PulseGTO
Getting started
What is PulseGTO?
PulseGTO is a solver for No-Limit Hold'em. It computes game-theory strategies: how often each hand should fold, call, bet or raise at every decision, and what each choice is worth in big blinds.
It has three parts:
- Blueprints. A blueprint is a whole game for 2 to 6 players, solved at once from the first preflop decision to the river. It holds the opening ranges, every 3-bet and 4-bet line, and an initial postflop strategy behind each line. Blueprint files have the extension
.pgb. - The desktop Engine. A Windows application that trains blueprints for your stacks, rake and bet sizes, and re-solves individual postflop spots on your own computer.
- pulsegto.com. The website, where you can browse the published strategy database without installing anything.
Blueprint, re-solve and database: the distinction to learn first
PulseGTO builds its strategies in two stages.
- Stage 1: the blueprint. The preflop engine trains opening ranges, later preflop lines and postflop play together. Postflop uses an adjustable abstraction level so a whole-game strategy can run on a desktop. Its postflop answers are approximate; the blueprint is a starting point for studying individual spots.
- Stage 2: re-solves. Postflop spots from the blueprint are solved again with RangeCFR, starting from the ranges with which the blueprint reaches the spot. The re-solve has no hand abstraction: every hand combination gets its own strategy. The result is saved into the
.pgbas an override for that node and board; the trained strategy stays in the file alongside it.
The strategy database you browse is the blueprint plus the re-solves saved into it. It improves gradually: spots that have been re-solved show the re-solve, all others still show the blueprint. PulseGTO re-solves spots on its own servers and updates the published database that Cloud users browse; Engine users can do the same for their own files. See Recalculating individual nodes for how to tell where a strategy comes from.
| Blueprint (training) | Node re-solve (RangeCFR) | |
|---|---|---|
| Scope | The whole game, preflop to river | One postflop node and the tree below it |
| Postflop hands | Abstraction-based strategy | Every combo separate (no hand abstraction) |
| Players | 2–6 | 2 or 3 |
| Accuracy is measured as | Convergence inside the abstraction | Exploitability from a best-response pass on that node's tree |
| Still limited by | Abstraction, iterations, the action tree | Iterations, the action tree, the input ranges, and available RAM; reduced turn coverage only when selected or necessary |
A re-solve improves one spot. It does not turn the whole database into a preflop-to-river equilibrium, and "no hand abstraction" does not mean "exact equilibrium": the re-solve still converges over a finite number of iterations inside the betting tree you chose.
Payoffs are chip EV in big blinds, with the rake you configure. There is no ICM model.