Rank #10 · Daily · AI
TradingAgents: Multi-Agent LLM Framework for Financial Trading Research
Spin up a multi-agent "trading firm" of LLMs that debate fundamentals, news, sentiment, and risk — then emit a research decision you can backtest.
What it does
TradingAgents is a multi-agent LLM financial trading framework (arXiv:2412.20138) that decomposes a trading decision into specialized roles:
- Analyst team — Fundamentals, Sentiment, News, and Technical analysts run in parallel with their own tools, then hand reports to research.
- Researcher team — Bullish and bearish researchers debate the analysts' findings.
- Trader agent — Turns the debate into a timing / magnitude proposal.
- Risk management + Portfolio Manager — Evaluate risk factors and approve or reject the proposal against an optional portfolio book; approved orders go to a simulated exchange.
Implementation notes from the README:
- Built on LangGraph; CLI (
tradingagents) plus a PythonTradingAgentsGraphAPI. - Multi-provider LLMs (OpenAI, Google, Anthropic, xAI, DeepSeek, Qwen, GLM, MiniMax, OpenRouter, Bedrock, Ollama, ChatGPT subscription login, OpenAI-compatible endpoints, and more). Deep and quick tiers can use different providers.
- Data path includes Yahoo Finance markets (US and exchange-suffixed tickers, crypto), SEC EDGAR point-in-time fundamentals for US filers, optional FRED / Alpha Vantage / social screening.
- Memory log settles past decisions with realized returns; optional checkpoint resume;
run_backtestscores decisions over a ticker/date grid. - Explicit disclaimer: research purposes only — not financial, investment, or trading advice. Performance varies with models, temperature, dates, and data quality.
Who it's for
Researchers and builders studying multi-agent finance workflows, backtesting scaffolds, or LLM debate patterns over market data. Not a ready-made live trading bot, and not suitable as a substitute for licensed advice.
Stars / language / license
120,363 stars · Python · Apache-2.0
Data verified: 2026-11-04 09:13 Asia/Shanghai (GitHub API get_repository + README get_file_contents)
Trending note: +3,336/wk (weekly Trending filler; selection captured 09:12 Asia/Shanghai)
Category tags
AI · multi-agent · finance · trading · LangGraph · research · LLM
How it differs from similar projects
Compared with single-agent "ask ChatGPT about a stock" scripts. TradingAgents hard-codes a desk-shaped role graph (analysts → debate → trader → risk → PM) with persistence and backtest hooks.
Compared with quant backtest libraries (zipline, backtrader, vectorbt). Those optimize classical strategy code and market simulators. TradingAgents is an LLM orchestration research scaffold; its own README warns that backtest figures are not a fixed, replicable strategy return.
Compared with proprietary AI trading products. This repo is Apache-2.0 source you run with your own keys / local models.
TidyHow does not endorse trading performance claims. Any returns in papers or READMEs are the authors' results under their settings — not independent verification.