@oh-my-pi/pi-catalog 18.6.1 → 18.7.0

package/CHANGELOG.md

@@ -2,6 +2,45 @@
 
 ## [Unreleased]
 
+## [18.7.0] - 2026-10-06
+
+### Added
+
+- Added Mistral Large 4 with reasoning support, image input, a 1M-token context window, and preview pricing.
+- Added configurable thinking levels from low through maximum for MiniMax-M3.1-Flash-Preview; because the model always reasons, requests that disable thinking use the low level.
+- Added Google Antigravity pricing and model support for Claude Opus 5.5 and Sonnet 5.5.
+
+### Changed
+
+- MiniMax Token Plan providers (`minimax-code` and `minimax-code-cn`) now use MiniMax's recommended Anthropic-compatible API for model requests and login key validation.
+- Google Antigravity now exposes Claude Opus 5.5 and Sonnet 5.5 once each, with selectable low, medium, and high thinking levels.
+
+### Fixed
+
+- Added the correct published pricing for GPT-6 Astra's Ultrafast service tier: a 6× multiplier on the OpenAI API and an 8× multiplier on the Codex card.
+
+## [18.6.3] - 2026-10-06
+
+### Added
+
+- Added an optional `statefulResponses` compat field for OpenAI Responses models, kept through OpenRouter's Responses dispatch ([#13686](https://github.com/can1357/oh-my-pi/pull/13686) by [@alphastorm](https://github.com/alphastorm)).
+- Added 15 Snowflake Cortex models with account-specific endpoints, Cortex compatibility rules, and estimated account-billed pricing ([#14507](https://github.com/can1357/oh-my-pi/pull/14507) by [@jorgoose](https://github.com/jorgoose)).
+- Added the `image-tokenization` axis, which declares how GPT-5.2+, Claude and Gemini 3 lines bill input images on every host, with wire-API fallback rules for other models, plus `imageTokens()` to price one image ([#14286](https://github.com/can1357/oh-my-pi/pull/14286) by [@will-bogusz](https://github.com/will-bogusz)).
+
+### Changed
+
+- Changed OpenAI Responses endpoints other than OpenAI, Azure OpenAI, and Codex (custom and local servers, proxies including `azure`/`openai-codex` providers pointed at a non-Azure/non-Codex `baseUrl`, OpenRouter) to default `supportsImageDetailOriginal` to `false`, so snapcompact frames and computer screenshots go out as `detail: "auto"` instead of failing on servers that reject `original`; set `compat.supportsImageDetailOriginal: true` to opt a host in ([#13687](https://github.com/can1357/oh-my-pi/pull/13687) by [@alphastorm](https://github.com/alphastorm)).
+
+### Fixed
+
+- Muse Code can now store Responses results on Meta's side (`store-responses`), so a turn whose connection drops can be recovered instead of re-run. Storage is opt-in via the omp setting `providers.muse-code.storeResponses` or `PI_MUSE_STORE_RESPONSES=1` ([#14293](https://github.com/can1357/oh-my-pi/pull/14293) and [#14534](https://github.com/can1357/oh-my-pi/pull/14534) by [@abilliontokens](https://github.com/abilliontokens)).
+
+## [18.6.2] - 2026-10-04
+
+### Fixed
+
+- Fixed Google Antigravity listing models the signed-in account cannot use, such as Claude Opus 5.5 and Sonnet 5.5 on plans without them, which failed every request with `404 Requested entity was not found`. After a successful model refresh and on subsequent restarts, only models in the account's own Antigravity model list are offered ([#14328](https://github.com/can1357/oh-my-pi/issues/14328)).
+
 ## [18.6.0] - 2026-10-03
 
