@tanstack/ai-grok 0.8.4 → 0.8.6

package/dist/esm/adapters/image.js

@@ -5,10 +5,11 @@ import { generateId } from "@tanstack/ai-utils";
 import { withGrokDefaults, getGrokApiKeyFromEnv } from "../utils/client.js";
 import { validatePrompt, validateImageSize, validateNumberOfImages } from "../image/image-provider-options.js";
 class GrokImageAdapter extends BaseImageAdapter {
+  kind = "image";
+  name = "grok";
+  client;
   constructor(config, model) {
     super(model, {});
-    this.kind = "image";
-    this.name = "grok";
     this.client = new OpenAI(withGrokDefaults(config));
   }
   async generateImages(options) {
@@ -16,11 +17,13 @@ class GrokImageAdapter extends BaseImageAdapter {
     validatePrompt({ prompt });
     validateImageSize(model, size);
     validateNumberOfImages(model, numberOfImages);
+    const resolvedSize = size;
     const request = {
       model,
       prompt,
       n: numberOfImages ?? 1,
-      size,
+      ...resolvedSize !== void 0 && { size: resolvedSize },
+      stream: false,
       ...modelOptions
     };
     try {
@@ -28,18 +31,25 @@ class GrokImageAdapter extends BaseImageAdapter {
         `activity=image provider=${this.name} model=${model} n=${request.n ?? 1} size=${request.size ?? "default"}`,
         { provider: this.name, model }
       );
-      const response = await this.client.images.generate({
-        ...request,
-        stream: false
-      });
+      const response = await this.client.images.generate(request);
       const images = (response.data ?? []).flatMap(
         (item) => {
           const revisedPrompt = item.revised_prompt;
           if (item.b64_json) {
-            return [{ b64Json: item.b64_json, revisedPrompt }];
+            return [
+              {
+                b64Json: item.b64_json,
+                ...revisedPrompt !== void 0 && { revisedPrompt }
+              }
+            ];
           }
           if (item.url) {
-            return [{ url: item.url, revisedPrompt }];
+            return [
+              {
+                url: item.url,
+                ...revisedPrompt !== void 0 && { revisedPrompt }
+              }
+            ];
           }
           return [];
         }
@@ -48,11 +58,13 @@ class GrokImageAdapter extends BaseImageAdapter {
         id: generateId(this.name),
         model,
         images,
-        usage: response.usage ? {
-          inputTokens: response.usage.input_tokens,
-          outputTokens: response.usage.output_tokens,
-          totalTokens: response.usage.total_tokens
-        } : void 0
+        ...response.usage && {
+          usage: {
+            inputTokens: response.usage.input_tokens,
+            outputTokens: response.usage.output_tokens,
+            totalTokens: response.usage.total_tokens
+          }
+        }
       };
     } catch (error) {
       options.logger.errors(`${this.name}.generateImages fatal`, {

package/dist/esm/adapters/image.js.map

@@ -1 +1 @@
-{"version":3,"file":"image.js","sources":["../../../src/adapters/image.ts"],"sourcesContent":["import OpenAI from 'openai'\nimport { BaseImageAdapter } from '@tanstack/ai/adapters'\nimport { toRunErrorPayload } from '@tanstack/ai/adapter-internals'\nimport { generateId } from '@tanstack/ai-utils'\nimport { getGrokApiKeyFromEnv, withGrokDefaults } from '../utils/client'\nimport {\n  validateImageSize,\n  validateNumberOfImages,\n  validatePrompt,\n} from '../image/image-provider-options'\nimport type {\n  GeneratedImage,\n  ImageGenerationOptions,\n  ImageGenerationResult,\n} from '@tanstack/ai'\nimport type OpenAI_SDK from 'openai'\nimport type { GrokImageModel } from '../model-meta'\nimport type {\n  GrokImageModelProviderOptionsByName,\n  GrokImageModelSizeByName,\n  GrokImageProviderOptions,\n} from '../image/image-provider-options'\nimport type { GrokClientConfig } from '../utils'\n\n/**\n * Configuration for Grok image adapter\n */\nexport interface GrokImageConfig extends GrokClientConfig {}\n\n/**\n * Grok Image Generation Adapter\n *\n * Tree-shakeable adapter for Grok image generation functionality.\n * Supports grok-2-image-1212 model.\n *\n * Features:\n * - Model-specific type-safe provider options\n * - Size validation per model\n * - Number of images validation\n */\nexport class GrokImageAdapter<\n  TModel extends GrokImageModel,\n> extends BaseImageAdapter<\n  TModel,\n  GrokImageProviderOptions,\n  GrokImageModelProviderOptionsByName,\n  GrokImageModelSizeByName\n> {\n  readonly kind = 'image' as const\n  readonly name = 'grok' as const\n\n  protected client: OpenAI\n\n  constructor(config: GrokImageConfig, model: TModel) {\n    super(model, {})\n    this.client = new OpenAI(withGrokDefaults(config))\n  }\n\n  async generateImages(\n    options: ImageGenerationOptions<GrokImageProviderOptions>,\n  ): Promise<ImageGenerationResult> {\n    const { model, prompt, numberOfImages, size, modelOptions } = options\n\n    validatePrompt({ prompt, model })\n    validateImageSize(model, size)\n    validateNumberOfImages(model, numberOfImages)\n\n    const request: OpenAI_SDK.Images.ImageGenerateParams = {\n      model,\n      prompt,\n      n: numberOfImages ?? 1,\n      size: size as OpenAI_SDK.Images.ImageGenerateParams['size'],\n      ...modelOptions,\n    }\n\n    try {\n      options.logger.request(\n        `activity=image provider=${this.name} model=${model} n=${request.n ?? 1} size=${request.size ?? 'default'}`,\n        { provider: this.name, model },\n      )\n      const response = await this.client.images.generate({\n        ...request,\n        stream: false,\n      })\n\n      const images: Array<GeneratedImage> = (response.data ?? []).flatMap(\n        (item): Array<GeneratedImage> => {\n          const revisedPrompt = item.revised_prompt\n          if (item.b64_json) {\n            return [{ b64Json: item.b64_json, revisedPrompt }]\n          }\n          if (item.url) {\n            return [{ url: item.url, revisedPrompt }]\n          }\n          return []\n        },\n      )\n\n      return {\n        id: generateId(this.name),\n        model,\n        images,\n        usage: response.usage\n          ? {\n              inputTokens: response.usage.input_tokens,\n              outputTokens: response.usage.output_tokens,\n              totalTokens: response.usage.total_tokens,\n            }\n          : undefined,\n      }\n    } catch (error: unknown) {\n      options.logger.errors(`${this.name}.generateImages fatal`, {\n        error: toRunErrorPayload(error, `${this.name}.generateImages failed`),\n        source: `${this.name}.generateImages`,\n      })\n      throw error\n    }\n  }\n}\n\n/**\n * Creates a Grok image adapter with explicit API key.\n * Type resolution happens here at the call site.\n *\n * @param model - The model name (e.g., 'grok-2-image-1212')\n * @param apiKey - Your xAI API key\n * @param config - Optional additional configuration\n * @returns Configured Grok image adapter instance with resolved types\n *\n * @example\n * ```typescript\n * const adapter = createGrokImage('grok-2-image-1212', \"xai-...\");\n *\n * const result = await generateImage({\n *   adapter,\n *   prompt: 'A cute baby sea otter'\n * });\n * ```\n */\nexport function createGrokImage<TModel extends GrokImageModel>(\n  model: TModel,\n  apiKey: string,\n  config?: Omit<GrokImageConfig, 'apiKey'>,\n): GrokImageAdapter<TModel> {\n  return new GrokImageAdapter({ apiKey, ...config }, model)\n}\n\n/**\n * Creates a Grok image adapter with automatic API key detection from environment variables.\n * Type resolution happens here at the call site.\n *\n * Looks for `XAI_API_KEY` in:\n * - `process.env` (Node.js)\n * - `window.env` (Browser with injected env)\n *\n * @param model - The model name (e.g., 'grok-2-image-1212')\n * @param config - Optional configuration (excluding apiKey which is auto-detected)\n * @returns Configured Grok image adapter instance with resolved types\n * @throws Error if XAI_API_KEY is not found in environment\n *\n * @example\n * ```typescript\n * // Automatically uses XAI_API_KEY from environment\n * const adapter = grokImage('grok-2-image-1212');\n *\n * const result = await generateImage({\n *   adapter,\n *   prompt: 'A beautiful sunset over mountains'\n * });\n * ```\n */\nexport function grokImage<TModel extends GrokImageModel>(\n  model: TModel,\n  config?: Omit<GrokImageConfig, 'apiKey'>,\n): GrokImageAdapter<TModel> {\n  const apiKey = getGrokApiKeyFromEnv()\n  return createGrokImage(model, apiKey, config)\n}\n"],"names":[],"mappings":";;;;;;AAwCO,MAAM,yBAEH,iBAKR;AAAA,EAMA,YAAY,QAAyB,OAAe;AAClD,UAAM,OAAO,EAAE;AANjB,SAAS,OAAO;AAChB,SAAS,OAAO;AAMd,SAAK,SAAS,IAAI,OAAO,iBAAiB,MAAM,CAAC;AAAA,EACnD;AAAA,EAEA,MAAM,eACJ,SACgC;AAChC,UAAM,EAAE,OAAO,QAAQ,gBAAgB,MAAM,iBAAiB;AAE9D,mBAAe,EAAE,OAAc,CAAC;AAChC,sBAAkB,OAAO,IAAI;AAC7B,2BAAuB,OAAO,cAAc;AAE5C,UAAM,UAAiD;AAAA,MACrD;AAAA,MACA;AAAA,MACA,GAAG,kBAAkB;AAAA,MACrB;AAAA,MACA,GAAG;AAAA,IAAA;AAGL,QAAI;AACF,cAAQ,OAAO;AAAA,QACb,2BAA2B,KAAK,IAAI,UAAU,KAAK,MAAM,QAAQ,KAAK,CAAC,SAAS,QAAQ,QAAQ,SAAS;AAAA,QACzG,EAAE,UAAU,KAAK,MAAM,MAAA;AAAA,MAAM;AAE/B,YAAM,WAAW,MAAM,KAAK,OAAO,OAAO,SAAS;AAAA,QACjD,GAAG;AAAA,QACH,QAAQ;AAAA,MAAA,CACT;AAED,YAAM,UAAiC,SAAS,QAAQ,CAAA,GAAI;AAAA,QAC1D,CAAC,SAAgC;AAC/B,gBAAM,gBAAgB,KAAK;AAC3B,cAAI,KAAK,UAAU;AACjB,mBAAO,CAAC,EAAE,SAAS,KAAK,UAAU,eAAe;AAAA,UACnD;AACA,cAAI,KAAK,KAAK;AACZ,mBAAO,CAAC,EAAE,KAAK,KAAK,KAAK,eAAe;AAAA,UAC1C;AACA,iBAAO,CAAA;AAAA,QACT;AAAA,MAAA;AAGF,aAAO;AAAA,QACL,IAAI,WAAW,KAAK,IAAI;AAAA,QACxB;AAAA,QACA;AAAA,QACA,OAAO,SAAS,QACZ;AAAA,UACE,aAAa,SAAS,MAAM;AAAA,UAC5B,cAAc,SAAS,MAAM;AAAA,UAC7B,aAAa,SAAS,MAAM;AAAA,QAAA,IAE9B;AAAA,MAAA;AAAA,IAER,SAAS,OAAgB;AACvB,cAAQ,OAAO,OAAO,GAAG,KAAK,IAAI,yBAAyB;AAAA,QACzD,OAAO,kBAAkB,OAAO,GAAG,KAAK,IAAI,wBAAwB;AAAA,QACpE,QAAQ,GAAG,KAAK,IAAI;AAAA,MAAA,CACrB;AACD,YAAM;AAAA,IACR;AAAA,EACF;AACF;AAqBO,SAAS,gBACd,OACA,QACA,QAC0B;AAC1B,SAAO,IAAI,iBAAiB,EAAE,QAAQ,GAAG,OAAA,GAAU,KAAK;AAC1D;AA0BO,SAAS,UACd,OACA,QAC0B;AAC1B,QAAM,SAAS,qBAAA;AACf,SAAO,gBAAgB,OAAO,QAAQ,MAAM;AAC9C;"}
+{"version":3,"file":"image.js","sources":["../../../src/adapters/image.ts"],"sourcesContent":["import OpenAI from 'openai'\nimport { BaseImageAdapter } from '@tanstack/ai/adapters'\nimport { toRunErrorPayload } from '@tanstack/ai/adapter-internals'\nimport { generateId } from '@tanstack/ai-utils'\nimport { getGrokApiKeyFromEnv, withGrokDefaults } from '../utils/client'\nimport {\n  validateImageSize,\n  validateNumberOfImages,\n  validatePrompt,\n} from '../image/image-provider-options'\nimport type {\n  GeneratedImage,\n  ImageGenerationOptions,\n  ImageGenerationResult,\n} from '@tanstack/ai'\nimport type OpenAI_SDK from 'openai'\nimport type { GrokImageModel } from '../model-meta'\nimport type {\n  GrokImageModelProviderOptionsByName,\n  GrokImageModelSizeByName,\n  GrokImageProviderOptions,\n} from '../image/image-provider-options'\nimport type { GrokClientConfig } from '../utils'\n\n/**\n * Configuration for Grok image adapter\n */\nexport interface GrokImageConfig extends GrokClientConfig {}\n\n/**\n * Grok Image Generation Adapter\n *\n * Tree-shakeable adapter for Grok image generation functionality.\n * Supports grok-2-image-1212 model.\n *\n * Features:\n * - Model-specific type-safe provider options\n * - Size validation per model\n * - Number of images validation\n */\nexport class GrokImageAdapter<\n  TModel extends GrokImageModel,\n> extends BaseImageAdapter<\n  TModel,\n  GrokImageProviderOptions,\n  GrokImageModelProviderOptionsByName,\n  GrokImageModelSizeByName\n> {\n  override readonly kind = 'image' as const\n  readonly name = 'grok' as const\n\n  protected client: OpenAI\n\n  constructor(config: GrokImageConfig, model: TModel) {\n    super(model, {})\n    this.client = new OpenAI(withGrokDefaults(config))\n  }\n\n  async generateImages(\n    options: ImageGenerationOptions<GrokImageProviderOptions>,\n  ): Promise<ImageGenerationResult> {\n    const { model, prompt, numberOfImages, size, modelOptions } = options\n\n    validatePrompt({ prompt, model })\n    validateImageSize(model, size)\n    validateNumberOfImages(model, numberOfImages)\n\n    const resolvedSize = size as OpenAI_SDK.Images.ImageGenerateParams['size']\n    const request: OpenAI_SDK.Images.ImageGenerateParamsNonStreaming = {\n      model,\n      prompt,\n      n: numberOfImages ?? 1,\n      ...(resolvedSize !== undefined && { size: resolvedSize }),\n      stream: false,\n      ...modelOptions,\n    }\n\n    try {\n      options.logger.request(\n        `activity=image provider=${this.name} model=${model} n=${request.n ?? 1} size=${request.size ?? 'default'}`,\n        { provider: this.name, model },\n      )\n      const response = await this.client.images.generate(request)\n\n      const images: Array<GeneratedImage> = (response.data ?? []).flatMap(\n        (item): Array<GeneratedImage> => {\n          const revisedPrompt = item.revised_prompt\n          if (item.b64_json) {\n            return [\n              {\n                b64Json: item.b64_json,\n                ...(revisedPrompt !== undefined && { revisedPrompt }),\n              },\n            ]\n          }\n          if (item.url) {\n            return [\n              {\n                url: item.url,\n                ...(revisedPrompt !== undefined && { revisedPrompt }),\n              },\n            ]\n          }\n          return []\n        },\n      )\n\n      return {\n        id: generateId(this.name),\n        model,\n        images,\n        ...(response.usage && {\n          usage: {\n            inputTokens: response.usage.input_tokens,\n            outputTokens: response.usage.output_tokens,\n            totalTokens: response.usage.total_tokens,\n          },\n        }),\n      }\n    } catch (error: unknown) {\n      options.logger.errors(`${this.name}.generateImages fatal`, {\n        error: toRunErrorPayload(error, `${this.name}.generateImages failed`),\n        source: `${this.name}.generateImages`,\n      })\n      throw error\n    }\n  }\n}\n\n/**\n * Creates a Grok image adapter with explicit API key.\n * Type resolution happens here at the call site.\n *\n * @param model - The model name (e.g., 'grok-2-image-1212')\n * @param apiKey - Your xAI API key\n * @param config - Optional additional configuration\n * @returns Configured Grok image adapter instance with resolved types\n *\n * @example\n * ```typescript\n * const adapter = createGrokImage('grok-2-image-1212', \"xai-...\");\n *\n * const result = await generateImage({\n *   adapter,\n *   prompt: 'A cute baby sea otter'\n * });\n * ```\n */\nexport function createGrokImage<TModel extends GrokImageModel>(\n  model: TModel,\n  apiKey: string,\n  config?: Omit<GrokImageConfig, 'apiKey'>,\n): GrokImageAdapter<TModel> {\n  return new GrokImageAdapter({ apiKey, ...config }, model)\n}\n\n/**\n * Creates a Grok image adapter with automatic API key detection from environment variables.\n * Type resolution happens here at the call site.\n *\n * Looks for `XAI_API_KEY` in:\n * - `process.env` (Node.js)\n * - `window.env` (Browser with injected env)\n *\n * @param model - The model name (e.g., 'grok-2-image-1212')\n * @param config - Optional configuration (excluding apiKey which is auto-detected)\n * @returns Configured Grok image adapter instance with resolved types\n * @throws Error if XAI_API_KEY is not found in environment\n *\n * @example\n * ```typescript\n * // Automatically uses XAI_API_KEY from environment\n * const adapter = grokImage('grok-2-image-1212');\n *\n * const result = await generateImage({\n *   adapter,\n *   prompt: 'A beautiful sunset over mountains'\n * });\n * ```\n */\nexport function grokImage<TModel extends GrokImageModel>(\n  model: TModel,\n  config?: Omit<GrokImageConfig, 'apiKey'>,\n): GrokImageAdapter<TModel> {\n  const apiKey = getGrokApiKeyFromEnv()\n  return createGrokImage(model, apiKey, config)\n}\n"],"names":[],"mappings":";;;;;;AAwCO,MAAM,yBAEH,iBAKR;AAAA,EACkB,OAAO;AAAA,EAChB,OAAO;AAAA,EAEN;AAAA,EAEV,YAAY,QAAyB,OAAe;AAClD,UAAM,OAAO,EAAE;AACf,SAAK,SAAS,IAAI,OAAO,iBAAiB,MAAM,CAAC;AAAA,EACnD;AAAA,EAEA,MAAM,eACJ,SACgC;AAChC,UAAM,EAAE,OAAO,QAAQ,gBAAgB,MAAM,iBAAiB;AAE9D,mBAAe,EAAE,OAAc,CAAC;AAChC,sBAAkB,OAAO,IAAI;AAC7B,2BAAuB,OAAO,cAAc;AAE5C,UAAM,eAAe;AACrB,UAAM,UAA6D;AAAA,MACjE;AAAA,MACA;AAAA,MACA,GAAG,kBAAkB;AAAA,MACrB,GAAI,iBAAiB,UAAa,EAAE,MAAM,aAAA;AAAA,MAC1C,QAAQ;AAAA,MACR,GAAG;AAAA,IAAA;AAGL,QAAI;AACF,cAAQ,OAAO;AAAA,QACb,2BAA2B,KAAK,IAAI,UAAU,KAAK,MAAM,QAAQ,KAAK,CAAC,SAAS,QAAQ,QAAQ,SAAS;AAAA,QACzG,EAAE,UAAU,KAAK,MAAM,MAAA;AAAA,MAAM;AAE/B,YAAM,WAAW,MAAM,KAAK,OAAO,OAAO,SAAS,OAAO;AAE1D,YAAM,UAAiC,SAAS,QAAQ,CAAA,GAAI;AAAA,QAC1D,CAAC,SAAgC;AAC/B,gBAAM,gBAAgB,KAAK;AAC3B,cAAI,KAAK,UAAU;AACjB,mBAAO;AAAA,cACL;AAAA,gBACE,SAAS,KAAK;AAAA,gBACd,GAAI,kBAAkB,UAAa,EAAE,cAAA;AAAA,cAAc;AAAA,YACrD;AAAA,UAEJ;AACA,cAAI,KAAK,KAAK;AACZ,mBAAO;AAAA,cACL;AAAA,gBACE,KAAK,KAAK;AAAA,gBACV,GAAI,kBAAkB,UAAa,EAAE,cAAA;AAAA,cAAc;AAAA,YACrD;AAAA,UAEJ;AACA,iBAAO,CAAA;AAAA,QACT;AAAA,MAAA;AAGF,aAAO;AAAA,QACL,IAAI,WAAW,KAAK,IAAI;AAAA,QACxB;AAAA,QACA;AAAA,QACA,GAAI,SAAS,SAAS;AAAA,UACpB,OAAO;AAAA,YACL,aAAa,SAAS,MAAM;AAAA,YAC5B,cAAc,SAAS,MAAM;AAAA,YAC7B,aAAa,SAAS,MAAM;AAAA,UAAA;AAAA,QAC9B;AAAA,MACF;AAAA,IAEJ,SAAS,OAAgB;AACvB,cAAQ,OAAO,OAAO,GAAG,KAAK,IAAI,yBAAyB;AAAA,QACzD,OAAO,kBAAkB,OAAO,GAAG,KAAK,IAAI,wBAAwB;AAAA,QACpE,QAAQ,GAAG,KAAK,IAAI;AAAA,MAAA,CACrB;AACD,YAAM;AAAA,IACR;AAAA,EACF;AACF;AAqBO,SAAS,gBACd,OACA,QACA,QAC0B;AAC1B,SAAO,IAAI,iBAAiB,EAAE,QAAQ,GAAG,OAAA,GAAU,KAAK;AAC1D;AA0BO,SAAS,UACd,OACA,QAC0B;AAC1B,QAAM,SAAS,qBAAA;AACf,SAAO,gBAAgB,OAAO,QAAQ,MAAM;AAC9C;"}

