New endpoints, parameters, and fixes. RSS remains a single English feed. · RSS (EN)
Last updated · Maintained by Pokerai API · Source OpenAPI snapshot
Cache metric endpoints — available in production since 2026-11-18. Cache EV/EQ endpoints are available in production since 2026-11-18. POST /v1/gto/postflop/cache/ev and /eq accept only a returned action token and concrete hole_cards. On /query, include_ev:true or include_eq:true returns 400 invalid_request before Auth validation, consumption or Native dispatch; omission/false remains ordinary query behavior. Each admitted independent EV/EQ request consumes 1 existing cache unit, including hit/join, and every admitted success carries that caller's cache_quota. Pre-admission inapplicable/unsupported missing consumes 0; after confirmation, failure, timeout or cancellation can still be charged. 503 cache_charge_unknown carries no quota and is not automatically retried; exhaustion returns 402 cache_quota_exceeded. The retained local-isolated capture is deployed:false, not a production three-street success or real-billing capture.
OpenAPI: POST /v1/gto/postflop/cache/ev · OpenAPI: POST /v1/gto/postflop/cache/eq · API reference · DocsCache EV/EQ endpoint migration is source-verified and pending deployment. Current production still uses the historical include_ev/include_eq query flags. In the future Relay release, true flags will return 400 invalid_request before validation, consumption or Native dispatch; omission/false remains an ordinary query. Use POST /v1/gto/postflop/cache/ev or /eq with only a returned action token and concrete hole_cards. Each admitted independent EV/EQ request consumes 1 existing cache unit, including hit/join; every admitted successful response carries that caller’s cache_quota. Pre-admission inapplicable or unsupported missing consumes 0; 503 cache_charge_unknown carries no quota and is not automatically retried. EV is raw_net_bb; EQ is root_range_joint, not a posterior. The verified 16-request, single-UID, 200NL 3BET replay is local-isolated/Auth-stub evidence, not production or customer charging.
Optional postflop cache EQ is live. Optional cache EQ is available in production since 2026-11-17. Display strategy first, then send include_eq=true with a returned action token and concrete hole_cards; it is mutually exclusive with EV. The ratio uses the effective cached Flop root opponent range, not a range conditioned on later actions. Node EQ reads are free; root and each Turn/River selection retain their existing charges. The original complete three-street examples remain local-isolated, not production captures.
Builder production success requests, responses and timings are now published; this is documentation of the existing release, not a new SDK or API release. Production examples: Docs · JSON · Python · JavaScript
Postflop cache EV/EQ budget is 15 seconds. Native single EV/EQ calculations have a 15-second hard cap and a 14,900ms soft budget. Service reads remain capped at 15 seconds and Relay requests at 20 seconds. Actual work can end sooner for the remaining deadline and cleanup reserve; these are implementation budgets, not a universal SLA. Linux processes remain limited to 2 GiB. Existing root/deal charges, free metric-node reads, fixed token TTL, mutually exclusive flags, and EV/EQ semantics are unchanged.
OpenAPI: POST /v1/gto/postflop/cache/ev · OpenAPI: POST /v1/gto/postflop/cache/eq · OpenAPI: POST /v1/gto/postflop/cache/query · API reference · DocsOptional postflop cache EV is live. Optional cache EV is available in production since 2026-11-16. Query and display strategy first, then request include_ev=true with a returned action token and concrete hole_cards. The EV request waits within its budget. Existing root and card selection charges apply; token continuation does not consume cache quota. The examples remain local-isolated, not production captures.
OpenAPI: POST /v1/gto/postflop/cache/query · API reference · DocsPersistent postflop cache query is live. Available in production since 2026-11-12. POST /v1/gto/postflop/cache/query queries preset or complete custom persistent caches. Follow returned token values through actions and use turn_card / river_card to cross streets. Only active API Builder/Pro subscriptions are eligible. An independent subscription-period quota applies: a root, each Turn, and each River selection each cost 1; ordinary reads with a valid token are free. Cache coverage may be incomplete.
Free preset-cache coverage is live. Available in production since 2026-11-13: GET /v1/gto/postflop/cache/coverage discovers published preset roots. Any valid API key, including Starter and exhausted accounts, can browse for free: no presolved, solve or cache quota is consumed. Strategy query still requires API Builder/Pro and follows the existing charges. No SDK or MCP package release is claimed. Every page requires board and a pack selection: preflop_pack, or its original flop_version input alias. Preserve the original board spelling and pack filters. Continue through empty items until next_cursor=null; limit defaults to 50 (1–200). The opaque cursor is not a strategy token; as_of is a page observation time, not a cross-page snapshot. Coverage does not guarantee that every listed cache can be read.
