GrindSense is a play-money Texas Hold'em trainer. You play against AI opponents that explain their thinking and adapt to exploit your habits — so you find and fix your leaks before real opponents do. No real money, no downloads.
New to poker? Start with how a hand works.
In Texas Hold'em each player gets two cards of their own, and five shared cards are dealt face up in the middle. You make the best five-card hand you can from any of those seven. You win the chips in the middle, the pot, by having the best hand at the end, or by betting so that everyone else gives up first.
A round of betting ends once everyone still in has had a turn and has put in the same amount, or all their chips. Then the button moves one seat to the left and the next hand begins.
At the table only the moves allowed right now are shown, so you cannot make an illegal one.
From best to worst. Suits are never ranked against each other.
When two players have the same kind of hand, the higher cards win: a pair of kings beats a pair of sevens. If those are equal too, the next highest card decides, the kicker. Only five cards ever count.
Where you sit relative to the button is your position. Going clockwise from the button at a six-player table:
Acting later is an advantage: you have seen what everyone before you did. That is why players in late seats can play more hands. Any other word you meet is in the Glossary.
Create a free account — by invitation while GrindSense is in its private test — then pick how you want to play:
Nine archetypes, and they are not hand-tuned guesses. Each one's VPIP (how often it voluntarily puts money in before the flop) and aggression are calibrated against 2,351 real players from a large hand-history database, so each sits at an estimated percentile of that sample of real 6-max play rather than at a number someone thought felt about right. The figures below are for a six-handed table; at a full nine-seat table every archetype plays about a quarter fewer hands, so the order holds and the numbers shrink.
| Archetype | Hands played (6-max) | Aggression |
|---|---|---|
| Tight-Passive | 17% | Low |
| Tight-Aggressive | 20% | High |
| Shark | 25% | High |
| Q-Learner (RL) | 25% | Medium |
| Neural Net (RL) | 26% | High |
| Loose-Aggressive | 35% | High |
| Loose-Passive | 48% | Low |
| Maniac | 66% | Extreme |
| Adaptive | 24% base | High |
Tap any opponent at the table to see how to play against them.
Which of them actually learned to play? Two. The Q-Learner plays from a table of values it learned by playing the other bots at cash tables, and it is the seat you are facing when the badge says Q. The Neural Net is a policy network — first copied from the house engine, then trained the same way. Both keep to the range their width promises: a hand outside it is checked or folded, and the explanation says so. Inside it, how to play the hand — call, raise, how much, and everything after the flop — is the model's own. Every other archetype is an expert-authored engine: explicit ranges and betting rules, tuned against a database of real hands. The Adaptive seat is one of those too — it reads your statistics and attacks your largest deviation, which is a strategy we wrote, not a model that trained.
Each learned seat is dealt only while it passes two tests: it has to win enough against the other bots, and it has to play the style its name claims. One that fails either, or that cannot be loaded, is not seated. A Tight-Aggressive bot takes that chair, and it tells you so — the RL badge disappears, the seat introduces itself as TAG, and the bot styles board marks the style as benched. A badge never implies a neural net you are not playing against, and neither does a name.
When it's your turn, your options (fold / check / call / bet / raise) appear — only the legal ones. A few things that speed you up:
In Practice, every AI decision comes with a plain-language explanation of why — the hand strength, pot odds, position, and read it's acting on. Crucially, the reasoning is honest: when a bot is bluffing, it says so. Read these to understand how strong players think through a spot. At Arena tables the bots keep their reasoning to themselves, as opponents do, with one exception: an Adaptive seat that has a read on you tells you which of your leaks it is attacking, in a direct message only you can see (the DM tab beside the chat). The rest of the coaching there comes from your own hands, in the review and the leak reports.
Coaching is woven through the app, not buried in a menu:
An opening range is the set of hands you raise with when nobody has entered the pot yet. It depends on your seat, because the earlier you act the more players are still to speak behind you, and the more of the hand you will play out of position. So the range starts tight under the gun and widens seat by seat towards the button.
This is the chart GrindSense teaches and grades against, as a share of all 1,326 two-card combinations (a pair is 6 of them, a suited hand 4, an offsuit hand 12):
Read the 13×13 grid on the Ranges page like this: pairs run down the diagonal, suited hands sit above it, offsuit hands below. The preflop drill under Drills marks you against exactly this chart, so what you study and what you are graded on are one thing.
Focused reps on one decision at a time, graded against a reference answer:
The Ranges page shows the same 13×13 opening chart the preflop drill grades you against, so the reference you study and the answer key you are marked against are never two different things.
Every hand commits a 256-bit seed before a single card exists, and the deck is shuffled from that seed with SHAKE256 — the extendable-output function from the SHA-3 family — using Fisher-Yates and rejection sampling. The shuffle is designed to avoid systematic bias toward any card or seat, and it is subjected to the statistical testing described below.
The shuffle is checked with the batteries used to test random number generators generally: dieharder's Diehard suite, NIST SP 800-22, and TestU01. Not only the underlying random source, but the dealt permutations themselves — a shuffle can be broken while its random source is perfect, so the cards are tested as cards.
This is not a gambling certification and does not claim to be. GrindSense is play-money and no regulator is involved. The testing exists so the question has an answer.
Your profile tracks lifetime stats and chip balance. Forgot your password? Use the reset link on the sign-in page. New to a term? The in-app Glossary defines it. You can delete your account from your profile at any time — see the privacy policy for what that removes and when.
Ready to find your leaks?
Sign in / Create accountPlay-money only — not a real-money gambling site.