@linxiraos/pi-catalog 1.1.26 → 1.2.0

package/dist/types/types.d.ts

@@ -14,7 +14,23 @@ export type KindApiKind = (typeof KIND_API_KINDS)[number];
 /** Grounding transport available to chat models selected by the web role. */
 export type WebSearchGrounding = "gemini" | "anthropic" | "codex" | "xai" | "openrouter" | "openai";
 /** Non-chat runner protocols accepted by catalog seeds, outside the chat dispatch union. */
-export declare const RUNNER_APIS: readonly ["local-inference", "web-search", "typesafe", "openrouter-decisions", "openai-images", "openrouter-images", "xai-tts", "openai-speech", "openai-embeddings", "openrouter-rerank", "openrouter-video", "openai-transcriptions"];
+export declare const RUNNER_APIS: readonly ["local-inference", "web-search", "typesafe", "openrouter-decisions", "openai-decisions", "workers-ai-systemone", "openai-images", "openrouter-images", "xai-tts", "openai-speech", "openai-embeddings", "openrouter-rerank", "openrouter-video", "openai-transcriptions"];
+/** Kind each single-purpose runner API serves; `local-inference` hosts several kinds. */
+export declare const RUNNER_API_KINDS: Record<Exclude<(typeof RUNNER_APIS)[number], "local-inference">, ModelKind>;
+/** Kind a runner API serves; `undefined` for chat transports and multi-kind `local-inference`. */
+export declare function runnerApiKind(api: Api): ModelKind | undefined;
+/** Catalog APIs `generate_image` runs through a pi-ai image client; the hosted Responses pair needs a carrier model. */
+export declare const IMAGE_GENERATION_APIS: readonly ["openai-images", "openrouter-images", "google-generative-ai", "google-gemini-cli", "openai-responses", "openai-codex-responses"];
+export type ImageGenerationApi = (typeof IMAGE_GENERATION_APIS)[number];
+/**
+ * Kinds a model on `api` may declare: a runner api serves its own kind; a chat
+ * transport serves `chat` and `tiny`, plus `image` when `generate_image` runs it
+ * (hosted Responses image tool, Gemini image models). `undefined` for
+ * `local-inference`, which hosts several kinds chosen by the model itself.
+ */
+export declare function servedKinds(api: Api): readonly ModelKind[] | undefined;
+/** Whether a model of `kind` can run on `api`. */
+export declare function apiServesKind(api: Api, kind: ModelKind): boolean;
 /** Resolve a model's kind while preserving chat semantics for existing catalog rows. */
 export declare function modelKind(model: Pick<Model, "kind">): ModelKind;
 /** Canonical thinking transport used by a model. */
@@ -176,6 +192,8 @@ export interface OpenAICompat {
     supportsMultipleSystemMessages?: boolean;
     /** Whether the provider supports `reasoning_effort`. Default: auto-detected from URL. */
     supportsReasoningEffort?: boolean;
+    /** Do not infer a thinking dial when discovery supplies no explicit thinking configuration. */
+    trustExplicitThinkingOnly?: boolean;
     /** Optional mapping from pi-ai reasoning levels to provider/model-specific `reasoning_effort` values. */
     reasoningEffortMap?: Partial<Record<Effort, string>>;
     /** Whether the provider supports `stream_options: { include_usage: true }` for token usage in streaming responses. Default: true. */
@@ -322,6 +340,8 @@ export interface OpenAICompat {
     extraBody?: Record<string, unknown>;
     /** Request-session header that should mirror the normalized prompt-cache key. Default: unset. */
     promptCacheSessionHeader?: "x-grok-conv-id";
+    /** Request header carrying the conversation session id, independent of prompt caching. Default: unset. */
+    sessionHeader?: "x-litellm-session-id";
     /** Whether chat-completions payloads should include provider-specific prompt-cache markers. */
     cacheControlFormat?: "anthropic" | undefined;
     /**
@@ -378,6 +398,8 @@ export interface OpenAICompat {
     supportsReasoningParams?: boolean;
     /** Whether Responses requests may include `reasoning.summary`. Default: true except on known incompatible hosts. */
     supportsReasoningSummary?: boolean;
+    /** Whether to chain OpenAI Responses turns using stored response ids. Unset uses the official-endpoint default; env and call options take precedence. */
+    statefulResponses?: boolean;
     /**
      * Whether the endpoint accepts explicit sampling parameters (`temperature`,
      * `top_p`, `top_k`, `min_p`, penalties). OpenAI proprietary reasoning models
@@ -398,13 +420,13 @@ export interface OpenAICompat {
     alwaysSendMaxTokens?: boolean;
     /** Whether Responses-API tool-call/result history must be strictly paired. Default: auto-detected (Azure OpenAI, GitHub Copilot). */
     strictResponsesPairing?: boolean;
-    /** Whether the Responses API accepts the `detail: "original"` image hint. Default: auto-detected (false for GitHub Copilot, which rejects it with a 400). */
+    /** Whether the Responses API accepts the `detail: "original"` image hint. Default: true for OpenAI, Azure OpenAI, and Codex; false for other hosts. Explicit overrides win. */
     supportsImageDetailOriginal?: boolean;