package/dist/esm/adapters/text.js

@@ -2,10 +2,10 @@ import OpenAI from "openai";
 import { OpenAIBaseChatCompletionsTextAdapter } from "@tanstack/openai-base";
 import { withGrokDefaults, getGrokApiKeyFromEnv } from "../utils/client.js";
 class GrokTextAdapter extends OpenAIBaseChatCompletionsTextAdapter {
+  kind = "text";
+  name = "grok";
   constructor(config, model) {
     super(model, "grok", new OpenAI(withGrokDefaults(config)));
-    this.kind = "text";
-    this.name = "grok";
   }
   /**
    * Surfaces xAI reasoning deltas on Grok reasoning models. The DeepSeek-style

package/dist/esm/adapters/text.js.map

@@ -1 +1 @@
-{"version":3,"file":"text.js","sources":["../../../src/adapters/text.ts"],"sourcesContent":["import OpenAI from 'openai'\nimport { OpenAIBaseChatCompletionsTextAdapter } from '@tanstack/openai-base'\nimport { getGrokApiKeyFromEnv, withGrokDefaults } from '../utils/client'\nimport type {\n  GROK_CHAT_MODELS,\n  GrokChatModelToolCapabilitiesByName,\n  ResolveInputModalities,\n  ResolveProviderOptions,\n} from '../model-meta'\nimport type { Modality } from '@tanstack/ai'\nimport type { GrokMessageMetadataByModality } from '../message-types'\nimport type { GrokClientConfig } from '../utils'\n\n/**\n * Resolve tool capabilities for a specific Grok model.\n */\ntype ResolveToolCapabilities<TModel extends string> =\n  TModel extends keyof GrokChatModelToolCapabilitiesByName\n    ? NonNullable<GrokChatModelToolCapabilitiesByName[TModel]>\n    : readonly []\n\n/**\n * Configuration for Grok text adapter\n */\nexport interface GrokTextConfig extends GrokClientConfig {}\n\n/**\n * Alias for TextProviderOptions for external use\n */\nexport type { ExternalTextProviderOptions as GrokTextProviderOptions } from '../text/text-provider-options'\n\n/**\n * Grok Text (Chat) Adapter\n *\n * Tree-shakeable adapter for Grok chat/text completion functionality.\n * Uses OpenAI-compatible Chat Completions API (not Responses API).\n *\n * Delegates implementation to {@link OpenAIBaseChatCompletionsTextAdapter}\n * from `@tanstack/openai-base` and threads Grok-specific tool-capability\n * typing through the 5th generic of the base class.\n */\nexport class GrokTextAdapter<\n  TModel extends (typeof GROK_CHAT_MODELS)[number],\n  TProviderOptions extends Record<string, any> = ResolveProviderOptions<TModel>,\n  TInputModalities extends ReadonlyArray<Modality> =\n    ResolveInputModalities<TModel>,\n  TToolCapabilities extends ReadonlyArray<string> =\n    ResolveToolCapabilities<TModel>,\n> extends OpenAIBaseChatCompletionsTextAdapter<\n  TModel,\n  TProviderOptions,\n  TInputModalities,\n  GrokMessageMetadataByModality,\n  TToolCapabilities\n> {\n  readonly kind = 'text' as const\n  readonly name = 'grok' as const\n\n  constructor(config: GrokTextConfig, model: TModel) {\n    super(model, 'grok', new OpenAI(withGrokDefaults(config)))\n  }\n\n  /**\n   * Surfaces xAI reasoning deltas on Grok reasoning models. The DeepSeek-style\n   * convention puts the chain-of-thought on `delta.reasoning_content`; some\n   * Grok variants also populate `delta.reasoning`. Reading both keeps\n   * reasoning flowing through the base's REASONING_* lifecycle for both\n   * `chatStream` and `structuredOutputStream`.\n   */\n  protected override extractReasoning(\n    chunk: OpenAI.Chat.Completions.ChatCompletionChunk,\n  ): { text: string } | undefined {\n    const delta = chunk.choices[0]?.delta as\n      | { reasoning?: unknown; reasoning_content?: unknown }\n      | undefined\n    const raw = delta?.reasoning_content ?? delta?.reasoning\n    if (typeof raw === 'string' && raw.length > 0) {\n      return { text: raw }\n    }\n    return undefined\n  }\n}\n\n/**\n * Creates a Grok text adapter with explicit API key.\n * Type resolution happens here at the call site.\n *\n * @param model - The model name (e.g., 'grok-3', 'grok-4')\n * @param apiKey - Your xAI API key\n * @param config - Optional additional configuration\n * @returns Configured Grok text adapter instance with resolved types\n *\n * @example\n * ```typescript\n * const adapter = createGrokText('grok-3', \"xai-...\");\n * // adapter has type-safe providerOptions for grok-3\n * ```\n */\nexport function createGrokText<\n  TModel extends (typeof GROK_CHAT_MODELS)[number],\n>(\n  model: TModel,\n  apiKey: string,\n  config?: Omit<GrokTextConfig, 'apiKey'>,\n): GrokTextAdapter<TModel> {\n  return new GrokTextAdapter({ apiKey, ...config }, model)\n}\n\n/**\n * Creates a Grok text adapter with automatic API key detection from environment variables.\n * Type resolution happens here at the call site.\n *\n * Looks for `XAI_API_KEY` in:\n * - `process.env` (Node.js)\n * - `window.env` (Browser with injected env)\n *\n * @param model - The model name (e.g., 'grok-3', 'grok-4')\n * @param config - Optional configuration (excluding apiKey which is auto-detected)\n * @returns Configured Grok text adapter instance with resolved types\n * @throws Error if XAI_API_KEY is not found in environment\n *\n * @example\n * ```typescript\n * // Automatically uses XAI_API_KEY from environment\n * const adapter = grokText('grok-3');\n *\n * const stream = chat({\n *   adapter,\n *   messages: [{ role: \"user\", content: \"Hello!\" }]\n * });\n * ```\n */\nexport function grokText<TModel extends (typeof GROK_CHAT_MODELS)[number]>(\n  model: TModel,\n  config?: Omit<GrokTextConfig, 'apiKey'>,\n): GrokTextAdapter<TModel> {\n  const apiKey = getGrokApiKeyFromEnv()\n  return createGrokText(model, apiKey, config)\n}\n"],"names":[],"mappings":";;;AAyCO,MAAM,wBAOH,qCAMR;AAAA,EAIA,YAAY,QAAwB,OAAe;AACjD,UAAM,OAAO,QAAQ,IAAI,OAAO,iBAAiB,MAAM,CAAC,CAAC;AAJ3D,SAAS,OAAO;AAChB,SAAS,OAAO;AAAA,EAIhB;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EASmB,iBACjB,OAC8B;AAC9B,UAAM,QAAQ,MAAM,QAAQ,CAAC,GAAG;AAGhC,UAAM,MAAM,OAAO,qBAAqB,OAAO;AAC/C,QAAI,OAAO,QAAQ,YAAY,IAAI,SAAS,GAAG;AAC7C,aAAO,EAAE,MAAM,IAAA;AAAA,IACjB;AACA,WAAO;AAAA,EACT;AACF;AAiBO,SAAS,eAGd,OACA,QACA,QACyB;AACzB,SAAO,IAAI,gBAAgB,EAAE,QAAQ,GAAG,OAAA,GAAU,KAAK;AACzD;AA0BO,SAAS,SACd,OACA,QACyB;AACzB,QAAM,SAAS,qBAAA;AACf,SAAO,eAAe,OAAO,QAAQ,MAAM;AAC7C;"}
+{"version":3,"file":"text.js","sources":["../../../src/adapters/text.ts"],"sourcesContent":["import OpenAI from 'openai'\nimport { OpenAIBaseChatCompletionsTextAdapter } from '@tanstack/openai-base'\nimport { getGrokApiKeyFromEnv, withGrokDefaults } from '../utils/client'\nimport type {\n  GROK_CHAT_MODELS,\n  GrokChatModelToolCapabilitiesByName,\n  ResolveInputModalities,\n  ResolveProviderOptions,\n} from '../model-meta'\nimport type { Modality } from '@tanstack/ai'\nimport type { GrokMessageMetadataByModality } from '../message-types'\nimport type { GrokClientConfig } from '../utils'\n\n/**\n * Resolve tool capabilities for a specific Grok model.\n */\ntype ResolveToolCapabilities<TModel extends string> =\n  TModel extends keyof GrokChatModelToolCapabilitiesByName\n    ? NonNullable<GrokChatModelToolCapabilitiesByName[TModel]>\n    : readonly []\n\n/**\n * Configuration for Grok text adapter\n */\nexport interface GrokTextConfig extends GrokClientConfig {}\n\n/**\n * Alias for TextProviderOptions for external use\n */\nexport type { ExternalTextProviderOptions as GrokTextProviderOptions } from '../text/text-provider-options'\n\n/**\n * Grok Text (Chat) Adapter\n *\n * Tree-shakeable adapter for Grok chat/text completion functionality.\n * Uses OpenAI-compatible Chat Completions API (not Responses API).\n *\n * Delegates implementation to {@link OpenAIBaseChatCompletionsTextAdapter}\n * from `@tanstack/openai-base` and threads Grok-specific tool-capability\n * typing through the 5th generic of the base class.\n */\nexport class GrokTextAdapter<\n  TModel extends (typeof GROK_CHAT_MODELS)[number],\n  TProviderOptions extends Record<string, any> = ResolveProviderOptions<TModel>,\n  TInputModalities extends ReadonlyArray<Modality> =\n    ResolveInputModalities<TModel>,\n  TToolCapabilities extends ReadonlyArray<string> =\n    ResolveToolCapabilities<TModel>,\n> extends OpenAIBaseChatCompletionsTextAdapter<\n  TModel,\n  TProviderOptions,\n  TInputModalities,\n  GrokMessageMetadataByModality,\n  TToolCapabilities\n> {\n  override readonly kind = 'text' as const\n  override readonly name = 'grok' as const\n\n  constructor(config: GrokTextConfig, model: TModel) {\n    super(model, 'grok', new OpenAI(withGrokDefaults(config)))\n  }\n\n  /**\n   * Surfaces xAI reasoning deltas on Grok reasoning models. The DeepSeek-style\n   * convention puts the chain-of-thought on `delta.reasoning_content`; some\n   * Grok variants also populate `delta.reasoning`. Reading both keeps\n   * reasoning flowing through the base's REASONING_* lifecycle for both\n   * `chatStream` and `structuredOutputStream`.\n   */\n  protected override extractReasoning(\n    chunk: OpenAI.Chat.Completions.ChatCompletionChunk,\n  ): { text: string } | undefined {\n    const delta = chunk.choices[0]?.delta as\n      | { reasoning?: unknown; reasoning_content?: unknown }\n      | undefined\n    const raw = delta?.reasoning_content ?? delta?.reasoning\n    if (typeof raw === 'string' && raw.length > 0) {\n      return { text: raw }\n    }\n    return undefined\n  }\n}\n\n/**\n * Creates a Grok text adapter with explicit API key.\n * Type resolution happens here at the call site.\n *\n * @param model - The model name (e.g., 'grok-3', 'grok-4')\n * @param apiKey - Your xAI API key\n * @param config - Optional additional configuration\n * @returns Configured Grok text adapter instance with resolved types\n *\n * @example\n * ```typescript\n * const adapter = createGrokText('grok-3', \"xai-...\");\n * // adapter has type-safe providerOptions for grok-3\n * ```\n */\nexport function createGrokText<\n  TModel extends (typeof GROK_CHAT_MODELS)[number],\n>(\n  model: TModel,\n  apiKey: string,\n  config?: Omit<GrokTextConfig, 'apiKey'>,\n): GrokTextAdapter<TModel> {\n  return new GrokTextAdapter({ apiKey, ...config }, model)\n}\n\n/**\n * Creates a Grok text adapter with automatic API key detection from environment variables.\n * Type resolution happens here at the call site.\n *\n * Looks for `XAI_API_KEY` in:\n * - `process.env` (Node.js)\n * - `window.env` (Browser with injected env)\n *\n * @param model - The model name (e.g., 'grok-3', 'grok-4')\n * @param config - Optional configuration (excluding apiKey which is auto-detected)\n * @returns Configured Grok text adapter instance with resolved types\n * @throws Error if XAI_API_KEY is not found in environment\n *\n * @example\n * ```typescript\n * // Automatically uses XAI_API_KEY from environment\n * const adapter = grokText('grok-3');\n *\n * const stream = chat({\n *   adapter,\n *   messages: [{ role: \"user\", content: \"Hello!\" }]\n * });\n * ```\n */\nexport function grokText<TModel extends (typeof GROK_CHAT_MODELS)[number]>(\n  model: TModel,\n  config?: Omit<GrokTextConfig, 'apiKey'>,\n): GrokTextAdapter<TModel> {\n  const apiKey = getGrokApiKeyFromEnv()\n  return createGrokText(model, apiKey, config)\n}\n"],"names":[],"mappings":";;;AAyCO,MAAM,wBAOH,qCAMR;AAAA,EACkB,OAAO;AAAA,EACP,OAAO;AAAA,EAEzB,YAAY,QAAwB,OAAe;AACjD,UAAM,OAAO,QAAQ,IAAI,OAAO,iBAAiB,MAAM,CAAC,CAAC;AAAA,EAC3D;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EASmB,iBACjB,OAC8B;AAC9B,UAAM,QAAQ,MAAM,QAAQ,CAAC,GAAG;AAGhC,UAAM,MAAM,OAAO,qBAAqB,OAAO;AAC/C,QAAI,OAAO,QAAQ,YAAY,IAAI,SAAS,GAAG;AAC7C,aAAO,EAAE,MAAM,IAAA;AAAA,IACjB;AACA,WAAO;AAAA,EACT;AACF;AAiBO,SAAS,eAGd,OACA,QACA,QACyB;AACzB,SAAO,IAAI,gBAAgB,EAAE,QAAQ,GAAG,OAAA,GAAU,KAAK;AACzD;AA0BO,SAAS,SACd,OACA,QACyB;AACzB,QAAM,SAAS,qBAAA;AACf,SAAO,eAAe,OAAO,QAAQ,MAAM;AAC7C;"}

