AI-Powered Poker Engine
Know the right
move, always
A 15.6M-parameter neural network backed by over a million solver-verified positions. Instant policy, depth-limited search, three streets deep — all running locally in your browser.
Tools
Four ways to
sharpen your edge
Engine
Built on real
poker science
GPU-accelerated CFR+ with a 15.6M-parameter neural network — 1,000,000+ spots pre-solved to 0.13% Nash distance, with depth-limited search for everything else.
Vectorized CFR+ / DCFR
GPU-accelerated counterfactual regret minimization — the same algorithm behind Libratus and Pluribus. 1,000,000+ spots solved to 0.13% Nash distance with multi-stream CUDA batching.
1,000,000+ Pre-Solved Spots
Flop, turn, and river solutions across dozens of scenarios and stack depths. Compressed to 5-15 KB per spot, served from CDN, and cached locally on your device for instant lookup.
Neural Network Policy
15.6M-parameter ensemble model trained on 303M solver rows. Four roles: standalone policy (<5 ms), solver warm-start, leaf node evaluator, and uncertainty router — all running in-browser via WASM.
Multi-Street Game Trees
Full flop-turn-river decision trees with card clustering (8 turn, 6 river clusters). NN-evaluated leaf nodes enable depth-limited search across all three streets with accurate EVs.
Preflop Solver
Heads-up CFR+ across 384 spots (24 matchups, 16 stack depths) plus GPU-accelerated 3-way preflop for 6-max and 9-max positions — 199 spots with mixed-precision CUDA.
Range Morphology
Quantifies polarization, nut advantage, range advantage, and detects capped ranges in real time. Blocker-aware evaluation across all 1,326 combos.
Exploitation Engine
Bayesian 5-axis imbalance tracker — betting volume, equity management, polarity, elasticity, board coverage. Finds leaks your opponents don't know they have.
ICM + Kelly
Malmuth-Harville equity, bubble factor, and fractional Kelly with risk-of-ruin analysis. The math that separates final table min-cashers from winners.
Why PO Solve
Desktop solver quality,
browser speed
PioSolver charges $250. GTO Wizard charges $50/month. PO Solve pairs a GPU-accelerated CFR+ solver with a 15.6M-parameter neural network — 1,000,000+ spots at 0.13% Nash distance, depth-limited search for everything else, all in your browser.
1,000,000+ solved spots
Flop, turn, and river solutions across dozens of scenarios and stack depths — pre-computed offline with GPU-accelerated CFR+. Compressed to 5-15 KB per spot, served from CDN, and cached locally on your device.
Multi-street game trees
Full flop-turn-river decision trees with card clustering (8 turn, 6 river clusters). Neural network evaluates leaf nodes for depth-limited search — every bet size, check-raise, and river shove mapped with accurate EVs.
0.13% Nash distance
GPU-accelerated CFR+ with DCFR discounting — the same algorithm that powered Libratus and Pluribus. Full traversal with zero-variance convergence, verified against known equilibria and calibrated across 24 gold-standard spots.
Exploitation tracking
A 5-axis Bayesian imbalance tracker monitors opponent tendencies in real time — betting volume, equity management, polarity, elasticity, and board coverage. It finds the leaks your opponents don't know they have.
Blocker-aware range analysis
Full 1,326-combo range morphology: polarization, nut advantage, and capped-range detection. Card removal handled correctly — if you hold the A♥, the solver knows villain can't.
No cloud, no subscription wall
The entire solver runs in your browser via WASM. Pre-solved strategies are cached in IndexedDB — once fetched, they work offline. No data leaves your device, no per-hand charges.
Every hand is a decision.
Make yours backed by 1,000,000+ solved spots and a neural network that never stops learning.
Ready?
Your edge starts now
Neural network policy in under 5 ms. 1,000,000+ pre-solved spots on demand. Works on any device — everything runs locally, no cloud dependency.