@oh-my-pi/pi-catalog 18.4.5 → 18.4.8

package/CHANGELOG.md

@@ -2,6 +2,17 @@
 
 ## [Unreleased]
 
+## [18.4.6] - 2026-10-01
+
+### Added
+
+- Added configurable catalog cache-warming lifetimes for cache-capable Claude models on Amazon Bedrock Converse, Bedrock Runtime, and Mantle, with 5-minute defaults and a 1-hour Converse option where supported.
+
+### Fixed
+
+- Fixed forced tool calls for Claude Opus 5.5 and Sonnet 5.5 on Amazon Bedrock Converse.
+- Fixed the thinking-off setting for Command Code models served through the Responses API so they no longer produce reasoning when disabled.
+
 ## [18.4.5] - 2026-09-30
 
 ### Added

package/dist/types/types.d.ts

@@ -651,6 +651,12 @@ export interface BedrockCompat {
      * `PI_OPENAI_STREAM_IDLE_TIMEOUT_MS` alias, then the 300s default.
      */
     streamIdleTimeoutMs?: number;
+    /**
+     * Whether the model accepts a forced `toolChoice` (`any` / `tool`). Claude
+     * Opus/Sonnet 5.5 reject it outright; the request builder downgrades forced
+     * choices to `auto` when this is false. Default: true.
+     */
+    supportsForcedToolChoice?: boolean;
 }
 /** Fully-resolved Bedrock Converse prompt-cache capabilities, materialized once by `buildModel`. */
 export interface ResolvedBedrockCompat {
@@ -660,6 +666,7 @@ export interface ResolvedBedrockCompat {
     supportsLongPromptCacheRetention: boolean;
     promptCacheMinimumTokens: number;
     promptCacheMaximumCheckpoints: number;
+    supportsForcedToolChoice: boolean;
     /**
      * Stream-watchdog idle-timeout fallback in ms for hosts with no keepalive
      * events; 0 disables the idle watchdog. Undefined defers to
@@ -1194,6 +1201,12 @@ export interface Model<TApi extends Api = Api> {
      * Custom models and provider overrides opt in via models.yml `promptCache`.
      */
     promptCache?: ModelPromptCache;
+    /**
+     * Verbatim configured lifetimes (models.yml, `modelOverrides`, runtime
+     * registrations). `buildModel` applies them over catalog `prompt-cache`
+     * rules on every rebuild; `{}` keeps warming disabled.
+     */
+    promptCacheConfig?: ModelPromptCache;
     /**
      * Interpretation of an all-zero token-rate card. Omitted zero-rate cards
      * are unknown; any non-zero rate is always treated as fixed pricing.

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@oh-my-pi/pi-catalog",
-	"version": "18.4.5",
+	"version": "18.4.8",
 	"description": "Model catalog for omp: bundled model database, provider discovery descriptors, model identity, classification, and equivalence",
 	"keywords": [
 		"ai",
@@ -93,12 +93,12 @@
 		"gen:proto": "bun scripts/generate-protocols.ts"
 	},
 	"dependencies": {
-		"@oh-my-pi/omptype": "18.4.5",
-		"@oh-my-pi/pi-utils": "18.4.5"
+		"@oh-my-pi/omptype": "18.4.8",
+		"@oh-my-pi/pi-utils": "18.4.8"
 	},
 	"devDependencies": {
 		"@bgotink/kdl": "0.4.0",
-		"@oh-my-pi/pi-ai": "18.4.5",
+		"@oh-my-pi/pi-ai": "18.4.8",
 		"@types/bun": "^1.3.14"
 	},
 	"engines": {

package/src/build.ts

@@ -363,5 +363,7 @@ export function buildModel<TApi extends Api>(spec: ModelSpec<TApi>): Model<TApi>
 	};
 	applyCatalogAssignments(model, policy.catalog);
 	applyCatalogCorrections(model, policy.catalog);
+	// Configured lifetimes replace catalog lifetimes rather than merging with them.
+	if (spec.promptCacheConfig !== undefined) model.promptCache = { ...spec.promptCacheConfig };
 	return model;
 }

