Poker AI that tells you
how to exploit your table.
We modeled how player types actually behave — then built a coach that grades your lines, names your leaks, and tells you how to attack the opponents in your game.
// $79/mo · full access · cancel anytime
Try an exploit read on your table →> coach
We modeled how players actually behave.
Thirty-six frontier models played millions of duplicated hands — we extracted archetypes, deception patterns, and exploit lines. The coach turns that research into plain-English reads on your table.
How do I beat a calling station?
Bluff less and value-bet thinner. Calling stations pay off too many marginal hands, so hands such as second pair can often bet for value instead of checking. Keep some bluffs when the spot supports them, but make your default adjustment simple: charge their calls rather than trying to force folds.
Generic LLMs hallucinate frequencies. Solvers are correct but unreadable. Poka explains what drives the decision, grounds answers in curated strategy sources, and reads your measured leaks when you have data.
Lab research. Player product.
Exploit the table. Find the leak. Drill the pattern. Pressure-test under fire.
Attack the opponents you actually face.
We modeled how player types behave from 6.6 million analyzed decisions. The coach turns that research into exploit lines — plus reads on your measured leaks when you have data.
Explore all features →> ask
How do I beat a calling station?
> coach
Value bet thinner and bigger — they pay off dominated pairs and draws. Cut bluffs unless you hold strong blockers; stations rarely fold. Size for extraction, not balance.
“Finally something that tells me *why* I bled chips — then makes me drill that exact mistake.”
“I stopped opening random solver trees. One leak, one queue, done.”
“The sim personas feel different. I actually notice when I auto-pilot against the station.”
Trainers wander. Solvers are cold. Poka models opponents.
Random spot packs don't fix patterns. Raw solver output doesn't tell you how to exploit a station. Poka connects archetype research to your leak profile.
The exploit models. Both of them.
Two adapter sets trained on 6.6 million analyzed decisions. Not a paper, not a demo — the trained weights, for sale.
Qwen3-4B per-street swarm
Four LoRA adapters — preflop, flop, turn, river — on Qwen3-4B-Instruct-2507. Rank 32, PEFT safetensors, ~126 MiB each. The deployed set that plays the sim table.
79.7% strict exact-match on the full 11,000-row PokerBench test (preflop 89.3 · flop 83.6 · turn 79.8 · river 77.4).
Qwen3-8B combined
One LoRA adapter on Qwen3-8B, rank 32, trained across all streets in a single model. PEFT safetensors (best_em checkpoint).
The clean single-model run: same street coverage, no per-street routing.
Adapters only — bases are the open Qwen3 weights. Delivery and licensing terms on contact.
// contact@pokagto.coach
Find what's costing you.
$79 a month, full access. Get the exploit read, drill the pattern, come back sharper than you left.