     /**
      * Whether the Responses endpoint accepts `configuration_update` input items
      * that change `reasoning.effort` mid-conversation while the request-level
-     * effort stays pinned for prompt caching (GPT-6 Astra). Default:
-     * rule-detected (`true` for `gpt-6-astra` on any host, `false` otherwise).
+     * effort stays pinned for prompt caching (GPT-6 family). Default:
+     * rule-detected (`true` for GPT-6 models on any host, `false` otherwise).
      * Set `false` for custom `openai-responses` / `openai-codex-responses`
      * endpoints that reject the item type with HTTP 400; effort changes are then
      * sent as the top-level `reasoning.effort`.
@@ -601,12 +623,31 @@ export interface AnthropicCompat {
     injectClaudeCodeInstruction?: boolean;
     /** Strip image inputs before encoding (text-only serving of a multimodal id). */
     stripImageInput?: boolean;
+    /**
+     * Largest width or height a single image block may carry. The canonical
+     * Anthropic API rejects anything larger with `At least one of the image
+     * dimensions exceed max allowed size: 8000 pixels`, and the rejection
+     * poisons every retry and model fallback because the block stays in
+     * context, so the request builder downscales to this value first. Hosts
+     * with a different image contract override it. Default: 8000.
+     */
+    maxImageDimension?: number;
+    /**
+     * Largest base64 payload a single image block may carry, in bytes. The
+     * canonical Anthropic API measures the encoded string, not the decoded
+     * bytes (`image exceeds 10 MB maximum: 14012300 bytes > 10485760 bytes`),
+     * and an image inside {@link maxImageDimension} can still cross it, so
+     * payload size is clamped on its own. Hosts with a different image
+     * contract override it. Default: 10485760.
+     */
+    maxImagePayloadBytes?: number;
     /** Thinking-loop watchdog guard family applied to streamed reasoning. */
     thinkingLoopGuard?: "gemini" | "deepseek" | "xai";
     /**
      * Drop the enabled `thinking` config and replayed thinking blocks when the
      * conversation is not thinking-led (a proxy contract for non-interleaved
-     * budget models). Default: false.
+     * budget models), and replay history without thinking blocks on every turn
+     * that sends `thinking: { type: "disabled" }`. Default: false.
      */
     stripThinkingHistory?: boolean;
     /**
@@ -616,10 +657,15 @@ export interface AnthropicCompat {
     effortBeta?: boolean;
     /**
      * Wire form of a disabled-thinking turn: omit the field, send
-     * `{ type: "disabled" }`, or keep adaptive thinking. Unset keeps the
-     * direct-provider behavior.
+     * `{ type: "disabled" }`, keep adaptive thinking, or send
+     * `{ type: "between_tools" }`. Unset keeps the direct-provider behavior.
+     */
+    disabledThinking?: "omit" | "disabled" | "adaptive" | "between-tools";
+    /**
+     * `output_config.effort` pinned on a `between_tools` turn. Unset leaves
+     * the turn's effort untouched.
      */
-    disabledThinking?: "omit" | "disabled" | "adaptive";
+    betweenToolsEffort?: "low" | "medium" | "high";
     /** The model is a fast-mode SKU: send `speed: "fast"` with the fast-mode beta. Default: false. */
     fastMode?: boolean;
 }
@@ -628,6 +674,8 @@ export interface AnthropicCompat {
  * deliberately not used to infer these request-shape capabilities.
  */
 export interface BedrockCompat {
+    /** Explicit disabled-thinking wire form; unset preserves the provider's existing behavior. */
+    disabledThinking?: AnthropicCompat["disabledThinking"];
     /** Whether this endpoint accepts no checkpoints, automatic caching, or explicit cachePoint blocks. */
     promptCacheMode?: "none" | "automatic" | "explicit";
     /** Whether this wire may revise already-streamed text (`stream-revision` axis). Unassigned: append-only. */
@@ -651,22 +699,42 @@ export interface BedrockCompat {
      * `PI_OPENAI_STREAM_IDLE_TIMEOUT_MS` alias, then the 300s default.
      */
     streamIdleTimeoutMs?: number;
+    /**
+     * Whether the served model accepts explicit sampling parameters
+     * (`temperature`, `topP`, …). Class rules set `false` for model lines that
+     * reject them on every host. Unassigned: accepted.
+     */
+    supportsSamplingParams?: boolean;
     /**
      * 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;
+    /**
+     * Whether the endpoint accepts thinking-binding controls (`thinking.block_binding`,
+     * the `thinking-binding-controls-2026-08-01` beta and the `/input_transformations`
+     * response path) on prefix-bound models. Only consulted when
+     * `thinking.prefixBinding` is set; `false` omits the controls while keeping the
+     * local prefix-binding safeguards. Default: true.
+     */
+    supportsThinkingBindingControls?: boolean;
 }
 /** Fully-resolved Bedrock Converse prompt-cache capabilities, materialized once by `buildModel`. */
