Kalshi trading bots: the loop that works, the ones that don't, and an open-source starting point

By Melvin Tercan, Octagon Research Editor

A Kalshi trading bot is software that trades event contracts through Kalshi's official API. The ones with a real chance run a three-stage loop — research a market, estimate an independent probability, execute only when the gap to the market price clears fees and spread — behind hard risk controls. Octagon maintains an open-source AI trading CLI for Kalshi that implements this whole stack, useful as a tool or as the reference architecture for your own.

Why "bot" isn't a strategy

Automation adds speed and discipline; it does not add edge. Event contracts are a zero-sum game minus fees, so a bot that trades patterns anyone can see is paying costs to flip coins. The question that decides profitability is the same one a human faces: do you have a better probability estimate than the price? Everything else — API plumbing, scheduling, order management — is necessary but not sufficient. Our Kalshi API guide covers the plumbing; this page is about the loop that makes it worth running.

The loop that works: research → edge → execution

• Research. For a candidate market, gather what actually drives it — the data release calendar, polls, filings, weather models, whatever the settlement source keys on — and produce an independent probability with stated reasoning. This is the stage AI genuinely changed: deep research per market used to be the unscalable part.
• Edge. Compare your probability against the live order book — not the last price. If you say 55% and the ask is 44¢, your gross edge is 11¢; subtract the round-trip fee and any spread you cross to get net edge. No net edge, no trade — most markets, most days.
• Execution with sizing. Kelly gives the bankroll-optimal fraction for a binary bet, and full Kelly is famously too aggressive for uncertain probability estimates — fractional Kelly (half or less) is the standard compromise. Then the risk engine: per-market caps, category exposure limits, daily loss kill switch.

An open-source reference: Octagon's trading CLI

Octagon's kalshi-trading-bot-cli implements the full loop as an AI-native terminal agent: deep fundamental research per market, an independent probability estimate, edge computed against the live order book, Kelly-based sizing, and execution gated behind a five-gate risk engine. It's open source — read it to steal the architecture, run it as your research-to-execution stack, or point its research layer at your own execution. The research it consumes is the same model-vs-market analysis Octagon publishes for every active Kalshi market and serves programmatically through its API and MCP server.

The failure modes that actually kill bots

• Fee-blind backtests. Strategies profitable before per-contract fees and spread are the default outcome of a naive backtest; add realistic round-trip costs — see Kalshi fees explained — before believing any curve.
• Settlement-rule surprises. A bot that trades headlines settles against rules; the wording is machine-readable and must be part of the research stage.
• Thin-book market impact. Sizing from your bankroll instead of the book's depth turns a good signal into a bad fill on long-tail contracts.
• Correlated exposure. Twenty positions on rate-sensitive markets is one position worn twenty ways; cap by theme, not just by market — the framework is in handling correlated markets.
• Skipping the demo. Kalshi's paper-money environment exists precisely so your first unattended week costs nothing. Use it until boredom, then promote.

Frequently asked questions

Are trading bots allowed on Kalshi?

Yes — Kalshi provides an official API with order placement, and API access tiers that scale with activity. Bots must be your own account acting under your control; the usual exchange rules about manipulation apply to automated trading exactly as to manual.

Do Kalshi trading bots make money?

Most naive ones don't. A bot has no edge just by being a bot: after per-contract fees and spreads, a strategy needs a genuine information or modeling advantage. The bots with a plausible edge are research-driven — they estimate an independent probability per market and trade only when it diverges enough from the price to clear costs.

What is the Octagon Kalshi trading bot CLI?

An open-source, AI-native command-line trading agent for Kalshi. It runs deep fundamental research on a market, generates an independent probability estimate, computes edge against the live order book, and can execute with Kelly-based sizing behind a multi-gate risk engine. It's on GitHub under OctagonAI as kalshi-trading-bot-cli — usable as a tool or as a reference architecture.

How much money do I need to run a Kalshi bot?

Contracts price under $1 each, though live accounts carry Kalshi's own small deposit minimum — contract pricing and account funding are separate floors — the real constraint is statistical: with small size, fees and spreads dominate, and with any size, you need enough independent positions for an edge to express itself. Start in Kalshi's demo environment with paper money; graduate only after the bot survives unattended weeks.

What risk controls should a Kalshi bot have?

Position caps per market and per category, exposure limits as a fraction of bankroll, fractional Kelly sizing rather than full Kelly, a max-daily-loss kill switch, and hard confirmation gates before orders. Binary settlement makes sizing errors unforgiving — the risk engine matters more than the alpha model.