@shop-os/modelkit 0.1.8 → 0.1.10

package/dist/catalog.d.cts

@@ -171,6 +171,11 @@ interface ModelConstraints {
     durationPresets?: number[];
     /** Curated quality choices for rows whose `supports` includes "quality". */
     qualityOptions?: string[];
+    /**
+     * Exact pixel size per aspect ratio for rows an app must size verbatim
+     * (gpt-image-2 keeps its historic WxH contract); callers pass it as `exactSize`.
+     */
+    exactSizes?: Record<string, string>;
     references?: ReferenceLimits;
     requires?: RequiredInput[];
     supports?: SupportedOption[];
@@ -212,6 +217,20 @@ interface CatalogDisplay {
     costTier?: 1 | 2 | 3;
     /** Chat picker highlight flag. */
     recommended?: boolean;
+    /**
+     * Picker position: lower comes first; rows without one keep catalog order
+     * after the ordered ones. Set it to surface a new model at the top.
+     */
+    sortOrder?: number;
+    /**
+     * Chat-surface stream tuning for text rows. Display data only: the kit never
+     * merges it into a call (row `providerOptions` is merged into every consumer's
+     * calls, which is why this lives here instead).
+     */
+    chat?: {
+        providerOptions?: ProviderOptions;
+        temperature?: number;
+    };
 }
 /** Capability flags a UI keys off. Absent means false / unknown. */
 interface CatalogCapabilities {
@@ -223,7 +242,8 @@ interface CatalogCapabilities {
     acceptsSourceVideo?: boolean;
     /** Video rows: first/last frames and reference images are mutually exclusive inputs. */
     framesExcludeReferences?: boolean;
-    /** Text rows. */
+    /** Text rows. `reasoning`: the model streams reasoning the UI should surface. */
+    reasoning?: boolean;
     vision?: boolean;
     tools?: boolean;
     structuredOutput?: boolean;
@@ -247,10 +267,18 @@ interface CatalogModel {
     exposure: "catalog" | "internal";
     constraints: ModelConstraints;
     defaults: Partial<Omit<ImageRequest, "model" | "prompt"> & Omit<VideoRequest, "model" | "prompt"> & Omit<AudioRequest, "model" | "prompt">>;
+    /** Provider options merged under the caller's on every call. Chat-only tuning belongs in display.chat. */
     providerOptions?: ProviderOptions;
+    /**
+     * `slug` and `providerLabel` keep an app's historic usage-event names for a
+     * row (e.g. cowork emits `image_gen_<slug>` from provider `<providerLabel>`);
+     * absent means the app derives them from the id and display.provider.
+     */
     usage: {
         unit: UsageUnit;
         modality: UsageModality;
+        slug?: string;
+        providerLabel?: string;
     };
     pricingKind: PricingKind;
     /** Required on every `active` row so a picker can render it without a hand-written catalog. */

package/dist/catalog.d.ts

@@ -171,6 +171,11 @@ interface ModelConstraints {
     durationPresets?: number[];
     /** Curated quality choices for rows whose `supports` includes "quality". */
     qualityOptions?: string[];
+    /**
+     * Exact pixel size per aspect ratio for rows an app must size verbatim
+     * (gpt-image-2 keeps its historic WxH contract); callers pass it as `exactSize`.
+     */
+    exactSizes?: Record<string, string>;
     references?: ReferenceLimits;
     requires?: RequiredInput[];
     supports?: SupportedOption[];
@@ -212,6 +217,20 @@ interface CatalogDisplay {
     costTier?: 1 | 2 | 3;
     /** Chat picker highlight flag. */
     recommended?: boolean;
+    /**
+     * Picker position: lower comes first; rows without one keep catalog order
+     * after the ordered ones. Set it to surface a new model at the top.
+     */
+    sortOrder?: number;
+    /**
+     * Chat-surface stream tuning for text rows. Display data only: the kit never
+     * merges it into a call (row `providerOptions` is merged into every consumer's
+     * calls, which is why this lives here instead).