 export interface ResolvedBedrockCompat {
+    /** See {@link BedrockCompat.disabledThinking}. */
+    disabledThinking?: BedrockCompat["disabledThinking"];
     promptCacheMode: NonNullable<BedrockCompat["promptCacheMode"]>;
     /** See {@link BedrockCompat.streamRevision}. */
     streamRevision?: BedrockCompat["streamRevision"];
     supportsLongPromptCacheRetention: boolean;
     promptCacheMinimumTokens: number;
     promptCacheMaximumCheckpoints: number;
+    /** See {@link BedrockCompat.supportsSamplingParams}. */
+    supportsSamplingParams?: boolean;
     supportsForcedToolChoice: boolean;
+    /** See {@link BedrockCompat.supportsThinkingBindingControls}. */
+    supportsThinkingBindingControls: boolean;
     /**
      * Stream-watchdog idle-timeout fallback in ms for hosts with no keepalive
      * events; 0 disables the idle watchdog. Undefined defers to
@@ -712,6 +780,7 @@ export interface ResolvedOpenAISharedCompat {
     supportsDeveloperRole: boolean;
     supportsStrictMode: boolean;
     supportsReasoningEffort: boolean;
+    trustExplicitThinkingOnly?: boolean;
     reasoningEffortMap: Partial<Record<Effort, string>>;
     supportsReasoningParams: boolean;
     supportsSamplingParams: boolean;
@@ -754,6 +823,7 @@ export interface ResolvedOpenAISharedCompat {
     emptyLengthFinishIsContextError: boolean;
     usesOpenAIToolCallIdLimit: boolean;
     promptCacheSessionHeader?: OpenAICompat["promptCacheSessionHeader"];
+    sessionHeader?: OpenAICompat["sessionHeader"];
     /**
      * Whether this model accepts explicit OpenAI prompt-cache breakpoints.
      * Built catalog models always materialize this false-by-default value;
@@ -788,7 +858,7 @@ export interface ResolvedOpenAISharedCompat {
  * `buildModel`; request handlers read fields and never detect, resolve, or
  * allocate.
  */
-export type ResolvedOpenAICompat = ResolvedOpenAISharedCompat & Required<Omit<OpenAICompat, "supportsDeveloperRole" | "supportsReasoningEffort" | "reasoningEffortMap" | "supportsReasoningParams" | "supportsReasoningSummary" | "supportsSamplingParams" | "supportsPenaltyAndStopParams" | "thinkingFormat" | "kimiApiFormat" | "reasoningDisableMode" | "omitReasoningEffort" | "includeEncryptedReasoning" | "filterReasoningHistory" | "disableReasoningOnForcedToolChoice" | "disableReasoningOnToolChoice" | "disableReasoningWithTools" | "supportsToolChoice" | "supportsForcedToolChoice" | "supportsNamedToolChoice" | "reasoningContentField" | "mistralReasoningContentParts" | "requiresReasoningContentForToolCalls" | "requiresReasoningContentForAllAssistantTurns" | "allowsSyntheticReasoningContentForToolCalls" | "syntheticReasoningContentFallback" | "replayReasoningContent" | "qwenPreserveThinking" | "qwenTemplateReasoningEffort" | "requiresThinkingAsText" | "requiresMistralToolIds" | "requiresToolResultName" | "requiresAssistantAfterToolResult" | "requiresAssistantContentForToolCalls" | "stripDeepseekSpecialTokens" | "streamMarkupHealingPattern" | "reasoningDeltasMayBeCumulative" | "emptyLengthFinishIsContextError" | "usesOpenAIToolCallIdLimit" | "promptCacheSessionHeader" | "supportsPromptCacheBreakpoints" | "promptCacheBreakpointTtl" | "openRouterRouting" | "isOpenRouterHost" | "supportsStrictMode" | "supportsLongPromptCacheRetention" | "alwaysSendMaxTokens" | "clampOutputToModelMax" | "wireModelIdMode" | "vercelGatewayRouting" | "extraBody" | "toolStrictMode" | "toolSchemaFlavor" | "streamFirstEventTimeoutMs" | "streamIdleTimeoutMs" | "streamRevision" | "cacheControlFormat" | "thinkingKeep" | "strictResponsesPairing" | "supportsImageDetailOriginal" | "supportsConfigurationUpdate" | "supportsSteering" | "stripImageInput" | "thinkingLoopGuard" | "whenThinking">> & {
+export type ResolvedOpenAICompat = ResolvedOpenAISharedCompat & Required<Omit<OpenAICompat, "supportsDeveloperRole" | "supportsReasoningEffort" | "trustExplicitThinkingOnly" | "reasoningEffortMap" | "supportsReasoningParams" | "supportsReasoningSummary" | "statefulResponses" | "supportsSamplingParams" | "supportsPenaltyAndStopParams" | "thinkingFormat" | "kimiApiFormat" | "reasoningDisableMode" | "omitReasoningEffort" | "includeEncryptedReasoning" | "filterReasoningHistory" | "disableReasoningOnForcedToolChoice" | "disableReasoningOnToolChoice" | "disableReasoningWithTools" | "supportsToolChoice" | "supportsForcedToolChoice" | "supportsNamedToolChoice" | "reasoningContentField" | "mistralReasoningContentParts" | "requiresReasoningContentForToolCalls" | "requiresReasoningContentForAllAssistantTurns" | "allowsSyntheticReasoningContentForToolCalls" | "syntheticReasoningContentFallback" | "replayReasoningContent" | "qwenPreserveThinking" | "qwenTemplateReasoningEffort" | "requiresThinkingAsText" | "requiresMistralToolIds" | "requiresToolResultName" | "requiresAssistantAfterToolResult" | "requiresAssistantContentForToolCalls" | "stripDeepseekSpecialTokens" | "streamMarkupHealingPattern" | "reasoningDeltasMayBeCumulative" | "emptyLengthFinishIsContextError" | "usesOpenAIToolCallIdLimit" | "promptCacheSessionHeader" | "sessionHeader" | "supportsPromptCacheBreakpoints" | "promptCacheBreakpointTtl" | "openRouterRouting" | "isOpenRouterHost" | "supportsStrictMode" | "supportsLongPromptCacheRetention" | "alwaysSendMaxTokens" | "clampOutputToModelMax" | "wireModelIdMode" | "vercelGatewayRouting" | "extraBody" | "toolStrictMode" | "toolSchemaFlavor" | "streamFirstEventTimeoutMs" | "streamIdleTimeoutMs" | "streamRevision" | "cacheControlFormat" | "thinkingKeep" | "strictResponsesPairing" | "supportsImageDetailOriginal" | "supportsConfigurationUpdate" | "supportsSteering" | "stripImageInput" | "thinkingLoopGuard" | "whenThinking">> & {
