About Arbitrage Agent

Founder

Diogo Gomes

Software engineer and prediction market trader. Built Arbitrage Agent after spending months manually scanning Polymarket and Kalshi for price discrepancies — and losing most of them to execution speed.

The first version was a Python script. Today it's a full platform monitoring 4,812 markets with AI-powered semantic matching, parallel execution, and automatic risk controls.

Your capital stays on-venue

Arbitrage Agent never holds your funds. Your capital stays in your Kalshi and Polymarket accounts at all times. We connect via API keys with read + trade permissions only — no withdrawal access, ever.

You can revoke API key access at any time from your Kalshi or Polymarket account settings. There is no lock-in and no custody risk.

API permissions — what the agent can and cannot do

Risk controls

Circuit breaker: If one leg of a trade fails to fill, the other is automatically unwound within 400ms. This ensures zero directional exposure — it's always both-or-neither.

Kelly criterion sizing: Position sizes are calculated using the Kelly criterion, optimizing for long-term growth while limiting drawdown risk.

Kill switch: Instantly exit all positions and stop the agent with one click from the dashboard.

Dry-run mode: Test the full system with paper trades before committing real capital. See exactly what the agent would do without any risk.

Semantic matching at 99.2% precision

Polymarket and Kalshi list the same events under completely different titles. Simple text matching fails — "Fed cuts rates in May FOMC" and "Will the Federal Reserve cut at May FOMC?" are the same event with zero keyword overlap.

Arbitrage Agent uses OpenAI embeddings combined with a custom reranker to semantically match events across platforms. This approach achieves 99.2% precision across all 4,812 monitored markets — linking events that text-based systems miss entirely.