How I Built an Automated Arbitrage Agent for Polymarket and Kalshi
--
Arbitrage is one of the oldest strategies in finance. Buy low on one exchange, sell high on another, pocket the difference. Simple in theory. Brutally hard to execute at scale.
When I started looking at prediction markets, specifically Polymarket and Kalshi, I noticed something most traders ignore: the same event, priced differently on two platforms simultaneously. Not sometimes. Constantly.
What Is Prediction Market Arbitrage?
If event X is priced at 48¢ on Polymarket and 50¢ on Kalshi, buying YES on one and NO on the other costs 98¢. The outcome pays $1.00 regardless, guaranteed 2% profit with zero directional risk.
No prediction needed. You’re not betting on what happens. You’re betting on the price difference disappearing, which it always does at settlement.
We wrote a deeper breakdown of how this works in our guide: How Prediction Market Arbitrage Works.
Why Manual Trading Doesn’t Work
I tried doing this manually for two weeks. The windows are short, often 30 to 90 seconds. By the time you spot an opportunity, check both platforms, calculate the spread and place both orders, the gap is already closing.
The only way to make this work is automation. Both legs executed simultaneously, spread validated before committing capital, scanning continuously across thousands of markets.
So I built Arbitrage Agent, an automated agent that does exactly this, 24/7.
How the Agent Works
The system has three layers:
1. Market matching: the same event exists on both platforms with different identifiers and phrasing. Getting the matching right is critical. A wrong pair looks like arbitrage but is actually directional exposure.
2. Opportunity detection: every matched pair gets a spread calculation on every price update. Candidates are filtered by liquidity depth, slippage estimate, time to expiry, and historical fill rate.
3. Execution: both orders sent in parallel via async tasks. If either leg fails to confirm, the other gets cancelled immediately. Current median execution time: under 400ms end to end.
For a more detailed look at the strategy and typical spreads, see our Polymarket Kalshi Arbitrage guide.
What I Learned the Hard Way
Fees matter more than the spread. A 2% gross spread can become 0.3% net after maker/taker fees on both sides.
Latency is everything. Being 200ms faster changes fill rate meaningfully at tight spreads.
Always build dry run mode first. Three weeks of simulation before going live caught bugs that would have caused real losses.
Want to go deeper on the strategy side? We covered the most common mistakes and how to avoid them on the Arbitrage Agent blog.
Where It Is Now
Arbitrage Agent scans 2,400+ markets simultaneously, 24/7. The matching accuracy is high, execution is fast enough to capture most flagged opportunities, and the risk controls have worked as designed.
We’re now in private beta, onboarding users who want to run the strategy without building the infrastructure themselves. The agent handles detection and execution, you control position sizes and risk limits. Check out the arbitrage calculator to see what the numbers look like with your own position size.
Join the waitlist at arbitrage agent.
Past performance is not indicative of future results. This is not financial advice.