     vercelGatewayRouting?: OpenAICompat["vercelGatewayRouting"];
     extraBody?: OpenAICompat["extraBody"];
     cacheControlFormat?: OpenAICompat["cacheControlFormat"];
@@ -818,8 +888,8 @@ export interface ResolvedOpenAIResponsesCompat extends ResolvedOpenAISharedCompa
     /**
      * Whether a `configuration_update` input item may change `reasoning.effort`
      * mid-conversation while the request-level effort stays byte-stable for
-     * prompt caching. Rule-owned: GPT-6 Astra only; every other model rejects
-     * the item type with 400.
+     * prompt caching. Rule-owned: the GPT-6 family; earlier models do not
+     * support the item type.
      */
     supportsConfigurationUpdate: boolean;
     /**
@@ -835,6 +905,23 @@ export interface ResolvedOpenAIResponsesCompat extends ResolvedOpenAISharedCompa
      * filling `"auto"`.
      */
     supportsReasoningSummary: boolean;
+    /** Optional chaining override; unset falls back to officialEndpoint at request time. */
+    statefulResponses?: boolean;
+    /**
+     * Whether the host can store Responses results server-side (`store: true`).
+     * Rule-owned: hosts that keep generating after a client disconnect so a
+     * dropped stream can resume via `GET /responses/{id}`. Storage is opt-in
+     * per request (`storeResponses` stream option, else
+     * `PI_MUSE_STORE_RESPONSES`); stored runs retain prompts and outputs on the provider.
+     */
+    storeResponses: boolean;
+    /**
+     * Whether the host binds native history items (`encrypted_content`, item
+     * ids) to the connection that issued them, so a new process must rebuild
+     * prior turns from message content until its first successful response.
+     * Rule-owned: GitHub Copilot.
+     */
+    connectionBoundNativeHistory: boolean;
     streamIdleTimeoutMs?: number;
     vercelGatewayRouting?: OpenAICompat["vercelGatewayRouting"];
     /** The model sits behind Vercel AI Gateway's Responses endpoint. */
@@ -858,15 +945,28 @@ export interface ResolvedOpenAIResponsesCompat extends ResolvedOpenAISharedCompa
  */
 export type ResolvedOpenRouterCompat = ResolvedOpenAICompat & ResolvedOpenAIResponsesCompat;
 /** Fully-resolved anthropic-messages compat view (same contract as `ResolvedOpenAICompat`). */
-export type ResolvedAnthropicCompat = Required<Omit<AnthropicCompat, "streamIdleTimeoutMs" | "thinkingLoopGuard" | "bedrockMessagesApi" | "effortBeta" | "disabledThinking" | "stripThinkingHistory" | "fastMode">> & {
+export type ResolvedAnthropicCompat = Required<Omit<AnthropicCompat, "streamIdleTimeoutMs" | "thinkingLoopGuard" | "bedrockMessagesApi" | "effortBeta" | "disabledThinking" | "betweenToolsEffort" | "stripThinkingHistory" | "fastMode" | "maxImageDimension" | "maxImagePayloadBytes">> & {
     /** Effort-beta override; undefined keeps the transport's legacy heuristic. */
     effortBeta?: AnthropicCompat["effortBeta"];
     /** Disabled-thinking wire form; undefined keeps the direct-provider behavior. */
     disabledThinking?: AnthropicCompat["disabledThinking"];
-    /** Strip thinking history on non-thinking-led turns; undefined behaves as false. */
+    /** Effort pinned on `between_tools` turns; undefined leaves effort untouched. */
+    betweenToolsEffort?: AnthropicCompat["betweenToolsEffort"];
+    /** Strip thinking history on non-thinking-led and disabled-thinking turns; undefined behaves as false. */
     stripThinkingHistory?: AnthropicCompat["stripThinkingHistory"];
     /** Fast-mode SKU; undefined behaves as false. */
     fastMode?: AnthropicCompat["fastMode"];
+    /**
+     * Largest width or height a single image block may carry. Undefined defers
+     * to the canonical Anthropic API limit applied by the request builder.
+     */
+    maxImageDimension?: number;
+    /**
+     * Largest base64 image payload a single block may carry, in bytes.
+     * Undefined defers to the canonical Anthropic API limit applied by the
+     * request builder.
+     */
+    maxImagePayloadBytes?: number;
     /** Thinking-loop watchdog guard family applied to streamed reasoning. */
     thinkingLoopGuard?: AnthropicCompat["thinkingLoopGuard"];
     /**
@@ -918,9 +1018,11 @@ export interface DevinCompat {
     modelRouter?: boolean;
     /** Whether the upstream model supports native parallel tool calls. */
     supportsParallelToolCalls?: boolean;