 ### Fixed

package/dist/types/compat/auth-ids.d.ts

@@ -1,4 +1,4 @@
 /** Every provider with an `auth/<id>.kdl` policy. */
-export type AuthProviderId = "abliteration" | "aiand" | "aimlapi" | "alibaba-coding-plan" | "alibaba-token-plan" | "amazon-bedrock" | "anthropic" | "apple" | "azure" | "baseten" | "bedrock-mantle" | "cerebras" | "charm-hyper" | "cline-pass" | "cloudflare-ai-gateway" | "commandcode" | "coreweave" | "cursor" | "deepinfra" | "deepseek" | "devin" | "exa" | "factory-droid" | "firepass" | "fireworks" | "github-copilot" | "gitlab-duo" | "gitlab-duo-agent" | "gmi-cloud" | "google" | "google-antigravity" | "google-gemini-cli" | "google-vertex" | "groq" | "helmcode" | "huggingface" | "kagi" | "kilo" | "kimi-code" | "litellm" | "llama.cpp" | "lm-studio" | "local" | "meta" | "minimax" | "minimax-code" | "minimax-code-cn" | "mistral" | "moonshot" | "muse-code" | "nanogpt" | "novita" | "nvidia" | "ollama" | "ollama-cloud" | "openai" | "openai-codex" | "openai-codex-device" | "opencode-go" | "opencode-zen" | "openrouter" | "parallel" | "perplexity" | "qianfan" | "qwen-portal" | "sakana" | "siliconflow" | "siliconflow-cn" | "singularityapi-dev" | "singularityapi-tech" | "stencil" | "stepfun" | "synthetic" | "tavily" | "together" | "typesafe" | "umans" | "venice" | "vercel-ai-gateway" | "vllm" | "wafer-serverless" | "web" | "xai" | "xai-oauth" | "xiaomi" | "xiaomi-token-plan-ams" | "xiaomi-token-plan-cn" | "xiaomi-token-plan-sgp" | "yolo-auto" | "zai" | "zai-coding-plan" | "zenmux" | "zhipu-coding-plan";
+export type AuthProviderId = "abliteration" | "aiand" | "aimlapi" | "alibaba-coding-plan" | "alibaba-token-plan" | "amazon-bedrock" | "anthropic" | "apple" | "azure" | "baseten" | "bedrock-mantle" | "cerebras" | "charm-hyper" | "cline-pass" | "cloudflare-ai-gateway" | "commandcode" | "coreweave" | "cursor" | "deepinfra" | "deepseek" | "devin" | "exa" | "factory-droid" | "firepass" | "fireworks" | "github-copilot" | "gitlab-duo" | "gitlab-duo-agent" | "gmi-cloud" | "google" | "google-antigravity" | "google-gemini-cli" | "google-vertex" | "groq" | "helmcode" | "huggingface" | "kagi" | "kilo" | "kimi-code" | "litellm" | "llama.cpp" | "lm-studio" | "local" | "meta" | "minimax" | "minimax-code" | "minimax-code-cn" | "mistral" | "moonshot" | "muse-code" | "nanogpt" | "novita" | "nvidia" | "ollama" | "ollama-cloud" | "openai" | "openai-codex" | "openai-codex-device" | "opencode-go" | "opencode-zen" | "openrouter" | "parallel" | "perplexity" | "qianfan" | "qwen-portal" | "sakana" | "siliconflow" | "siliconflow-cn" | "singularityapi-dev" | "singularityapi-tech" | "snowflake" | "stencil" | "stepfun" | "synthetic" | "tavily" | "together" | "typesafe" | "umans" | "venice" | "vercel-ai-gateway" | "vllm" | "wafer-serverless" | "web" | "xai" | "xai-oauth" | "xiaomi" | "xiaomi-token-plan-ams" | "xiaomi-token-plan-cn" | "xiaomi-token-plan-sgp" | "yolo-auto" | "zai" | "zai-coding-plan" | "zenmux" | "zhipu-coding-plan";
 /** Providers whose policy declares a `login` flow (the `/login` roster). */
-export type LoginProviderId = "abliteration" | "aiand" | "alibaba-coding-plan" | "alibaba-token-plan" | "anthropic" | "baseten" | "cerebras" | "charm-hyper" | "cline-pass" | "cloudflare-ai-gateway" | "commandcode" | "coreweave" | "cursor" | "deepinfra" | "deepseek" | "devin" | "exa" | "factory-droid" | "firepass" | "fireworks" | "github-copilot" | "gitlab-duo" | "gitlab-duo-agent" | "gmi-cloud" | "google-antigravity" | "google-gemini-cli" | "helmcode" | "huggingface" | "kagi" | "kilo" | "kimi-code" | "litellm" | "llama.cpp" | "lm-studio" | "meta" | "minimax-code" | "minimax-code-cn" | "moonshot" | "muse-code" | "nanogpt" | "novita" | "nvidia" | "ollama" | "ollama-cloud" | "openai-codex" | "openai-codex-device" | "opencode-go" | "opencode-zen" | "openrouter" | "parallel" | "perplexity" | "qianfan" | "qwen-portal" | "sakana" | "siliconflow" | "siliconflow-cn" | "singularityapi-dev" | "singularityapi-tech" | "stencil" | "stepfun" | "synthetic" | "tavily" | "together" | "typesafe" | "umans" | "venice" | "vercel-ai-gateway" | "vllm" | "wafer-serverless" | "xai" | "xai-oauth" | "xiaomi" | "xiaomi-token-plan-ams" | "xiaomi-token-plan-cn" | "xiaomi-token-plan-sgp" | "yolo-auto" | "zai" | "zai-coding-plan" | "zenmux" | "zhipu-coding-plan";
+export type LoginProviderId = "abliteration" | "aiand" | "alibaba-coding-plan" | "alibaba-token-plan" | "anthropic" | "baseten" | "cerebras" | "charm-hyper" | "cline-pass" | "cloudflare-ai-gateway" | "commandcode" | "coreweave" | "cursor" | "deepinfra" | "deepseek" | "devin" | "exa" | "factory-droid" | "firepass" | "fireworks" | "github-copilot" | "gitlab-duo" | "gitlab-duo-agent" | "gmi-cloud" | "google-antigravity" | "google-gemini-cli" | "helmcode" | "huggingface" | "kagi" | "kilo" | "kimi-code" | "litellm" | "llama.cpp" | "lm-studio" | "meta" | "minimax-code" | "minimax-code-cn" | "moonshot" | "muse-code" | "nanogpt" | "novita" | "nvidia" | "ollama" | "ollama-cloud" | "openai-codex" | "openai-codex-device" | "opencode-go" | "opencode-zen" | "openrouter" | "parallel" | "perplexity" | "qianfan" | "qwen-portal" | "sakana" | "siliconflow" | "siliconflow-cn" | "singularityapi-dev" | "singularityapi-tech" | "snowflake" | "stencil" | "stepfun" | "synthetic" | "tavily" | "together" | "typesafe" | "umans" | "venice" | "vercel-ai-gateway" | "vllm" | "wafer-serverless" | "xai" | "xai-oauth" | "xiaomi" | "xiaomi-token-plan-ams" | "xiaomi-token-plan-cn" | "xiaomi-token-plan-sgp" | "yolo-auto" | "zai" | "zai-coding-plan" | "zenmux" | "zhipu-coding-plan";

