Pot-Limit Omaha, played precisely.

Six players, six hole cards. Every opponent runs a trained near-equilibrium policy served from a Python inference API — the same engine that powers the state builder.

Play the table Open the builder

fold

check

call

25%

pot

allin

The model outputs a distribution over actions.

Two ways in

Play Sit at the table Take a seat against five neural-network bots. Pot-limit betting, side pots and showdowns — full hands dealt and resolved in the browser. Play now Build Craft a game state Set the hole cards, board, stacks and action line by hand, then read off the exact ApiGameState the engine builds — and query the policy for live action probabilities. Open the builder

Under the hood

POLICY Deep-CFR near-equilibrium Bots sample from a policy trained over the 13-action pot-limit abstraction, served from a Python inference API.

ENGINE Pot-limit, chip-exact Blinds, side pots and pot-limit sizing are computed in integer engine units — the same units the model is trained on.

BUILDER Feature-faithful Hand-craft any spot and read the exact feature payload the policy receives, chip-for-chip with the live table.