+    /** See {@link BedrockCompat.supportsSamplingParams}. */
+    supportsSamplingParams?: boolean;
 }
 /** Fully-resolved devin-agent compat view. */
-export type ResolvedDevinCompat = Required<DevinCompat>;
+export type ResolvedDevinCompat = Required<Omit<DevinCompat, "supportsSamplingParams">> & Pick<DevinCompat, "supportsSamplingParams">;
 /**
  * Compatibility settings for the Google API family (google-generative-ai,
  * google-vertex, google-gemini-cli). Class-driven defaults come from the
@@ -955,13 +1057,16 @@ export interface GoogleCompat {
     stripImageInput?: boolean;
     /** Thinking-loop watchdog guard family applied to streamed reasoning. */
     thinkingLoopGuard?: "gemini" | "deepseek" | "xai";
+    /** See {@link BedrockCompat.supportsSamplingParams}. */
+    supportsSamplingParams?: boolean;
 }
 /** Fully-resolved google-API compat view, materialized once by `buildModel`. */
-export type ResolvedGoogleCompat = Required<Omit<GoogleCompat, "streamFirstEventTimeoutMs" | "streamIdleTimeoutMs" | "thinkingLoopGuard" | "antigravityUsageLabel">> & {
+export type ResolvedGoogleCompat = Required<Omit<GoogleCompat, "streamFirstEventTimeoutMs" | "streamIdleTimeoutMs" | "thinkingLoopGuard" | "antigravityUsageLabel" | "supportsSamplingParams">> & {
     streamFirstEventTimeoutMs?: number;
     streamIdleTimeoutMs?: number;
     thinkingLoopGuard?: GoogleCompat["thinkingLoopGuard"];
     antigravityUsageLabel?: string;
+    supportsSamplingParams?: boolean;
 };
 /** Sparse, user-authored compat overrides for a given API (models.json / config vocabulary). */
 export type CompatConfigOf<TApi extends Api> = TApi extends "openai-completions" | "openrouter" | "openai-responses" | "azure-openai-responses" | "openai-codex-responses" ? OpenAICompat : TApi extends "anthropic-messages" ? AnthropicCompat : TApi extends "bedrock-converse-stream" ? BedrockCompat : TApi extends "devin-agent" ? DevinCompat : TApi extends "google-generative-ai" | "google-vertex" | "google-gemini-cli" ? GoogleCompat : undefined;
@@ -1065,6 +1170,12 @@ export interface Model<TApi extends Api = Api> {
     id: string;
     /** Role-specific runner capability; omitted for ordinary chat models. */
     kind?: ModelKind;
+    /**
+     * Verbatim configured kind (models.yml, `modelOverrides`, runtime
+     * registrations). `buildModel` applies it over catalog `kind` rules on every
+     * rebuild, so a configured runner keeps its role.
+     */
+    kindConfig?: ModelKind;
     /** Grounding transport supported by this chat model. */
     webSearch?: WebSearchGrounding;
     /** Cheaper same-provider model to run hosted web search in this model's place (model id or provider/id). */
@@ -1162,11 +1273,11 @@ export interface Model<TApi extends Api = Api> {
     cursorMaxModeRoutes?: Readonly<Record<string, boolean>>;
     /**
      * Per-account availability recorded by multi-account discovery: provider
-     * account id (Codex: ChatGPT `chatgpt_account_id`) → that account's
-     * entitlements on this model. An account appears only when its own catalog
-     * lists the model, so credential selection can route account-gated models
-     * (e.g. `gpt-daybreak-blue-latest`) straight to eligible accounts. Absent on
-     * bundled/config rows and on single-account discovery.
+     * account key (Codex: ChatGPT `chatgpt_account_id`; Antigravity: login
+     * email) → that account's entitlements on this model. An account appears
+     * only when its own catalog lists the model, so credential selection can
+     * route account-gated models (e.g. `gpt-daybreak-blue-latest`, Antigravity
+     * Claude 5.5) straight to eligible accounts. Absent on bundled/config rows.
      */
     accountAccess?: Readonly<Record<string, ModelAccountAccess>>;
     /** Cursor `RequestedModel.parameters` for this model's default variant. */
@@ -1277,9 +1388,9 @@ export interface Model<TApi extends Api = Api> {
     /**
      * Per-service-tier cost multipliers baked from the `service-tier-cost`
      * catalog axis (e.g. `{ priority: 2.5 }`). Absent tiers use the API-generic
-     * defaults.
+     * defaults, and a tier with no published price stays at 1x.
      */
-    serviceTierCost?: Readonly<Partial<Record<"flex" | "priority", number>>>;
+    serviceTierCost?: Readonly<Partial<Record<"flex" | "priority" | "ultrafast", number>>>;
     /**
      * Provider-supplied one-line blurb for this model. Set only when an upstream
      * ships one (Devin's `GetCliModelConfigs`); never synthesized locally.
@@ -1289,7 +1400,7 @@ export interface Model<TApi extends Api = Api> {
     isNew?: boolean;
     /** Upstream marks this model as beta / preview quality. */
     isBeta?: boolean;
-    /** Authenticated provider catalog marks this as the account's default model. */
+    /** Authenticated catalog marks this as the account's default; the startup pick prefers it over `default-model`. */
     isProviderDefault?: boolean;
     /** Upstream marks this model as one of its recommended picks. */
     isRecommended?: boolean;

