cusPokerTools: The GTO Trees Build
What we have today
cusPokerTools (cuspokertools.netlify.app) is a live, plain-English poker equity calculator. You type a spot the way you'd say it at the table ("KK78 double-suited vs any set or big wrap on 5h4d2c") and it computes exact multi-way equity — NL Hold'em, PLO, and double-board bomb pots. The math engine has been verified against independent evaluators across 24+ test scenarios to within 0.25 points. No other free tool takes plain English.
What we're adding
GTO answers: "what should I DO here" — bet/check/raise frequencies and EVs, not just equity. Two layers:
- A precomputed library. 184 strategically distinct flops × 6 common preflop configurations ≈ 1,100 solved spots (expandable to ~5,000 with turn subtrees). Covers the most common NLHE postflop situations.
- Live solve-on-request. If a user asks a spot NOT in the library, we solve it on demand through the DeepSolver cloud API and cache the result forever. Second person asking the same spot gets it instantly, free.
How fast is a solve?
| Scenario | Time |
|---|---|
| One spot via DeepSolver API (their published latency; GPU cloud) | 2 to 20 seconds |
| A complex live spot we've never seen, end to end including polling — then cached permanently | ~5 to 25 seconds |
| The entire 1,100-spot library via API (they queue 10,000+ jobs) | One afternoon |
| Same library self-solved on a rented 16-core server | 3 to 5 days continuous |
| Our current server | Physically cannot |
Not minutes, not hours: a live solve lands in seconds. The only thing our existing box can't do is the heavy lifting itself — measured, the solver needs more RAM than the box has.
The cost, exactly
| Item | Cost | What it buys |
|---|---|---|
| DeepSolver on-demand pack | $250 one-time | 10,000 solves at $0.025 each. The full 1,100-spot library costs $27.50 of it; the other ~8,900 solves power live user queries for months. |
| Hetzner server month (optional, recommended) | ~$160 one-time | Self-solve the same library with an open-source solver as an independent cross-check. Provable accuracy, and we own those outputs outright with zero vendor dependency. |
| MonkerSolver license + big-RAM rental (later, PLO) | ~$660 one-time | The only realistic path to PLO trees. No free PLO solver exists anywhere. Optional phase 2. |
Split two ways: ~$205 each recommended, ~$550 each full.
For scale: GTOWizard costs $26-206/month forever and has no API. PioSolver's business tier is €5,000/month. We get our own asset for a one-time few hundred dollars.
Why this is legitimate
GTO solutions are mathematics — facts are not copyrightable. No scraping, no reverse engineering. DeepSolver's terms state verbatim:
"Can I resell or white-label the API? Yes, no restrictions."
The only bar: no real-time assistance at live tables, which we don't do. Every solve result gets archived to our own storage, so the library is ours permanently regardless of any vendor.
Honest risks
- Vendor size. DeepSolver is a small Polish company (~$265K revenue 2023, 6 years old, founder-run). If they vanish, only the live-solve feature pauses; the cached library survives.
- Accuracy is estimated, not proven. Their neural-net solutions land an estimated 0.3-0.7% of pot from Nash, not mathematically proven. The $160 self-solve cross-check covers this.
- Scope today. API is heads-up NLHE postflop only: no multiway, no preflop trees, no ICM. PLO exists in their stack but needs a commercial conversation.