+     */
+    chat?: {
+        providerOptions?: ProviderOptions;
+        temperature?: number;
+    };
 }
 /** Capability flags a UI keys off. Absent means false / unknown. */
 interface CatalogCapabilities {
@@ -223,7 +242,8 @@ interface CatalogCapabilities {
     acceptsSourceVideo?: boolean;
     /** Video rows: first/last frames and reference images are mutually exclusive inputs. */
     framesExcludeReferences?: boolean;
-    /** Text rows. */
+    /** Text rows. `reasoning`: the model streams reasoning the UI should surface. */
+    reasoning?: boolean;
     vision?: boolean;
     tools?: boolean;
     structuredOutput?: boolean;
@@ -247,10 +267,18 @@ interface CatalogModel {
     exposure: "catalog" | "internal";
     constraints: ModelConstraints;
     defaults: Partial<Omit<ImageRequest, "model" | "prompt"> & Omit<VideoRequest, "model" | "prompt"> & Omit<AudioRequest, "model" | "prompt">>;
+    /** Provider options merged under the caller's on every call. Chat-only tuning belongs in display.chat. */
     providerOptions?: ProviderOptions;
+    /**
+     * `slug` and `providerLabel` keep an app's historic usage-event names for a
+     * row (e.g. cowork emits `image_gen_<slug>` from provider `<providerLabel>`);
+     * absent means the app derives them from the id and display.provider.
+     */
     usage: {
         unit: UsageUnit;
         modality: UsageModality;
+        slug?: string;
+        providerLabel?: string;
     };
     pricingKind: PricingKind;
     /** Required on every `active` row so a picker can render it without a hand-written catalog. */

package/dist/catalog.js

@@ -127,7 +127,7 @@ function geminiImage(id, label, aliases, sizes, display, exposure) {
       supports: ["safetySettings", "temperature"]
     },
     defaults: { size: "1K", aspectRatio: "1:1" },
-    usage: { ...IMAGE_USAGE },
+    usage: { ...IMAGE_USAGE, slug: "gemini" },
     pricingKind: "image_size",
     display,
     capabilities: { ...IMAGE_GENERATE_EDIT }
@@ -195,6 +195,15 @@ var IMAGE_MODELS = [
     lifecycle: "active",
     exposure: "catalog",
     constraints: {
+      exactSizes: {
+        "1:1": "1024x1024",
+        "3:2": "1536x1024",
+        "2:3": "1024x1536",
+        "16:9": "2048x1152",
+        "9:16": "1152x2048",
+        "4:5": "1024x1280",
+        "5:4": "1280x1024"
+      },
       aspectRatios: ["1:1", "3:2", "2:3", "16:9", "9:16", "4:5", "5:4", "adaptive"],
       sizes: ["1K", "2K", "4K"],
       qualityOptions: ["auto", "low", "medium", "high"],
@@ -202,7 +211,7 @@ var IMAGE_MODELS = [
       supports: ["quality", "count"]
     },
     defaults: { size: "1K", aspectRatio: "adaptive", quality: "auto" },
-    usage: { ...IMAGE_USAGE },
+    usage: { ...IMAGE_USAGE, slug: "gpt" },
     pricingKind: "image_size",
     display: {
       provider: "openai",
@@ -222,6 +231,15 @@ var IMAGE_MODELS = [
     lifecycle: "active",
     exposure: "internal",
     constraints: {
+      exactSizes: {
+        "1:1": "1024x1024",
+        "3:2": "1536x1024",
+        "2:3": "1024x1536",
+        "16:9": "2048x1152",
+        "9:16": "1152x2048",
+        "4:5": "1024x1280",
+        "5:4": "1280x1024"
+      },
       aspectRatios: ["1:1", "3:2", "2:3", "16:9", "9:16", "4:5", "5:4", "adaptive"],
       sizes: ["1K", "2K", "4K"],
       qualityOptions: ["auto", "low", "medium", "high"],
@@ -248,6 +266,15 @@ var IMAGE_MODELS = [
     lifecycle: "active",
     exposure: "internal",
     constraints: {
+      exactSizes: {
+        "1:1": "1024x1024",
+        "3:2": "1536x1024",
+        "2:3": "1024x1536",
+        "16:9": "2048x1152",
+        "9:16": "1152x2048",
+        "4:5": "1024x1280",
+        "5:4": "1280x1024"
+      },
       aspectRatios: ["1:1", "3:2", "2:3", "16:9", "9:16", "4:5", "5:4", "adaptive"],
       sizes: ["1K", "2K", "4K"],
       qualityOptions: ["auto", "low", "medium", "high"],
@@ -293,7 +320,7 @@ var IMAGE_MODELS = [
       supports: ["count", "outputFormat"]
     },
     defaults: { size: "1K", aspectRatio: "1:1", outputFormat: "png", count: 1 },
-    usage: { ...IMAGE_USAGE },
+    usage: { ...IMAGE_USAGE, slug: "muse" },
     pricingKind: "image_size",
     display: {
       provider: "meta",
@@ -324,6 +351,7 @@ var IMAGE_MODELS = [
     pricingKind: "image_size",
     display: {
       provider: "bytedance",
+      logo: "https://d31vvfpk2ppdyq.cloudfront.net/assets/image/3bba5d81-4e57-4e71-911b-cdfa05acca2c_compressed.png",
       description: "ByteDance Seedream 5.0 Pro text-to-image and multi-reference image-to-image (up to 10 inputs) with region editing; outputs 1K, 1.5K, or 2K."
