█

AI/TLDR

TradingAgents

An LLM trading firm: analysts, bull-bear researchers, trader and risk agents

GitHub★ 103kWebsitetauric.ai
Multi-Agent SystemsOpen source
Language
Python
License
Apache-2.0
$git clone https://github.com/TauricResearch/TradingAgents.git

Overview

TradingAgents is a multi-agent framework from Tauric Research that models the division of labour inside a real trading firm. Specialised LLM agents each take one role — fundamentals, sentiment, news and technical analysts, bullish and bearish researchers, a trader, a risk-management team and a portfolio manager — and they debate each other before a decision is produced, rather than a single prompt answering the whole question.

The analyst team gathers the inputs: company financials, aggregated news headlines and social chatter, macro indicators, and technical indicators such as MACD and RSI. The researcher team then argues both sides of the resulting thesis, the trader composes those reports into a call, and the risk team and portfolio manager review it against volatility and liquidity before the final decision is recorded.

It is built on LangGraph and is deliberately provider-agnostic: OpenAI, Google, Anthropic, xAI, DeepSeek, Qwen, GLM, MiniMax, OpenRouter, Azure OpenAI and Amazon Bedrock are supported, as are local models through Ollama or any OpenAI-compatible server such as vLLM or LM Studio. The framework is Apache-2.0, has an accompanying paper (arXiv 2412.20138), and its authors state plainly that it is built for research and is not financial, investment or trading advice.

What it does

  • Role-specialised agents: fundamentals, sentiment, news and technical analysts, plus trader, risk team and portfolio manager
  • Structured bull-versus-bear researcher debate before a position is proposed
  • Built on LangGraph, with checkpoint resume and a persistent decision log
  • Wide provider support — OpenAI, Google, Anthropic, xAI, DeepSeek, Qwen, GLM, MiniMax, OpenRouter, Azure and Bedrock — plus Ollama and OpenAI-compatible local servers
  • Interactive CLI for picking tickers, analysis date, provider and research depth, and a Python package for use in your own code
  • Works with any market Yahoo Finance covers via exchange-suffixed tickers (0700.HK, 7203.T, RELIANCE.NS, 600519.SS) and crypto pairs like BTC-USD

Getting started

Clone the repo, install it into a fresh environment, set the API key for whichever LLM provider you want to drive the agents, then launch the CLI.

Clone and create an environment

The README uses conda, but any environment manager works.

bashbash
git clone https://github.com/TauricResearch/TradingAgents.git
cd TradingAgents
conda create -n tradingagents python=3.12
conda activate tradingagents

Install the package

Install TradingAgents and its dependencies from the checkout.

bashbash
pip install .

Set your provider keys

Export the key for the provider you plan to use; Alpha Vantage supplies market data. You can also copy .env.example to .env and fill it in there.

bashbash
export OPENAI_API_KEY=...
export ANTHROPIC_API_KEY=...
export ALPHA_VANTAGE_API_KEY=...

Run the CLI

The interactive CLI asks for tickers, an analysis date, the LLM provider and research depth, then streams each agent's progress as it runs. A Docker path (docker compose run --rm tradingagents) is available too.

bashbash
tradingagents

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

When to use it

  • Studying how role-specialised agents and structured debate change an LLM's decision quality
  • A worked reference for multi-agent design: analyst fan-out, adversarial review, then a single decision-maker
  • Backtesting research on how different backbone models reason over the same market inputs
  • Prototyping finance-flavoured agent pipelines over live news, fundamentals and technical indicators

How TradingAgents compares

TradingAgents alongside other open-source multi-agent systems tools AI/TLDR tracks, ranked by GitHub stars.

ToolStarsWhat it does
TradingAgents★ 103kAn LLM trading firm: analysts, bull-bear researchers, trader and risk agents
Ruflo★ 70.9kAgent meta-harness that wraps Claude Code and Codex with 100+ specialized agents, swarm coordination, vector memory, background workers and cross-machine agent federation.
MetaGPT★ 70.2kA multi-agent framework that models a software company, assigning roles like product manager, architect, and engineer to generate code from a single prompt.
AutoGen★ 60.8kMicrosoft Research's framework for building applications where multiple agents converse with each other and with tools to solve tasks.
CrewAI★ 58.2kA framework for assembling teams ('crews') of role-playing agents that divide tasks and collaborate to complete a goal.
AgentScope★ 30.9kA framework for building multi-agent applications with message passing, visual debugging tools, and distributed execution.
OpenAI Agents SDK★ 29.2kOpenAI's lightweight Python SDK for building multi-agent workflows using explicit handoffs, tools, and guardrails.
A2A★ 25.7kOpen Agent2Agent protocol from Google (now in the Linux Foundation) that lets agents from different frameworks discover each other and collaborate over JSON-RPC.