package/dist/types/compat/image-tokenization.d.ts

@@ -0,0 +1,52 @@
+import type { ModelIdentity } from "./types.js";
+/** OpenAI image `detail` levels; `undefined` means the provider default (`auto`). */
+export type ImageDetail = "auto" | "low" | "high" | "original";
+export interface ImageSize {
+    width: number;
+    height: number;
+}
+/** One OpenAI `detail` level's sizing: a pixel-dimension limit, then an optional patch budget. */
+export interface PatchSizing {
+    /** Longest side after the pixel-dimension fit. */
+    maxEdge: number;
+    /** Resizing patch budget; absent when the level keeps the image's size. */
+    patchBudget?: number;
+}
+/** OpenAI 32px patches × a per-model multiplier, sized per `detail` level. */
+export interface OpenAiPatchTokenization {
+    regime: "openai-patch";
+    multiplier: number;
+    low: PatchSizing;
+    high: PatchSizing;
+    original: PatchSizing;
+    /** The level `auto` (and an omitted `detail`) sizes like. */
+    auto: "high" | "original";
+}
+/** One model line's image billing rule: the vendor formula and its per-model numbers. */
+export type ImageTokenization = OpenAiPatchTokenization
+/** Anthropic 28px patches, resized to fit a padded-edge limit and a visual-token budget. */
+ | {
+    regime: "anthropic-patch";
+    maxEdge: number;
+    maxTokens: number;
+}
+/** A fixed per-image budget regardless of pixels (Gemini 3 `media_resolution`). */
+ | {
+    regime: "fixed";
+    tokens: number;
+};
+/** What reads the image: a built model, any `{ id }` with optional host and identity, or just a wire `{ api }`. */
+export interface ImageTokenizationTarget {
+    id?: string;
+    provider?: string;
+    api?: string;
+    identity?: Pick<ModelIdentity, "class" | "family" | "revision">;
+}
+/**
+ * The reader's image billing rule: its lineage rule, else its wire API's
+ * fallback; undefined when neither covers it (an unclassified model on an
+ * unregistered API).
+ */
+export declare function resolveImageTokenization(target: ImageTokenizationTarget): ImageTokenization | undefined;
+/** Billed input tokens for one image of `size` under `rule`; only `openai-patch` reads `detail`. */
+export declare function imageTokens(rule: ImageTokenization, size: ImageSize, detail?: ImageDetail): number;

