Forezai has released TradingAgents, an Apache-2.0 open-source research framework that models a trading desk using multiple AI agents, including analyst agents, bull and bear researchers, a trader and a risk manager that can veto proposed actions.

The project was introduced by Thorsten Meyer AI as part of a Built in Public series and is available through Forezai’s website and GitHub, according to the source material. The system is described as experimental and is not presented as a trading recommendation, investment product or profit-generating tool.

TradingAgents is designed around role separation. Specialized analyst agents gather different types of signals, including fundamentals, news or sentiment, and technical price action. A bull researcher builds the case for action, while a bear researcher challenges it. A trader then proposes an action, which a risk manager can reject, reduce or approve within limits.

The source material says the framework records reasoning at each stage and treats “no trade” as a common possible outcome. Its stated purpose is to use structured disagreement and oversight to reduce the risk of a single AI model producing an overconfident market call.

AI Trading Gets A Risk Gate

The release matters because it points to a broader shift in AI tooling: away from single-model answers and toward systems that assign different roles to different agents. In financial analysis, that distinction matters because a fluent answer can sound persuasive even when the underlying judgment is weak.

Forezai’s claim is that a simulated desk can make AI-assisted decision-making more inspectable by forcing arguments to pass through debate and risk review. That does not mean the output is correct, profitable or lawful to use in live markets. The source material repeatedly warns that automated trading carries a substantial risk of loss, including loss of all capital.

For readers following AI software development, TradingAgents is also part of a larger question: whether open-source agent frameworks can make high-risk decision processes more auditable. The framework may be most relevant as a research pattern for decision-making under uncertainty, rather than as a system for placing trades.

Forezai Completes Its Markets Pair

The announcement follows an earlier Forezai item about Polybot, described in the source material as a single AI forecaster comparing one estimate with one market price. TradingAgents is presented as the companion system: not one forecaster, but a simulated firm with internal disagreement.

The project is part of an 18-product operator portfolio described by Thorsten Meyer AI. Within that portfolio, Polybot and TradingAgents make up the Markets family. The source material says the broader foundation is local-first and provider-agnostic, meaning the systems are intended to run on owned compute and support swappable AI models across roles.

The release also echoes another project in the series, IdeaClyst, which the source material describes as using a council-style approach. In TradingAgents, that pattern is applied to market analysis, with debate and risk review built into the process.

“This is not financial advice, and nothing here recommends trading, investing, or using this software.”

— Thorsten Meyer AI

Performance Evidence Is Still Missing

It is not yet clear how TradingAgents performs in live or backtested market settings. The source material does not provide audited performance results, benchmark data or evidence that the framework improves outcomes compared with single-model systems or conventional trading tools.

It is also unclear how users would need to adapt the framework for specific markets, brokers, compliance rules or model providers. The source material says market and trading-software access may be regulated or restricted in some jurisdictions, leaving legal and operational use cases dependent on where and how the software is used.

Users Can Inspect The Code

The next step is public inspection and experimentation by developers, researchers and market-technology observers. Because the framework is open source under Apache-2.0, users can review the code, test the agent structure and evaluate whether its role-based design is useful for research.

Any move from research use toward live trading would require separate validation, legal review, risk controls and professional advice. The source material states that the software is provided without warranty of accuracy or profitability.

Key Questions

What is Forezai TradingAgents?

TradingAgents is an experimental open-source research framework that uses multiple AI agents to simulate parts of a trading desk, including analysis, debate, trade proposal and risk review.

Is TradingAgents financial advice?

No. The source material states that it is not financial, investment, legal or tax advice and is not a recommendation to trade, invest or use the software.

What makes TradingAgents different from a single AI forecaster?

The framework separates roles. Instead of relying on one model answer, it has analyst agents, opposing bull and bear researchers, a trader and a risk manager that can reject or limit proposed actions.

Is there proof that TradingAgents is profitable?

No audited performance evidence is provided in the source material. The project is described as a research framework, and the author says there is no guarantee of accuracy or profit.

Where does this fit in Forezai’s portfolio?

TradingAgents completes Forezai’s Markets family alongside Polybot, according to the source material. Polybot is described as a lone forecaster, while TradingAgents is described as a simulated firm of debating agents.

Source: Thorsten Meyer AI