openfi-dao/kalshi-trading-bot

This is a TypeScript bot that trades on Kalshi, a US-regulated prediction market where people bet on the outcomes of real-world events such as election results, economic indicators, or weather. The bot uses five AI agents working in parallel to analyze each available market, debate the outcome, and agree on whether to place a bet and how large it should be. The five agents each play a different role: one produces a baseline probability estimate, one analyzes recent news, one argues the case for a market going up, one argues the case for it going down, and one manages risk and can veto any position. Their outputs are combined using a weighted formula, and if fewer than three agents agree or confidence is too low, no trade is placed. Position sizes are calculated using the Kelly criterion, a well-known mathematical approach to sizing bets based on estimated probability and payoff, with hard daily loss limits applied on top. The bot runs in two modes. Paper mode executes the full analysis and decision pipeline but does not place real trades, which is the recommended starting point. Live mode requires explicitly setting a flag in the configuration file. Every decision is logged to a local SQLite database so trades can be reviewed or debugged later. A command-line dashboard lets the user check current status, agent performance scores, and trade history. Setup requires Node.js 22.5 or newer, a Kalshi account with API credentials, and an OpenRouter account for LLM access. Configuration is done through a single environment file that also sets safety limits like a daily cap on AI API spending. Four trading strategies are included with different approaches: one focused on compounding conservative positions, one that quotes on both sides of a market, one for short-hold directional bets, and one that scores markets by category. The README advises treating LLM outputs as probabilistic rather than reliable predictions and consulting local regulations before trading.