package/src/compat/axes.ts

@@ -292,7 +292,7 @@ export const AXES: Readonly<Record<string, AxisDef>> = {
 	"stream-first-event-timeout-ms": wire("streamFirstEventTimeoutMs", [...OAI, "google"]),
 	"stream-idle-timeout-ms": wire("streamIdleTimeoutMs", [...OAI, "anthropic", "bedrock", "google"]),
 	"strip-image-input": wire("stripImageInput", [...OAI, "anthropic", "google"]),
-	"supports-forced-tool-choice": wire("supportsForcedToolChoice", [...OAI, "anthropic"]),
+	"supports-forced-tool-choice": wire("supportsForcedToolChoice", [...OAI, "anthropic", "bedrock"]),
 	"supports-sampling-params": wire("supportsSamplingParams", [...OAI, "anthropic"]),
 	"thinking-loop-guard": wire("thinkingLoopGuard", [...OAI, "anthropic", "google"], "scalar", [
 		"gemini",

package/src/compat/resolve.ts

@@ -935,6 +935,7 @@ function resolveBedrockPolicy(spec: ModelSpec<"bedrock-converse-stream">, axes:
 		supportsLongPromptCacheRetention: false,
 		promptCacheMinimumTokens: 0,
 		promptCacheMaximumCheckpoints: 0,
+		supportsForcedToolChoice: true,
 	};
 	// Reasoning capability is a mechanism gate; adaptive-lineage duration is rule-owned.
 	compat.streamIdleTimeoutMs = compatReasoning(spec, axes) ? BEDROCK_REASONING_STREAM_IDLE_TIMEOUT_MS : undefined;

package/src/compat/rules/README.md

@@ -183,8 +183,8 @@ provider "openrouter" {
 | ---------- | ---------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
 | `class`    | `class "id" { ... }`                     | Exact class ID. At document root it may contain `on`, `on-api`, `family`, `revision`, and `models`. Under `provider` it may contain `family`, `revision`, and `models`.        |
 | `provider` | `provider "id" { ... }`                  | Exact provider ID. It is root-only and may contain `class`, `on-api`, and `models`.                                                                                           |
-| `on`       | `on "provider-a" "provider-b" { ... }`   | One or more provider IDs, combined as OR. It is allowed only under a root `class`, and may contain `family`, `revision`, and `models`.                                         |
-| `on-api`   | `on-api "adapter-a" "adapter-b" { ... }` | One or more request adapter IDs, combined as OR. At document root it may contain `class` and `models`; under a root `class` or `provider` it may contain `family`, `revision`, and `models`. |
+| `on`       | `on "provider-a" "provider-b" { ... }`   | One or more provider IDs, combined as OR. It is allowed under a root `class`, and may contain `on-api` to conjoin an API scope plus `family`, `revision`, and `models`. |
+| `on-api`   | `on-api "adapter-a" "adapter-b" { ... }` | One or more request adapter IDs, combined as OR. At document root it may contain `class` and `models`; under a root `class` (directly or nested inside `on`) or `provider` it may contain `family`, `revision`, and `models`. |
 | `family`   | `family "id" { ... }`                    | Exact classified family ID. It may contain `revision` and `models`. A target with no family does not match.                                                                    |
 | `revision` | `revision ">=2.5 <4" { ... }`            | A non-empty, whitespace-separated conjunction of comparisons. It may contain `models`. A target with no revision does not match.                                               |
 | `models`   | `models "id" "vendor/*" { ... }`         | One or more alternatives, combined as OR. It may contain only `on-upstream`. `token="name"` matches an ASCII-case-insensitive token bounded by non-alphanumerics.                 |
@@ -192,6 +192,8 @@ provider "openrouter" {
 
 Every selector scope may additionally contain `on-upstream`; it cannot replace an already constrained upstream. Class, provider/`on`, `on-api`, `on-upstream`, and family values are compared exactly and case-sensitively to the structured resolve target. Revision operators are `>=`, `>`, `<=`, `<`, and `=`; operands have one to three dot-separated unsigned 8-bit components, omitted components zero.
 