     },
     capabilities: { ...IMAGE_GENERATE_EDIT }
@@ -529,6 +557,7 @@ var IMAGE_MODELS = [
 ];
 
 // src/catalog/other.ts
+var GOOGLE_LOGO = "https://d31vvfpk2ppdyq.cloudfront.net/assets/image/192485d2-a76a-4657-9c06-930b2c70badd_compressed.png";
 var OTHER_MODELS = [
   {
     id: "lyria-3-clip-preview",
@@ -549,6 +578,7 @@ var OTHER_MODELS = [
     pricingKind: "flat",
     display: {
       provider: "google",
+      logo: GOOGLE_LOGO,
       description: "Google Lyria 3 Clip short music clips, loops, and previews; takes text and up to 10 image prompts and returns a fixed 30-second MP3 plus lyrics."
     },
     capabilities: {}
@@ -571,6 +601,7 @@ var OTHER_MODELS = [
     pricingKind: "flat",
     display: {
       provider: "google",
+      logo: GOOGLE_LOGO,
       description: "Google Lyria 3 Pro flagship music model for full-length songs with verses, choruses, and bridges; takes text and up to 10 image prompts and returns MP3 plus lyrics."
     },
     capabilities: {}
@@ -710,6 +741,14 @@ function text(id, label, providerModelId, extra) {
     ...extra
   };
 }
+var ANTHROPIC_THINKING = {
+  anthropic: { thinking: { type: "enabled", budgetTokens: 4e3 }, sendReasoning: true }
+};
+var ANTHROPIC_ADAPTIVE = {
+  anthropic: { thinking: { type: "adaptive", display: "summarized" }, sendReasoning: true }
+};
+var OPENAI_REASONING = { openai: { reasoningEffort: "medium", reasoningSummary: "auto" } };
+var DEEPSEEK_THINKING = { deepseek: { thinking: { type: "enabled" } } };
 var TEXT_MODELS = [
   text("gpt-4o", "GPT-4o", "gpt-4o", {
     exposure: "internal",
@@ -790,11 +829,13 @@ var TEXT_MODELS = [
   text("gpt-5.6-luna", "GPT-5.6 Luna", "gpt-5.6-luna", {
     exposure: "catalog",
     display: {
+      chat: { providerOptions: OPENAI_REASONING },
       provider: "openai",
       costTier: 1,
       description: "OpenAI GPT-5.6 Luna cost-optimized tier of the GPT-5.6 reasoning family; text and image input, 1.05M-token context."
     },
     capabilities: {
+      reasoning: true,
       vision: true,
       tools: true,
       structuredOutput: true,
@@ -804,11 +845,13 @@ var TEXT_MODELS = [
   text("gpt-5.6-terra", "GPT-5.6 Terra", "gpt-5.6-terra", {
     exposure: "catalog",
     display: {
+      chat: { providerOptions: OPENAI_REASONING },
       provider: "openai",
       costTier: 2,
       description: "OpenAI GPT-5.6 Terra balanced tier of the GPT-5.6 reasoning family; text and image input, 1.05M-token context."