package/dist/types/wire/cloudflare-ai-gateway.d.ts

@@ -6,13 +6,3 @@ export declare const CLOUDFLARE_AI_GATEWAY_ANTHROPIC_BASE_URL = "https://gateway
 export declare const CLOUDFLARE_AI_GATEWAY_OPENAI_BASE_URL = "https://gateway.ai.cloudflare.com/v1/<account>/<gateway>/openai";
 /** OpenAI-compatible Workers AI endpoint. */
 export declare const CLOUDFLARE_AI_GATEWAY_COMPAT_BASE_URL = "https://gateway.ai.cloudflare.com/v1/<account>/<gateway>/compat";
-/** Opaque API-key credential payload persisted by Cloudflare AI Gateway login. */
-export interface CloudflareAiGatewayCredential {
-    token: string;
-    accountId?: string;
-    gatewayId?: string;
-}
-/** Parse both structured login credentials and legacy plain gateway tokens. */
-export declare function parseCloudflareAiGatewayCredential(value: string): CloudflareAiGatewayCredential | null;
-/** Serialize the gateway token with the routing identifiers collected during login. */
-export declare function serializeCloudflareAiGatewayCredential(token: string, accountId: string, gatewayId: string): string;

package/dist/types/wire/cloudflare.d.ts

@@ -0,0 +1,10 @@
+/** Cloudflare credential persisted by the AI Gateway and Workers AI logins: a bearer token plus the routing ids collected at login. */
+export interface CloudflareCredential {
+    token: string;
+    accountId?: string;
+    gatewayId?: string;
+}
+/** Parses a structured login credential or a legacy plain token; null when empty or malformed. */
+export declare function parseCloudflareCredential(value: string): CloudflareCredential | null;
+/** Serializes a login credential, trimming fields and omitting empty optional ids. */
+export declare function serializeCloudflareCredential(credential: CloudflareCredential): string;

