AI-powered PLO training
Train double-board PLO bomb pots against a neural network.
A model trained on billions of self-play decisions, in a GTO-style trainer. Drill random spots and get every decision graded, see what each one cost you in EV, study the hands you played — and run private home games with your friends.
Sign in with GoogleFree while the models are in development · no card, no daily limit · why free?
What you get
A trainer, a study tool and a review, built for one format.
Trainer
Random bomb pots dealt instantly, any stack depth, two- to six-handed. Play them against the network and every move is graded — Best move, Correct, Inaccuracy, Wrong, Blunder.
Review
After each hand, step through every decision — yours and theirs — see what each of yours cost in EV, and swap a card to see how the answer changes.
Study
Enter a hand you played: seats, stacks, cards and actions. The network answers at every decision, with the bet sizes it prefers. Share any spot with a link.
Home games
Play with your friends — on the same tables.
Start a private club, send your friends an invite link, and deal double-board PLO5, PLO6 or PLO67 bomb pots on any phone or computer.
- Private clubs. Only members see your tables, hand histories and numbers.
- A ledger, not a cashier. WrapGTO keeps score of who's up and who owes whom; no money ever moves through the site.
- A shuffle you can check. Every player's device helps cut the deck, and anyone can verify a deal afterwards.
- Every hand kept. A hand replayer, club stats, and the same move-by-move grades as the trainer.
Why a neural network
Classical solvers can't touch this format. A network can.
Any stack depth
Real games don't come in preset configurations. The model trains across two- to six-handed pots with stacks from short to 300bb deep — and reads the exact stacks in front of it, not the nearest template.
Any bet size
No fixed bet menu. The network converges on its own preferred size for each spot — a continuous choice from min-bet to pot — and grades your sizing against that judgment, not a coarse grid.
Every multiway spot
Bomb pots are multiway by nature, which is exactly where game trees explode. The network doesn't enumerate the tree — it plays the spot in front of it, however many players are in.
What it is — and isn't
Not a true solver. Honestly better than pretending.
WrapGTO does not compute a Nash equilibrium. The model approximates GTO play, learned through self-play — that's a real limitation, and I'd rather state it than market around it.
The alternative doesn't exist: a true CFR solver for double-board PLO5 bomb pots would require crushing abstractions — pre-configured seat and stack layouts, a handful of allowed bet sizes, and heavy card bucketing — before the tree fits in any computer. What you'd gain in theoretical purity, you'd lose in relevance to the hand you actually played.
I'm also building a best-response probe that measures how much a perfect adversary could win against the live model. The number will be published here when it's done — whatever it says.
Why it's free
The models are still in development, so the whole site is free.
The networks behind WrapGTO get stronger with every training run, but they aren't yet at the level I'd be comfortable charging for. Until they are, everything is open: unlimited trainer hands, full Study, home games. New weights reach the site regularly as training goes on.
If paid plans ever arrive, anyone using the site will hear about it well in advance — priced to cover GPU time, not to turn a profit.
Sign in and startA Google sign-in keeps your hands and stats yours. Nothing else is asked for.
Why this exists
A study tool first. A business barely.
WrapGTO started as a personal project: I'm a professional player who wanted to study double-board bomb pots and found that nothing for the format existed — no solver, no trainer, no library. So I built the model, the engine, and this site to study against myself. Each new format is weeks of training at around $50 a day, funded out of my own poker winnings. I'd be running it either way.