@tanstack/ai-grok 0.14.8 → 0.14.10

package/dist/esm/adapters/tts.js

@@ -1,142 +1,159 @@
-import { BaseTTSAdapter } from "@tanstack/ai/adapters";
-import { generateId } from "@tanstack/ai-utils";
 import { getGrokApiKeyFromEnv } from "../utils/client.js";
-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.apiKey = config.apiKey;
-    this.baseURL = (config.baseURL ?? DEFAULT_GROK_BASE_URL).replace(/\/+$/, "");
-    this.defaultHeaders = config.defaultHeaders ?? {};
-  }
-  async generateSpeech(options) {
-    const { logger } = options;
-    const { model, text, voice, format, modelOptions } = options;
-    logger.request(`activity=generateSpeech provider=grok model=${model}`, {
-      provider: "grok",
-      model
-    });
-    const { body, codec, sampleRateForContentType } = buildTTSRequestBody({
-      text,
-      voice,
-      format,
-      modelOptions
-    });
-    try {
-      const response = await fetch(`${this.baseURL}/tts`, {
-        method: "POST",
-        headers: {
-          // `defaultHeaders` first so the adapter's Authorization / Content-Type
-          // always win — otherwise a caller-supplied `Authorization` header
-          // could silently clobber the bearer token.
-          ...this.defaultHeaders,
-          Authorization: `Bearer ${this.apiKey}`,
-          "Content-Type": "application/json"
-        },
-        body: JSON.stringify(body)
-      });
-      if (!response.ok) {
-        const errorText = await response.text();
-        throw new Error(
-          `Grok TTS request failed: ${response.status} ${errorText}`
-        );
-      }
-      const arrayBuffer = await response.arrayBuffer();
-      const audio = arrayBufferToBase64(arrayBuffer);
-      return {
-        id: generateId(this.name),
-        model,
-        audio,
-        format: codec,
-        contentType: getContentType(codec, sampleRateForContentType)
-      };
-    } catch (error) {
-      logger.errors("grok.generateSpeech fatal", {
-        error,
-        source: "grok.generateSpeech"
-      });
-      throw error;
-    }
-  }
-}
+import { generateId } from "../utils/index.js";
+import { BaseTTSAdapter } from "@tanstack/ai/adapters";
+//#region src/adapters/tts.ts
+var DEFAULT_GROK_BASE_URL = "https://api.x.ai/v1";
+/**
+* Grok Text-to-Speech Adapter.
+*
+* Talks to `POST {baseURL}/tts` per
+* https://docs.x.ai/developers/model-capabilities/audio/text-to-speech
+*/
+var GrokSpeechAdapter = class extends BaseTTSAdapter {
+	name = "grok";
+	apiKey;
+	baseURL;
+	defaultHeaders;
+	constructor(config, model) {
+		super(model, config);
+		this.apiKey = config.apiKey;
+		this.baseURL = (config.baseURL ?? DEFAULT_GROK_BASE_URL).replace(/\/+$/, "");
+		this.defaultHeaders = config.defaultHeaders ?? {};
+	}
+	async generateSpeech(options) {
+		const { logger } = options;
+		const { model, text, voice, format, modelOptions } = options;
+		logger.request(`activity=generateSpeech provider=grok model=${model}`, {
+			provider: "grok",
+			model
+		});
+		const { body, codec, sampleRateForContentType } = buildTTSRequestBody({
+			text,
+			voice,
+			format,
+			modelOptions
+		});
+		try {
+			const response = await fetch(`${this.baseURL}/tts`, {
+				method: "POST",
+				headers: {
+					...this.defaultHeaders,
+					Authorization: `Bearer ${this.apiKey}`,
+					"Content-Type": "application/json"
+				},
+				body: JSON.stringify(body)
+			});
+			if (!response.ok) {
+				const errorText = await response.text();
+				throw new Error(`Grok TTS request failed: ${response.status} ${errorText}`);
+			}
+			const audio = arrayBufferToBase64(await response.arrayBuffer());
+			return {
+				id: generateId(this.name),
+				model,
+				audio,
+				format: codec,
+				contentType: getContentType(codec, sampleRateForContentType)
+			};
+		} catch (error) {
+			logger.errors("grok.generateSpeech fatal", {
+				error,
+				source: "grok.generateSpeech"
+			});
+			throw error;
+		}
+	}
+};
+/**
+* Build the JSON body for `POST /v1/tts`, resolving codec / sample-rate / voice
+* defaults in one place.