package/dist/esm/adapters/transcription.js

@@ -5,9 +5,12 @@ import "@tanstack/openai-base";
 import { toAudioFile } from "../utils/audio.js";
 const DEFAULT_GROK_BASE_URL = "https://api.x.ai/v1";
 class GrokTranscriptionAdapter extends BaseTranscriptionAdapter {
+  name = "grok";
+  apiKey;
+  baseURL;
+  defaultHeaders;
   constructor(config, model) {
     super(model, config);
-    this.name = "grok";
     this.apiKey = config.apiKey;
     this.baseURL = (config.baseURL ?? DEFAULT_GROK_BASE_URL).replace(/\/+$/, "");
     this.defaultHeaders = config.defaultHeaders ?? {};
@@ -50,13 +53,14 @@ class GrokTranscriptionAdapter extends BaseTranscriptionAdapter {
           return tw;
         }
       );
+      const resolvedLanguage = data.language ?? language;
       return {
         id: generateId(this.name),
         model,
         text: data.text,
-        language: data.language ?? language,
+        ...resolvedLanguage !== void 0 && { language: resolvedLanguage },
         duration: data.duration,
-        words
+        ...words !== void 0 && { words }
       };
     } catch (error) {
       logger.errors("grok.transcribe fatal", {

package/dist/esm/adapters/transcription.js.map

@@ -1 +1 @@
-{"version":3,"file":"transcription.js","sources":["../../../src/adapters/transcription.ts"],"sourcesContent":["import { BaseTranscriptionAdapter } from '@tanstack/ai/adapters'\nimport { generateId, getGrokApiKeyFromEnv, toAudioFile } from '../utils'\nimport type {\n  TranscriptionOptions,\n  TranscriptionResult,\n  TranscriptionWord,\n} from '@tanstack/ai'\nimport type { GrokTranscriptionModel } from '../model-meta'\nimport type { GrokTranscriptionProviderOptions } from '../audio/transcription-provider-options'\n\n/**\n * Grok-specific extension of `TranscriptionWord` that surfaces the extra\n * fields xAI returns when diarization / confidence are enabled. The base\n * cross-provider `TranscriptionWord` contract doesn't include these, so\n * callers who know they're using Grok can narrow with:\n *\n * ```ts\n * const words = result.words as Array<GrokTranscriptionWord> | undefined\n * ```\n */\nexport interface GrokTranscriptionWord extends TranscriptionWord {\n  /** Model confidence for the word, when xAI returns one. */\n  confidence?: number\n  /** Speaker index, populated when `modelOptions.diarize === true`. */\n  speaker?: number\n}\n\nconst DEFAULT_GROK_BASE_URL = 'https://api.x.ai/v1'\n\n/**\n * Configuration for the Grok transcription adapter.\n *\n * Uses direct `fetch` rather than the OpenAI SDK because xAI's `/v1/stt`\n * endpoint is not OpenAI-compatible.\n */\nexport interface GrokTranscriptionConfig {\n  apiKey: string\n  baseURL?: string\n  /** Additional headers to merge into every request (e.g., test IDs). */\n  defaultHeaders?: Record<string, string>\n}\n\n/**\n * xAI STT response shape from `POST /v1/stt`.\n * Grok returns word-level timestamps only; no segment array.\n */\ninterface GrokSTTWord {\n  text: string\n  start: number\n  end: number\n  confidence?: number\n  speaker?: number\n}\n\ninterface GrokSTTResponse {\n  text: string\n  language?: string\n  duration?: number\n  words?: Array<GrokSTTWord>\n  channels?: Array<unknown>\n}\n\n/**\n * Grok Speech-to-Text Adapter.\n *\n * Talks to `POST {baseURL}/stt` per\n * https://docs.x.ai/developers/rest-api-reference/inference/voice\n */\nexport class GrokTranscriptionAdapter<\n  TModel extends GrokTranscriptionModel,\n> extends BaseTranscriptionAdapter<TModel, GrokTranscriptionProviderOptions> {\n  readonly name = 'grok' as const\n\n  private readonly apiKey: string\n  private readonly baseURL: string\n  private readonly defaultHeaders: Record<string, string>\n\n  constructor(config: GrokTranscriptionConfig, model: TModel) {\n    super(model, config)\n    this.apiKey = config.apiKey\n    this.baseURL = (config.baseURL ?? DEFAULT_GROK_BASE_URL).replace(/\\/+$/, '')\n    this.defaultHeaders = config.defaultHeaders ?? {}\n  }\n\n  async transcribe(\n    options: TranscriptionOptions<GrokTranscriptionProviderOptions>,\n  ): Promise<TranscriptionResult> {\n    const { logger } = options\n    const { model, audio, language, modelOptions } = options\n\n    logger.request(\n      `activity=generateTranscription provider=grok model=${model}`,\n      { provider: 'grok', model },\n    )\n\n    const file = toAudioFile(audio, modelOptions?.audio_format)\n    const form = buildTranscriptionFormData({ file, language, modelOptions })\n\n    try {\n      const response = await fetch(`${this.baseURL}/stt`, {\n        method: 'POST',\n        headers: {\n          // `defaultHeaders` first so Authorization always wins.\n          ...this.defaultHeaders,\n          Authorization: `Bearer ${this.apiKey}`,\n        },\n        body: form,\n      })\n\n      if (!response.ok) {\n        const errorText = await response.text()\n        throw new Error(\n          `Grok transcription request failed: ${response.status} ${errorText}`,\n        )\n      }\n\n      const data = (await response.json()) as GrokSTTResponse\n\n      const words: Array<TranscriptionWord> | undefined = data.words?.map(\n        (w) => {\n          // Construct a GrokTranscriptionWord so that `confidence` and\n          // `speaker` (when xAI returns them under `diarize` / confidence\n          // mode) are preserved on the result. The returned array is typed\n          // as `Array<TranscriptionWord>` per the cross-provider contract;\n          // callers who want the extras narrow via `as Array<GrokTranscriptionWord>`.\n          const tw: GrokTranscriptionWord = {\n            word: w.text,\n            start: w.start,\n            end: w.end,\n          }\n          if (w.confidence !== undefined) tw.confidence = w.confidence\n          if (w.speaker !== undefined) tw.speaker = w.speaker\n          return tw\n        },\n      )\n\n      return {\n        id: generateId(this.name),\n        model,\n        text: data.text,\n        language: data.language ?? language,\n        duration: data.duration,\n        words,\n      }\n    } catch (error) {\n      logger.errors('grok.transcribe fatal', {\n        error,\n        source: 'grok.transcribe',\n      })\n      throw error\n    }\n  }\n}\n\n/**\n * Build the multipart/form-data body for `POST /v1/stt`, coercing SDK-level\n * model options into xAI's wire format (booleans as `'true'`/`'false'`\n * strings, numeric fields stringified, etc.).\n *\n * Wire-field mapping:\n *   - `modelOptions.inverse_text_normalization` → `format` (xAI's chosen\n *     wire-field name for the ITN boolean; the SDK surfaces it under the\n *     clearer `inverse_text_normalization` key).\n *   - `modelOptions.audio_format`, `sample_rate`, `multichannel`, `channels`,\n *     `diarize` map to same-named form fields.\n */\nexport function buildTranscriptionFormData(options: {\n  file: File\n  language: string | undefined\n  modelOptions: GrokTranscriptionProviderOptions | undefined\n}): FormData {\n  const { file, language, modelOptions } = options\n  const form = new FormData()\n  form.set('file', file)\n  if (language) form.set('language', language)\n  if (modelOptions?.audio_format !== undefined) {\n    form.set('audio_format', modelOptions.audio_format)\n  }\n  if (modelOptions?.sample_rate !== undefined) {\n    form.set('sample_rate', String(modelOptions.sample_rate))\n  }\n  if (modelOptions?.inverse_text_normalization !== undefined) {\n    form.set(\n      'format',\n      modelOptions.inverse_text_normalization ? 'true' : 'false',\n    )\n  }\n  if (modelOptions?.multichannel !== undefined) {\n    form.set('multichannel', modelOptions.multichannel ? 'true' : 'false')\n  }\n  if (modelOptions?.channels !== undefined) {\n    form.set('channels', String(modelOptions.channels))\n  }\n  if (modelOptions?.diarize !== undefined) {\n    form.set('diarize', modelOptions.diarize ? 'true' : 'false')\n  }\n  return form\n}\n\n/**\n * Creates a Grok transcription adapter with an explicit API key.\n *\n * @example\n * ```typescript\n * const adapter = createGrokTranscription('grok-stt', 'xai-...')\n * const result = await generateTranscription({\n *   adapter,\n *   audio: audioFile,\n *   language: 'en',\n * })\n * ```\n */\nexport function createGrokTranscription<TModel extends GrokTranscriptionModel>(\n  model: TModel,\n  apiKey: string,\n  config?: Omit<GrokTranscriptionConfig, 'apiKey'>,\n): GrokTranscriptionAdapter<TModel> {\n  return new GrokTranscriptionAdapter({ apiKey, ...config }, model)\n}\n\n/**\n * Creates a Grok transcription adapter, reading the API key from\n * `XAI_API_KEY` in the environment.\n *\n * @throws Error if `XAI_API_KEY` is not set.\n */\nexport function grokTranscription<TModel extends GrokTranscriptionModel>(\n  model: TModel,\n  config?: Omit<GrokTranscriptionConfig, 'apiKey'>,\n): GrokTranscriptionAdapter<TModel> {\n  const apiKey = getGrokApiKeyFromEnv()\n  return createGrokTranscription(model, apiKey, config)\n}\n"],"names":[],"mappings":";;;;;AA2BA,MAAM,wBAAwB;AAyCvB,MAAM,iCAEH,yBAAmE;AAAA,EAO3E,YAAY,QAAiC,OAAe;AAC1D,UAAM,OAAO,MAAM;AAPrB,SAAS,OAAO;AAQd,SAAK,SAAS,OAAO;AACrB,SAAK,WAAW,OAAO,WAAW,uBAAuB,QAAQ,QAAQ,EAAE;AAC3E,SAAK,iBAAiB,OAAO,kBAAkB,CAAA;AAAA,EACjD;AAAA,EAEA,MAAM,WACJ,SAC8B;AAC9B,UAAM,EAAE,WAAW;AACnB,UAAM,EAAE,OAAO,OAAO,UAAU,iBAAiB;AAEjD,WAAO;AAAA,MACL,sDAAsD,KAAK;AAAA,MAC3D,EAAE,UAAU,QAAQ,MAAA;AAAA,IAAM;AAG5B,UAAM,OAAO,YAAY,OAAO,cAAc,YAAY;AAC1D,UAAM,OAAO,2BAA2B,EAAE,MAAM,UAAU,cAAc;AAExE,QAAI;AACF,YAAM,WAAW,MAAM,MAAM,GAAG,KAAK,OAAO,QAAQ;AAAA,QAClD,QAAQ;AAAA,QACR,SAAS;AAAA;AAAA,UAEP,GAAG,KAAK;AAAA,UACR,eAAe,UAAU,KAAK,MAAM;AAAA,QAAA;AAAA,QAEtC,MAAM;AAAA,MAAA,CACP;AAED,UAAI,CAAC,SAAS,IAAI;AAChB,cAAM,YAAY,MAAM,SAAS,KAAA;AACjC,cAAM,IAAI;AAAA,UACR,sCAAsC,SAAS,MAAM,IAAI,SAAS;AAAA,QAAA;AAAA,MAEtE;AAEA,YAAM,OAAQ,MAAM,SAAS,KAAA;AAE7B,YAAM,QAA8C,KAAK,OAAO;AAAA,QAC9D,CAAC,MAAM;AAML,gBAAM,KAA4B;AAAA,YAChC,MAAM,EAAE;AAAA,YACR,OAAO,EAAE;AAAA,YACT,KAAK,EAAE;AAAA,UAAA;AAET,cAAI,EAAE,eAAe,OAAW,IAAG,aAAa,EAAE;AAClD,cAAI,EAAE,YAAY,OAAW,IAAG,UAAU,EAAE;AAC5C,iBAAO;AAAA,QACT;AAAA,MAAA;AAGF,aAAO;AAAA,QACL,IAAI,WAAW,KAAK,IAAI;AAAA,QACxB;AAAA,QACA,MAAM,KAAK;AAAA,QACX,UAAU,KAAK,YAAY;AAAA,QAC3B,UAAU,KAAK;AAAA,QACf;AAAA,MAAA;AAAA,IAEJ,SAAS,OAAO;AACd,aAAO,OAAO,yBAAyB;AAAA,QACrC;AAAA,QACA,QAAQ;AAAA,MAAA,CACT;AACD,YAAM;AAAA,IACR;AAAA,EACF;AACF;AAcO,SAAS,2BAA2B,SAI9B;AACX,QAAM,EAAE,MAAM,UAAU,aAAA,IAAiB;AACzC,QAAM,OAAO,IAAI,SAAA;AACjB,OAAK,IAAI,QAAQ,IAAI;AACrB,MAAI,SAAU,MAAK,IAAI,YAAY,QAAQ;AAC3C,MAAI,cAAc,iBAAiB,QAAW;AAC5C,SAAK,IAAI,gBAAgB,aAAa,YAAY;AAAA,EACpD;AACA,MAAI,cAAc,gBAAgB,QAAW;AAC3C,SAAK,IAAI,eAAe,OAAO,aAAa,WAAW,CAAC;AAAA,EAC1D;AACA,MAAI,cAAc,+BAA+B,QAAW;AAC1D,SAAK;AAAA,MACH;AAAA,MACA,aAAa,6BAA6B,SAAS;AAAA,IAAA;AAAA,EAEvD;AACA,MAAI,cAAc,iBAAiB,QAAW;AAC5C,SAAK,IAAI,gBAAgB,aAAa,eAAe,SAAS,OAAO;AAAA,EACvE;AACA,MAAI,cAAc,aAAa,QAAW;AACxC,SAAK,IAAI,YAAY,OAAO,aAAa,QAAQ,CAAC;AAAA,EACpD;AACA,MAAI,cAAc,YAAY,QAAW;AACvC,SAAK,IAAI,WAAW,aAAa,UAAU,SAAS,OAAO;AAAA,EAC7D;AACA,SAAO;AACT;AAeO,SAAS,wBACd,OACA,QACA,QACkC;AAClC,SAAO,IAAI,yBAAyB,EAAE,QAAQ,GAAG,OAAA,GAAU,KAAK;AAClE;AAQO,SAAS,kBACd,OACA,QACkC;AAClC,QAAM,SAAS,qBAAA;AACf,SAAO,wBAAwB,OAAO,QAAQ,MAAM;AACtD;"}
+{"version":3,"file":"transcription.js","sources":["../../../src/adapters/transcription.ts"],"sourcesContent":["import { BaseTranscriptionAdapter } from '@tanstack/ai/adapters'\nimport { generateId, getGrokApiKeyFromEnv, toAudioFile } from '../utils'\nimport type {\n  TranscriptionOptions,\n  TranscriptionResult,\n  TranscriptionWord,\n} from '@tanstack/ai'\nimport type { GrokTranscriptionModel } from '../model-meta'\nimport type { GrokTranscriptionProviderOptions } from '../audio/transcription-provider-options'\n\n/**\n * Grok-specific extension of `TranscriptionWord` that surfaces the extra\n * fields xAI returns when diarization / confidence are enabled. The base\n * cross-provider `TranscriptionWord` contract doesn't include these, so\n * callers who know they're using Grok can narrow with:\n *\n * ```ts\n * const words = result.words as Array<GrokTranscriptionWord> | undefined\n * ```\n */\nexport interface GrokTranscriptionWord extends TranscriptionWord {\n  /** Model confidence for the word, when xAI returns one. */\n  confidence?: number\n  /** Speaker index, populated when `modelOptions.diarize === true`. */\n  speaker?: number\n}\n\nconst DEFAULT_GROK_BASE_URL = 'https://api.x.ai/v1'\n\n/**\n * Configuration for the Grok transcription adapter.\n *\n * Uses direct `fetch` rather than the OpenAI SDK because xAI's `/v1/stt`\n * endpoint is not OpenAI-compatible.\n */\nexport interface GrokTranscriptionConfig {\n  apiKey: string\n  baseURL?: string\n  /** Additional headers to merge into every request (e.g., test IDs). */\n  defaultHeaders?: Record<string, string>\n}\n\n/**\n * xAI STT response shape from `POST /v1/stt`.\n * Grok returns word-level timestamps only; no segment array.\n */\ninterface GrokSTTWord {\n  text: string\n  start: number\n  end: number\n  confidence?: number\n  speaker?: number\n}\n\ninterface GrokSTTResponse {\n  text: string\n  language?: string\n  duration?: number\n  words?: Array<GrokSTTWord>\n  channels?: Array<unknown>\n}\n\n/**\n * Grok Speech-to-Text Adapter.\n *\n * Talks to `POST {baseURL}/stt` per\n * https://docs.x.ai/developers/rest-api-reference/inference/voice\n */\nexport class GrokTranscriptionAdapter<\n  TModel extends GrokTranscriptionModel,\n> extends BaseTranscriptionAdapter<TModel, GrokTranscriptionProviderOptions> {\n  readonly name = 'grok' as const\n\n  private readonly apiKey: string\n  private readonly baseURL: string\n  private readonly defaultHeaders: Record<string, string>\n\n  constructor(config: GrokTranscriptionConfig, model: TModel) {\n    super(model, config)\n    this.apiKey = config.apiKey\n    this.baseURL = (config.baseURL ?? DEFAULT_GROK_BASE_URL).replace(/\\/+$/, '')\n    this.defaultHeaders = config.defaultHeaders ?? {}\n  }\n\n  async transcribe(\n    options: TranscriptionOptions<GrokTranscriptionProviderOptions>,\n  ): Promise<TranscriptionResult> {\n    const { logger } = options\n    const { model, audio, language, modelOptions } = options\n\n    logger.request(\n      `activity=generateTranscription provider=grok model=${model}`,\n      { provider: 'grok', model },\n    )\n\n    const file = toAudioFile(audio, modelOptions?.audio_format)\n    const form = buildTranscriptionFormData({ file, language, modelOptions })\n\n    try {\n      const response = await fetch(`${this.baseURL}/stt`, {\n        method: 'POST',\n        headers: {\n          // `defaultHeaders` first so Authorization always wins.\n          ...this.defaultHeaders,\n          Authorization: `Bearer ${this.apiKey}`,\n        },\n        body: form,\n      })\n\n      if (!response.ok) {\n        const errorText = await response.text()\n        throw new Error(\n          `Grok transcription request failed: ${response.status} ${errorText}`,\n        )\n      }\n\n      const data = (await response.json()) as GrokSTTResponse\n\n      const words: Array<TranscriptionWord> | undefined = data.words?.map(\n        (w) => {\n          // Construct a GrokTranscriptionWord so that `confidence` and\n          // `speaker` (when xAI returns them under `diarize` / confidence\n          // mode) are preserved on the result. The returned array is typed\n          // as `Array<TranscriptionWord>` per the cross-provider contract;\n          // callers who want the extras narrow via `as Array<GrokTranscriptionWord>`.\n          const tw: GrokTranscriptionWord = {\n            word: w.text,\n            start: w.start,\n            end: w.end,\n          }\n          if (w.confidence !== undefined) tw.confidence = w.confidence\n          if (w.speaker !== undefined) tw.speaker = w.speaker\n          return tw\n        },\n      )\n\n      const resolvedLanguage = data.language ?? language\n      return {\n        id: generateId(this.name),\n        model,\n        text: data.text,\n        ...(resolvedLanguage !== undefined && { language: resolvedLanguage }),\n        duration: data.duration,\n        ...(words !== undefined && { words }),\n      }\n    } catch (error) {\n      logger.errors('grok.transcribe fatal', {\n        error,\n        source: 'grok.transcribe',\n      })\n      throw error\n    }\n  }\n}\n\n/**\n * Build the multipart/form-data body for `POST /v1/stt`, coercing SDK-level\n * model options into xAI's wire format (booleans as `'true'`/`'false'`\n * strings, numeric fields stringified, etc.).\n *\n * Wire-field mapping:\n *   - `modelOptions.inverse_text_normalization` → `format` (xAI's chosen\n *     wire-field name for the ITN boolean; the SDK surfaces it under the\n *     clearer `inverse_text_normalization` key).\n *   - `modelOptions.audio_format`, `sample_rate`, `multichannel`, `channels`,\n *     `diarize` map to same-named form fields.\n */\nexport function buildTranscriptionFormData(options: {\n  file: File\n  language: string | undefined\n  modelOptions: GrokTranscriptionProviderOptions | undefined\n}): FormData {\n  const { file, language, modelOptions } = options\n  const form = new FormData()\n  form.set('file', file)\n  if (language) form.set('language', language)\n  if (modelOptions?.audio_format !== undefined) {\n    form.set('audio_format', modelOptions.audio_format)\n  }\n  if (modelOptions?.sample_rate !== undefined) {\n    form.set('sample_rate', String(modelOptions.sample_rate))\n  }\n  if (modelOptions?.inverse_text_normalization !== undefined) {\n    form.set(\n      'format',\n      modelOptions.inverse_text_normalization ? 'true' : 'false',\n    )\n  }\n  if (modelOptions?.multichannel !== undefined) {\n    form.set('multichannel', modelOptions.multichannel ? 'true' : 'false')\n  }\n  if (modelOptions?.channels !== undefined) {\n    form.set('channels', String(modelOptions.channels))\n  }\n  if (modelOptions?.diarize !== undefined) {\n    form.set('diarize', modelOptions.diarize ? 'true' : 'false')\n  }\n  return form\n}\n\n/**\n * Creates a Grok transcription adapter with an explicit API key.\n *\n * @example\n * ```typescript\n * const adapter = createGrokTranscription('grok-stt', 'xai-...')\n * const result = await generateTranscription({\n *   adapter,\n *   audio: audioFile,\n *   language: 'en',\n * })\n * ```\n */\nexport function createGrokTranscription<TModel extends GrokTranscriptionModel>(\n  model: TModel,\n  apiKey: string,\n  config?: Omit<GrokTranscriptionConfig, 'apiKey'>,\n): GrokTranscriptionAdapter<TModel> {\n  return new GrokTranscriptionAdapter({ apiKey, ...config }, model)\n}\n\n/**\n * Creates a Grok transcription adapter, reading the API key from\n * `XAI_API_KEY` in the environment.\n *\n * @throws Error if `XAI_API_KEY` is not set.\n */\nexport function grokTranscription<TModel extends GrokTranscriptionModel>(\n  model: TModel,\n  config?: Omit<GrokTranscriptionConfig, 'apiKey'>,\n): GrokTranscriptionAdapter<TModel> {\n  const apiKey = getGrokApiKeyFromEnv()\n  return createGrokTranscription(model, apiKey, config)\n}\n"],"names":[],"mappings":";;;;;AA2BA,MAAM,wBAAwB;AAyCvB,MAAM,iCAEH,yBAAmE;AAAA,EAClE,OAAO;AAAA,EAEC;AAAA,EACA;AAAA,EACA;AAAA,EAEjB,YAAY,QAAiC,OAAe;AAC1D,UAAM,OAAO,MAAM;AACnB,SAAK,SAAS,OAAO;AACrB,SAAK,WAAW,OAAO,WAAW,uBAAuB,QAAQ,QAAQ,EAAE;AAC3E,SAAK,iBAAiB,OAAO,kBAAkB,CAAA;AAAA,EACjD;AAAA,EAEA,MAAM,WACJ,SAC8B;AAC9B,UAAM,EAAE,WAAW;AACnB,UAAM,EAAE,OAAO,OAAO,UAAU,iBAAiB;AAEjD,WAAO;AAAA,MACL,sDAAsD,KAAK;AAAA,MAC3D,EAAE,UAAU,QAAQ,MAAA;AAAA,IAAM;AAG5B,UAAM,OAAO,YAAY,OAAO,cAAc,YAAY;AAC1D,UAAM,OAAO,2BAA2B,EAAE,MAAM,UAAU,cAAc;AAExE,QAAI;AACF,YAAM,WAAW,MAAM,MAAM,GAAG,KAAK,OAAO,QAAQ;AAAA,QAClD,QAAQ;AAAA,QACR,SAAS;AAAA;AAAA,UAEP,GAAG,KAAK;AAAA,UACR,eAAe,UAAU,KAAK,MAAM;AAAA,QAAA;AAAA,QAEtC,MAAM;AAAA,MAAA,CACP;AAED,UAAI,CAAC,SAAS,IAAI;AAChB,cAAM,YAAY,MAAM,SAAS,KAAA;AACjC,cAAM,IAAI;AAAA,UACR,sCAAsC,SAAS,MAAM,IAAI,SAAS;AAAA,QAAA;AAAA,MAEtE;AAEA,YAAM,OAAQ,MAAM,SAAS,KAAA;AAE7B,YAAM,QAA8C,KAAK,OAAO;AAAA,QAC9D,CAAC,MAAM;AAML,gBAAM,KAA4B;AAAA,YAChC,MAAM,EAAE;AAAA,YACR,OAAO,EAAE;AAAA,YACT,KAAK,EAAE;AAAA,UAAA;AAET,cAAI,EAAE,eAAe,OAAW,IAAG,aAAa,EAAE;AAClD,cAAI,EAAE,YAAY,OAAW,IAAG,UAAU,EAAE;AAC5C,iBAAO;AAAA,QACT;AAAA,MAAA;AAGF,YAAM,mBAAmB,KAAK,YAAY;AAC1C,aAAO;AAAA,QACL,IAAI,WAAW,KAAK,IAAI;AAAA,QACxB;AAAA,QACA,MAAM,KAAK;AAAA,QACX,GAAI,qBAAqB,UAAa,EAAE,UAAU,iBAAA;AAAA,QAClD,UAAU,KAAK;AAAA,QACf,GAAI,UAAU,UAAa,EAAE,MAAA;AAAA,MAAM;AAAA,IAEvC,SAAS,OAAO;AACd,aAAO,OAAO,yBAAyB;AAAA,QACrC;AAAA,QACA,QAAQ;AAAA,MAAA,CACT;AACD,YAAM;AAAA,IACR;AAAA,EACF;AACF;AAcO,SAAS,2BAA2B,SAI9B;AACX,QAAM,EAAE,MAAM,UAAU,aAAA,IAAiB;AACzC,QAAM,OAAO,IAAI,SAAA;AACjB,OAAK,IAAI,QAAQ,IAAI;AACrB,MAAI,SAAU,MAAK,IAAI,YAAY,QAAQ;AAC3C,MAAI,cAAc,iBAAiB,QAAW;AAC5C,SAAK,IAAI,gBAAgB,aAAa,YAAY;AAAA,EACpD;AACA,MAAI,cAAc,gBAAgB,QAAW;AAC3C,SAAK,IAAI,eAAe,OAAO,aAAa,WAAW,CAAC;AAAA,EAC1D;AACA,MAAI,cAAc,+BAA+B,QAAW;AAC1D,SAAK;AAAA,MACH;AAAA,MACA,aAAa,6BAA6B,SAAS;AAAA,IAAA;AAAA,EAEvD;AACA,MAAI,cAAc,iBAAiB,QAAW;AAC5C,SAAK,IAAI,gBAAgB,aAAa,eAAe,SAAS,OAAO;AAAA,EACvE;AACA,MAAI,cAAc,aAAa,QAAW;AACxC,SAAK,IAAI,YAAY,OAAO,aAAa,QAAQ,CAAC;AAAA,EACpD;AACA,MAAI,cAAc,YAAY,QAAW;AACvC,SAAK,IAAI,WAAW,aAAa,UAAU,SAAS,OAAO;AAAA,EAC7D;AACA,SAAO;AACT;AAeO,SAAS,wBACd,OACA,QACA,QACkC;AAClC,SAAO,IAAI,yBAAyB,EAAE,QAAQ,GAAG,OAAA,GAAU,KAAK;AAClE;AAQO,SAAS,kBACd,OACA,QACkC;AAClC,QAAM,SAAS,qBAAA;AACf,SAAO,wBAAwB,OAAO,QAAQ,MAAM;AACtD;"}

