Built for the moments your tracker can't reach.
Live cash. Mixed games. Private home games. Casino lineups you've seen twice in three months. Trackers don't help here. PokerReads does — opponent recall written like notes a serious player would actually take, table-side prep that fits in 30 seconds, bankroll-level decisions that don't pretend variance is theoretical, and a measured synthesis-quality engine whose numbers are public.
You sit at a 9-handed table at 9 PM. You've played two of these guys before, three are unfamiliar, four have shifted seats since last week.
What did Liam do on the river last time? Does the kid in seat 4 fold or jam to a 3-bet? Your tracker has none of this. Your memory has some of it but not in time for the first orbit.
You just lost a pot you should have won. Your hands are warm. The next decision matters more than the last one did.
You don't need a solver, you need a 60-second protocol. Title, three steps, decision rule. Before any model output streams.
You closed the session up. Or down. Either way you went loose somewhere in the middle.
You can't process it now. You'll forget by tomorrow. By Sunday it's all blurred into "I ran bad" or "I played well" — neither of which is a decision you can act on.
You're 50 buy-ins into a downswing. Your stake says "you're under-rolled." Your gut says "grind through."
Generic risk-of-ruin formulas don't model your actual swings. They don't see the account drift, the cashflow churn, the four months of your real session sigma.
Your study group is in three different chats. Half the reads are in WhatsApp, half are in Telegram, the screenshots got dropped in Discord by the guy who deleted his account.
You re-tell the same hand to four people. The reads decay. Someone misses the lineup tonight because the chat moved on at 2 AM.
"Trust me bro, the AI got it right."
Every other tool says some version of this. None of them prove it. None of them gate prompt changes on poker-correctness regressions. None of them publish their numbers.
What you actually use, day to day.
Every feature below is shipped today. Nothing on this page is "coming soon" or behind a wait list. Sign up, set a starting bankroll, and you're in.
Bankroll Operating Brief
Readiness score 0–100, color-coded. Five live metrics: true bankroll, main-stake cover, ruin risk modeled against your real sigma, account trust score, monthly pace. Three concrete next moves before you sit.
Table Scan + Live-Read Prewarm
Point your phone at the table. Names extracted. Reads pre-warmed in the background while you stack chips. Sub-second tap-to-render on warm tables once you sit down.
60-Second Reset
Tilt-trigger words detected on-device. Title, three steps, decision rule — surfaced before any AI output streams. Steady, urgent, and protect tones matched to the moment.
Tap any seat — get the read
Streaming synthesis with cited hands and exploit lines. Fact / inference distinction explicit. Confidence map matched to sample size. Quick Reference for the table-side glance.
Last Session Command Review
One screen, five bullets: what went well, what bled, one play to study, mood pattern, bankroll move. Push notification when the session ends so the lesson lands before sleep.
Bankroll Weekly Review
Sunday-night recap. Net result, four-week trend, biggest swing, biggest leak, best game, account drift, recommended next-week plan. AI coach can pre-load the summary as context.
Pattern Map + Edge Map
Deterministic patterns: venue concentration, drift, stale check-ins, unassigned ledger rows, cashflow churn, expense drag, reconciliation cadence. Where your bankroll edge actually lives.
Groups + Scout Links
Private read pool with your study crew. Hands, notes, screenshots, and reads stay scoped to the group. Share-a-Read scout links scrub names before they leave the app.
What it costs.
$19.99/month for everything above. No tiered features. No "starter" trap. No data lock-in (export at any time, CSV included).
Free includes the demo, three players with up to five hands each, and the first read on those players. Enough to know whether the synthesis layer is worth paying for. If it isn't, you owe us nothing.
For context: at $5/$10 NL, $19.99/month is 4% of one buy-in. If PokerReads helps you avoid one bad call per month, the math is done. If it doesn't, cancel.
How to know we're not faking it.
Every read goes through a 75-fixture cross-model evaluation arena before any prompt change touches production. Deterministic graders for poker correctness, a paired GPT-4o judge, a held-out predictive lane that scores how well each read compresses behavior into action prediction, and a self-check retry gate that runs the deterministic graders against the model's output before saving. Internally we call it the Queen of Hearts — only one queen sits at the table at a time.
The numbers are public and updated as we run them. See the live quality scorecard →
Skeptical is the right setting. Try the demo without signing up. If the read on a synthetic opponent doesn't read like a serious player wrote it, the rest of this page didn't matter.