package/dist/types/wire/factory-droid.d.ts

@@ -3,7 +3,7 @@
  * native request corpus with `packages/ai/scripts/capture-factory-droid-native.ts`
  * against the matching CLI binary and make `factory-droid-native-parity.test.ts` pass.
  */
-export declare const FACTORY_DROID_CLIENT_VERSION = "0.230.0";
+export declare const FACTORY_DROID_CLIENT_VERSION = "0.237.0";
 /** Wire protocols the Factory proxy multiplexes; `api-routes provider="factory-droid"` picks one per model. */
 export declare const FACTORY_DROID_WIRES: readonly ["openai-completions", "openai-responses", "anthropic-messages", "google-generate"];
 /** Wire protocol used by the Factory proxy for each model. */

package/dist/types/wire/gemini-headers.d.ts

@@ -24,7 +24,7 @@ export declare const getGeminiCliHeaders: (modelId?: string) => {
  * client the version and manifest are captured from, independent of the host
  * platform. Overrides: PI_AI_ANTIGRAVITY_VERSION / _CL / _OS / _ARCH.
  */
-export declare const DEFAULT_ANTIGRAVITY_VERSION = "2.8.0";
+export declare const DEFAULT_ANTIGRAVITY_VERSION = "2.19.1";
 /** Current Antigravity client version: env override → manifest-discovered → pinned fallback. */
 export declare function getAntigravityVersion(): string;
 /**
@@ -35,8 +35,11 @@ export declare function parseAntigravityManifestVersion(yamlText: string): strin
 /**
  * Resolves the latest Antigravity release from the official update manifest.
  * Success is cached for the process lifetime; failures are silent (the pinned
- * fallback stays valid) and clear the in-flight cache so a later call retries.
- * Skipped entirely when PI_AI_ANTIGRAVITY_VERSION is set.
+ * fallback stays valid) and suppress further lookups for
+ * `ANTIGRAVITY_VERSION_RETRY_MS`. Skipped entirely when PI_AI_ANTIGRAVITY_VERSION is set.
+ *
+ * The lookup is shared by concurrent callers, so it is bounded only by its own
+ * timeout; `signal` ends this caller's wait without cancelling the lookup.
  */
 export declare function ensureAntigravityVersion(fetcher?: FetchImpl, signal?: AbortSignal): Promise<void>;
 /** Antigravity `User-Agent` header value; rebuilt when the discovered version changes. */

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@linxiraos/pi-catalog",
-	"version": "1.1.26",
+	"version": "1.2.0",
 	"description": "Model catalog for omp: bundled model database, provider discovery descriptors, model identity, classification, and equivalence",
 	"keywords": [
 		"ai",
@@ -90,12 +90,12 @@
 		"gen:proto": "bun scripts/generate-protocols.ts"
 	},
 	"dependencies": {
-		"@linxiraos/pi-omptype": "1.1.26",
-		"@linxiraos/pi-utils": "1.1.26"
+		"@linxiraos/pi-omptype": "1.2.0",
+		"@linxiraos/pi-utils": "1.2.0"
 	},
 	"devDependencies": {
 		"@bgotink/kdl": "0.4.0",
-		"@linxiraos/pi-ai": "1.1.26",
+		"@linxiraos/pi-ai": "1.2.0",
 		"@types/bun": "^1.3.14"
 	},
 	"engines": {