     },
     capabilities: {
+      reasoning: true,
       vision: true,
       tools: true,
       structuredOutput: true,
@@ -818,11 +861,13 @@ var TEXT_MODELS = [
   text("gpt-5.6-sol", "GPT-5.6 Sol", "gpt-5.6-sol", {
     exposure: "catalog",
     display: {
+      chat: { providerOptions: OPENAI_REASONING },
       provider: "openai",
       costTier: 3,
       description: "OpenAI GPT-5.6 Sol flagship tier of the GPT-5.6 reasoning family; text and image input, 1.05M-token context."
     },
     capabilities: {
+      reasoning: true,
       vision: true,
       tools: true,
       structuredOutput: true,
@@ -850,11 +895,13 @@ var TEXT_MODELS = [
       "claude-sonnet-4-20250514"
     ],
     display: {
+      chat: { providerOptions: ANTHROPIC_THINKING },
       provider: "anthropic",
       costTier: 2,
       description: "Anthropic Claude Sonnet 4.6 (legacy) balanced model; text and image input, 1M-token context."
     },
     capabilities: {
+      reasoning: true,
       vision: true,
       tools: true,
       structuredOutput: true,
@@ -864,11 +911,13 @@ var TEXT_MODELS = [
   text("claude-sonnet-5", "Claude Sonnet 5", "claude-sonnet-5", {
     exposure: "catalog",
     display: {
+      chat: { providerOptions: ANTHROPIC_ADAPTIVE },
       provider: "anthropic",
       costTier: 3,
       description: "Anthropic Claude Sonnet 5 current balanced model; text and image input, 1M-token context."
     },
     capabilities: {
+      reasoning: true,
       vision: true,
       tools: true,
       structuredOutput: true,
@@ -879,11 +928,13 @@ var TEXT_MODELS = [
     exposure: "catalog",
     aliases: ["claude-opus-4.8", "claude-opus-4", "claude-opus-4.0", "claude-opus-4-20250514"],
     display: {
+      chat: { providerOptions: ANTHROPIC_ADAPTIVE },
       provider: "anthropic",
       costTier: 3,
       description: "Anthropic Claude Opus 4.8 (legacy) frontier model; text and image input, 1M-token context."
     },
     capabilities: {
+      reasoning: true,
       vision: true,
       tools: true,
       structuredOutput: true,
@@ -894,11 +945,18 @@ var TEXT_MODELS = [
     exposure: "catalog",
     aliases: ["claude-haiku-4.5", "claude-haiku-4-5-20251001", "claude-haiku-4"],
     display: {
+      chat: { providerOptions: ANTHROPIC_THINKING },
       provider: "anthropic",
       costTier: 1,
       description: "Anthropic Claude Haiku 4.5 fast model with near-frontier intelligence; text and image input, 200K-token context."
     },
-    capabilities: { vision: true, tools: true, structuredOutput: true, contextWindow: 2e5 }
+    capabilities: {
+      reasoning: true,
+      vision: true,
+      tools: true,
+      structuredOutput: true,
+      contextWindow: 2e5
+    }
   }),
   text("claude-opus-5", "Claude Opus 5", "claude-opus-5", {
     exposure: "internal",
@@ -1090,15 +1148,21 @@ var TEXT_MODELS = [
       contextWindow: 128e3
     }
   }),
-  text("deepseek-v4-flash", "DeepSeek V4 Flash", "deepseek-v4-flash", {
+  // DeepSeek retired V4 Flash upstream and routes the old id to V4.1 Flash at
+  // Flash rates (https://api-docs.deepseek.com/news/news260910/); the old id
+  // stays an alias so stored selections keep resolving.
+  text("deepseek-flash", "DeepSeek V4.1 Flash", "deepseek-flash", {
+    aliases: ["deepseek-v4-flash"],
     exposure: "catalog",
     display: {
+      chat: { providerOptions: DEEPSEEK_THINKING },
       provider: "deepseek",
       costTier: 1,
       recommended: true,
-      description: "DeepSeek V4 Flash thinking/non-thinking model with vision, tool calls, and JSON output; 1M-token context and up to 384K output tokens."
+      description: "DeepSeek V4.1 Flash thinking/non-thinking model with vision, tool calls, and JSON output; 1M-token context and up to 384K output tokens."