+*
+* Returns the request `body`, the resolved `codec`, and the `sampleRateForContentType`
+* used by the caller to label the response via `getContentType`.
+*/
 function buildTTSRequestBody(options) {
-  const { text, voice, format, modelOptions } = options;
-  const codec = pickCodec(modelOptions?.codec, format);
-  const callerSampleRate = modelOptions?.sample_rate;
-  const pcmDefault = 24e3;
-  const needsRateInContentType = codec === "pcm";
-  const outputFormat = { codec };
-  if (callerSampleRate !== void 0) {
-    outputFormat.sample_rate = callerSampleRate;
-  } else if (needsRateInContentType) {
-    outputFormat.sample_rate = pcmDefault;
-  }
-  if (codec === "mp3" && modelOptions?.bit_rate !== void 0) {
-    outputFormat.bit_rate = modelOptions.bit_rate;
-  }
-  const sampleRateForContentType = callerSampleRate ?? pcmDefault;
-  const body = {
-    text,
-    voice_id: voice ?? "eve",
-    language: modelOptions?.language ?? "en",
-    output_format: outputFormat
-  };
-  if (modelOptions?.optimize_streaming_latency !== void 0) {
-    body.optimize_streaming_latency = modelOptions.optimize_streaming_latency;
-  }
-  if (modelOptions?.text_normalization !== void 0) {
-    body.text_normalization = modelOptions.text_normalization;
-  }
-  return { body, codec, sampleRateForContentType };
+	const { text, voice, format, modelOptions } = options;
+	const codec = pickCodec(modelOptions?.codec, format);
+	const callerSampleRate = modelOptions?.sample_rate;
+	const pcmDefault = 24e3;
+	const needsRateInContentType = codec === "pcm";
+	const outputFormat = { codec };
+	if (callerSampleRate !== void 0) outputFormat.sample_rate = callerSampleRate;
+	else if (needsRateInContentType) outputFormat.sample_rate = pcmDefault;
+	if (codec === "mp3" && modelOptions?.bit_rate !== void 0) outputFormat.bit_rate = modelOptions.bit_rate;
+	const sampleRateForContentType = callerSampleRate ?? pcmDefault;
+	const body = {
+		text,
+		voice_id: voice ?? "eve",
+		language: modelOptions?.language ?? "en",
+		output_format: outputFormat
+	};
+	if (modelOptions?.optimize_streaming_latency !== void 0) body.optimize_streaming_latency = modelOptions.optimize_streaming_latency;
+	if (modelOptions?.text_normalization !== void 0) body.text_normalization = modelOptions.text_normalization;
+	return {
+		body,
+		codec,
+		sampleRateForContentType
+	};
 }
+/**
+* Maps the cross-provider `TTSOptions.format` onto Grok's supported codecs.
+* `opus`, `aac`, and `flac` are not supported by xAI TTS (which only exposes
+* mp3/wav/pcm/mulaw/alaw) — we fall back to mp3. An explicit
+* `modelOptions.codec` always wins.