package/dist/esm/adapters/tts.js

@@ -5,9 +5,12 @@ import "@tanstack/openai-base";
 import { arrayBufferToBase64 } from "../utils/audio.js";
 const DEFAULT_GROK_BASE_URL = "https://api.x.ai/v1";
 class GrokSpeechAdapter extends BaseTTSAdapter {
+  name = "grok";
+  apiKey;
+  baseURL;
+  defaultHeaders;
   constructor(config, model) {
     super(model, config);
-    this.name = "grok";
     this.apiKey = config.apiKey;
     this.baseURL = (config.baseURL ?? DEFAULT_GROK_BASE_URL).replace(/\/+$/, "");
     this.defaultHeaders = config.defaultHeaders ?? {};

package/dist/esm/adapters/tts.js.map

@@ -1 +1 @@
-{"version":3,"file":"tts.js","sources":["../../../src/adapters/tts.ts"],"sourcesContent":["import { BaseTTSAdapter } from '@tanstack/ai/adapters'\nimport { arrayBufferToBase64, generateId, getGrokApiKeyFromEnv } from '../utils'\nimport type { TTSOptions, TTSResult } from '@tanstack/ai'\nimport type { GrokTTSModel } from '../model-meta'\nimport type {\n  GrokTTSCodec,\n  GrokTTSProviderOptions,\n  GrokTTSVoice,\n} from '../audio/tts-provider-options'\n\nconst DEFAULT_GROK_BASE_URL = 'https://api.x.ai/v1'\n\n/**\n * Configuration for the Grok TTS adapter.\n *\n * Unlike chat/image/summarize adapters, TTS does not use the OpenAI SDK\n * because xAI's `/v1/tts` endpoint is not OpenAI-compatible. This config\n * is a minimal subset suitable for direct `fetch` calls.\n */\nexport interface GrokSpeechConfig {\n  apiKey: string\n  baseURL?: string\n  /** Additional headers to merge into every request (e.g., test IDs). */\n  defaultHeaders?: Record<string, string>\n}\n\n/**\n * Grok Text-to-Speech Adapter.\n *\n * Talks to `POST {baseURL}/tts` per\n * https://docs.x.ai/developers/model-capabilities/audio/text-to-speech\n */\nexport class GrokSpeechAdapter<\n  TModel extends GrokTTSModel,\n> extends BaseTTSAdapter<TModel, GrokTTSProviderOptions> {\n  readonly name = 'grok' as const\n\n  private readonly apiKey: string\n  private readonly baseURL: string\n  private readonly defaultHeaders: Record<string, string>\n\n  constructor(config: GrokSpeechConfig, model: TModel) {\n    super(model, config)\n    this.apiKey = config.apiKey\n    this.baseURL = (config.baseURL ?? DEFAULT_GROK_BASE_URL).replace(/\\/+$/, '')\n    this.defaultHeaders = config.defaultHeaders ?? {}\n  }\n\n  async generateSpeech(\n    options: TTSOptions<GrokTTSProviderOptions>,\n  ): Promise<TTSResult> {\n    const { logger } = options\n    const { model, text, voice, format, modelOptions } = options\n\n    logger.request(`activity=generateSpeech provider=grok model=${model}`, {\n      provider: 'grok',\n      model,\n    })\n\n    const { body, codec, sampleRateForContentType } = buildTTSRequestBody({\n      text,\n      voice,\n      format,\n      modelOptions,\n    })\n\n    try {\n      const response = await fetch(`${this.baseURL}/tts`, {\n        method: 'POST',\n        headers: {\n          // `defaultHeaders` first so the adapter's Authorization / Content-Type\n          // always win — otherwise a caller-supplied `Authorization` header\n          // could silently clobber the bearer token.\n          ...this.defaultHeaders,\n          Authorization: `Bearer ${this.apiKey}`,\n          'Content-Type': 'application/json',\n        },\n        body: JSON.stringify(body),\n      })\n\n      if (!response.ok) {\n        const errorText = await response.text()\n        throw new Error(\n          `Grok TTS request failed: ${response.status} ${errorText}`,\n        )\n      }\n\n      const arrayBuffer = await response.arrayBuffer()\n      const audio = arrayBufferToBase64(arrayBuffer)\n\n      return {\n        id: generateId(this.name),\n        model,\n        audio,\n        format: codec,\n        contentType: getContentType(codec, sampleRateForContentType),\n      }\n    } catch (error) {\n      logger.errors('grok.generateSpeech fatal', {\n        error,\n        source: 'grok.generateSpeech',\n      })\n      throw error\n    }\n  }\n}\n\n/**\n * Build the JSON body for `POST /v1/tts`, resolving codec / sample-rate / voice\n * defaults in one place.\n *\n * Returns the request `body`, the resolved `codec`, and the `sampleRateForContentType`\n * used by the caller to label the response via `getContentType`.\n */\nexport function buildTTSRequestBody(options: {\n  text: string\n  voice: string | undefined\n  format: TTSOptions['format'] | undefined\n  modelOptions: GrokTTSProviderOptions | undefined\n}): {\n  body: Record<string, unknown>\n  codec: GrokTTSCodec\n  sampleRateForContentType: number\n} {\n  const { text, voice, format, modelOptions } = options\n\n  const codec = pickCodec(modelOptions?.codec, format)\n\n  // Only forward `sample_rate` when either:\n  //   - the caller explicitly set `modelOptions.sample_rate`, or\n  //   - the codec's Content-Type carries the rate (pcm → audio/L16;rate=…).\n  // For mp3/wav/opus/aac/flac we leave sample_rate unset so xAI's server\n  // default applies.\n  const callerSampleRate = modelOptions?.sample_rate\n  // Default sample rate documented in GrokTTSProviderOptions is 24000 Hz —\n  // used only when we MUST attach a rate to the contentType (pcm) and the\n  // caller didn't pick one.\n  const pcmDefault = 24000\n  const needsRateInContentType = codec === 'pcm'\n\n  const outputFormat: Record<string, unknown> = { codec }\n  if (callerSampleRate !== undefined) {\n    outputFormat.sample_rate = callerSampleRate\n  } else if (needsRateInContentType) {\n    outputFormat.sample_rate = pcmDefault\n  }\n  if (codec === 'mp3' && modelOptions?.bit_rate !== undefined) {\n    outputFormat.bit_rate = modelOptions.bit_rate\n  }\n\n  // pcm embeds the rate in `audio/L16;rate=…`; mulaw/alaw embed it in\n  // `audio/PCMU;rate=…` / `audio/PCMA;rate=…` when non-default. mp3/wav\n  // don't carry a rate parameter so the value is unused for those.\n  const sampleRateForContentType = callerSampleRate ?? pcmDefault\n\n  const body: Record<string, unknown> = {\n    text,\n    voice_id: (voice as GrokTTSVoice | undefined) ?? 'eve',\n    language: modelOptions?.language ?? 'en',\n    output_format: outputFormat,\n  }\n  if (modelOptions?.optimize_streaming_latency !== undefined) {\n    body.optimize_streaming_latency = modelOptions.optimize_streaming_latency\n  }\n  if (modelOptions?.text_normalization !== undefined) {\n    body.text_normalization = modelOptions.text_normalization\n  }\n\n  return { body, codec, sampleRateForContentType }\n}\n\n/**\n * Maps the cross-provider `TTSOptions.format` onto Grok's supported codecs.\n * `opus`, `aac`, and `flac` are not supported by xAI TTS (which only exposes\n * mp3/wav/pcm/mulaw/alaw) — we fall back to mp3. An explicit\n * `modelOptions.codec` always wins.\n */\nfunction pickCodec(\n  codecOverride: GrokTTSCodec | undefined,\n  format: TTSOptions['format'] | undefined,\n): GrokTTSCodec {\n  if (codecOverride) return codecOverride\n  if (!format) return 'mp3'\n  switch (format) {\n    case 'mp3':\n    case 'wav':\n    case 'pcm':\n      return format\n    case 'flac':\n    case 'opus':\n    case 'aac':\n      return 'mp3'\n    default:\n      return 'mp3'\n  }\n}\n\nexport function getContentType(\n  codec: GrokTTSCodec,\n  sampleRate: number,\n): string {\n  switch (codec) {\n    case 'mp3':\n      return 'audio/mpeg'\n    case 'wav':\n      return 'audio/wav'\n    case 'pcm':\n      // `audio/L16` requires a `rate` parameter per RFC 3551/3555.\n      return `audio/L16;rate=${sampleRate}`\n    case 'mulaw':\n      // `audio/basic` is 8 kHz mono by RFC 2046 registration. For non-8kHz\n      // streams xAI still produces mulaw-encoded bytes at the requested\n      // rate, but the registered MIME can't carry that rate — so we use\n      // the non-standard but commonly-supported `audio/PCMU;rate=…` (RFC 3551\n      // RTP payload name) whenever the caller asked for a rate other than\n      // 8000, and keep `audio/basic` for the standard 8kHz case.\n      return sampleRate === 8000\n        ? 'audio/basic'\n        : `audio/PCMU;rate=${sampleRate}`\n    case 'alaw':\n      return sampleRate === 8000\n        ? 'audio/x-alaw-basic'\n        : `audio/PCMA;rate=${sampleRate}`\n  }\n}\n\n/**\n * Creates a Grok speech (TTS) adapter with an explicit API key.\n *\n * @example\n * ```typescript\n * const adapter = createGrokSpeech('grok-tts', 'xai-...')\n * const result = await generateSpeech({\n *   adapter,\n *   text: 'Hello from Grok',\n *   voice: 'eve',\n * })\n * ```\n */\nexport function createGrokSpeech<TModel extends GrokTTSModel>(\n  model: TModel,\n  apiKey: string,\n  config?: Omit<GrokSpeechConfig, 'apiKey'>,\n): GrokSpeechAdapter<TModel> {\n  return new GrokSpeechAdapter({ apiKey, ...config }, model)\n}\n\n/**\n * Creates a Grok speech (TTS) adapter, reading the API key from\n * `XAI_API_KEY` in the environment.\n *\n * @throws Error if `XAI_API_KEY` is not set.\n */\nexport function grokSpeech<TModel extends GrokTTSModel>(\n  model: TModel,\n  config?: Omit<GrokSpeechConfig, 'apiKey'>,\n): GrokSpeechAdapter<TModel> {\n  const apiKey = getGrokApiKeyFromEnv()\n  return createGrokSpeech(model, apiKey, config)\n}\n"],"names":[],"mappings":";;;;;AAUA,MAAM,wBAAwB;AAsBvB,MAAM,0BAEH,eAA+C;AAAA,EAOvD,YAAY,QAA0B,OAAe;AACnD,UAAM,OAAO,MAAM;AAPrB,SAAS,OAAO;AAQd,SAAK,SAAS,OAAO;AACrB,SAAK,WAAW,OAAO,WAAW,uBAAuB,QAAQ,QAAQ,EAAE;AAC3E,SAAK,iBAAiB,OAAO,kBAAkB,CAAA;AAAA,EACjD;AAAA,EAEA,MAAM,eACJ,SACoB;AACpB,UAAM,EAAE,WAAW;AACnB,UAAM,EAAE,OAAO,MAAM,OAAO,QAAQ,iBAAiB;AAErD,WAAO,QAAQ,+CAA+C,KAAK,IAAI;AAAA,MACrE,UAAU;AAAA,MACV;AAAA,IAAA,CACD;AAED,UAAM,EAAE,MAAM,OAAO,yBAAA,IAA6B,oBAAoB;AAAA,MACpE;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,IAAA,CACD;AAED,QAAI;AACF,YAAM,WAAW,MAAM,MAAM,GAAG,KAAK,OAAO,QAAQ;AAAA,QAClD,QAAQ;AAAA,QACR,SAAS;AAAA;AAAA;AAAA;AAAA,UAIP,GAAG,KAAK;AAAA,UACR,eAAe,UAAU,KAAK,MAAM;AAAA,UACpC,gBAAgB;AAAA,QAAA;AAAA,QAElB,MAAM,KAAK,UAAU,IAAI;AAAA,MAAA,CAC1B;AAED,UAAI,CAAC,SAAS,IAAI;AAChB,cAAM,YAAY,MAAM,SAAS,KAAA;AACjC,cAAM,IAAI;AAAA,UACR,4BAA4B,SAAS,MAAM,IAAI,SAAS;AAAA,QAAA;AAAA,MAE5D;AAEA,YAAM,cAAc,MAAM,SAAS,YAAA;AACnC,YAAM,QAAQ,oBAAoB,WAAW;AAE7C,aAAO;AAAA,QACL,IAAI,WAAW,KAAK,IAAI;AAAA,QACxB;AAAA,QACA;AAAA,QACA,QAAQ;AAAA,QACR,aAAa,eAAe,OAAO,wBAAwB;AAAA,MAAA;AAAA,IAE/D,SAAS,OAAO;AACd,aAAO,OAAO,6BAA6B;AAAA,QACzC;AAAA,QACA,QAAQ;AAAA,MAAA,CACT;AACD,YAAM;AAAA,IACR;AAAA,EACF;AACF;AASO,SAAS,oBAAoB,SASlC;AACA,QAAM,EAAE,MAAM,OAAO,QAAQ,iBAAiB;AAE9C,QAAM,QAAQ,UAAU,cAAc,OAAO,MAAM;AAOnD,QAAM,mBAAmB,cAAc;AAIvC,QAAM,aAAa;AACnB,QAAM,yBAAyB,UAAU;AAEzC,QAAM,eAAwC,EAAE,MAAA;AAChD,MAAI,qBAAqB,QAAW;AAClC,iBAAa,cAAc;AAAA,EAC7B,WAAW,wBAAwB;AACjC,iBAAa,cAAc;AAAA,EAC7B;AACA,MAAI,UAAU,SAAS,cAAc,aAAa,QAAW;AAC3D,iBAAa,WAAW,aAAa;AAAA,EACvC;AAKA,QAAM,2BAA2B,oBAAoB;AAErD,QAAM,OAAgC;AAAA,IACpC;AAAA,IACA,UAAW,SAAsC;AAAA,IACjD,UAAU,cAAc,YAAY;AAAA,IACpC,eAAe;AAAA,EAAA;AAEjB,MAAI,cAAc,+BAA+B,QAAW;AAC1D,SAAK,6BAA6B,aAAa;AAAA,EACjD;AACA,MAAI,cAAc,uBAAuB,QAAW;AAClD,SAAK,qBAAqB,aAAa;AAAA,EACzC;AAEA,SAAO,EAAE,MAAM,OAAO,yBAAA;AACxB;AAQA,SAAS,UACP,eACA,QACc;AACd,MAAI,cAAe,QAAO;AAC1B,MAAI,CAAC,OAAQ,QAAO;AACpB,UAAQ,QAAA;AAAA,IACN,KAAK;AAAA,IACL,KAAK;AAAA,IACL,KAAK;AACH,aAAO;AAAA,IACT,KAAK;AAAA,IACL,KAAK;AAAA,IACL,KAAK;AACH,aAAO;AAAA,IACT;AACE,aAAO;AAAA,EAAA;AAEb;AAEO,SAAS,eACd,OACA,YACQ;AACR,UAAQ,OAAA;AAAA,IACN,KAAK;AACH,aAAO;AAAA,IACT,KAAK;AACH,aAAO;AAAA,IACT,KAAK;AAEH,aAAO,kBAAkB,UAAU;AAAA,IACrC,KAAK;AAOH,aAAO,eAAe,MAClB,gBACA,mBAAmB,UAAU;AAAA,IACnC,KAAK;AACH,aAAO,eAAe,MAClB,uBACA,mBAAmB,UAAU;AAAA,EAAA;AAEvC;AAeO,SAAS,iBACd,OACA,QACA,QAC2B;AAC3B,SAAO,IAAI,kBAAkB,EAAE,QAAQ,GAAG,OAAA,GAAU,KAAK;AAC3D;AAQO,SAAS,WACd,OACA,QAC2B;AAC3B,QAAM,SAAS,qBAAA;AACf,SAAO,iBAAiB,OAAO,QAAQ,MAAM;AAC/C;"}
+{"version":3,"file":"tts.js","sources":["../../../src/adapters/tts.ts"],"sourcesContent":["import { BaseTTSAdapter } from '@tanstack/ai/adapters'\nimport { arrayBufferToBase64, generateId, getGrokApiKeyFromEnv } from '../utils'\nimport type { TTSOptions, TTSResult } from '@tanstack/ai'\nimport type { GrokTTSModel } from '../model-meta'\nimport type {\n  GrokTTSCodec,\n  GrokTTSProviderOptions,\n} from '../audio/tts-provider-options'\n\nconst DEFAULT_GROK_BASE_URL = 'https://api.x.ai/v1'\n\n/**\n * Configuration for the Grok TTS adapter.\n *\n * Unlike chat/image/summarize adapters, TTS does not use the OpenAI SDK\n * because xAI's `/v1/tts` endpoint is not OpenAI-compatible. This config\n * is a minimal subset suitable for direct `fetch` calls.\n */\nexport interface GrokSpeechConfig {\n  apiKey: string\n  baseURL?: string\n  /** Additional headers to merge into every request (e.g., test IDs). */\n  defaultHeaders?: Record<string, string>\n}\n\n/**\n * Grok Text-to-Speech Adapter.\n *\n * Talks to `POST {baseURL}/tts` per\n * https://docs.x.ai/developers/model-capabilities/audio/text-to-speech\n */\nexport class GrokSpeechAdapter<\n  TModel extends GrokTTSModel,\n> extends BaseTTSAdapter<TModel, GrokTTSProviderOptions> {\n  readonly name = 'grok' as const\n\n  private readonly apiKey: string\n  private readonly baseURL: string\n  private readonly defaultHeaders: Record<string, string>\n\n  constructor(config: GrokSpeechConfig, model: TModel) {\n    super(model, config)\n    this.apiKey = config.apiKey\n    this.baseURL = (config.baseURL ?? DEFAULT_GROK_BASE_URL).replace(/\\/+$/, '')\n    this.defaultHeaders = config.defaultHeaders ?? {}\n  }\n\n  async generateSpeech(\n    options: TTSOptions<GrokTTSProviderOptions>,\n  ): Promise<TTSResult> {\n    const { logger } = options\n    const { model, text, voice, format, modelOptions } = options\n\n    logger.request(`activity=generateSpeech provider=grok model=${model}`, {\n      provider: 'grok',\n      model,\n    })\n\n    const { body, codec, sampleRateForContentType } = buildTTSRequestBody({\n      text,\n      voice,\n      format,\n      modelOptions,\n    })\n\n    try {\n      const response = await fetch(`${this.baseURL}/tts`, {\n        method: 'POST',\n        headers: {\n          // `defaultHeaders` first so the adapter's Authorization / Content-Type\n          // always win — otherwise a caller-supplied `Authorization` header\n          // could silently clobber the bearer token.\n          ...this.defaultHeaders,\n          Authorization: `Bearer ${this.apiKey}`,\n          'Content-Type': 'application/json',\n        },\n        body: JSON.stringify(body),\n      })\n\n      if (!response.ok) {\n        const errorText = await response.text()\n        throw new Error(\n          `Grok TTS request failed: ${response.status} ${errorText}`,\n        )\n      }\n\n      const arrayBuffer = await response.arrayBuffer()\n      const audio = arrayBufferToBase64(arrayBuffer)\n\n      return {\n        id: generateId(this.name),\n        model,\n        audio,\n        format: codec,\n        contentType: getContentType(codec, sampleRateForContentType),\n      }\n    } catch (error) {\n      logger.errors('grok.generateSpeech fatal', {\n        error,\n        source: 'grok.generateSpeech',\n      })\n      throw error\n    }\n  }\n}\n\n/**\n * Build the JSON body for `POST /v1/tts`, resolving codec / sample-rate / voice\n * defaults in one place.\n *\n * Returns the request `body`, the resolved `codec`, and the `sampleRateForContentType`\n * used by the caller to label the response via `getContentType`.\n */\nexport function buildTTSRequestBody(options: {\n  text: string\n  voice: string | undefined\n  format: TTSOptions['format'] | undefined\n  modelOptions: GrokTTSProviderOptions | undefined\n}): {\n  body: Record<string, unknown>\n  codec: GrokTTSCodec\n  sampleRateForContentType: number\n} {\n  const { text, voice, format, modelOptions } = options\n\n  const codec = pickCodec(modelOptions?.codec, format)\n\n  // Only forward `sample_rate` when either:\n  //   - the caller explicitly set `modelOptions.sample_rate`, or\n  //   - the codec's Content-Type carries the rate (pcm → audio/L16;rate=…).\n  // For mp3/wav/opus/aac/flac we leave sample_rate unset so xAI's server\n  // default applies.\n  const callerSampleRate = modelOptions?.sample_rate\n  // Default sample rate documented in GrokTTSProviderOptions is 24000 Hz —\n  // used only when we MUST attach a rate to the contentType (pcm) and the\n  // caller didn't pick one.\n  const pcmDefault = 24000\n  const needsRateInContentType = codec === 'pcm'\n\n  const outputFormat: Record<string, unknown> = { codec }\n  if (callerSampleRate !== undefined) {\n    outputFormat.sample_rate = callerSampleRate\n  } else if (needsRateInContentType) {\n    outputFormat.sample_rate = pcmDefault\n  }\n  if (codec === 'mp3' && modelOptions?.bit_rate !== undefined) {\n    outputFormat.bit_rate = modelOptions.bit_rate\n  }\n\n  // pcm embeds the rate in `audio/L16;rate=…`; mulaw/alaw embed it in\n  // `audio/PCMU;rate=…` / `audio/PCMA;rate=…` when non-default. mp3/wav\n  // don't carry a rate parameter so the value is unused for those.\n  const sampleRateForContentType = callerSampleRate ?? pcmDefault\n\n  const body: Record<string, unknown> = {\n    text,\n    voice_id: voice ?? 'eve',\n    language: modelOptions?.language ?? 'en',\n    output_format: outputFormat,\n  }\n  if (modelOptions?.optimize_streaming_latency !== undefined) {\n    body.optimize_streaming_latency = modelOptions.optimize_streaming_latency\n  }\n  if (modelOptions?.text_normalization !== undefined) {\n    body.text_normalization = modelOptions.text_normalization\n  }\n\n  return { body, codec, sampleRateForContentType }\n}\n\n/**\n * Maps the cross-provider `TTSOptions.format` onto Grok's supported codecs.\n * `opus`, `aac`, and `flac` are not supported by xAI TTS (which only exposes\n * mp3/wav/pcm/mulaw/alaw) — we fall back to mp3. An explicit\n * `modelOptions.codec` always wins.\n */\nfunction pickCodec(\n  codecOverride: GrokTTSCodec | undefined,\n  format: TTSOptions['format'] | undefined,\n): GrokTTSCodec {\n  if (codecOverride) return codecOverride\n  if (!format) return 'mp3'\n  switch (format) {\n    case 'mp3':\n    case 'wav':\n    case 'pcm':\n      return format\n    case 'flac':\n    case 'opus':\n    case 'aac':\n      return 'mp3'\n    default:\n      return 'mp3'\n  }\n}\n\nexport function getContentType(\n  codec: GrokTTSCodec,\n  sampleRate: number,\n): string {\n  switch (codec) {\n    case 'mp3':\n      return 'audio/mpeg'\n    case 'wav':\n      return 'audio/wav'\n    case 'pcm':\n      // `audio/L16` requires a `rate` parameter per RFC 3551/3555.\n      return `audio/L16;rate=${sampleRate}`\n    case 'mulaw':\n      // `audio/basic` is 8 kHz mono by RFC 2046 registration. For non-8kHz\n      // streams xAI still produces mulaw-encoded bytes at the requested\n      // rate, but the registered MIME can't carry that rate — so we use\n      // the non-standard but commonly-supported `audio/PCMU;rate=…` (RFC 3551\n      // RTP payload name) whenever the caller asked for a rate other than\n      // 8000, and keep `audio/basic` for the standard 8kHz case.\n      return sampleRate === 8000\n        ? 'audio/basic'\n        : `audio/PCMU;rate=${sampleRate}`\n    case 'alaw':\n      return sampleRate === 8000\n        ? 'audio/x-alaw-basic'\n        : `audio/PCMA;rate=${sampleRate}`\n  }\n}\n\n/**\n * Creates a Grok speech (TTS) adapter with an explicit API key.\n *\n * @example\n * ```typescript\n * const adapter = createGrokSpeech('grok-tts', 'xai-...')\n * const result = await generateSpeech({\n *   adapter,\n *   text: 'Hello from Grok',\n *   voice: 'eve',\n * })\n * ```\n */\nexport function createGrokSpeech<TModel extends GrokTTSModel>(\n  model: TModel,\n  apiKey: string,\n  config?: Omit<GrokSpeechConfig, 'apiKey'>,\n): GrokSpeechAdapter<TModel> {\n  return new GrokSpeechAdapter({ apiKey, ...config }, model)\n}\n\n/**\n * Creates a Grok speech (TTS) adapter, reading the API key from\n * `XAI_API_KEY` in the environment.\n *\n * @throws Error if `XAI_API_KEY` is not set.\n */\nexport function grokSpeech<TModel extends GrokTTSModel>(\n  model: TModel,\n  config?: Omit<GrokSpeechConfig, 'apiKey'>,\n): GrokSpeechAdapter<TModel> {\n  const apiKey = getGrokApiKeyFromEnv()\n  return createGrokSpeech(model, apiKey, config)\n}\n"],"names":[],"mappings":";;;;;AASA,MAAM,wBAAwB;AAsBvB,MAAM,0BAEH,eAA+C;AAAA,EAC9C,OAAO;AAAA,EAEC;AAAA,EACA;AAAA,EACA;AAAA,EAEjB,YAAY,QAA0B,OAAe;AACnD,UAAM,OAAO,MAAM;AACnB,SAAK,SAAS,OAAO;AACrB,SAAK,WAAW,OAAO,WAAW,uBAAuB,QAAQ,QAAQ,EAAE;AAC3E,SAAK,iBAAiB,OAAO,kBAAkB,CAAA;AAAA,EACjD;AAAA,EAEA,MAAM,eACJ,SACoB;AACpB,UAAM,EAAE,WAAW;AACnB,UAAM,EAAE,OAAO,MAAM,OAAO,QAAQ,iBAAiB;AAErD,WAAO,QAAQ,+CAA+C,KAAK,IAAI;AAAA,MACrE,UAAU;AAAA,MACV;AAAA,IAAA,CACD;AAED,UAAM,EAAE,MAAM,OAAO,yBAAA,IAA6B,oBAAoB;AAAA,MACpE;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,IAAA,CACD;AAED,QAAI;AACF,YAAM,WAAW,MAAM,MAAM,GAAG,KAAK,OAAO,QAAQ;AAAA,QAClD,QAAQ;AAAA,QACR,SAAS;AAAA;AAAA;AAAA;AAAA,UAIP,GAAG,KAAK;AAAA,UACR,eAAe,UAAU,KAAK,MAAM;AAAA,UACpC,gBAAgB;AAAA,QAAA;AAAA,QAElB,MAAM,KAAK,UAAU,IAAI;AAAA,MAAA,CAC1B;AAED,UAAI,CAAC,SAAS,IAAI;AAChB,cAAM,YAAY,MAAM,SAAS,KAAA;AACjC,cAAM,IAAI;AAAA,UACR,4BAA4B,SAAS,MAAM,IAAI,SAAS;AAAA,QAAA;AAAA,MAE5D;AAEA,YAAM,cAAc,MAAM,SAAS,YAAA;AACnC,YAAM,QAAQ,oBAAoB,WAAW;AAE7C,aAAO;AAAA,QACL,IAAI,WAAW,KAAK,IAAI;AAAA,QACxB;AAAA,QACA;AAAA,QACA,QAAQ;AAAA,QACR,aAAa,eAAe,OAAO,wBAAwB;AAAA,MAAA;AAAA,IAE/D,SAAS,OAAO;AACd,aAAO,OAAO,6BAA6B;AAAA,QACzC;AAAA,QACA,QAAQ;AAAA,MAAA,CACT;AACD,YAAM;AAAA,IACR;AAAA,EACF;AACF;AASO,SAAS,oBAAoB,SASlC;AACA,QAAM,EAAE,MAAM,OAAO,QAAQ,iBAAiB;AAE9C,QAAM,QAAQ,UAAU,cAAc,OAAO,MAAM;AAOnD,QAAM,mBAAmB,cAAc;AAIvC,QAAM,aAAa;AACnB,QAAM,yBAAyB,UAAU;AAEzC,QAAM,eAAwC,EAAE,MAAA;AAChD,MAAI,qBAAqB,QAAW;AAClC,iBAAa,cAAc;AAAA,EAC7B,WAAW,wBAAwB;AACjC,iBAAa,cAAc;AAAA,EAC7B;AACA,MAAI,UAAU,SAAS,cAAc,aAAa,QAAW;AAC3D,iBAAa,WAAW,aAAa;AAAA,EACvC;AAKA,QAAM,2BAA2B,oBAAoB;AAErD,QAAM,OAAgC;AAAA,IACpC;AAAA,IACA,UAAU,SAAS;AAAA,IACnB,UAAU,cAAc,YAAY;AAAA,IACpC,eAAe;AAAA,EAAA;AAEjB,MAAI,cAAc,+BAA+B,QAAW;AAC1D,SAAK,6BAA6B,aAAa;AAAA,EACjD;AACA,MAAI,cAAc,uBAAuB,QAAW;AAClD,SAAK,qBAAqB,aAAa;AAAA,EACzC;AAEA,SAAO,EAAE,MAAM,OAAO,yBAAA;AACxB;AAQA,SAAS,UACP,eACA,QACc;AACd,MAAI,cAAe,QAAO;AAC1B,MAAI,CAAC,OAAQ,QAAO;AACpB,UAAQ,QAAA;AAAA,IACN,KAAK;AAAA,IACL,KAAK;AAAA,IACL,KAAK;AACH,aAAO;AAAA,IACT,KAAK;AAAA,IACL,KAAK;AAAA,IACL,KAAK;AACH,aAAO;AAAA,IACT;AACE,aAAO;AAAA,EAAA;AAEb;AAEO,SAAS,eACd,OACA,YACQ;AACR,UAAQ,OAAA;AAAA,IACN,KAAK;AACH,aAAO;AAAA,IACT,KAAK;AACH,aAAO;AAAA,IACT,KAAK;AAEH,aAAO,kBAAkB,UAAU;AAAA,IACrC,KAAK;AAOH,aAAO,eAAe,MAClB,gBACA,mBAAmB,UAAU;AAAA,IACnC,KAAK;AACH,aAAO,eAAe,MAClB,uBACA,mBAAmB,UAAU;AAAA,EAAA;AAEvC;AAeO,SAAS,iBACd,OACA,QACA,QAC2B;AAC3B,SAAO,IAAI,kBAAkB,EAAE,QAAQ,GAAG,OAAA,GAAU,KAAK;AAC3D;AAQO,SAAS,WACd,OACA,QAC2B;AAC3B,QAAM,SAAS,qBAAA;AACf,SAAO,iBAAiB,OAAO,QAAQ,MAAM;AAC/C;"}

