What is Pokerai API?Pokerai API 是什么?
The first self-serve, solver-grade GTO strategy API for 6-max No-Limit Hold'em. It returns the mixed-frequency strategy for every street — preflop and flop from millions of presolved solutions (milliseconds), turn and river solved on demand. Every request is stateless and self-contained.首个自助式、solver 级的 6-max 无限德州 GTO 策略 API。返回每一街的混合频率策略——preflop 与 flop 来自数百万预解方案(毫秒级),turn/river 按需实时求解。每个请求无状态、自包含。
What is GTO?什么是 GTO?
GTO (game-theory-optimal) is a strategy that can't be exploited no matter how your opponents adjust. Instead of one “best” move, each spot has a balanced mix of actions at specific frequencies — which is exactly what Pokerai API returns.GTO(博弈论最优)是一种无论对手怎么调整都无法被剥削的策略。它不给单一「最优手」,而是每个局面给一组按特定频率平衡的动作——这正是 Pokerai API 返回的。
What is a solver?什么是求解器(solver)?
Software that computes the GTO strategy for a spot by iterating (counterfactual regret minimization, CFR) until both players’ strategies stop improving — i.e. converge to equilibrium. Solving is compute-heavy, so Pokerai serves millions of presolved spots instantly and solves the rest on demand.通过迭代(反事实遗憾最小化 CFR)计算某局面 GTO 策略的软件,直到双方策略不再改进、收敛到均衡。求解很耗算力,所以 Pokerai 把数百万预解局面即时返回,其余按需求解。
Why frequencies instead of one “best” move?为什么给频率、而不是单一「最优手」?
Because a single fixed action is exploitable. GTO stays balanced by mixing — e.g. 99 facing an open might raise 64%, call 11%, fold 25%. You randomize per those frequencies (a clock, a die, your own scheme) so opponents can't read you.因为固定单一动作会被剥削。GTO 靠混合保持平衡——比如 99 面对开局,可能 raise 64%、call 11%、fold 25%。你按这些频率随机化(看秒表、掷骰子、自定规则),对手就读不出你。
What’s an “isomorphic board”?什么是「同构牌面」(isomorphic board)?
Boards identical up to suit relabeling — A♠K♠2♠ and A♥K♥2♥ play the same. Collapsing these suit-symmetric duplicates to one canonical board is why ~66K isomorphic flop spots cover every real flop, keeping the presolved database small and fast.花色重命名后等价的牌面——A♠K♠2♠ 和 A♥K♥2♥ 打法完全一样。把这些花色对称的重复折叠成一个规范牌面,就是为什么约 66K 个同构 flop 局面能覆盖所有真实 flop,让预解库又小又快。
Which formats and streets does it cover?覆盖哪些格式和街?
6-max NLHE, all four streets — done to solver depth, not a shallow multi-format menu. Preflop across multiple stack depths (100bb + 40bb), flop from 5.8M+ presolved solutions across 66K isomorphic-board spots, turn/river solved on demand. PLO and ICM aren't covered yet.6-max 无限德州,四街全覆盖——做到 solver 深度,不是浅尝辄止的多格式菜单。preflop 多深度(100bb + 40bb)、flop 66K 同构面局面 / 5.8M+ 预解方案、turn/river 按需实时求解。PLO 与 ICM 暂未覆盖。
What stack depths are available?有哪些筹码深度?
Multiple depths (e.g. 100bb, 40bb), selected via the preflop chart version. Call GET /v1/gto/preflop/versions to list them.多种深度(如 100bb、40bb),按 preflop 图表版本选择。GET /v1/gto/preflop/versions 可列出。
How do I authenticate?如何鉴权?
Send an Authorization: Bearer <API_KEY> header. Get a free key at /login — no credit card.带 Authorization: Bearer <API_KEY> 头。在 /login 免费拿 Key,免信用卡。
How does the quota work?配额怎么算?
Two independent monthly counters per key: a presolved counter (preflop/flop lookups) and a solve counter (real-time turn/river). When one is exhausted the API returns HTTP 402 quota_exceeded. See pricing.每个 Key 两个独立的逐月计数:预解(preflop/flop 查询)与求解(实时 turn/river)。某个耗尽时返回 HTTP 402 quota_exceeded。见定价。
Which calls are free?哪些调用免费?
Reading a solved spot's tree/nodes (/solver/tree, /solver/node), node EVs (/evs), a flop node (/flop/node), and reusing a cached solve. /v1/gto/solver only charges when it triggers a new solve.读已解局面的树/节点(/solver/tree、/solver/node)、节点 EV(/evs)、flop 节点(/flop/node),以及复用已缓存的解。/v1/gto/solver 仅在触发新求解时扣费。
Does it return a single recommended action?它返回单一推荐动作吗?
No. It returns the GTO frequency of each action (check/call/bet/raise/fold), so you randomize per your own strategy rather than always taking one move.不。它返回每个动作的 GTO 频率(check/call/bet/raise/fold),你按自己的策略随机化,而非永远取某一手。
Are requests stateless? Do you store my hands?请求无状态吗?会存我的牌吗?
Requests are stateless and self-contained. The API does not store the hands/spots you send — only your account email and per-key usage counts.无状态、自包含。不存你发送的牌局/局面——只保留账户邮箱与每个 Key 的用量计数。
Can I use it for real-time assistance (RTA)?能用于实时辅助(RTA)吗?
No — RTA at real-money tables is prohibited. Pokerai API is for training, coaching, hand review, study, and research. See terms.不能——禁止真钱牌桌实时辅助。Pokerai API 面向陪练、教学、复盘、学习与研究。见条款。
Is there a free tier?有免费层吗?
Yes — 1,000 presolved lookups and 25 real-time solves per month, no credit card. Enough to build a full trainer and run every tutorial.有——每月 1,000 次预解查询 + 25 次实时求解,免信用卡。够搭完整陪练并跑完所有教程。
How fast is it?有多快?
Preflop and flop are served in milliseconds. Turn/river are solved on demand: submit to /v1/gto/solver, poll /v1/gto/solver/tree until queryable, then read — an available → computing → queryable state machine.preflop 与 flop 毫秒级返回。turn/river 按需求解:提交到 /v1/gto/solver,轮询 /v1/gto/solver/tree 到 queryable 再读——available → computing → queryable 状态机。
Is there an OpenAPI spec and interactive reference?有 OpenAPI 与交互式参考吗?
Yes: /openapi.yaml, an interactive reference, and a full LLM context file at /llms-full.txt.有:/openapi.yaml、交互式参考,以及给 LLM 的完整上下文 /llms-full.txt。
Do you have SDKs or an MCP server?有 SDK 或 MCP server 吗?
Yes — official SDKs are published: pokerai-bet (Python, pip install pokerai-bet), @pokerai/client (TypeScript, npm install @pokerai/client), and an MCP server for AI agents (npx @pokerai/mcp). You can also generate a client in any language from the OpenAPI spec.有——官方 SDK 已发布:pokerai-bet(Python,pip install pokerai-bet)、@pokerai/client(TypeScript,npm install @pokerai/client),以及面向 AI 代理的 MCP server(npx @pokerai/mcp)。也可以用 OpenAPI 规范生成任意语言的客户端。