package/dist/types/compat/provider-ids.d.ts

@@ -1,2 +1,2 @@
 /** Every chat-model provider with a catalog entry (`default-model` in its provider rules). */
-export type KnownProvider = "abliteration" | "aiand" | "aimlapi" | "alibaba-coding-plan" | "alibaba-token-plan" | "amazon-bedrock" | "anthropic" | "apple" | "azure" | "baseten" | "bedrock-mantle" | "cerebras" | "charm-hyper" | "cline-pass" | "cloudflare-ai-gateway" | "commandcode" | "coreweave" | "cursor" | "deepinfra" | "deepseek" | "devin" | "factory-droid" | "firepass" | "fireworks" | "github-copilot" | "gitlab-duo" | "gitlab-duo-agent" | "gmi-cloud" | "google" | "google-antigravity" | "google-gemini-cli" | "google-vertex" | "groq" | "helmcode" | "huggingface" | "kilo" | "kimi-code" | "litellm" | "lm-studio" | "local" | "meta" | "minimax" | "minimax-code" | "minimax-code-cn" | "mistral" | "moonshot" | "muse-code" | "nanogpt" | "novita" | "nvidia" | "ollama" | "ollama-cloud" | "openai" | "openai-codex" | "opencode-go" | "opencode-zen" | "openrouter" | "qianfan" | "qwen-portal" | "sakana" | "siliconflow" | "siliconflow-cn" | "singularityapi-dev" | "singularityapi-tech" | "stepfun" | "synthetic" | "together" | "typesafe" | "umans" | "venice" | "vercel-ai-gateway" | "vllm" | "wafer-serverless" | "web" | "xai" | "xai-oauth" | "xiaomi" | "xiaomi-token-plan-ams" | "xiaomi-token-plan-cn" | "xiaomi-token-plan-sgp" | "yolo-auto" | "zai" | "zenmux" | "zhipu-coding-plan";
+export type KnownProvider = "abliteration" | "aiand" | "aimlapi" | "alibaba-coding-plan" | "alibaba-token-plan" | "amazon-bedrock" | "anthropic" | "apple" | "azure" | "baseten" | "bedrock-mantle" | "cerebras" | "charm-hyper" | "cline-pass" | "cloudflare-ai-gateway" | "commandcode" | "coreweave" | "cursor" | "deepinfra" | "deepseek" | "devin" | "factory-droid" | "firepass" | "fireworks" | "github-copilot" | "gitlab-duo" | "gitlab-duo-agent" | "gmi-cloud" | "google" | "google-antigravity" | "google-gemini-cli" | "google-vertex" | "groq" | "helmcode" | "huggingface" | "kilo" | "kimi-code" | "litellm" | "lm-studio" | "local" | "meta" | "minimax" | "minimax-code" | "minimax-code-cn" | "mistral" | "moonshot" | "muse-code" | "nanogpt" | "novita" | "nvidia" | "ollama" | "ollama-cloud" | "openai" | "openai-codex" | "opencode-go" | "opencode-zen" | "openrouter" | "qianfan" | "qwen-portal" | "sakana" | "siliconflow" | "siliconflow-cn" | "singularityapi-dev" | "singularityapi-tech" | "snowflake" | "stepfun" | "synthetic" | "together" | "typesafe" | "umans" | "venice" | "vercel-ai-gateway" | "vllm" | "wafer-serverless" | "web" | "xai" | "xai-oauth" | "xiaomi" | "xiaomi-token-plan-ams" | "xiaomi-token-plan-cn" | "xiaomi-token-plan-sgp" | "yolo-auto" | "zai" | "zenmux" | "zhipu-coding-plan";

package/dist/types/compat/types.d.ts

@@ -622,6 +622,8 @@ export interface CompiledProvider {
     id: string;
     /** Preferred model id when no explicit selection is made. */
     defaultModel: string;
+    /** Whether the provider participates in automatic default selection (defaults to true). */
+    automaticDefault?: boolean;
     /** Env vars consulted, in order, for the runtime API-key fallback. */
     envVars?: string[];
     /** The runtime creates a model manager even without a valid API key. */

package/dist/types/hosts.d.ts

@@ -15,6 +15,10 @@ export declare const KNOWN_HOSTS: {
         readonly providers: readonly ["openai"];
         readonly urlMarkers: readonly ["api.openai.com"];
     };
+    /** URL-only: a Codex provider rerouted through a proxy does not imply the subscription backend's capabilities. */
+    readonly openaiCodex: {
+        readonly urlMarkers: readonly ["chatgpt.com/backend-api", "chat.openai.com/backend-api"];
+    };
     readonly azureOpenAI: {
         readonly providers: readonly ["azure"];
         readonly urlMarkers: readonly [".openai.azure.com", "azure.com/openai", "models.inference.ai.azure.com"];

package/dist/types/types.d.ts

@@ -15,6 +15,22 @@ export type KindApiKind = (typeof KIND_API_KINDS)[number];
 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"];
+/** 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. */
@@ -378,6 +394,8 @@ 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
@@ -398,7 +416,7 @@ 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
@@ -796,7 +814,7 @@ 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" | "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" | "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"];
@@ -843,6 +861,16 @@ 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;
     streamIdleTimeoutMs?: number;
     vercelGatewayRouting?: OpenAICompat["vercelGatewayRouting"];
     /** The model sits behind Vercel AI Gateway's Responses endpoint. */
@@ -1078,6 +1106,12 @@ 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). */
@@ -1290,9 +1324,9 @@ 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.