package/dist/esm/model-meta.d.ts

@@ -229,7 +229,26 @@ declare const GROK_4_3: {
         };
     };
 };
-export declare const GROK_CHAT_MODELS: readonly ["grok-4-1-fast-reasoning", "grok-4-1-fast-non-reasoning", "grok-code-fast-1", "grok-4-fast-reasoning", "grok-4-fast-non-reasoning", "grok-4", "grok-3", "grok-3-mini", "grok-2-vision-1212", "grok-4.20", "grok-4.20-multi-agent", "grok-4.3"];
+declare const GROK_BUILD_0_1: {
+    readonly name: "grok-build-0.1";
+    readonly context_window: 256000;
+    readonly supports: {
+        readonly input: ["text", "image"];
+        readonly output: ["text"];
+        readonly capabilities: ["reasoning", "structured_outputs", "tool_calling"];
+        readonly tools: readonly [];
+    };
+    readonly pricing: {
+        readonly input: {
+            readonly normal: 1;
+            readonly cached: 0.2;
+        };
+        readonly output: {
+            readonly normal: 2;
+        };
+    };
+};
+export declare const GROK_CHAT_MODELS: readonly ["grok-4-1-fast-reasoning", "grok-4-1-fast-non-reasoning", "grok-code-fast-1", "grok-4-fast-reasoning", "grok-4-fast-non-reasoning", "grok-4", "grok-3", "grok-3-mini", "grok-2-vision-1212", "grok-4.20", "grok-4.20-multi-agent", "grok-4.3", "grok-build-0.1"];
 /**
  * Grok Image Generation Models
  */
@@ -259,6 +278,7 @@ export type GrokModelInputModalitiesByName = {
     [GROK_4_20.name]: typeof GROK_4_20.supports.input;
     [GROK_4_20_MULTI_AGENT.name]: typeof GROK_4_20_MULTI_AGENT.supports.input;
     [GROK_4_3.name]: typeof GROK_4_3.supports.input;
+    [GROK_BUILD_0_1.name]: typeof GROK_BUILD_0_1.supports.input;
 };
 /**
  * Type-only map from Grok chat model name to its provider options type.

package/dist/esm/model-meta.js

@@ -37,6 +37,9 @@ const GROK_4_20_MULTI_AGENT = {
 const GROK_4_3 = {
   name: "grok-4.3"
 };
+const GROK_BUILD_0_1 = {
+  name: "grok-build-0.1"
+};
 const GROK_CHAT_MODELS = [
   GROK_4_1_FAST_REASONING.name,
   GROK_4_1_FAST_NON_REASONING.name,
@@ -49,7 +52,8 @@ const GROK_CHAT_MODELS = [
   GROK_2_VISION.name,
   GROK_4_20.name,
   GROK_4_20_MULTI_AGENT.name,
-  GROK_4_3.name
+  GROK_4_3.name,
+  GROK_BUILD_0_1.name
 ];
 const GROK_IMAGE_MODELS = [GROK_2_IMAGE.name];
 const GROK_TTS = {