+*/
 function pickCodec(codecOverride, format) {
-  if (codecOverride) return codecOverride;
-  if (!format) return "mp3";
-  switch (format) {
-    case "mp3":
-    case "wav":
-    case "pcm":
-      return format;
-    case "flac":
-    case "opus":
-    case "aac":
-      return "mp3";
-    default:
-      return "mp3";
-  }
+	if (codecOverride) return codecOverride;
+	if (!format) return "mp3";
+	switch (format) {
+		case "mp3":
+		case "wav":
+		case "pcm": return format;
+		case "flac":
+		case "opus":
+		case "aac": return "mp3";
+		default: return "mp3";
+	}
 }
 function getContentType(codec, sampleRate) {
-  switch (codec) {
-    case "mp3":
-      return "audio/mpeg";
-    case "wav":
-      return "audio/wav";
-    case "pcm":
-      return `audio/L16;rate=${sampleRate}`;
-    case "mulaw":
-      return sampleRate === 8e3 ? "audio/basic" : `audio/PCMU;rate=${sampleRate}`;
-    case "alaw":
-      return sampleRate === 8e3 ? "audio/x-alaw-basic" : `audio/PCMA;rate=${sampleRate}`;
-  }
+	switch (codec) {
+		case "mp3": return "audio/mpeg";
+		case "wav": return "audio/wav";
+		case "pcm": return `audio/L16;rate=${sampleRate}`;
+		case "mulaw": return sampleRate === 8e3 ? "audio/basic" : `audio/PCMU;rate=${sampleRate}`;
+		case "alaw": return sampleRate === 8e3 ? "audio/x-alaw-basic" : `audio/PCMA;rate=${sampleRate}`;
+	}
 }
+/**
+* Creates a Grok speech (TTS) adapter with an explicit API key.
+*
+* @example
+* ```typescript
+* const adapter = createGrokSpeech('grok-tts', 'xai-...')
+* const result = await generateSpeech({
+*   adapter,
+*   text: 'Hello from Grok',
+*   voice: 'eve',
+* })
+* ```
+*/
 function createGrokSpeech(model, apiKey, config) {
-  return new GrokSpeechAdapter({ apiKey, ...config }, model);
+	return new GrokSpeechAdapter({
+		apiKey,
+		...config
+	}, model);
 }
+/**
+* Creates a Grok speech (TTS) adapter, reading the API key from
+* `XAI_API_KEY` in the environment.
+*
+* @throws Error if `XAI_API_KEY` is not set.
+*/
 function grokSpeech(model, config) {
-  const apiKey = getGrokApiKeyFromEnv();
-  return createGrokSpeech(model, apiKey, config);
+	return createGrokSpeech(model, getGrokApiKeyFromEnv(), config);
 }
-export {
-  GrokSpeechAdapter,
-  buildTTSRequestBody,
-  createGrokSpeech,
-  getContentType,
-  grokSpeech
-};
-//# sourceMappingURL=tts.js.map
+//#endregion
+export { GrokSpeechAdapter, buildTTSRequestBody, createGrokSpeech, getContentType, grokSpeech };
