openthomas
github.com/PredictionMarketTrader/openthomas
Autonomous agent that forecasts station temperatures and trades them on Kalshi and Polymarket, publishing every claim before settlement with falsification prices.
Numerical weather prediction + Kelly-criterion position sizing across Kalshi and Polymarket temperature markets, with full transparency of forecasts and market conditions at trade time
A fully autonomous weather trading agent that operates across both Kalshi and Polymarket. It generates station-level temperature forecasts using numerical weather prediction models, compares these to market prices, and executes trades using Kelly criterion position sizing when it detects edge. What distinguishes this bot is its transparency mandate: every forecast and trade is published on openthomas.com before settlement, along with the market price at formation and the conditions that would falsify the prediction. The 2.1MB Python codebase is substantial and production-quality, including a MCP server for external integrations, database persistence for trade history, and a web dashboard (referenced as openthomas.com). The architecture shows separation between the weather forecaster, market price fetcher, Kelly calculator, and trade executor. Created July 3, 2026, it's very new but shows rapid development. Code quality is high with extensive type annotations, async operations, and config-driven design. No malware concerns. This is a reference implementation for anyone building cross-exchange prediction market bots, particularly for weather markets. The MCP server integration makes it extensible for other domains. Best suited for developers who want to study a complete, transparent trading system architecture.