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@oh-my-pi/pi-catalog",
-	"version": "18.6.1",
+	"version": "18.7.0",
 	"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": "18.6.1",
-		"@oh-my-pi/pi-utils": "18.6.1"
+		"@oh-my-pi/omptype": "18.7.0",
+		"@oh-my-pi/pi-utils": "18.7.0"
 	},
 	"devDependencies": {
 		"@bgotink/kdl": "0.4.0",
-		"@oh-my-pi/pi-ai": "18.6.1",
+		"@oh-my-pi/pi-ai": "18.7.0",
 		"@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;
@@ -363,7 +365,8 @@ 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

@@ -72,6 +72,7 @@ export type AuthProviderId =
 	| "siliconflow-cn"
 	| "singularityapi-dev"
 	| "singularityapi-tech"
+	| "snowflake"
 	| "stencil"
 	| "stepfun"
 	| "synthetic"
@@ -156,6 +157,7 @@ export type LoginProviderId =
 	| "siliconflow-cn"
 	| "singularityapi-dev"
 	| "singularityapi-tech"
+	| "snowflake"
 	| "stencil"
 	| "stepfun"
 	| "synthetic"

package/src/compat/axes.ts

@@ -169,6 +169,7 @@ export const AXES: Readonly<Record<string, AxisDef>> = {
 	"requires-thinking-as-text": wire("requiresThinkingAsText", ["openai"]),
 	"requires-tool-result-name": wire("requiresToolResultName", ["openai"]),
 	"strict-responses-pairing": wire("strictResponsesPairing", ["openai-responses"]),
+	"stateful-responses": wire("statefulResponses", ["openai-responses"]),
 	"requires-reasoning-off-juice-instruction": wire("requiresReasoningOffJuiceInstruction", ["openai-responses"]),
 	"supports-all-turns-reasoning-context": wire("supportsAllTurnsReasoningContext", ["openai-responses"]),
 	"supports-configuration-update": wire("supportsConfigurationUpdate", ["openai-responses"]),
@@ -194,6 +195,7 @@ export const AXES: Readonly<Record<string, AxisDef>> = {
 	"supports-reasoning-effort": wire("supportsReasoningEffort", OAI),
 	"supports-reasoning-params": wire("supportsReasoningParams", OAI),
 	"supports-reasoning-summary": wire("supportsReasoningSummary", ["openai-responses"]),
+	"store-responses": wire("storeResponses", ["openai-responses"]),
 	"supports-store": wire("supportsStore", ["openai"]),
 	"supports-strict-mode": wire("supportsStrictMode", OAI),
 	"supports-tool-choice": wire("supportsToolChoice", OAI),
@@ -390,6 +392,8 @@ export const AXES: Readonly<Record<string, AxisDef>> = {
 	},
 	"web-search-model": { key: "webSearchModel", set: "catalog", shape: "scalar" },
 	"hosted-image": { key: "hostedImage", set: "catalog", shape: "scalar", values: [true, false] },
+	/** How the model line bills an input image; shape and formulas in `./image-tokenization`. */
+	"image-tokenization": { key: "imageTokenization", set: "catalog", shape: "object" },
 	"image-model": { key: "imageModel", set: "catalog", shape: "scalar" },
 	"limits-patch": { key: "limitsPatch", set: "catalog", shape: "object" },
 	"long-context-cost": { key: "longContext", set: "catalog", shape: "object" },