package/dist/esm/model-meta.js.map

@@ -1 +1 @@
-{"version":3,"file":"model-meta.js","sources":["../../src/model-meta.ts"],"sourcesContent":["/**\n * Model metadata interface for documentation and type inference\n */\ninterface ModelMeta {\n  name: string\n  supports: {\n    input: Array<'text' | 'image' | 'audio' | 'video' | 'document'>\n    output: Array<'text' | 'image' | 'audio' | 'video'>\n    capabilities?: Array<'reasoning' | 'tool_calling' | 'structured_outputs'>\n    tools?: ReadonlyArray<never>\n  }\n  max_input_tokens?: number\n  max_output_tokens?: number\n  context_window?: number\n  knowledge_cutoff?: string\n  pricing?: {\n    input: {\n      normal: number\n      cached?: number\n    }\n    output: {\n      normal: number\n    }\n  }\n}\n\nconst GROK_4_1_FAST_REASONING = {\n  name: 'grok-4-1-fast-reasoning',\n  context_window: 2_000_000,\n  supports: {\n    input: ['text', 'image'],\n    output: ['text'],\n    capabilities: ['reasoning', 'structured_outputs', 'tool_calling'],\n    tools: [] as const,\n  },\n  pricing: {\n    input: {\n      normal: 0.2,\n      cached: 0.05,\n    },\n    output: {\n      normal: 0.5,\n    },\n  },\n} as const satisfies ModelMeta\n\nconst GROK_4_1_FAST_NON_REASONING = {\n  name: 'grok-4-1-fast-non-reasoning',\n  context_window: 2_000_000,\n  supports: {\n    input: ['text', 'image'],\n    output: ['text'],\n    capabilities: ['structured_outputs', 'tool_calling'],\n    tools: [] as const,\n  },\n  pricing: {\n    input: {\n      normal: 0.2,\n      cached: 0.05,\n    },\n    output: {\n      normal: 0.5,\n    },\n  },\n} as const satisfies ModelMeta\n\nconst GROK_CODE_FAST_1 = {\n  name: 'grok-code-fast-1',\n  context_window: 256_000,\n  supports: {\n    input: ['text'],\n    output: ['text'],\n    capabilities: ['reasoning', 'structured_outputs', 'tool_calling'],\n    tools: [] as const,\n  },\n  pricing: {\n    input: {\n      normal: 0.2,\n      cached: 0.02,\n    },\n    output: {\n      normal: 1.5,\n    },\n  },\n} as const satisfies ModelMeta\n\nconst GROK_4_FAST_REASONING = {\n  name: 'grok-4-fast-reasoning',\n  context_window: 2_000_000,\n  supports: {\n    input: ['text', 'image'],\n    output: ['text'],\n    capabilities: ['reasoning', 'structured_outputs', 'tool_calling'],\n    tools: [] as const,\n  },\n  pricing: {\n    input: {\n      normal: 0.2,\n      cached: 0.05,\n    },\n    output: {\n      normal: 0.5,\n    },\n  },\n} as const satisfies ModelMeta\n\nconst GROK_4_FAST_NON_REASONING = {\n  name: 'grok-4-fast-non-reasoning',\n  context_window: 2_000_000,\n  supports: {\n    input: ['text', 'image'],\n    output: ['text'],\n    capabilities: ['structured_outputs', 'tool_calling'],\n    tools: [] as const,\n  },\n  pricing: {\n    input: {\n      normal: 0.2,\n      cached: 0.05,\n    },\n    output: {\n      normal: 0.5,\n    },\n  },\n} as const satisfies ModelMeta\n\nconst GROK_4 = {\n  name: 'grok-4',\n  context_window: 256_000,\n  supports: {\n    input: ['text', 'image'],\n    output: ['text'],\n    capabilities: ['reasoning', 'structured_outputs', 'tool_calling'],\n    tools: [] as const,\n  },\n  pricing: {\n    input: {\n      normal: 3,\n      cached: 0.75,\n    },\n    output: {\n      normal: 15,\n    },\n  },\n} as const satisfies ModelMeta\n\nconst GROK_3_MINI = {\n  name: 'grok-3-mini',\n  context_window: 131_072,\n  supports: {\n    input: ['text'],\n    output: ['text'],\n    capabilities: ['reasoning', 'structured_outputs', 'tool_calling'],\n    tools: [] as const,\n  },\n  pricing: {\n    input: {\n      normal: 0.3,\n      cached: 0.075,\n    },\n    output: {\n      normal: 0.5,\n    },\n  },\n} as const satisfies ModelMeta\n\nconst GROK_3 = {\n  name: 'grok-3',\n  context_window: 131_072,\n  supports: {\n    input: ['text'],\n    output: ['text'],\n    capabilities: ['structured_outputs', 'tool_calling'],\n    tools: [] as const,\n  },\n  pricing: {\n    input: {\n      normal: 3,\n      cached: 0.75,\n    },\n    output: {\n      normal: 15,\n    },\n  },\n} as const satisfies ModelMeta\n\nconst GROK_2_VISION = {\n  name: 'grok-2-vision-1212',\n  context_window: 32_768,\n  supports: {\n    input: ['text', 'image'],\n    output: ['text'],\n    capabilities: ['structured_outputs', 'tool_calling'],\n    tools: [] as const,\n  },\n  pricing: {\n    input: {\n      normal: 2,\n    },\n    output: {\n      normal: 10,\n    },\n  },\n} as const satisfies ModelMeta\n\nconst GROK_2_IMAGE = {\n  name: 'grok-2-image-1212',\n  supports: {\n    input: ['text'],\n    output: ['image'],\n  },\n  pricing: {\n    input: {\n      normal: 0.07,\n    },\n    output: {\n      normal: 0.07,\n    },\n  },\n} as const satisfies ModelMeta\n\n/**\n * Grok Chat Models\n * Based on xAI's available models as of 2025\n */\nconst GROK_4_20 = {\n  name: 'grok-4.20',\n  context_window: 2_000_000,\n  supports: {\n    input: ['text', 'image', 'document'],\n    output: ['text'],\n    capabilities: ['reasoning', 'structured_outputs', 'tool_calling'],\n    tools: [] as const,\n  },\n  pricing: {\n    input: {\n      normal: 2,\n      cached: 0.2,\n    },\n    output: {\n      normal: 6,\n    },\n  },\n} as const satisfies ModelMeta\n\nconst GROK_4_20_MULTI_AGENT = {\n  name: 'grok-4.20-multi-agent',\n  context_window: 2_000_000,\n  supports: {\n    input: ['text', 'image', 'document'],\n    output: ['text'],\n    capabilities: ['reasoning', 'structured_outputs', 'tool_calling'],\n    tools: [] as const,\n  },\n  pricing: {\n    input: {\n      normal: 2,\n      cached: 0.2,\n    },\n    output: {\n      normal: 6,\n    },\n  },\n} as const satisfies ModelMeta\n\nconst GROK_4_3 = {\n  name: 'grok-4.3',\n  context_window: 1_000_000,\n  supports: {\n    input: ['text', 'image'],\n    output: ['text'],\n    capabilities: ['reasoning', 'structured_outputs', 'tool_calling'],\n    tools: [],\n  },\n  pricing: {\n    input: {\n      normal: 1.25,\n      cached: 0.2,\n    },\n    output: {\n      normal: 2.5,\n    },\n  },\n} as const satisfies ModelMeta\n\nexport const GROK_CHAT_MODELS = [\n  GROK_4_1_FAST_REASONING.name,\n  GROK_4_1_FAST_NON_REASONING.name,\n  GROK_CODE_FAST_1.name,\n  GROK_4_FAST_REASONING.name,\n  GROK_4_FAST_NON_REASONING.name,\n  GROK_4.name,\n  GROK_3.name,\n  GROK_3_MINI.name,\n  GROK_2_VISION.name,\n\n  GROK_4_20.name,\n  GROK_4_20_MULTI_AGENT.name,\n\n  GROK_4_3.name,\n] as const\n\n/**\n * Grok Image Generation Models\n */\nexport const GROK_IMAGE_MODELS = [GROK_2_IMAGE.name] as const\n\n// xAI's `/v1/tts` endpoint is endpoint-addressed and does not take a `model`\n// parameter. This synthetic identifier satisfies the SDK's `TTSOptions.model`\n// contract and provides a stable value for logging and fixture matching.\nconst GROK_TTS = {\n  name: 'grok-tts',\n  supports: {\n    input: ['text'],\n    output: ['audio'],\n  },\n} as const satisfies ModelMeta\n\n// xAI's `/v1/stt` endpoint is endpoint-addressed and does not take a `model`\n// parameter. This synthetic identifier satisfies the SDK's\n// `TranscriptionOptions.model` contract.\nconst GROK_STT = {\n  name: 'grok-stt',\n  supports: {\n    input: ['audio'],\n    output: ['text'],\n  },\n} as const satisfies ModelMeta\n\nconst GROK_VOICE_FAST_1 = {\n  name: 'grok-voice-fast-1.0',\n  supports: {\n    input: ['audio', 'text'],\n    output: ['audio', 'text'],\n    capabilities: ['tool_calling'],\n    tools: [] as const,\n  },\n} as const satisfies ModelMeta\n\nconst GROK_VOICE_THINK_FAST_1 = {\n  name: 'grok-voice-think-fast-1.0',\n  supports: {\n    input: ['audio', 'text'],\n    output: ['audio', 'text'],\n    capabilities: ['reasoning', 'tool_calling'],\n    tools: [] as const,\n  },\n} as const satisfies ModelMeta\n\nexport const GROK_TTS_MODELS = [GROK_TTS.name] as const\n\nexport const GROK_TRANSCRIPTION_MODELS = [GROK_STT.name] as const\n\nexport const GROK_REALTIME_MODELS = [\n  GROK_VOICE_FAST_1.name,\n  GROK_VOICE_THINK_FAST_1.name,\n] as const\n\nexport type GrokChatModel = (typeof GROK_CHAT_MODELS)[number]\nexport type GrokImageModel = (typeof GROK_IMAGE_MODELS)[number]\nexport type GrokTTSModel = (typeof GROK_TTS_MODELS)[number]\nexport type GrokTranscriptionModel = (typeof GROK_TRANSCRIPTION_MODELS)[number]\nexport type GrokRealtimeModel = (typeof GROK_REALTIME_MODELS)[number]\n\n/**\n * Type-only map from Grok chat model name to its supported input modalities.\n * Used for type inference when constructing multimodal messages.\n */\nexport type GrokModelInputModalitiesByName = {\n  [GROK_4_1_FAST_REASONING.name]: typeof GROK_4_1_FAST_REASONING.supports.input\n  [GROK_4_1_FAST_NON_REASONING.name]: typeof GROK_4_1_FAST_NON_REASONING.supports.input\n  [GROK_CODE_FAST_1.name]: typeof GROK_CODE_FAST_1.supports.input\n  [GROK_4_FAST_REASONING.name]: typeof GROK_4_FAST_REASONING.supports.input\n  [GROK_4_FAST_NON_REASONING.name]: typeof GROK_4_FAST_NON_REASONING.supports.input\n  [GROK_4.name]: typeof GROK_4.supports.input\n  [GROK_3.name]: typeof GROK_3.supports.input\n  [GROK_3_MINI.name]: typeof GROK_3_MINI.supports.input\n  [GROK_2_VISION.name]: typeof GROK_2_VISION.supports.input\n  [GROK_4_20.name]: typeof GROK_4_20.supports.input\n  [GROK_4_20_MULTI_AGENT.name]: typeof GROK_4_20_MULTI_AGENT.supports.input\n  [GROK_4_3.name]: typeof GROK_4_3.supports.input\n}\n\n/**\n * Type-only map from Grok chat model name to its provider options type.\n * Since Grok uses OpenAI-compatible API, we reuse OpenAI provider options.\n */\nexport type GrokChatModelProviderOptionsByName = {\n  [K in (typeof GROK_CHAT_MODELS)[number]]: GrokProviderOptions\n}\n\n/**\n * Type-only map from Grok chat model name to its supported provider tools.\n * Grok exposes no provider-specific tool factories, so every model gets an\n * empty tuple. This ensures that passing an Anthropic/OpenAI ProviderTool to\n * a Grok adapter produces a compile-time type error.\n */\nexport type GrokChatModelToolCapabilitiesByName = {\n  [GROK_4_1_FAST_REASONING.name]: typeof GROK_4_1_FAST_REASONING.supports.tools\n  [GROK_4_1_FAST_NON_REASONING.name]: typeof GROK_4_1_FAST_NON_REASONING.supports.tools\n  [GROK_CODE_FAST_1.name]: typeof GROK_CODE_FAST_1.supports.tools\n  [GROK_4_FAST_REASONING.name]: typeof GROK_4_FAST_REASONING.supports.tools\n  [GROK_4_FAST_NON_REASONING.name]: typeof GROK_4_FAST_NON_REASONING.supports.tools\n  [GROK_4.name]: typeof GROK_4.supports.tools\n  [GROK_3.name]: typeof GROK_3.supports.tools\n  [GROK_3_MINI.name]: typeof GROK_3_MINI.supports.tools\n  [GROK_2_VISION.name]: typeof GROK_2_VISION.supports.tools\n  [GROK_4_20.name]: typeof GROK_4_20.supports.tools\n  [GROK_4_20_MULTI_AGENT.name]: typeof GROK_4_20_MULTI_AGENT.supports.tools\n}\n\n/**\n * Grok-specific provider options\n * Based on OpenAI-compatible API options\n */\nexport interface GrokProviderOptions {\n  /** Temperature for response generation (0-2) */\n  temperature?: number\n  /** Maximum tokens in the response */\n  max_tokens?: number\n  /** Top-p sampling parameter */\n  top_p?: number\n  /** Frequency penalty (-2.0 to 2.0) */\n  frequency_penalty?: number\n  /** Presence penalty (-2.0 to 2.0) */\n  presence_penalty?: number\n  /** Stop sequences */\n  stop?: string | Array<string>\n  /** A unique identifier representing your end-user */\n  user?: string\n}\n\n// ===========================\n// Type Resolution Helpers\n// ===========================\n\n/**\n * Resolve provider options for a specific model.\n * If the model has explicit options in the map, use those; otherwise use base options.\n */\nexport type ResolveProviderOptions<TModel extends string> =\n  TModel extends keyof GrokChatModelProviderOptionsByName\n    ? GrokChatModelProviderOptionsByName[TModel]\n    : GrokProviderOptions\n\n/**\n * Resolve input modalities for a specific model.\n * If the model has explicit modalities in the map, use those; otherwise use text only.\n */\nexport type ResolveInputModalities<TModel extends string> =\n  TModel extends keyof GrokModelInputModalitiesByName\n    ? GrokModelInputModalitiesByName[TModel]\n    : readonly ['text']\n"],"names":[],"mappings":"AA0BA,MAAM,0BAA0B;AAAA,EAC9B,MAAM;AAiBR;AAEA,MAAM,8BAA8B;AAAA,EAClC,MAAM;AAiBR;AAEA,MAAM,mBAAmB;AAAA,EACvB,MAAM;AAiBR;AAEA,MAAM,wBAAwB;AAAA,EAC5B,MAAM;AAiBR;AAEA,MAAM,4BAA4B;AAAA,EAChC,MAAM;AAiBR;AAEA,MAAM,SAAS;AAAA,EACb,MAAM;AAiBR;AAEA,MAAM,cAAc;AAAA,EAClB,MAAM;AAiBR;AAEA,MAAM,SAAS;AAAA,EACb,MAAM;AAiBR;AAEA,MAAM,gBAAgB;AAAA,EACpB,MAAM;AAgBR;AAEA,MAAM,eAAe;AAAA,EACnB,MAAM;AAaR;AAMA,MAAM,YAAY;AAAA,EAChB,MAAM;AAiBR;AAEA,MAAM,wBAAwB;AAAA,EAC5B,MAAM;AAiBR;AAEA,MAAM,WAAW;AAAA,EACf,MAAM;AAiBR;AAEO,MAAM,mBAAmB;AAAA,EAC9B,wBAAwB;AAAA,EACxB,4BAA4B;AAAA,EAC5B,iBAAiB;AAAA,EACjB,sBAAsB;AAAA,EACtB,0BAA0B;AAAA,EAC1B,OAAO;AAAA,EACP,OAAO;AAAA,EACP,YAAY;AAAA,EACZ,cAAc;AAAA,EAEd,UAAU;AAAA,EACV,sBAAsB;AAAA,EAEtB,SAAS;AACX;AAKO,MAAM,oBAAoB,CAAC,aAAa,IAAI;AAKnD,MAAM,WAAW;AAAA,EACf,MAAM;AAKR;AAKA,MAAM,WAAW;AAAA,EACf,MAAM;AAKR;AAEA,MAAM,oBAAoB;AAAA,EACxB,MAAM;AAOR;AAEA,MAAM,0BAA0B;AAAA,EAC9B,MAAM;AAOR;AAEO,MAAM,kBAAkB,CAAC,SAAS,IAAI;AAEtC,MAAM,4BAA4B,CAAC,SAAS,IAAI;AAEhD,MAAM,uBAAuB;AAAA,EAClC,kBAAkB;AAAA,EAClB,wBAAwB;AAC1B;"}