+An `on-api` nested inside `on` under a class requires both the selected provider and request adapter; values within each selector remain alternatives.
+
 A `models` string without `*` is an exact, case-sensitive match against the provider-relative model identifier. A string containing `*` is an anchored, ASCII-case-insensitive wildcard match. Prefer taxonomy ranks; retain exact/glob lists only when they isolate the census member set exactly, and keep a `// residue:` comment explaining why ranks do not.
 
 `priority=N` is an optional signed integer property on the block that owns axis assignments. Its default is zero. Use it only to resolve an intentional equal-specificity overlap; do not use it to encode declaration order.

package/src/compat/rules/classes/anthropic.kdl

@@ -49,8 +49,9 @@ class "anthropic" {
 	// Opus 5.5 rejects forced tool selection the way Fable/Mythos do — a
 	// `tool_choice` of `tool` or `any` fails the request with 400
 	// `invalid_request_error`: `tool_choice: type "tool" and "any" are not
-	// supported for this model`. The transport downgrades the selector to
-	// `auto` so the tool stays offered instead of the turn dying.
+	// supported for this model`. This holds on every route, including Bedrock
+	// Converse (`toolChoice: {any: {}}`). The transports downgrade the selector
+	// to `auto` so the tool stays offered instead of the turn dying.
 	//
 	// Upper-bounded on purpose: separator-collapsed ids parse their revision
 	// as a whole number (`claude-opus-45` is Opus 4.5, revision 45), so an
@@ -281,27 +282,11 @@ class "anthropic" {
 			}
 		}
 	}
-	// AWS Bedrock Converse prompt-cache checkpoints, from the AWS model cards
-	// (docs.aws.amazon.com/bedrock/latest/userguide/model-cards.html); cache
-	// pricing is deliberately not used to infer these request shapes.
 	on "amazon-bedrock" {
+		// Model metadata applies to every Bedrock request adapter.
+		// Upstream metadata is stale for cache pricing and context.
 		family "opus" {
-			revision ">=4 <4.5" {
-				prompt-cache-mode "explicit"
-				prompt-cache-minimum-tokens 1024
-				prompt-cache-maximum-checkpoints 4
-			}
-			revision "=4.5" {
-				prompt-cache-mode "explicit"
-				prompt-cache-minimum-tokens 4096
-				prompt-cache-maximum-checkpoints 4
-				supports-long-prompt-cache-retention #true
-			}
 			revision "=4.6" {
-				prompt-cache-mode "explicit"
-				prompt-cache-minimum-tokens 4096
-				prompt-cache-maximum-checkpoints 4
-				// Upstream metadata is stale for cache pricing and context.
 				cost-patch {
 					cache-read 0.5
 					cache-write 6.25
@@ -311,74 +296,196 @@ class "anthropic" {
 					max-tokens 128000
 				}
 			}
-			revision ">=4.7 <5" {
-				prompt-cache-mode "explicit"
-				prompt-cache-minimum-tokens 4096
-				prompt-cache-maximum-checkpoints 4
-				supports-long-prompt-cache-retention #true
-			}
-			revision ">=5" {
-				prompt-cache-mode "explicit"
-				prompt-cache-minimum-tokens 512
-				prompt-cache-maximum-checkpoints 4
-				supports-long-prompt-cache-retention #true
-			}
 		}
-		family "sonnet" {
-			revision "=3.5" {
-				models "*3-5-sonnet-20241022-v2*" {
-				prompt-cache-mode "explicit"
-				prompt-cache-minimum-tokens 1024
-				prompt-cache-maximum-checkpoints 4
-			}
+		// AWS Bedrock Converse prompt-cache checkpoints, from the AWS model cards
+		// (docs.aws.amazon.com/bedrock/latest/userguide/model-cards.html); cache
+		// pricing is deliberately not used to infer these request shapes. Keep the
+		// provider and adapter conjunctive so this policy stays on Converse.
+		on-api "bedrock-converse-stream" {
+			family "opus" {
+				revision ">=4 <4.5" {
+					prompt-cache-mode "explicit"
+					prompt-cache-minimum-tokens 1024
+					prompt-cache-maximum-checkpoints 4
+					prompt-cache {
+						short 300
+					}
+				}
+				revision "=4.5" {
+					prompt-cache-mode "explicit"
+					prompt-cache-minimum-tokens 4096
+					prompt-cache-maximum-checkpoints 4
+					prompt-cache {
+						short 300
+						long 3600
+					}
+					supports-long-prompt-cache-retention #true
+				}
+				revision "=4.6" {
+					prompt-cache-mode "explicit"
+					prompt-cache-minimum-tokens 4096
+					prompt-cache-maximum-checkpoints 4
+					prompt-cache {
+						short 300
+					}
+				}
+				revision ">=4.7 <5" {
+					prompt-cache-mode "explicit"
+					prompt-cache-minimum-tokens 4096
+					prompt-cache-maximum-checkpoints 4
+					supports-long-prompt-cache-retention #true
+					prompt-cache {
+						short 300
+						long 3600
+					}
+				}
+				revision ">=5" {
+					prompt-cache-mode "explicit"
+					prompt-cache-minimum-tokens 512