     },
     capabilities: {
+      reasoning: true,
       vision: true,
       tools: true,
       structuredOutput: true,
@@ -1108,11 +1172,13 @@ var TEXT_MODELS = [
   text("deepseek-v4-pro", "DeepSeek V4 Pro", "deepseek-v4-pro", {
     exposure: "catalog",
     display: {
+      chat: { providerOptions: DEEPSEEK_THINKING },
       provider: "deepseek",
       costTier: 2,
       description: "DeepSeek V4 Pro thinking/non-thinking text model with tool calls and JSON output (no vision); 1M-token context and up to 384K output tokens."
     },
     capabilities: {
+      reasoning: true,
       vision: false,
       tools: true,
       structuredOutput: true,
@@ -1138,6 +1204,7 @@ var TEXT_MODELS = [
   text("kimi-k2.5", "Kimi K2.5", "kimi-k2.5", {
     exposure: "catalog",
     display: {
+      chat: { temperature: 1 },
       provider: "moonshot",
       costTier: 2,
       description: "Moonshot Kimi K2.5; the provider retired the model on 2026-08-31 and calls now return a model-not-found error."
@@ -1147,11 +1214,13 @@ var TEXT_MODELS = [
   text("kimi-k3", "Kimi K3", "kimi-k3", {
     exposure: "catalog",
     display: {
+      chat: { temperature: 1 },
       provider: "moonshot",
       costTier: 2,
       description: "Moonshot Kimi K3 flagship model with native visual understanding and always-on thinking; text, image, and video input, 1M-token context."
     },
     capabilities: {
+      reasoning: true,
       vision: true,
       tools: true,
       structuredOutput: true,
@@ -1182,7 +1251,8 @@ var H3_RATIOS = ["21:9", "16:9", "4:3", "1:1", "3:4", "9:16"];
 var H3_REFERENCE_RATIOS = ["adaptive", ...H3_RATIOS];
 var NEGATIVE_PROMPT = "blur, distort, and low quality";
 var FRAMES_EXCLUDE_REFERENCES = { framesExcludeReferences: true };
-var GOOGLE_LOGO = "https://d31vvfpk2ppdyq.cloudfront.net/assets/image/192485d2-a76a-4657-9c06-930b2c70badd_compressed.png";
+var BYTEDANCE_LOGO = "https://d31vvfpk2ppdyq.cloudfront.net/assets/image/3bba5d81-4e57-4e71-911b-cdfa05acca2c_compressed.png";
+var GOOGLE_LOGO2 = "https://d31vvfpk2ppdyq.cloudfront.net/assets/image/192485d2-a76a-4657-9c06-930b2c70badd_compressed.png";
 var VIDEO_MODELS = [
   {
     id: "veo-3.1",
@@ -1206,12 +1276,12 @@ var VIDEO_MODELS = [
       supports: ["lastFrame"]
     },
     defaults: { durationSeconds: 8, resolution: "720p", aspectRatio: "16:9" },
-    usage: { ...VIDEO_USAGE },
+    usage: { ...VIDEO_USAGE, slug: "veo" },
     pricingKind: "video_per_second",
     display: {
       provider: "google",
       description: "Google's highest-quality video generation model. Supports up to 4K resolution and 8-second clips with reference image conditioning.",
-      logo: GOOGLE_LOGO
+      logo: GOOGLE_LOGO2
     },
     capabilities: { ...FRAMES_EXCLUDE_REFERENCES }
   },
@@ -1298,12 +1368,12 @@ var VIDEO_MODELS = [
       supports: []
     },
     defaults: {},
-    usage: { ...VIDEO_USAGE },
+    usage: { ...VIDEO_USAGE, slug: "gemini_omni" },
     pricingKind: "video_per_second",
     display: {
       provider: "google",
       description: "Google's conversational video model with native audio, human support, and natural language editing. Aspect ratio only \u2014 resolution and duration are model-determined.",
-      logo: GOOGLE_LOGO
+      logo: GOOGLE_LOGO2
     },
     capabilities: { acceptsSourceVideo: true }
   },
@@ -1328,6 +1398,7 @@ var VIDEO_MODELS = [
     pricingKind: "video_per_second",
     display: {
       provider: "google",
+      logo: GOOGLE_LOGO2,
       description: "Google Gemini Omni Flash 1.1 conversational video generation and editing; text, image, and up to 10-second video input, 3-10 second output at up to 4K and 24 FPS."
