Knowledge Base
Everything you need to know about the TradingAgents platform, its data sources, and the autonomous AI agents powering the research pipeline.
TradingAgents pulls from 12 distinct public data channels across two layers. The first layer is our proprietary OSINT sweep (Colonel Wolfe), which queries:
The second layer is the Tauric Research framework, where four specialized AI research agents independently process this data across technical, sentiment, news, and fundamental verticals. No paid data subscriptions are required. Everything runs through publicly available government and financial APIs.
TradingAgents is an autonomous multi-agent research platform that deploys 9 AI agents across 6 analytical stages to produce institutional-grade stock analysis. It extends the open-source TradingAgents framework from Tauric Research with a proprietary OSINT intelligence layer, custom agent architectures, and a full-stack analytical dashboard. The platform researches, debates, and assesses risk without human intervention, delivering a structured investment recommendation to the final human decision-maker.
The platform produces research and recommendations only. It does not execute real trades or connect to any brokerage. Our SimTrader module (currently in beta) allows paper trading at live market prices to benchmark the quality of agent analysis over time. No real money is ever at risk through the platform itself. The final investment decision always rests with the human portfolio manager.
The platform supports multiple LLM providers including Anthropic (Claude), OpenAI (GPT), Google Gemini, DeepSeek, and xAI Grok. Each analysis run can be configured with different model tiers. The default configuration uses DeepSeek for cost-efficient deep reasoning, but the architecture is model-agnostic. Any agent can be swapped to a different provider without changing the pipeline logic.
A complete 6-stage analysis for a single ticker typically takes 3 to 8 minutes, depending on the model tier selected and the complexity of the adversarial debate. This includes the full OSINT intelligence sweep, all four research verticals, multiple rounds of bull/bear debate, judicial verdict, three-perspective risk assessment, and final consensus synthesis. Every step produces documented deliverables stored in the database.
Yes. The platform supports any publicly traded US equity with a valid ticker symbol. You can add tickers to your watchlist, set analysis frequency, and run on-demand or scheduled analysis. The OSINT layer adapts its intelligence sweep based on the company's sector, whether it holds government contracts, and the availability of options data for that symbol.
The Crypto Intelligence Analyzer is a specialized pipeline that activates whenever a digital asset or cryptocurrency is the target. Instead of the traditional equity OSINT sweep, Agent Jake Summers executes a crypto-native reconnaissance protocol that includes:
Jake's briefing replaces Colonel Wolfe's equity-focused intelligence when the target is crypto, ensuring the downstream research analysts, debate advocates, and risk assessors are working with data sources relevant to decentralized markets rather than traditional financial instruments.
The Polymarket Intelligence Analyzer is a completely separate 6-agent adversarial pipeline purpose-built for binary prediction markets. It does not use the equity or crypto pipeline. Instead, it deploys its own specialized agents:
The output is a structured analysis with a consensus probability, the calculated edge against the market, and a recommendation ranging from Strong Buy YES to Strong Buy NO.
DayTrader is a compressed, speed-optimized version of the full TradingAgents pipeline designed specifically for intraday trading. While a standard analysis run takes 3 to 8 minutes across 6 stages, DayTrader condenses the analytical process to generate actionable intelligence within minutes.
DayTrader focuses on time-sensitive signals that matter for intraday positioning:
DayTrader is designed for traders who need the intelligence quality of the full pipeline but cannot wait for a multi-stage adversarial debate when markets are moving in real time.
Col. Don Wolfe (RET) is the Intelligence Officer and the first agent in the pipeline. Before any research analyst touches a ticker, he executes a full-spectrum open-source intelligence sweep across every available data channel.
His reconnaissance protocol includes:
The Colonel compiles everything into a classified-style intelligence briefing document that serves as the foundational input for all downstream agents. No research analyst, debate advocate, or risk assessor operates without first reviewing his findings.
Marcus Chen is the Chief Technical Analyst operating in Stage 1. He specializes in quantitative price action analysis and multi-timeframe technical indicator synthesis.
Marcus processes the technical data from Colonel Wolfe's briefing and applies his own analysis layer:
His output is a structured technical probability framework that quantifies the statistical likelihood of upside vs downside price movement based purely on price action data.
Sarah Mitchell is the Sentiment Intelligence Lead. She operates the sentiment quantification engine that processes both institutional and retail positioning data.
Sarah constructs a composite sentiment score that captures conviction levels across both retail and institutional participants, fear/greed indicators, and short interest dynamics.
James Rivera is the News and Media Analyst. He monitors and processes real-time information streams to extract actionable intelligence from unstructured media data.
James flags material catalysts, sector rotation triggers, and geopolitical risk vectors. His analysis identifies time-sensitive information that could alter the investment thesis before the market fully prices it in.
Elena Kowalski is the Fundamentals Research Lead. She conducts deep-dive financial statement analysis and valuation modeling.
Elena produces a composite fundamentals score using proprietary scoring methodologies that weight each factor based on the company's sector and growth stage.
David Park is the Bull Case Advocate operating in Stage 2 (Investment Debate). His role is to construct the strongest possible investment thesis in favor of each position.
David aggregates the most compelling evidence from all four research verticals (technical, sentiment, news, fundamentals) and builds a rigorous case for upside potential. He identifies:
David is not permitted to equivocate or present balanced analysis. His mandate is to present the most aggressive, evidence-backed bull thesis possible, which is then stress-tested by Catherine Walsh (Bear Advocate).
Catherine Walsh is the Bear Case Advocate. She systematically deconstructs the investment thesis, stress-testing every assumption against downside scenarios.
Catherine forces the pipeline to confront uncomfortable data points that the bull thesis may have downplayed. This adversarial structure ensures that no investment recommendation passes through without surviving rigorous scrutiny from both sides.
Michael Torres is the Research Director, operating in Stage 3. He presides over the adversarial debate between the bull and bear advocates like a judicial figure.
Michael does not have a predetermined bias. His role is to:
His verdict determines the informational foundation upon which the final risk assessment and portfolio decision are built. The debate transcript and judicial decision are both preserved in the audit trail.
The Risk Committee is a three-member evaluation board in Stage 4 that examines every position through three distinct analytical lenses simultaneously:
A Risk Judge then synthesizes all three inputs into a unified risk verdict that directly informs the final portfolio directive. The risk assessment considers not just the individual position, but how it interacts with the broader portfolio context.
Brian Galvan is the Portfolio Manager and the only human in the entire pipeline. He operates at Stage 5, the final decision stage, where every analysis run terminates.
By the time a recommendation reaches Brian, it has already passed through:
Brian reviews all 14 documented deliverables produced by the pipeline and makes the final call: buy, sell, hold, or pass. He evaluates the consensus score, weighs the strength of the debate verdict against the risk assessment, and considers his own portfolio-level context that the agents cannot see, including cash allocation, existing position sizes, and personal risk tolerance.
When Brian decides to act on a recommendation, he uses SimTrader to execute paper trades at live market prices, benchmarking whether the agents' analysis translates into profitable positioning. Over time, this creates a trackable performance record that feeds back into refining the agent pipeline. He can enter trades at market price or at custom prices to mirror his real brokerage positions.
The key principle: AI agents research, debate, and assess risk. The human makes the final portfolio decision. No trade is ever executed without Brian's explicit approval. The platform is a research tool, not an autonomous trading system.