management, offering a structured approach to evaluating market conditions through simulated roles.

Project type: AI Agent · Workflow · Data Processing
Use cases: Search & Research · Data Analysis
Deployment: Refer to project documentation
License: Apache-2.0

Best for

Researchers, developers, and data teams seeking a multi-agent LLM framework for analyzing market conditions through simulated trading firm roles.
Users needing structured debates between bullish and bearish agents to balance potential gains and risks before generating a simulated transaction proposal.

Key capabilities

Deploys specialized agents to evaluate company financials, aggregate sentiment from social and news sources, monitor macro indicators, and detect trading patterns.
Comprises bullish and bearish researchers who critically assess insights through structured debates to balance potential gains against risks.
Composes reports from analysts and researchers to inform trading decisions, determining the timing and magnitude of trades.
Continuously evaluates portfolio risk and adjusts strategies, with the Portfolio Manager approving or rejecting transaction proposals for the simulated exchange.
Works with any market Yahoo Finance covers using exchange-suffixed tickers, and automatically resolves company identity and alpha benchmarks.
Supports multiple LLM providers, with named examples including OpenAI, Google, Anthropic, xAI, DeepSeek, Qwen, GLM, MiniMax, OpenRouter, Ollama, Azure OpenAI, and AWS Bedrock.
Maintains an always-on persistent log that appends completed decisions, realized returns, and reflections to memory for future runs.
Provides opt-in saving of state after each node so that a crashed or interrupted run resumes from the last successful step instead of starting over.

Limitations and risks

LLM-driven sampling is non-deterministic, meaning two runs of the same ticker and date can produce different results.
Live social and news data changes over time, causing runs on the same historical date to see different inputs.
Backtest results are not guaranteed to match any published figure due to various model and data factors.
The framework is designed for research purposes and is not intended as financial, investment, or trading advice.

Getting started

Setup is rated medium difficulty, requiring users to create a conda environment, install dependencies via pip, and configure environment variables with API keys for selected LLM providers.
After cloning and configuration, users launch the interactive CLI, select a ticker, choose an analysis date, and specify the analysis depth to initiate the workflow.

Evidence and sources

GitHub project description: TradingAgents: Multi-Agents LLM Financial Trading Framework
README: TradingAgents is a multi-agent trading framework that mirrors the dynamics of real-world trading firms.
README: By deploying specialized LLM-powered agents: from fundamental analysts, sentiment experts, and technical analysts, to trader, risk management team, the platform collaboratively ev…
README: - Fundamentals Analyst: Evaluates company financials and performance metrics, identifying intrinsic values and potential red flags.
README: TradingAgents officially released! We have received numerous inquiries about the work, and we would like to express our thanks for the enthusiasm in our community.

Reviewed at: 2026-07-17T04:23:29Z

Open original GitHub project