+{"version":3,"file":"model-meta.js","sources":["../../src/model-meta.ts"],"sourcesContent":["/**\n * Model metadata interface for documentation and type inference\n */\ninterface ModelMeta {\n  name: string\n  supports: {\n    input: Array<'text' | 'image' | 'audio' | 'video' | 'document'>\n    output: Array<'text' | 'image' | 'audio' | 'video'>\n    capabilities?: Array<'reasoning' | 'tool_calling' | 'structured_outputs'>\n    tools?: ReadonlyArray<never>\n  }\n  max_input_tokens?: number\n  max_output_tokens?: number\n  context_window?: number\n  knowledge_cutoff?: string\n  pricing?: {\n    input: {\n      normal: number\n      cached?: number\n    }\n    output: {\n      normal: number\n    }\n  }\n}\n\nconst GROK_4_1_FAST_REASONING = {\n  name: 'grok-4-1-fast-reasoning',\n  context_window: 2_000_000,\n  supports: {\n    input: ['text', 'image'],\n    output: ['text'],\n    capabilities: ['reasoning', 'structured_outputs', 'tool_calling'],\n    tools: [] as const,\n  },\n  pricing: {\n    input: {\n      normal: 0.2,\n      cached: 0.05,\n    },\n    output: {\n      normal: 0.5,\n    },\n  },\n} as const satisfies ModelMeta\n\nconst GROK_4_1_FAST_NON_REASONING = {\n  name: 'grok-4-1-fast-non-reasoning',\n  context_window: 2_000_000,\n  supports: {\n    input: ['text', 'image'],\n    output: ['text'],\n    capabilities: ['structured_outputs', 'tool_calling'],\n    tools: [] as const,\n  },\n  pricing: {\n    input: {\n      normal: 0.2,\n      cached: 0.05,\n    },\n    output: {\n      normal: 0.5,\n    },\n  },\n} as const satisfies ModelMeta\n\nconst GROK_CODE_FAST_1 = {\n  name: 'grok-code-fast-1',\n  context_window: 256_000,\n  supports: {\n    input: ['text'],\n    output: ['text'],\n    capabilities: ['reasoning', 'structured_outputs', 'tool_calling'],\n    tools: [] as const,\n  },\n  pricing: {\n    input: {\n      normal: 0.2,\n      cached: 0.02,\n    },\n    output: {\n      normal: 1.5,\n    },\n  },\n} as const satisfies ModelMeta\n\nconst GROK_4_FAST_REASONING = {\n  name: 'grok-4-fast-reasoning',\n  context_window: 2_000_000,\n  supports: {\n    input: ['text', 'image'],\n    output: ['text'],\n    capabilities: ['reasoning', 'structured_outputs', 'tool_calling'],\n    tools: [] as const,\n  },\n  pricing: {\n    input: {\n      normal: 0.2,\n      cached: 0.05,\n    },\n    output: {\n      normal: 0.5,\n    },\n  },\n} as const satisfies ModelMeta\n\nconst GROK_4_FAST_NON_REASONING = {\n  name: 'grok-4-fast-non-reasoning',\n  context_window: 2_000_000,\n  supports: {\n    input: ['text', 'image'],\n    output: ['text'],\n    capabilities: ['structured_outputs', 'tool_calling'],\n    tools: [] as const,\n  },\n  pricing: {\n    input: {\n      normal: 0.2,\n      cached: 0.05,\n    },\n    output: {\n      normal: 0.5,\n    },\n  },\n} as const satisfies ModelMeta\n\nconst GROK_4 = {\n  name: 'grok-4',\n  context_window: 256_000,\n  supports: {\n    input: ['text', 'image'],\n    output: ['text'],\n    capabilities: ['reasoning', 'structured_outputs', 'tool_calling'],\n    tools: [] as const,\n  },\n  pricing: {\n    input: {\n      normal: 3,\n      cached: 0.75,\n    },\n    output: {\n      normal: 15,\n    },\n  },\n} as const satisfies ModelMeta\n\nconst GROK_3_MINI = {\n  name: 'grok-3-mini',\n  context_window: 131_072,\n  supports: {\n    input: ['text'],\n    output: ['text'],\n    capabilities: ['reasoning', 'structured_outputs', 'tool_calling'],\n    tools: [] as const,\n  },\n  pricing: {\n    input: {\n      normal: 0.3,\n      cached: 0.075,\n    },\n    output: {\n      normal: 0.5,\n    },\n  },\n} as const satisfies ModelMeta\n\nconst GROK_3 = {\n  name: 'grok-3',\n  context_window: 131_072,\n  supports: {\n    input: ['text'],\n    output: ['text'],\n    capabilities: ['structured_outputs', 'tool_calling'],\n    tools: [] as const,\n  },\n  pricing: {\n    input: {\n      normal: 3,\n      cached: 0.75,\n    },\n    output: {\n      normal: 15,\n    },\n  },\n} as const satisfies ModelMeta\n\nconst GROK_2_VISION = {\n  name: 'grok-2-vision-1212',\n  context_window: 32_768,\n  supports: {\n    input: ['text', 'image'],\n    output: ['text'],\n    capabilities: ['structured_outputs', 'tool_calling'],\n    tools: [] as const,\n  },\n  pricing: {\n    input: {\n      normal: 2,\n    },\n    output: {\n      normal: 10,\n    },\n  },\n} as const satisfies ModelMeta\n\nconst GROK_2_IMAGE = {\n  name: 'grok-2-image-1212',\n  supports: {\n    input: ['text'],\n    output: ['image'],\n  },\n  pricing: {\n    input: {\n      normal: 0.07,\n    },\n    output: {\n      normal: 0.07,\n    },\n  },\n} as const satisfies ModelMeta\n\n/**\n * Grok Chat Models\n * Based on xAI's available models as of 2025\n */\nconst GROK_4_20 = {\n  name: 'grok-4.20',\n  context_window: 2_000_000,\n  supports: {\n    input: ['text', 'image', 'document'],\n    output: ['text'],\n    capabilities: ['reasoning', 'structured_outputs', 'tool_calling'],\n    tools: [] as const,\n  },\n  pricing: {\n    input: {\n      normal: 2,\n      cached: 0.2,\n    },\n    output: {\n      normal: 6,\n    },\n  },\n} as const satisfies ModelMeta\n\nconst GROK_4_20_MULTI_AGENT = {\n  name: 'grok-4.20-multi-agent',\n  context_window: 2_000_000,\n  supports: {\n    input: ['text', 'image', 'document'],\n    output: ['text'],\n    capabilities: ['reasoning', 'structured_outputs', 'tool_calling'],\n    tools: [] as const,\n  },\n  pricing: {\n    input: {\n      normal: 2,\n      cached: 0.2,\n    },\n    output: {\n      normal: 6,\n    },\n  },\n} as const satisfies ModelMeta\n\nconst GROK_4_3 = {\n  name: 'grok-4.3',\n  context_window: 1_000_000,\n  supports: {\n    input: ['text', 'image'],\n    output: ['text'],\n    capabilities: ['reasoning', 'structured_outputs', 'tool_calling'],\n    tools: [],\n  },\n  pricing: {\n    input: {\n      normal: 1.25,\n      cached: 0.2,\n    },\n    output: {\n      normal: 2.5,\n    },\n  },\n} as const satisfies ModelMeta\n\nconst GROK_BUILD_0_1 = {\n  name: 'grok-build-0.1',\n  context_window: 256_000,\n  supports: {\n    input: ['text', 'image'],\n    output: ['text'],\n    capabilities: ['reasoning', 'structured_outputs', 'tool_calling'],\n    tools: [],\n  },\n  pricing: {\n    input: {\n      normal: 1,\n      cached: 0.2,\n    },\n    output: {\n      normal: 2,\n    },\n  },\n} as const satisfies ModelMeta\n\nexport const GROK_CHAT_MODELS = [\n  GROK_4_1_FAST_REASONING.name,\n  GROK_4_1_FAST_NON_REASONING.name,\n  GROK_CODE_FAST_1.name,\n  GROK_4_FAST_REASONING.name,\n  GROK_4_FAST_NON_REASONING.name,\n  GROK_4.name,\n  GROK_3.name,\n  GROK_3_MINI.name,\n  GROK_2_VISION.name,\n\n  GROK_4_20.name,\n  GROK_4_20_MULTI_AGENT.name,\n\n  GROK_4_3.name,\n\n  GROK_BUILD_0_1.name,\n] as const\n\n/**\n * Grok Image Generation Models\n */\nexport const GROK_IMAGE_MODELS = [GROK_2_IMAGE.name] as const\n\n// xAI's `/v1/tts` endpoint is endpoint-addressed and does not take a `model`\n// parameter. This synthetic identifier satisfies the SDK's `TTSOptions.model`\n// contract and provides a stable value for logging and fixture matching.\nconst GROK_TTS = {\n  name: 'grok-tts',\n  supports: {\n    input: ['text'],\n    output: ['audio'],\n  },\n} as const satisfies ModelMeta\n\n// xAI's `/v1/stt` endpoint is endpoint-addressed and does not take a `model`\n// parameter. This synthetic identifier satisfies the SDK's\n// `TranscriptionOptions.model` contract.\nconst GROK_STT = {\n  name: 'grok-stt',\n  supports: {\n    input: ['audio'],\n    output: ['text'],\n  },\n} as const satisfies ModelMeta\n\nconst GROK_VOICE_FAST_1 = {\n  name: 'grok-voice-fast-1.0',\n  supports: {\n    input: ['audio', 'text'],\n    output: ['audio', 'text'],\n    capabilities: ['tool_calling'],\n    tools: [] as const,\n  },\n} as const satisfies ModelMeta\n\nconst GROK_VOICE_THINK_FAST_1 = {\n  name: 'grok-voice-think-fast-1.0',\n  supports: {\n    input: ['audio', 'text'],\n    output: ['audio', 'text'],\n    capabilities: ['reasoning', 'tool_calling'],\n    tools: [] as const,\n  },\n} as const satisfies ModelMeta\n\nexport const GROK_TTS_MODELS = [GROK_TTS.name] as const\n\nexport const GROK_TRANSCRIPTION_MODELS = [GROK_STT.name] as const\n\nexport const GROK_REALTIME_MODELS = [\n  GROK_VOICE_FAST_1.name,\n  GROK_VOICE_THINK_FAST_1.name,\n] as const\n\nexport type GrokChatModel = (typeof GROK_CHAT_MODELS)[number]\nexport type GrokImageModel = (typeof GROK_IMAGE_MODELS)[number]\nexport type GrokTTSModel = (typeof GROK_TTS_MODELS)[number]\nexport type GrokTranscriptionModel = (typeof GROK_TRANSCRIPTION_MODELS)[number]\nexport type GrokRealtimeModel = (typeof GROK_REALTIME_MODELS)[number]\n\n/**\n * Type-only map from Grok chat model name to its supported input modalities.\n * Used for type inference when constructing multimodal messages.\n */\nexport type GrokModelInputModalitiesByName = {\n  [GROK_4_1_FAST_REASONING.name]: typeof GROK_4_1_FAST_REASONING.supports.input\n  [GROK_4_1_FAST_NON_REASONING.name]: typeof GROK_4_1_FAST_NON_REASONING.supports.input\n  [GROK_CODE_FAST_1.name]: typeof GROK_CODE_FAST_1.supports.input\n  [GROK_4_FAST_REASONING.name]: typeof GROK_4_FAST_REASONING.supports.input\n  [GROK_4_FAST_NON_REASONING.name]: typeof GROK_4_FAST_NON_REASONING.supports.input\n  [GROK_4.name]: typeof GROK_4.supports.input\n  [GROK_3.name]: typeof GROK_3.supports.input\n  [GROK_3_MINI.name]: typeof GROK_3_MINI.supports.input\n  [GROK_2_VISION.name]: typeof GROK_2_VISION.supports.input\n  [GROK_4_20.name]: typeof GROK_4_20.supports.input\n  [GROK_4_20_MULTI_AGENT.name]: typeof GROK_4_20_MULTI_AGENT.supports.input\n  [GROK_4_3.name]: typeof GROK_4_3.supports.input\n  [GROK_BUILD_0_1.name]: typeof GROK_BUILD_0_1.supports.input\n}\n\n/**\n * Type-only map from Grok chat model name to its provider options type.\n * Since Grok uses OpenAI-compatible API, we reuse OpenAI provider options.\n */\nexport type GrokChatModelProviderOptionsByName = {\n  [K in (typeof GROK_CHAT_MODELS)[number]]: GrokProviderOptions\n}\n\n/**\n * Type-only map from Grok chat model name to its supported provider tools.\n * Grok exposes no provider-specific tool factories, so every model gets an\n * empty tuple. This ensures that passing an Anthropic/OpenAI ProviderTool to\n * a Grok adapter produces a compile-time type error.\n */\nexport type GrokChatModelToolCapabilitiesByName = {\n  [GROK_4_1_FAST_REASONING.name]: typeof GROK_4_1_FAST_REASONING.supports.tools\n  [GROK_4_1_FAST_NON_REASONING.name]: typeof GROK_4_1_FAST_NON_REASONING.supports.tools\n  [GROK_CODE_FAST_1.name]: typeof GROK_CODE_FAST_1.supports.tools\n  [GROK_4_FAST_REASONING.name]: typeof GROK_4_FAST_REASONING.supports.tools\n  [GROK_4_FAST_NON_REASONING.name]: typeof GROK_4_FAST_NON_REASONING.supports.tools\n  [GROK_4.name]: typeof GROK_4.supports.tools\n  [GROK_3.name]: typeof GROK_3.supports.tools\n  [GROK_3_MINI.name]: typeof GROK_3_MINI.supports.tools\n  [GROK_2_VISION.name]: typeof GROK_2_VISION.supports.tools\n  [GROK_4_20.name]: typeof GROK_4_20.supports.tools\n  [GROK_4_20_MULTI_AGENT.name]: typeof GROK_4_20_MULTI_AGENT.supports.tools\n}\n\n/**\n * Grok-specific provider options\n * Based on OpenAI-compatible API options\n */\nexport interface GrokProviderOptions {\n  /** Temperature for response generation (0-2) */\n  temperature?: number\n  /** Maximum tokens in the response */\n  max_tokens?: number\n  /** Top-p sampling parameter */\n  top_p?: number\n  /** Frequency penalty (-2.0 to 2.0) */\n  frequency_penalty?: number\n  /** Presence penalty (-2.0 to 2.0) */\n  presence_penalty?: number\n  /** Stop sequences */\n  stop?: string | Array<string>\n  /** A unique identifier representing your end-user */\n  user?: string\n}\n\n// ===========================\n// Type Resolution Helpers\n// ===========================\n\n/**\n * Resolve provider options for a specific model.\n * If the model has explicit options in the map, use those; otherwise use base options.\n */\nexport type ResolveProviderOptions<TModel extends string> =\n  TModel extends keyof GrokChatModelProviderOptionsByName\n    ? GrokChatModelProviderOptionsByName[TModel]\n    : GrokProviderOptions\n\n/**\n * Resolve input modalities for a specific model.\n * If the model has explicit modalities in the map, use those; otherwise use text only.\n */\nexport type ResolveInputModalities<TModel extends string> =\n  TModel extends keyof GrokModelInputModalitiesByName\n    ? GrokModelInputModalitiesByName[TModel]\n    : readonly ['text']\n"],"names":[],"mappings":"AA0BA,MAAM,0BAA0B;AAAA,EAC9B,MAAM;AAiBR;AAEA,MAAM,8BAA8B;AAAA,EAClC,MAAM;AAiBR;AAEA,MAAM,mBAAmB;AAAA,EACvB,MAAM;AAiBR;AAEA,MAAM,wBAAwB;AAAA,EAC5B,MAAM;AAiBR;AAEA,MAAM,4BAA4B;AAAA,EAChC,MAAM;AAiBR;AAEA,MAAM,SAAS;AAAA,EACb,MAAM;AAiBR;AAEA,MAAM,cAAc;AAAA,EAClB,MAAM;AAiBR;AAEA,MAAM,SAAS;AAAA,EACb,MAAM;AAiBR;AAEA,MAAM,gBAAgB;AAAA,EACpB,MAAM;AAgBR;AAEA,MAAM,eAAe;AAAA,EACnB,MAAM;AAaR;AAMA,MAAM,YAAY;AAAA,EAChB,MAAM;AAiBR;AAEA,MAAM,wBAAwB;AAAA,EAC5B,MAAM;AAiBR;AAEA,MAAM,WAAW;AAAA,EACf,MAAM;AAiBR;AAEA,MAAM,iBAAiB;AAAA,EACrB,MAAM;AAiBR;AAEO,MAAM,mBAAmB;AAAA,EAC9B,wBAAwB;AAAA,EACxB,4BAA4B;AAAA,EAC5B,iBAAiB;AAAA,EACjB,sBAAsB;AAAA,EACtB,0BAA0B;AAAA,EAC1B,OAAO;AAAA,EACP,OAAO;AAAA,EACP,YAAY;AAAA,EACZ,cAAc;AAAA,EAEd,UAAU;AAAA,EACV,sBAAsB;AAAA,EAEtB,SAAS;AAAA,EAET,eAAe;AACjB;AAKO,MAAM,oBAAoB,CAAC,aAAa,IAAI;AAKnD,MAAM,WAAW;AAAA,EACf,MAAM;AAKR;AAKA,MAAM,WAAW;AAAA,EACf,MAAM;AAKR;AAEA,MAAM,oBAAoB;AAAA,EACxB,MAAM;AAOR;AAEA,MAAM,0BAA0B;AAAA,EAC9B,MAAM;AAOR;AAEO,MAAM,kBAAkB,CAAC,SAAS,IAAI;AAEtC,MAAM,4BAA4B,CAAC,SAAS,IAAI;AAEhD,MAAM,uBAAuB;AAAA,EAClC,kBAAkB;AAAA,EAClB,wBAAwB;AAC1B;"}

