Start with real decisions.
Hand histories provide examples of actions and bet sizes in context. They are the training evidence behind the models; API access does not include raw hand-history downloads.
POPULATION MODELS FOR DEVELOPERS
Real poker behavior.
Built into your app.
Build poker experiences around how real players actually play. Build bots, practice environments, and analysis tools with population-based poker models trained on real hand histories. Bring human tendencies into your product with context-aware action probabilities and conditional bet-sizing distributions.
Try HoldemAPI free for 7 days. Choose a plan and start building.
Explore the demonstrations. Play BlitzFold ↗ · Explore a hand ↗
7-day free trial. Card required. One trial per eligible account.
After the trial, your selected plan automatically renews at $30 USD/month for Starter or $90 USD/month for Pro until canceled. Cancel before the trial ends to avoid the first charge.
Creating an account is free. Your trial starts only when you complete Stripe Checkout.
DECISION PREVIEW01 / FLOP
YOUR HAND
2 ♣7 ♦J ♠
Human tendencies. Estimated for this decision.
ACTION PROBABILITIESILLUSTRATIVE
Fold
2%
Call
26%
Raise
72%
Sampled action: Raise. Your engine brings the move to life.
POPULATION BEHAVIOR MODELSHAND-AWARE PROBABILITIESCONDITIONAL BET SIZINGBUILT FOR DEVELOPERS
FROM OBSERVED PLAY TO YOUR PRODUCT
One foundation.
Many poker experiences.
Access a reusable model of population behavior: the tendencies, sizing patterns, and imperfect decisions that make human poker different from an optimal strategy.
Estimate how players behave.
Models estimate action frequencies and conditional sizing tendencies from the board, betting history, position, pot, and stacks. Responses are model estimates, rather than direct counts of matching hands in a database.
Build on those estimates.
Use the outputs in your own game, simulation, or analysis interface. Your application supplies game rules, manages state, and turns predictions into an experience.
WHAT YOU CAN BUILD
Human tendencies.
Your application.
These are applications you can develop using the model outputs.
Bots and simulated opponents
Drive opponents with estimated population decisions and sizing patterns. Give simulations behavior grounded in observed play.
Practice games and training experiences
Build repeatable situations against modeled opponents. Add your own lessons, feedback, and progression.
Tendency explorers and hand review
Show how predicted action mixes change across situations. Put recorded decisions beside population estimates for context.
Strategy experiments
Test hypotheses against modeled populations, including exploitative strategy research. You supply the experiment and evaluation; the model does not promise optimal strategy or profitable play.
HOW WE APPROACH REALISM
Learned tendencies.
Estimated hand behavior.
Population first, then the hand.
The public model estimates the overall action mix for a situation. The equity gate assigns hand-dependent probabilities while numerically targeting those learned aggregate frequencies across the weighted candidate hands. Postflop allocation also uses estimated improvement potential.
A constraint, with limits.
Matching the aggregate mix does not prove hidden-card ranges are correct or guarantee realism throughout a simulated hand. Range assumptions, missing private cards, and model coverage matter. Your integration carries actor ranges forward after the API’s sampled action.
Read the methodology and assumptions ↗POPULATION ESTIMATES IN YOUR WORKFLOW
A clear contract.
A reusable foundation.
The actions endpoint combines learned game-context decisions with hand strength and draw potential. Bets and raises can represent value, bluffs, or semi-bluffs.
POST
/api/actions- Exact hole cards and optional actor-range weights
- A sampled action, probabilities, and conditional sizing distributions
- Optional style presets with a compatible configured model
JAVASCRIPT & TYPESCRIPT SDK
Our server-side SDK is being updated for the new API. The existing SDK requires Node.js 20+; public installation instructions and actions support are coming with the release.
SDK availability & API documentation ↗actions.response.json
// Illustrative response · facing a bet
{
"ok": true,
"hand": "AsAh",
"equity": 0.84,
"chosenAction": "raise",
"output": {
"fold": 0.02,
"call": 0.26,
"raise": 0.72
}
}Example values only. Full responses also include population targets and sizing weights.
CURRENT COVERAGE
Know the population
behind the model.
Available foundation
The current released action models were trained on 50,000 hands from PokerStars USD $0.10/$0.25 regular no-limit Hold’em cash games, using a five/six-player cohort. The research archive contains over six million hands; that archive size is not the current model’s training count.
Modeled player styles
Optional tight-passive, tight-aggressive, loose-passive, and loose-aggressive presets condition the action and sizing model on historical style features. They require a compatible configured style model. These are descriptive behavior models, not validated personalities for every situation.
Expansion takes new evidence
Accepting a game state or hand history from another stake does not select a separately trained stake model. Broader populations, formats, and our own analysis products are future directions, requiring further training and validation.
BUILT ON THE SAME FOUNDATION
Explore the demonstrations.
BlitzFold
A playable poker experience using HoldemAPI. See one way population models can become opponents in a game.
Play BlitzFold ↗Hand analyzer
Paste or upload a PokerStars USD cash hand, replay its decisions, and explore public-state population action predictions and conditional sizing. Hole cards are shown for reference; they do not condition these predictions.
Explore a hand ↗A PLAN FOR YOUR NEXT BUILD
Start small.
Build your way up.
Choose your monthly allowance. Create an account, pick your plan, and manage your API keys in one place.
Starter
For prototypes, personal projects, and your first integration.
$30USD / month
1,000 requests per month
- Population modeling via Actions API
- Secure, revocable API keys
- Usage and billing dashboard
- Shared allowance across your keys
7-day free trial. Card required. One trial per eligible account.
After the trial, your selected plan automatically renews at $30 USD/month until canceled. Cancel before the trial ends to avoid the first charge.
Choose your plan after signup. Cancel renewal anytime.
Pro
10× the requestsMore room for active development and growing applications.
$90USD / month
10,000 requests per month
- Population modeling via Actions API
- Secure, revocable API keys
- Usage and billing dashboard
- Shared allowance across your keys
7-day free trial. Card required. One trial per eligible account.
After the trial, your selected plan automatically renews at $90 USD/month until canceled. Cancel before the trial ends to avoid the first charge.
Choose your plan after signup. Cancel renewal anytime.
Monthly allowances renew with your billing period. No automatic overage charges.
READY FOR YOUR NEXT MOVE
Build around real player behavior.
Create an account, choose a plan, and start building with population models.
Building something with poker? Let’s talk about it ↗