+
+//# sourceMappingURL=tts.js.map

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} 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;"}
+{"version":3,"file":"tts.js","names":[],"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"],"mappings":";;;;;AASA,IAAM,wBAAwB;;;;;;;AAsB9B,IAAa,oBAAb,cAEU,eAA+C;CACvD,OAAgB;CAEhB;CACA;CACA;CAEA,YAAY,QAA0B,OAAe;EACnD,MAAM,OAAO,MAAM;EACnB,KAAK,SAAS,OAAO;EACrB,KAAK,WAAW,OAAO,WAAW,sBAAA,CAAuB,QAAQ,QAAQ,EAAE;EAC3E,KAAK,iBAAiB,OAAO,kBAAkB,CAAC;CAClD;CAEA,MAAM,eACJ,SACoB;EACpB,MAAM,EAAE,WAAW;EACnB,MAAM,EAAE,OAAO,MAAM,OAAO,QAAQ,iBAAiB;EAErD,OAAO,QAAQ,+CAA+C,SAAS;GACrE,UAAU;GACV;EACF,CAAC;EAED,MAAM,EAAE,MAAM,OAAO,6BAA6B,oBAAoB;GACpE;GACA;GACA;GACA;EACF,CAAC;EAED,IAAI;GACF,MAAM,WAAW,MAAM,MAAM,GAAG,KAAK,QAAQ,OAAO;IAClD,QAAQ;IACR,SAAS;KAIP,GAAG,KAAK;KACR,eAAe,UAAU,KAAK;KAC9B,gBAAgB;IAClB;IACA,MAAM,KAAK,UAAU,IAAI;GAC3B,CAAC;GAED,IAAI,CAAC,SAAS,IAAI;IAChB,MAAM,YAAY,MAAM,SAAS,KAAK;IACtC,MAAM,IAAI,MACR,4BAA4B,SAAS,OAAO,GAAG,WACjD;GACF;GAGA,MAAM,QAAQ,oBAAoB,MADR,SAAS,YAAY,CACF;GAE7C,OAAO;IACL,IAAI,WAAW,KAAK,IAAI;IACxB;IACA;IACA,QAAQ;IACR,aAAa,eAAe,OAAO,wBAAwB;GAC7D;EACF,SAAS,OAAO;GACd,OAAO,OAAO,6BAA6B;IACzC;IACA,QAAQ;GACV,CAAC;GACD,MAAM;EACR;CACF;AACF;;;;;;;;AASA,SAAgB,oBAAoB,SASlC;CACA,MAAM,EAAE,MAAM,OAAO,QAAQ,iBAAiB;CAE9C,MAAM,QAAQ,UAAU,cAAc,OAAO,MAAM;CAOnD,MAAM,mBAAmB,cAAc;CAIvC,MAAM,aAAa;CACnB,MAAM,yBAAyB,UAAU;CAEzC,MAAM,eAAwC,EAAE,MAAM;CACtD,IAAI,qBAAqB,KAAA,GACvB,aAAa,cAAc;MACtB,IAAI,wBACT,aAAa,cAAc;CAE7B,IAAI,UAAU,SAAS,cAAc,aAAa,KAAA,GAChD,aAAa,WAAW,aAAa;CAMvC,MAAM,2BAA2B,oBAAoB;CAErD,MAAM,OAAgC;EACpC;EACA,UAAU,SAAS;EACnB,UAAU,cAAc,YAAY;EACpC,eAAe;CACjB;CACA,IAAI,cAAc,+BAA+B,KAAA,GAC/C,KAAK,6BAA6B,aAAa;CAEjD,IAAI,cAAc,uBAAuB,KAAA,GACvC,KAAK,qBAAqB,aAAa;CAGzC,OAAO;EAAE;EAAM;EAAO;CAAyB;AACjD;;;;;;;AAQA,SAAS,UACP,eACA,QACc;CACd,IAAI,eAAe,OAAO;CAC1B,IAAI,CAAC,QAAQ,OAAO;CACpB,QAAQ,QAAR;EACE,KAAK;EACL,KAAK;EACL,KAAK,OACH,OAAO;EACT,KAAK;EACL,KAAK;EACL,KAAK,OACH,OAAO;EACT,SACE,OAAO;CACX;AACF;AAEA,SAAgB,eACd,OACA,YACQ;CACR,QAAQ,OAAR;EACE,KAAK,OACH,OAAO;EACT,KAAK,OACH,OAAO;EACT,KAAK,OAEH,OAAO,kBAAkB;EAC3B,KAAK,SAOH,OAAO,eAAe,MAClB,gBACA,mBAAmB;EACzB,KAAK,QACH,OAAO,eAAe,MAClB,uBACA,mBAAmB;CAC3B;AACF;;;;;;;;;;;;;;AAeA,SAAgB,iBACd,OACA,QACA,QAC2B;CAC3B,OAAO,IAAI,kBAAkB;EAAE;EAAQ,GAAG;CAAO,GAAG,KAAK;AAC3D;;;;;;;AAQA,SAAgB,WACd,OACA,QAC2B;CAE3B,OAAO,iBAAiB,OADT,qBACgB,GAAQ,MAAM;AAC/C"}