What is kalshi-trading-bot?

This is a TypeScript-based automated trading bot for Kalshi, a regulated U.S. prediction market platform where you bet real money on the outcomes of events like elections, economic indicators, and sports results. What distinguishes this bot from simpler trading bots is its multi-AI-agent architecture. Instead of a single AI making a buy/sell decision, it runs five specialized AI agents in parallel through OpenRouter (a service that routes requests to multiple AI providers with one API key): a forecaster, a news analyst, a bull researcher arguing for a position, a bear researcher arguing against it, and a risk manager. The bull and bear agents debate the trade, and the bot only proceeds if disagreement is below a threshold. The five agents' probability estimates are combined using confidence-weighted consensus, a voting-like approach where models with better historical calibration have more influence. Position sizing uses the Kelly Criterion, a mathematical formula that bets a fraction proportional to your estimated edge to maximize long-run growth without risking ruin. Safety features include hard position size limits, a daily dollar loss cap, a built-in daily AI cost cap (default ten USD) that physically prevents LLM calls when the budget runs out, and paper trading mode (simulated trading with fake money) that is enabled by default so you must explicitly flip a flag to use real money. Trade history, model performance, and category scores are stored in SQLite. Built on Node.js 22.5 or newer, TypeScript, and Vitest for testing.