+					prompt-cache-maximum-checkpoints 4
+					supports-long-prompt-cache-retention #true
+					prompt-cache {
+						short 300
+						long 3600
+					}
+				}
 			}
-			revision "=3.7" {
-				prompt-cache-mode "explicit"
-				prompt-cache-minimum-tokens 1024
-				prompt-cache-maximum-checkpoints 4
+			family "sonnet" {
+				revision "=3.5" {
+					models "*3-5-sonnet-20241022-v2*" {
+						prompt-cache-mode "explicit"
+						prompt-cache-minimum-tokens 1024
+						prompt-cache-maximum-checkpoints 4
+						prompt-cache {
+							short 300
+						}
+					}
+				}
+				revision "=3.7" {
+					prompt-cache-mode "explicit"
+					prompt-cache-minimum-tokens 1024
+					prompt-cache-maximum-checkpoints 4
+					prompt-cache {
+						short 300
+					}
+				}
+				revision ">=4 <4.5" {
+					prompt-cache-mode "explicit"
+					prompt-cache-minimum-tokens 1024
+					prompt-cache-maximum-checkpoints 4
+					prompt-cache {
+						short 300
+					}
+				}
+				revision "=4.5" {
+					prompt-cache-mode "explicit"
+					prompt-cache-minimum-tokens 4096
+					prompt-cache-maximum-checkpoints 4
+					supports-long-prompt-cache-retention #true
+					prompt-cache {
+						short 300
+						long 3600
+					}
+				}
+				revision "=4.6" {
+					prompt-cache-mode "explicit"
+					prompt-cache-minimum-tokens 1024
+					prompt-cache-maximum-checkpoints 4
+					prompt-cache {
+						short 300
+					}
+				}
+				revision ">=5" {
+					prompt-cache-mode "explicit"
+					prompt-cache-minimum-tokens 4096
+					prompt-cache-maximum-checkpoints 4
+					supports-long-prompt-cache-retention #true
+					prompt-cache {
+						short 300
+						long 3600
+					}
+				}
 			}
-			revision ">=4 <4.5" {
-				prompt-cache-mode "explicit"
-				prompt-cache-minimum-tokens 1024
-				prompt-cache-maximum-checkpoints 4
+			family "haiku" {
+				revision "=3.5" {
+					prompt-cache-mode "explicit"
+					prompt-cache-minimum-tokens 2048
+					prompt-cache-maximum-checkpoints 4
+					prompt-cache {
+						short 300
+					}
+				}
+				revision ">=4.5" {
+					prompt-cache-mode "explicit"
+					prompt-cache-minimum-tokens 4096
+					prompt-cache-maximum-checkpoints 4
+					supports-long-prompt-cache-retention #true
+					prompt-cache {
+						short 300
+						long 3600
+					}
+				}
 			}
-			revision "=4.5" {
-				prompt-cache-mode "explicit"
-				prompt-cache-minimum-tokens 4096
-				prompt-cache-maximum-checkpoints 4
-				supports-long-prompt-cache-retention #true
+			family "fable" {
+				revision ">=5" {
+					prompt-cache-mode "explicit"
+					prompt-cache-minimum-tokens 1024
+					prompt-cache-maximum-checkpoints 4
+					supports-long-prompt-cache-retention #true
+					prompt-cache {
+						short 300
+						long 3600
+					}
+				}
 			}
-			revision "=4.6" {
-				prompt-cache-mode "explicit"
-				prompt-cache-minimum-tokens 1024
-				prompt-cache-maximum-checkpoints 4
+		}
+	}
+	// Messages routes on Bedrock Runtime and Mantle use the Anthropic five-minute
+	// cache lifetime. Require both the AWS provider and Messages API: model IDs
+	// alone do not distinguish Bedrock from gateways or other API lanes.
+	on "amazon-bedrock" "bedrock-mantle" {
+		on-api "anthropic-messages" {
+			family "opus" {
+				revision ">=4.7" {
+					prompt-cache {
+						short 300
+					}
+				}
 			}
-			revision ">=5" {
-				prompt-cache-mode "explicit"
-				prompt-cache-minimum-tokens 4096
-				prompt-cache-maximum-checkpoints 4
-				supports-long-prompt-cache-retention #true
+			family "sonnet" {
+				revision ">=4.6" {
+					prompt-cache {
+						short 300
+					}
+				}
 			}
-		}
-		family "haiku" {
-			revision "=3.5" {
-				prompt-cache-mode "explicit"
-				prompt-cache-minimum-tokens 2048
-				prompt-cache-maximum-checkpoints 4
+			family "haiku" {
+				revision ">=4.5" {
+					prompt-cache {
+						short 300
+					}
+				}
 			}
-			revision ">=4.5" {
-				prompt-cache-mode "explicit"
-				prompt-cache-minimum-tokens 4096
-				prompt-cache-maximum-checkpoints 4
-				supports-long-prompt-cache-retention #true
+			family "fable" {
+				revision ">=5" {
+					prompt-cache {
+						short 300
+					}
+				}
 			}
-		}
-		family "fable" {
-			revision ">=5" {
-				prompt-cache-mode "explicit"
-				prompt-cache-minimum-tokens 1024
-				prompt-cache-maximum-checkpoints 4
-				supports-long-prompt-cache-retention #true
+			family "mythos" {
+				// The Mythos Preview id has no numeric revision to select.
+				models "*mythos-preview*" {
+					prompt-cache {
+						short 300
+					}
+				}
+
+				revision ">=5" {
+					prompt-cache {
+						short 300
+					}
+				}
 			}
 		}
 	}