     },
     capabilities: { acceptsSourceVideo: true }
@@ -1357,12 +1428,12 @@ var VIDEO_MODELS = [
       aspectRatio: "16:9",
       generateAudio: true
     },
-    usage: { ...VIDEO_USAGE },
+    usage: { ...VIDEO_USAGE, slug: "seedance2", providerLabel: "byteplus" },
     pricingKind: "video_per_second",
     display: {
       provider: "bytedance",
       description: "High-quality AI video generation with reference image conditioning. Supports 4-15 second clips with human preprocessing.",
-      logo: "https://d31vvfpk2ppdyq.cloudfront.net/assets/image/3bba5d81-4e57-4e71-911b-cdfa05acca2c_compressed.png"
+      logo: BYTEDANCE_LOGO
     },
     capabilities: { acceptsSourceVideo: true, ...FRAMES_EXCLUDE_REFERENCES }
   },
@@ -1393,6 +1464,7 @@ var VIDEO_MODELS = [
     pricingKind: "video_per_second",
     display: {
       provider: "bytedance",
+      logo: BYTEDANCE_LOGO,
       description: "Speed- and cost-optimized ByteDance Seedance 2.0 Fast; text plus image, video, and audio references to generate or edit 4-15 second clips."
     },
     capabilities: { acceptsSourceVideo: true, ...FRAMES_EXCLUDE_REFERENCES }
@@ -1425,6 +1497,7 @@ var VIDEO_MODELS = [
     pricingKind: "video_per_second",
     display: {
       provider: "bytedance",
+      logo: BYTEDANCE_LOGO,
       description: "ByteDance Seedance 2.5 for long-form storytelling and omni-reference generation/editing; up to 50 reference assets and 4-30 second clips at up to 1080p."
     },
     capabilities: { acceptsSourceVideo: true, ...FRAMES_EXCLUDE_REFERENCES }
@@ -1449,7 +1522,7 @@ var VIDEO_MODELS = [
     },
     // No resolution default: xAI picks its own unless the caller asks.
     defaults: { durationSeconds: 8, aspectRatio: "16:9" },
-    usage: { unit: "reported_cost_usd", modality: "VIDEO" },
+    usage: { unit: "reported_cost_usd", modality: "VIDEO", slug: "grok" },
     pricingKind: "video_per_second",
     display: {
       provider: "xai",
@@ -1930,6 +2003,40 @@ function resolveVideoOptions(model, request, onUnsupported = "error") {
         model: model.id
       });
   }
+  if (merged.sourceVideo !== void 0 && !model.capabilities?.acceptsSourceVideo) {
+    if (onUnsupported === "error") {
+      throw new ModelKitError({
+        code: "unsupported_option",
+        message: `${model.id} does not take a source video`,
+        model: model.id
+      });
+    }
+    warnings.push({
+      code: "option_ignored",
+      field: "sourceVideo",
+      message: `${model.id} does not take a source video; dropped`
+    });
+    delete merged.sourceVideo;
+  }
+  const hasFrames = merged.firstFrame !== void 0 || merged.lastFrame !== void 0;
+  const hasRefs = compact(merged.referenceImages).length + compact(merged.referenceVideos).length + compact(merged.referenceAudios).length > 0;
+  if (hasFrames && hasRefs && model.capabilities?.framesExcludeReferences) {
+    if (onUnsupported === "error") {
+      throw new ModelKitError({
+        code: "unsupported_option",
+        message: `${model.id} takes frames or references, not both`,
+        model: model.id
+      });
+    }
+    warnings.push({
+      code: "option_ignored",
+      field: "referenceImages",
+      message: `${model.id} takes frames or references, not both; references dropped`
+    });
+    delete merged.referenceImages;
+    delete merged.referenceVideos;
+    delete merged.referenceAudios;
+  }
   for (const input of c.ratioIgnoredWith ?? []) {
     if (merged[input] !== void 0 && merged.aspectRatio !== void 0) {
       if (request.aspectRatio !== void 0) {