TradingAgents
Multi-agent framework where language-model agents act as analysts, researchers, a trader and a risk team.
GitHub statistics as of 2026-11-10. Status: active = commits in the last 90 days, slowing = in the last year, inactive = older, archived = read-only.
Overview
TradingAgents splits a trading decision across several LLM agents: analysts read fundamentals, news, sentiment and technicals, bullish and bearish researchers debate, a trader proposes a position and a risk team reviews it. It runs on commercial or local language-model APIs, so each decision costs tokens, and it is presented by its authors as a research framework rather than investment advice.
Best for: Exploring how LLM agents reason about stocks; research and demos.
Strengths
- Readable reasoning for every decision
- Modular agent roles and model choice
- Huge community interest
Limitations
- Hard to backtest over long periods: LLM calls are slow and costly, and models may already know the outcome of past events
- Results vary between runs and models
- Not a substitute for a systematic, testable rule
From our own use
The look-ahead problem is real: a language model trained on data up to a recent date may already know what happened in the period you backtest. Any LLM strategy result needs a test period after the model's training cutoff.
Alternatives in AI and machine learning
All frameworks · Descriptions are our own summary and may lag behind each project's documentation. Check the project's license before using it in a product. Nothing here is investment advice.