Thorsten Meyer AI has released Forezai Polybot, an MIT-licensed open-source experiment that uses AI to compare its own probability estimates with Polymarket prices, a development that puts automated forecasting, market pricing and trading risk into one public software project.

According to the project material, Polybot is a trading bot for Polymarket that asks whether an AI agent can read public information, form a probability estimate and identify a meaningful gap between that estimate and a market price. The release is available through forezai.com/polybot.html and GitHub, and it opens the portfolio’s Markets family.

The stated design is conservative: the system compares an AI estimate with a market price, records the reasoning behind each estimate and defaults to no trade unless the gap clears cost, confidence and risk limits. The examples in the source material are described as illustrative logic, not a performance record.

The release also carries direct warnings. Thorsten Meyer AI says Polybot is not financial, investment, legal or tax advice, is not a recommendation to trade or use the software, and has no guarantee of accuracy or profitability. The material adds that prediction-market participation is restricted or prohibited in some jurisdictions, including for U.S. persons.

AI Forecasting Meets Market Prices

Polybot matters because it tests a live question in AI and finance: whether a model-generated forecast can add anything useful when markets already price future outcomes through traders with money at risk. A prediction-market contract trading at 62 cents is commonly read as a market-implied probability of about 62 percent, though fees, liquidity and market structure can affect that reading.

The project is not presented as proof that AI can beat markets. Its own framing starts from the opposite point: markets are difficult to beat because prices already collect information, opinions and incentives from many participants. The release treats any apparent edge as a hypothesis that can fail once costs, slippage, limited liquidity and market adaptation enter the picture.

For readers following AI agents, Polybot is also a practical example of agentic software being used in a setting where errors can have direct financial consequences. The project’s emphasis on recorded reasoning and rare action makes auditability part of the announcement, though it does not remove the risk of loss.

Built In Public Day 13

Polybot was introduced as Day 13 of 19 in Thorsten Meyer AI’s Built in Public series and is described as the first node in the operator portfolio’s Markets layer. The broader portfolio material lists several product areas, with Polybot placed under forecasting and trading.

The source material says the project follows the series’ broader themes: local-first operation, provider-agnostic models and software that can be inspected. In Polybot’s case, that means the forecasting model is meant to be swappable, the code is open source under the MIT license and each estimate is supposed to include reasoning that can be reviewed after the fact.

The project’s stated trading stance is narrow. The default action is no trade, with activity reserved for the strongest disagreements between model estimate and market price. The material says even those cases can be wrong.

“This is not financial advice.”

— Thorsten Meyer AI project material

No Track Record Yet

It is not yet clear whether Polybot can produce forecasts that are more accurate than market prices after fees, execution costs and changing market behavior. The source material does not provide audited live trading results, and it says the figures shown are illustrative rather than a record of performance.

Legal access is also unresolved for many potential users. The project warns that prediction-market participation is restricted or prohibited in some jurisdictions, including for U.S. persons. Readers would need jurisdiction-specific legal guidance before using any related trading system.

It is also unclear how the software performs across different model providers, market categories, liquidity levels and news conditions. Open-source code can be inspected, but inspection alone does not establish reliability, profitability or compliance.

Testing Moves To Users

The next step is public review of the code, assumptions and risk controls by users who can inspect the GitHub repository. Any meaningful judgment about the project will depend on transparent testing, careful logs and evidence from real or simulated conditions that include costs and failed calls.

For now, the confirmed development is the release of an experimental open-source tool. Its practical value remains to be shown, and its own materials tell readers to treat automated trading as high-risk and to seek qualified professional guidance before making financial decisions.

Key Questions

What is Forezai Polybot?

Forezai Polybot is an open-source experiment from Thorsten Meyer AI that compares AI-generated probability estimates with Polymarket prices and can be used to study possible disagreements between a model and a prediction market.

Is Polybot a recommendation to trade?

No. The source material says it is not financial advice, not a recommendation to trade, invest or use the software, and not a promise of profit or accuracy.

What makes the project newsworthy?

The release puts AI forecasting against live market pricing in public code, with an explicit focus on auditability, risk limits and the difficulty of finding any real edge against markets.

Can U.S. users use Polybot on Polymarket?

The project warns that prediction-market participation is restricted or prohibited in some jurisdictions, including for U.S. persons. Users need to know the law that applies to them before taking any action.

Has Polybot proved it can beat prediction markets?

No. The source material does not provide audited live performance results and says its examples are illustrative. Its own framing treats any apparent edge as uncertain.

Source: Thorsten Meyer AI