Methodology: how our analyses are produced
How our AI agent team works: analysts, bull/bear debate, risk management, quality checks and limitations.
A team of AI agents
Every analysis is produced by TradingAgents, an open-source framework (Apache-2.0) that simulates an investment firm. Four specialised analysts —technical, fundamental, news and sentiment— study the stock using public market data.
Bull vs bear debate
A bullish and a bearish researcher debate the analysts' reports. A research manager weighs the debate and proposes an investment plan, which a trader turns into a concrete proposal.
Risk management and final rating
Three risk profiles (aggressive, neutral and conservative) challenge the proposal and a portfolio manager issues the final rating on a five-level scale (Buy, Overweight, Hold, Underweight, Sell), shown in simplified form as BUY / HOLD / SELL.
Automatic verification before publication
Every analysis must pass a battery of automatic checks before it is published. On the decision: the analysts' figures must come from the primary data, the data must be recent and match market prices and metrics, the rating must be unambiguous and the final decision must not use unsupported figures or sources. On each article: the rating must match the decision, there must be no invented figures or sources, the text must be entirely in its language, the structure must be complete and promotional language is banned. If a check fails, the system corrects the text automatically and verifies it again; if it still fails, the analysis is not published. Every report shows the detail of the checks it passed.
Models and data
The agents run mainly on open language models on our own infrastructure; for the ten largest companies we also run a weekly deep analysis with a commercial model for the debate and decision stages. We cover the 25 largest S&P 500 companies by market cap and refresh every analysis at least every two days. Price and metric data come from public sources and may contain errors or delays.
Limitations
AI models can be wrong, miss relevant information or misread data. Analyses do not consider your personal situation and are not financial advice. We publish the full rating history of every stock so you can judge its track record.
TradingAgents © Tauric Research, Apache License 2.0 — github.com/TauricResearch/TradingAgents
TradingView Lightweight Charts™ © TradingView, Inc., Apache License 2.0 — tradingview.com
This product uses the FRED® API but is not endorsed or certified by the Federal Reserve Bank of St. Louis.
Company logos: Wikimedia Commons, cropped and adapted by us (public domain; NVDA: Nvidia, Apache License 2.0)