package/src/compat/rules/providers/commandcode.kdl

@@ -54,6 +54,24 @@ provider "commandcode" {
 		supports-eager-tool-input-streaming #false
 		supports-long-cache-retention #false
 	}
+	// Endpoint contract for every id `api-routes provider="commandcode"` sends to
+	// Responses. Responses accepts `reasoning.effort: "none"` while chat
+	// completions rejects it (low|medium|high|xhigh|max only), so thinking-off
+	// disables reasoning instead of clamping to the lowest effort. Ids without
+	// an effort ladder stay inert through provider-level
+	// `supports-reasoning-effort #false`. The Responses endpoint also rejects an
+	// object `tool_choice` (verified 2026-09-29), so named choices degrade to
+	// `"required"`.
+	on-api "openai-responses" {
+		reasoning-disable-mode "none-effort"
+		supports-named-tool-choice #false
+	}
+	// First-party OpenAI grants `none` only to the 5.6 generation
+	// (`providers/openai.kdl`); the gateway may forward `reasoning.effort`
+	// unchanged, so the other GPT ids keep the first-party lowest-effort off.
+	models "gpt-5.3-codex" "gpt-5.4" "gpt-5.4-mini" "gpt-5.5" "gpt-6-astra" "gpt-6-luna" "gpt-6-sol" {
+		reasoning-disable-mode "lowest-effort"
+	}
 	models "typesafe/jev" {
 		kind "judge"
 	}
@@ -154,15 +172,13 @@ provider "commandcode" {
 			max-tokens 65536
 		}
 	}
-	// Command Code's Responses endpoint rejects an object `tool_choice` and returns
-	// no bytes from `image_generation` (verified 2026-09-29), so the openai
-	// class's hosted-image and image-model grants do not apply and named choices
-	// degrade to `"required"`. Hosted web search still works with a string `tool_choice`.
+	// Command Code's Responses endpoint returns no bytes from `image_generation`
+	// (verified 2026-09-29), so the openai class's hosted-image and image-model
+	// grants do not apply. Hosted web search still works with a string `tool_choice`.
 	models "gpt-5.3-codex" "gpt-5.4" "gpt-5.4-mini" "gpt-5.5" "gpt-5.6-luna" "gpt-5.6-sol" \
 		"gpt-5.6-terra" "gpt-6-astra" "gpt-6-luna" "gpt-6-sol" {
 		hosted-image #false
 		image-model #false
-		supports-named-tool-choice #false
 	}
 	// ---------------------------------------------------------------------------
 	// This table prices every model id that GET /provider/v1/models serves.