package/src/build.ts

@@ -101,9 +101,11 @@ function applyCatalogAssignments<TApi extends Api>(model: Model<TApi>, catalog:
 	if (serviceTierCost !== undefined) {
 		const flex = numberField(serviceTierCost, "flex");
 		const priorityTier = numberField(serviceTierCost, "priority");
+		const ultrafast = numberField(serviceTierCost, "ultrafast");
 		model.serviceTierCost = {
 			...(flex !== undefined && { flex }),
 			...(priorityTier !== undefined && { priority: priorityTier }),
+			...(ultrafast !== undefined && { ultrafast }),
 		};
 	}
 	const priority = catalog.priority;
@@ -174,40 +176,6 @@ export function applyCatalogCorrections(
 	model: Pick<ModelSpec<Api>, "cost" | "contextWindow" | "maxTokens" | "input">,
 	catalog: Record<string, unknown>,
 ): void {
-	const longContext = objectPayload(catalog.longContext);
-	if (longContext !== undefined) {
-		const inputThreshold = numberField(longContext, "inputThreshold");
-		const inclusive = Reflect.get(longContext, "inputThresholdInclusive") === true;
-		const multiplier = numberField(longContext, "multiplier");
-		const input = numberField(longContext, "input");
-		const output = numberField(longContext, "output");
-		const cacheRead = numberField(longContext, "cacheRead");
-		const cacheWrite = numberField(longContext, "cacheWrite");
-		const base = model.cost;
-		const hasTokenPrice = base.input !== 0 || base.output !== 0 || base.cacheRead !== 0 || base.cacheWrite !== 0;
-		if (inputThreshold !== undefined && multiplier !== undefined && hasTokenPrice) {
-			// Multiplier form: tier rates derive from the row's live list price.
-			model.cost = {
-				...base,
-				longContext: {
-					inputThreshold,
-					...(inclusive && { inputThresholdInclusive: true }),
-					input: base.input * multiplier,
-					output: base.output * multiplier,
-					cacheRead: base.cacheRead * multiplier,
-					cacheWrite: base.cacheWrite * multiplier,
-				},
-			};
-		} else if (
-			inputThreshold !== undefined &&
-			input !== undefined &&
-			output !== undefined &&
-			cacheRead !== undefined &&
-			cacheWrite !== undefined
-		) {
-			model.cost = { ...model.cost, longContext: { inputThreshold, input, output, cacheRead, cacheWrite } };
-		}
-	}
 	const patch = objectPayload(catalog.costPatch);
 	if (patch !== undefined) {
 		model.cost = { ...model.cost };
@@ -244,6 +212,41 @@ export function applyCatalogCorrections(
 			applyEffectiveFallbackRates(model.cost, Reflect.get(fallback, "effectiveRates"));
 		}
 	}
+	const longContext = objectPayload(catalog.longContext);
+	if (longContext !== undefined) {
+		const inputThreshold = numberField(longContext, "inputThreshold");
+		const inclusive = Reflect.get(longContext, "inputThresholdInclusive") === true;
+		const multiplier = numberField(longContext, "multiplier");
+		const input = numberField(longContext, "input");
+		const output = numberField(longContext, "output");
+		const cacheRead = numberField(longContext, "cacheRead");
+		const cacheWrite = numberField(longContext, "cacheWrite");
+		const base = model.cost;
+		const hasTokenPrice = base.input !== 0 || base.output !== 0 || base.cacheRead !== 0 || base.cacheWrite !== 0;
+		if (inputThreshold !== undefined && multiplier !== undefined && hasTokenPrice) {
+			// Multiplier form: tier rates derive from the row's final base card,
+			// after `cost-patch`/`cost-fallback` corrected the live list price.
+			model.cost = {
+				...base,
+				longContext: {
+					inputThreshold,
+					...(inclusive && { inputThresholdInclusive: true }),
+					input: base.input * multiplier,
+					output: base.output * multiplier,
+					cacheRead: base.cacheRead * multiplier,
+					cacheWrite: base.cacheWrite * multiplier,
+				},
+			};
+		} else if (
+			inputThreshold !== undefined &&
+			input !== undefined &&
+			output !== undefined &&
+			cacheRead !== undefined &&
+			cacheWrite !== undefined
+		) {
+			model.cost = { ...model.cost, longContext: { inputThreshold, input, output, cacheRead, cacheWrite } };
+		}
+	}
 	if (catalog.cacheReadAtInputRate === true) {
 		const { longContext } = model.cost;
 		model.cost = {
@@ -363,7 +366,8 @@ 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.
+	// Configured kinds and lifetimes replace catalog values rather than merging with them.
+	if (spec.kindConfig !== undefined) model.kind = spec.kindConfig;
 	if (spec.promptCacheConfig !== undefined) model.promptCache = { ...spec.promptCacheConfig };
 	return model;
 }

package/src/compat/auth-ids.ts

@@ -17,7 +17,9 @@ export type AuthProviderId =
 	| "charm-hyper"
 	| "cline-pass"
 	| "cloudflare-ai-gateway"
+	| "cloudflare-workers-ai"
 	| "commandcode"
+	| "coralbricks"
 	| "coreweave"
 	| "cursor"
 	| "deepinfra"
@@ -72,6 +74,7 @@ export type AuthProviderId =
 	| "siliconflow-cn"
 	| "singularityapi-dev"
 	| "singularityapi-tech"
+	| "snowflake"
 	| "stencil"
 	| "stepfun"
 	| "synthetic"
@@ -108,7 +111,9 @@ export type LoginProviderId =
 	| "charm-hyper"
 	| "cline-pass"
 	| "cloudflare-ai-gateway"
+	| "cloudflare-workers-ai"
 	| "commandcode"
+	| "coralbricks"
 	| "coreweave"
 	| "cursor"
 	| "deepinfra"
@@ -156,6 +161,7 @@ export type LoginProviderId =
 	| "siliconflow-cn"
 	| "singularityapi-dev"
 	| "singularityapi-tech"
+	| "snowflake"
 	| "stencil"
 	| "stepfun"
 	| "synthetic"