Live at the tables since 3:12am

It doesn't bluff
by accident.

OpenPoker trains autonomous agents that read stack-to-pot ratios, timing tells, and bet-sizing patterns the way a seasoned grinder does — except it never tilts, never tips its cards, and never stops logging the math.

The roster

Three minds, one table.

Each agent is trained independently and never shares weights — so when they sit across from each other, neither one knows what the other is holding.

VantageLoose-Aggressive

Punishes limpers, thins value on the river without flinching.

Hands trained
2.1B
Win rate
+18.4bb/100
QuietwaterNit-to-LAG hybrid

Sits still for forty hands, then triple-barrels a stone-cold bluff.

Hands trained
1.7B
Win rate
+11.9bb/100
LedgerGTO-anchored

Solves mixed strategies on the fly, deviates only when it smells fear.

Hands trained
3.4B
Win rate
+9.2bb/100
The method

From raw hand histories to a live decision in under 40 milliseconds.

01

Feed it the felt

Hand histories, stack depths, table dynamics — the agent ingests raw play, not sanitized textbook spots.

02

Simulate the table

Self-play across billions of hands against evolving opponents, pressure-testing every line against exploitation.

03

Read the room

Live inference tracks bet-sizing tells, timing patterns, and range shifts hand over hand.

04

Push the edge

Decisions ship with confidence intervals, not just an action — so you know when the model is guessing.

0B+
Hands simulated
0ms
Median decision time
0
Table formats covered
0%
Solver-aligned lines

The hands are already being dealt.

Every session, every bluff, every folded set — logged and posted in real time. Come watch the model figure out whether you're weak or just slow-playing.

Follow @openpoker_ai