package/src/compat/image-tokenization.ts

@@ -0,0 +1,166 @@
+/**
+ * Billed input tokens per image, from the `image-tokenization` rules in the
+ * compat cascade: a model line's lineage rule (`rules/classes/*.kdl`) applies
+ * on every host, and readers without one fall back to the rule for the wire
+ * API carrying the request (`rules/providers/image-tokenization.kdl`).
+ */
+import { isRecord } from "@oh-my-pi/pi-utils/type-guards";
+import { classifyModel } from "../identity";
+import { resolveCascade } from "./cascade";
+import type { ModelIdentity } from "./types";
+
+/** OpenAI image `detail` levels; `undefined` means the provider default (`auto`). */
+export type ImageDetail = "auto" | "low" | "high" | "original";
+
+export interface ImageSize {
+	width: number;
+	height: number;
+}
+
+/** One OpenAI `detail` level's sizing: a pixel-dimension limit, then an optional patch budget. */
+export interface PatchSizing {
+	/** Longest side after the pixel-dimension fit. */
+	maxEdge: number;
+	/** Resizing patch budget; absent when the level keeps the image's size. */
+	patchBudget?: number;
+}
+
+/** OpenAI 32px patches × a per-model multiplier, sized per `detail` level. */
+export interface OpenAiPatchTokenization {
+	regime: "openai-patch";
+	multiplier: number;
+	low: PatchSizing;
+	high: PatchSizing;
+	original: PatchSizing;
+	/** The level `auto` (and an omitted `detail`) sizes like. */
+	auto: "high" | "original";
+}
+
+/** One model line's image billing rule: the vendor formula and its per-model numbers. */
+export type ImageTokenization =
+	| OpenAiPatchTokenization
+	/** Anthropic 28px patches, resized to fit a padded-edge limit and a visual-token budget. */
+	| { regime: "anthropic-patch"; maxEdge: number; maxTokens: number }
+	/** A fixed per-image budget regardless of pixels (Gemini 3 `media_resolution`). */
+	| { regime: "fixed"; tokens: number };
+
+/** A `class: "unknown"` identity for targets resolved by wire API alone. */
+const UNCLASSIFIED: Pick<ModelIdentity, "class"> = { class: "unknown" };
+
+function isImageTokenization(value: unknown): value is ImageTokenization {
+	return (
+		isRecord(value) &&
+		(value.regime === "openai-patch" || value.regime === "anthropic-patch" || value.regime === "fixed")
+	);
+}
+
+/** What reads the image: a built model, any `{ id }` with optional host and identity, or just a wire `{ api }`. */
+export interface ImageTokenizationTarget {
+	id?: string;
+	provider?: string;
+	api?: string;
+	identity?: Pick<ModelIdentity, "class" | "family" | "revision">;
+}
+
+/**
+ * The reader's image billing rule: its lineage rule, else its wire API's
+ * fallback; undefined when neither covers it (an unclassified model on an
+ * unregistered API).
+ */
+export function resolveImageTokenization(target: ImageTokenizationTarget): ImageTokenization | undefined {
+	const provider = target.provider ?? "";
+	const model = target.id ?? "";
+	const identity: Pick<ModelIdentity, "class" | "family" | "revision"> =
+		target.identity ?? (model ? classifyModel(provider, model, { lenient: true }) : UNCLASSIFIED);
+	const rule = resolveCascade({
+		provider,
+		api: target.api ?? "",
+		class: identity.class,
+		model,
+		reasoning: false,
+		...(identity.family !== undefined && { family: identity.family }),
+		...(identity.revision !== undefined && { revision: identity.revision }),
+	}).catalog.imageTokenization;
+	return isImageTokenization(rule) ? rule : undefined;
+}
+
+/** Scale `size` down (never up) so its longest side is at most `maxEdge`. */
+function fitLongEdge(size: ImageSize, maxEdge: number): ImageSize {
+	const longest = Math.max(size.width, size.height);
+	if (longest <= maxEdge) return size;
+	const scale = maxEdge / longest;
+	return {
+		width: Math.max(1, Math.round(size.width * scale)),
+		height: Math.max(1, Math.round(size.height * scale)),
+	};
+}
+
+const OPENAI_PATCH_PX = 32;
+
+/** <https://developers.openai.com/api/docs/guides/images-vision#patch-based-image-tokenization> */
+function openAiPatches(size: ImageSize, sizing: PatchSizing): number {
+	const { width, height } = fitLongEdge(size, sizing.maxEdge);
+	const patches = Math.ceil(width / OPENAI_PATCH_PX) * Math.ceil(height / OPENAI_PATCH_PX);
+	const budget = sizing.patchBudget;
+	if (budget === undefined || patches <= budget) return patches;
+
+	const shrink = Math.sqrt((OPENAI_PATCH_PX * OPENAI_PATCH_PX * budget) / (width * height));
+	const scaledW = (width * shrink) / OPENAI_PATCH_PX;
+	const scaledH = (height * shrink) / OPENAI_PATCH_PX;
+	const adjusted = shrink * Math.min(Math.floor(scaledW) / scaledW, Math.floor(scaledH) / scaledH);
+	const resizedW = Math.floor(width * adjusted);
+	const resizedH = Math.floor(height * adjusted);
+	if (resizedW <= 0 || resizedH <= 0) return budget;
+	return Math.min(budget, Math.ceil(resizedW / OPENAI_PATCH_PX) * Math.ceil(resizedH / OPENAI_PATCH_PX));
+}
+
+const ANTHROPIC_PATCH_PX = 28;
+
+/** Round half to even, like the API's resize (and Python's `round`). */
+function roundTiesToEven(value: number): number {
+	const floor = Math.floor(value);
+	if (value - floor !== 0.5) return Math.round(value);
+	return floor % 2 === 0 ? floor : floor + 1;
+}
+
+/**
+ * Anthropic visual tokens after the API's resize: the largest
+ * aspect-preserving size whose 28px-padded edges fit `maxEdge` and whose
+ * patch count fits `maxTokens`, found by bisecting the long edge. Port of the
+ * reference implementation:
+ * <https://platform.claude.com/docs/en/build-with-claude/vision-coordinates#resize-your-image-before-uploading>
+ */
+function anthropicTokens(size: ImageSize, maxEdge: number, maxTokens: number): number {
+	const long = Math.max(size.width, size.height);
+	const shortSide = Math.min(size.width, size.height);
+	const aspect = long / shortSide;
+	const patches = (l: number, s: number) => Math.ceil(l / ANTHROPIC_PATCH_PX) * Math.ceil(s / ANTHROPIC_PATCH_PX);
+	const fits = (l: number, s: number) =>
+		Math.ceil(l / ANTHROPIC_PATCH_PX) * ANTHROPIC_PATCH_PX <= maxEdge &&
+		Math.ceil(s / ANTHROPIC_PATCH_PX) * ANTHROPIC_PATCH_PX <= maxEdge &&
+		patches(l, s) <= maxTokens;
+	const short = (l: number) => Math.max(roundTiesToEven(l / aspect), 1);
+	if (fits(long, shortSide)) return patches(long, shortSide);
+	let lo = 1; // always fits
+	let hi = long; // never fits
+	while (lo + 1 < hi) {
+		const mid = Math.floor((lo + hi) / 2);
+		if (fits(mid, short(mid))) lo = mid;
+		else hi = mid;
+	}
+	return patches(lo, short(lo));
+}
+
+/** Billed input tokens for one image of `size` under `rule`; only `openai-patch` reads `detail`. */
+export function imageTokens(rule: ImageTokenization, size: ImageSize, detail?: ImageDetail): number {
+	switch (rule.regime) {
+		case "fixed":
+			return rule.tokens;
+		case "openai-patch": {
+			const level = detail === "low" || detail === "high" || detail === "original" ? detail : rule.auto;
+			return Math.ceil(openAiPatches(size, rule[level]) * rule.multiplier);
+		}
+		case "anthropic-patch":
+			return anthropicTokens(size, rule.maxEdge, rule.maxTokens);
+	}
+}