package/src/compat/rules/runtime/behavior.kdl

@@ -212,14 +212,31 @@ behavior {
 		route "openai-responses" prefix="grok-4." prefix="gpt-5" prefix="gpt-6" prefix="oswe" prefix="mai-"
 	}
 	// Command Code serves Claude ids through its Anthropic-compatible Messages
-	// endpoint, its ten GPT ids (the same set command-code@1.67.0 marks
-	// `spec: "responses"`) through its Responses endpoint, and every other id
-	// through chat completions.
+	// endpoint and everything else through chat completions or Responses
+	// (`supported_endpoints` on /provider/v1/models). Chat completions rejects
+	// `reasoning_effort: "none"` (low|medium|high|xhigh|max only), while
+	// Responses accepts `reasoning.effort: "none"`. Every id with a reviewed
+	// effort ladder that lists `/responses` therefore rides Responses, so
+	// thinking-off actually disables reasoning instead of clamping to the
+	// lowest effort. Ladder ids without `/responses` (deepseek-v4-flash-fast,
+	// gemini-3.7-flash, Qwen3.8-Flash, Qwen3.8-Max-0902, hy4-preview,
+	// space-bunny-alpha), the Qwen3.8 ids (chat already disables them through
+	// `enable_thinking: false`), and non-reasoning ids stay on chat completions.
 	api-routes provider="commandcode" default="openai-completions" {
 		route "anthropic-messages" prefix="claude-"
 		route "openai-responses" exact="gpt-5.3-codex" exact="gpt-5.4" exact="gpt-5.4-mini" exact="gpt-5.5" \
 			exact="gpt-5.6-luna" exact="gpt-5.6-sol" exact="gpt-5.6-terra" exact="gpt-6-astra" \
-			exact="gpt-6-luna" exact="gpt-6-sol"
+			exact="gpt-6-luna" exact="gpt-6-sol" \
+			exact="deepseek/deepseek-v4-flash" exact="deepseek/deepseek-v4-flash-vision-exp" \
+			exact="deepseek/deepseek-v4-pro" exact="deepseek/deepseek-v4.1-flash" \
+			exact="deepseek/deepseek-v4.1-flash-fast" exact="moonshotai/Kimi-K3" \
+			exact="z-ai/glm-5.3-flash" exact="z-ai/glm-5.3-flashx" exact="zai-org/GLM-5.2" \
+			exact="zai-org/GLM-5.3" exact="google/gemini-3.1-flash-lite" exact="google/gemini-3.5-flash" \
+			exact="google/gemini-3.5-flash-lite" exact="google/gemini-3.6-flash" \
+			exact="google/gemini-3.8-flash" exact="MiniMaxAI/MiniMax-M3" exact="stepfun/Step-5-Preview" \
+			exact="xai/grok-4.5" exact="xai/grok-4.6" exact="xai/grok-4.7" exact="meta/muse-spark-1.1" \
+			exact="meta/muse-spark-1.2" exact="meta/muse-spark-1.2-contributor" exact="meta/muse-spark-1.3" \
+			exact="meta/muse-spark-1.3-contributor" exact="stealth/pixel-canary" exact="sakana/fugu-ultra"
 	}
 	// ZenMux fronts Anthropic ids on a dedicated Messages endpoint; every
 	// other namespace stays on chat completions.