package/dist/esm/realtime/adapter.js

@@ -49,7 +49,8 @@ async function createWebRTCConnection(token, logger) {
   let outputSource = null;
   let localStream = null;
   let audioElement = null;
-  let dataChannel = null;
+  const channel = pc.createDataChannel("oai-events");
+  let dataChannel = channel;
   let currentMode = "idle";
   let currentMessageId = null;
   let isTornDown = false;
@@ -66,7 +67,6 @@ async function createWebRTCConnection(token, logger) {
       }
     }
   }
-  dataChannel = pc.createDataChannel("oai-events");
   let dataChannelOpened = false;
   let rejectDataChannelReady = null;
   let dataChannelReadyTimeout = null;
@@ -88,7 +88,7 @@ async function createWebRTCConnection(token, logger) {
         );
       }
     }, 15e3);
-    dataChannel.onopen = () => {
+    channel.onopen = () => {
       dataChannelOpened = true;
       if (dataChannelReadyTimeout !== null) {
         clearTimeout(dataChannelReadyTimeout);
@@ -100,7 +100,7 @@ async function createWebRTCConnection(token, logger) {
       resolve();
     };
   });
-  dataChannel.onmessage = (event) => {
+  channel.onmessage = (event) => {
     try {
       const message = JSON.parse(event.data);
       const messageRecord = message !== null && typeof message === "object" ? message : {};
@@ -119,7 +119,7 @@ async function createWebRTCConnection(token, logger) {
       });
     }
   };
-  dataChannel.onerror = (error) => {
+  channel.onerror = (error) => {
     if (isTornDown) return;
     logger.errors("grok.realtime fatal", {
       error,
@@ -134,7 +134,7 @@ async function createWebRTCConnection(token, logger) {
     }
     emit("error", { error: dcErr });
   };
-  dataChannel.onclose = () => {
+  channel.onclose = () => {
     if (isTornDown) return;
     if (!dataChannelOpened) {
       rejectDataChannelReady?.(new Error("Data channel closed before opening"));
@@ -461,7 +461,9 @@ async function createWebRTCConnection(token, logger) {
           currentMode = "listening";
           emit("mode_change", { mode: "listening" });
         }
-        emit("interrupted", { messageId: currentMessageId ?? void 0 });
+        emit("interrupted", {
+          ...currentMessageId !== null && { messageId: currentMessageId }
+        });
         break;
       case "error": {
         const errorObj = readObject(event, "error") ?? {};
@@ -480,6 +482,7 @@ async function createWebRTCConnection(token, logger) {
         emit("error", { error: err });
         break;
       }
+      case void 0:
       default:
         logger.provider("grok.realtime unhandled server event", {
           type: event.type
@@ -592,7 +595,7 @@ async function createWebRTCConnection(token, logger) {
           `provider=grok direction=out type=${readString(event, "type") ?? "<unknown>"}`,
           { frame: event }
         );
-        dataChannel.send(JSON.stringify(event));
+        channel.send(JSON.stringify(event));
       }
       pendingEvents.length = 0;
     } catch (error) {
@@ -733,13 +736,17 @@ async function createWebRTCConnection(token, logger) {
       sendEvent({ type: "response.cancel" });
       currentMode = "listening";
       emit("mode_change", { mode: "listening" });
-      emit("interrupted", { messageId: currentMessageId ?? void 0 });
+      emit("interrupted", {
+        ...currentMessageId !== null && { messageId: currentMessageId }
+      });
     },
     on(event, handler) {
-      if (!eventHandlers.has(event)) {
-        eventHandlers.set(event, /* @__PURE__ */ new Set());
+      let handlers = eventHandlers.get(event);
+      if (!handlers) {
+        handlers = /* @__PURE__ */ new Set();
+        eventHandlers.set(event, handlers);
       }
-      eventHandlers.get(event).add(handler);
+      handlers.add(handler);
       return () => {
         eventHandlers.get(event)?.delete(handler);
       };