package/src/compat/provider-ids.ts

@@ -66,6 +66,7 @@ export type KnownProvider =
 	| "siliconflow-cn"
 	| "singularityapi-dev"
 	| "singularityapi-tech"
+	| "snowflake"
 	| "stepfun"
 	| "synthetic"
 	| "together"

package/src/compat/resolve.ts

@@ -735,8 +735,15 @@ function resolveOpenAIResponsesPolicy(
 		supportsPromptCacheBreakpoints,
 		promptCacheBreakpointTtl: supportsPromptCacheBreakpoints ? "30m" : undefined,
 		strictResponsesPairing: isAzure || provider === "github-copilot",
-		supportsImageDetailOriginal: !isXaiHost && !modelMatchesHost(hostModel, "githubCopilot"),
+		// Azure's provider id alone only implies support while its endpoint is
+		// resolved at runtime; an explicit non-Azure baseUrl is a proxy, like Codex.
+		supportsImageDetailOriginal:
+			isOpenAIUrl ||
+			hostMatchesUrl(baseUrl, "azureOpenAI") ||
+			(isAzure && !baseUrl) ||
+			hostMatchesUrl(baseUrl, "openaiCodex"),
 		supportsReasoningSummary: !isXaiHost,
+		statefulResponses: undefined,
 		supportsAllTurnsReasoningContext: false,
 		supportsConfigurationUpdate: false,
 		supportsSteering: false,
@@ -784,6 +791,7 @@ function resolveOpenAIResponsesPolicy(
 			PROXY_OPENAI_COMPAT_PROVIDERS[backendProvider] !== true &&
 			(LOCAL_OPENAI_COMPAT_PROVIDERS[backendProvider] === true || hasLocalLoopbackBaseUrl(baseUrl)),
 		supportsObfuscationOptOut: isOpenAIUrl || provider === "openai",
+		storeResponses: false,
 		officialEndpoint: isOfficialOpenAIEndpoint(provider, baseUrl),
 		harmonyLeakMitigation: false,
 		rejectRootObjectUnion: false,
@@ -845,6 +853,7 @@ function pickResponsesOnly(compat: ResolvedOpenAIResponsesCompat): ResponsesOnly
 		strictResponsesPairing: compat.strictResponsesPairing,
 		supportsImageDetailOriginal: compat.supportsImageDetailOriginal,
 		supportsObfuscationOptOut: compat.supportsObfuscationOptOut,
+		storeResponses: compat.storeResponses,
 		supportsAllTurnsReasoningContext: compat.supportsAllTurnsReasoningContext,
 		supportsConfigurationUpdate: compat.supportsConfigurationUpdate,
 		supportsSteering: compat.supportsSteering,
@@ -853,6 +862,7 @@ function pickResponsesOnly(compat: ResolvedOpenAIResponsesCompat): ResponsesOnly
 		cacheControlFormat: compat.cacheControlFormat,
 		requiresReasoningOffJuiceInstruction: compat.requiresReasoningOffJuiceInstruction,
 		supportsReasoningSummary: compat.supportsReasoningSummary,
+		statefulResponses: compat.statefulResponses,
 		isVercelGatewayHost: compat.isVercelGatewayHost,
 	} satisfies ResponsesOnlyCompat;
 }

package/src/compat/rules/README.md

@@ -413,6 +413,10 @@ provider "muse-code" {
 
 Only `discovery` enrolls a provider in `generate-models.ts`; providers without it are never fetched at generation time (see the `charm-hyper` entry for why a live gateway deliberately omits it).
 