preflop_pack field migration is live. The public standard input is preflop_pack; an invalid explicit value never falls back to a legacy selector or default. Field migration available in production since 2026-11-13. Optional preflop pack for the chart. Omit both selectors for the platform default (6max). When explicitly present, preflop_pack must be a non-empty string from this pack catalog, without trimming or case conversion. A valid effective legacy alias may accompany it only with the same value; conflicting selectors fail with 400 before consumption. Legacy-only requests retain their original omission, null/empty, coercion, errors, defaults and fallback behavior; these rules do not relax the strict cache or coverage inputs. With preflop_pack=A, the chart uses A and an enabled guarantee also uses A. Legacy-only preflop_version=A and flop_version=B remain independent (each defaults to 6max): preflop_version=A alone with the flag true gives chart A / guarantee 6max, whereas preflop_pack=A gives A / A. Mixed inputs compare only effective synonymous selectors. When the flag is off, legacy flop_version is ignored even if invalid. The guarantee response publishes preflop_pack only; manifest_version is unchanged. Use preflop_pack inside request for type preflop or flop; the original type-specific legacy input alias remains accepted. requested_preflop_pack is the requested pack, resolved_preflop_pack is the actual hit (including fallback). Solver has no pack selector or echo. Old output names are not duplicated; old-input compatibility does not make old-output readers compatible. Input aliases only: preflop/preflop-range accept deprecated preflop_version; Flop tree/projected-range and cache preset accept deprecated flop_version; availability uses the alias of request.type inside request. Only those original entry points accept them. Solver, custom cache roots, node requests and App-specific /v1/apps/solver/ranges gain no pack override. Pure legacy server selectors preserve null/empty omission and coercion; strict cache/coverage parsing remains strict. Canonical preflop_pack errors use 400 invalid_request; legacy unsupported_preflop_version/unsupported_flop_version errors remain unchanged. The pack catalog still returns versions/id/default, and APIs without a pack echo gain none. Canonical output uses preflop_pack only; availability uses requested_preflop_pack and resolved_preflop_pack. Old input compatibility does not guarantee compatibility for readers of old output fields. URLs are unchanged.
Optional EV source has passed source acceptance but is not deployed. To show strategy first, query normally, then request include_ev=true with an action token and hole_cards; this request waits for calculation within its budget. Use response token, not the display node path. EV is raw_net_bb in bb from the quantized cached fixed strategy, not solver live EV or EQ times pot. First release supports proven preset non-ARP configurations; custom and unsupported configurations return missing. Values follow strategy order; zero and negative values are valid, missing is never zero-filled, and any missing action makes combo missing. status is ok, partial, or missing. Non-Hero action returns the existing view and empty action_values with missing/not_applicable. EV uses one slot per owner, busy returns 200 strategy plus missing without queue; normal reader busy is 429, request timeout 504, identity drift 409. The 10-second and 2-GiB implementation budgets are not a universal SLA. Omission or false retains old wire, billing and fixed token TTL.
OpenAPI: POST /v1/gto/postflop/cache/query · OpenAPI: GET /v1/gto/postflop/cache/coverage · OpenAPI: POST /v1/gto/preflop · OpenAPI: POST /v1/gto/preflop/range · OpenAPI: POST /v1/gto/flop/tree · OpenAPI: POST /v1/gto/flop/projected-range · OpenAPI: POST /v1/gto/availability · API reference · DocsPOST /v1/gto/postflop/cache/query queries preset or complete custom persistent caches. Follow returned token values through actions and use turn_card / river_card to cross streets. Only active API Builder/Pro subscriptions are eligible. An independent subscription-period quota applies: a root, each Turn, and each River selection each cost 1; ordinary reads with a valid token are free. Cache coverage may be incomplete. Online availability depends on deployment. Subsequently available in production since 2026-11-12. OpenAPI: POST /v1/gto/postflop/cache/query · API reference · DocsPOST /v1/gto/availability is an authenticated, free preflight for a matching preflop, flop, or solver request. It returns a point-in-time availability snapshot without consuming quota; it does not reserve capacity or guarantee a later cache load or result request, which continues under its existing billing rules. OpenAPI: POST /v1/gto/availability · API reference · DocsPOST /v1/gto/flop/tree now returns ranges.requested, ranges.effective, ranges.adjusted, adjustment_summary, range_provenance, and optional hero_hand_coverage. Top-level ranges are the effective aliases used by the tree; requested ranges are audit-only. A removed hand must not continue to /flop/node, which returns typed HTTP 422 hand_not_in_effective_range. POST /v1/gto/preflop/range can return an optional offline immutable pruning guarantee; missing or incomplete coverage fails closed. OpenAPI: POST /v1/gto/flop/tree · OpenAPI: POST /v1/gto/preflop/range · API reference · Docs/v1/gto/solver now accepts independent raise_sizes, donk_sizes, and raise_limit fields in addition to bet_sizes. OpenAPI: POST /v1/gto/solver · API reference · DocsPOST /v1/gto/solver/release — release a solve's pool port early so it returns to the pool immediately instead of waiting out the cache TTL (optional, free; call after your last /tree / /node query for that solve). OpenAPI: POST /v1/gto/solver/release · API reference · DocsPOST /v1/gto/evs — per-hand, per-action expected values at a node of a completed solve (free). OpenAPI: POST /v1/gto/evs · API reference · Docs/v1/gto/turn/projected-range now supports partner_hands (block partners' dead cards from the villain range). OpenAPI: POST /v1/gto/turn/projected-range · API reference · Docs/v1/gto/flop/projected-range now honors bluff_discount_ratio / bluff_combos_ratio from the request, injects Hero's own hand into Hero's range, and supports partner_hands. OpenAPI: POST /v1/gto/flop/projected-range · API reference · Docs/v1/pokerkit/*): board/hand semantics, range equity, eval/ICM, game simulation — same key & quota. OpenAPI: /v1/pokerkit/* · API reference · DocsPOST /v1/gto/preflop/range — the whole 13×13 preflop range in one call. OpenAPI: POST /v1/gto/preflop/range · API reference · Docs/v1/gto/flop/tree (fetch tree) + /v1/gto/flop/node (fetch node, free), aligned with the solver's tree/node model. OpenAPI: POST /v1/gto/flop/tree · OpenAPI: POST /v1/gto/flop/node · API reference · Docs/v1/gto/flop/projected-range + /v1/gto/turn/projected-range, the OpenAPI 3.0 spec, real-time flop solving, and a unified board/hand format. OpenAPI: POST /v1/gto/flop/projected-range · OpenAPI: POST /v1/gto/turn/projected-range · API reference · Docs