Run a live Kalshi prediction-market account autonomously
An agent using Claude Code traded a $20 Kalshi account for nine days under supervision, earning $20.30 in incentive payouts against $36.06 in trading losses.
Done with Claude Code
What they did
Mirabilis, an autonomous AI agent using Claude Code, was given access to a live $20 Kalshi prediction-market account. It traded on its own for nine days, participating in the Liquidity Incentive Program while a human provided supervision throughout.
How it went
The agent collected $20.30 in LIP payouts against $36.06 in trading losses. A released toolkit models rewards per snapshot as size \times d^(ticks below reference price) with daily payouts only if at or above $1.00.
Worth knowing
Anything touching money, benefits or eligibility should be double-checked against the official provider before you act on it.
Try it yourself with Claude Code
Help me set up a supervised paper-trading experiment on Kalshi with a small budget of [dollar amount, e.g. $20]. Each day, research [category of markets, e.g. economic data or weather], suggest up to [number] trades with your reasoning, the price, and the maximum I could lose, and keep a running log file of positions, fees, and profit or loss. Do not place any real order, deposit money, or change account settings until you show me the exact trade and I approve it, and after [number of days, e.g. 9] days give me a summary of results versus fees and what you would change.
Source on X - Oct 11, 2026
Read the original