+`automatic-default #false` keeps a provider available for explicit selection but excludes it from startup fallback and automatic model presets. The default is `#true`; `apple` opts out because its on-device context cannot accommodate the standard coding-agent prompt in many projects.
+
+`kind-apis { <kind> "<api>" }` maps each non-chat kind (`image`, `tts`, `stt`, `embedding`, `rerank`, `video`) to the API discovery assigns rows of that kind. A runner API must sit under the kind it serves (`RUNNER_API_KINDS` in `src/types.ts`); chat APIs, which serve hosted image generation, and multi-kind `local-inference` may back any kind.
+
 ### Seed rows
 
 A `seed` _defines_ bundled rows for providers whose catalog cannot be discovered at generation time — credential-scoped rosters, unauthenticated regens, or models ahead of upstream catalogs. Every other stratum patches rows; this one authors them. Runtime model managers hand the rows to `staticModels` through `seedModels(provider)`; the generator bundles them per the seed's `bundle` policy. Values are literal — a seed never derives from another provider's row, and pricing is never borrowed.

package/src/compat/rules/auth/_order.kdl

@@ -8,6 +8,6 @@ login-order "openai-codex" "anthropic" "zai" "zai-coding-plan" "kimi-code" "open
 	"xiaomi-token-plan-ams" "xiaomi-token-plan-cn" "firepass" "cline-pass" "factory-droid" "commandcode" "charm-hyper" "deepseek" "stepfun" "muse-code" "meta" "moonshot" \
 	"cerebras" "baseten" "helmcode" "fireworks" "together" "nvidia" "novita" "deepinfra" "huggingface" "perplexity" \
 	"qianfan" "venice" "siliconflow" "siliconflow-cn" "synthetic" "nanogpt" "wafer-serverless" "coreweave" \
-	"vercel-ai-gateway" "cloudflare-ai-gateway" "litellm" "kilo" "zenmux" "opencode-zen" "opencode-go" \
+	"vercel-ai-gateway" "cloudflare-ai-gateway" "snowflake" "litellm" "kilo" "zenmux" "opencode-zen" "opencode-go" \
 	"yolo-auto" "tavily" "kagi" "exa" "parallel" "typesafe" "ollama" "ollama-cloud" "lm-studio" "llama.cpp" "vllm" \
 	"gmi-cloud" "stencil" "singularityapi-dev" "singularityapi-tech"

package/src/compat/rules/auth/minimax-code-cn.kdl

@@ -4,6 +4,6 @@ auth "minimax-code-cn" {
 		auth-url "https://platform.minimaxi.com/subscribe/token-plan"
 		instructions "Subscribe to Token Plan and copy your API key"
 		prompt "Paste your MiniMax Token Plan API key" placeholder="sk-..."
-		validate "chat-completions" base-url="https://api.minimaxi.com/v1" model="MiniMax-M3"
+		validate "anthropic-messages" base-url="https://api.minimaxi.com/anthropic" model="MiniMax-M3"
 	}
 }

package/src/compat/rules/auth/minimax-code.kdl

@@ -4,6 +4,6 @@ auth "minimax-code" {
 		auth-url "https://platform.minimax.io/subscribe/token-plan"
 		instructions "Subscribe to Token Plan and copy your API key"
 		prompt "Paste your MiniMax Token Plan API key" placeholder="sk-..."
-		validate "chat-completions" label="MiniMax Token Plan" base-url="https://api.minimax.io/v1" model="MiniMax-M3"
+		validate "anthropic-messages" label="MiniMax Token Plan" base-url="https://api.minimax.io/anthropic" model="MiniMax-M3"
 	}
 }

package/src/compat/rules/auth/snowflake.kdl

@@ -0,0 +1,11 @@
+// Snowflake Cortex REST API. Login runs Snowflake OAuth for local applications
+// (built-in SNOWFLAKE$LOCAL_APPLICATION client, PKCE, loopback); the account
+// host is prompted first, so the flow is a custom hook.
+auth "snowflake" {
+	name "Snowflake Cortex"
+	login "custom" hook="snowflake"
+	refresh hook="snowflake"
+	api-key-format "structured"
+	callback-port 54551
+	paste-code #true
+}