@tanstack/ai-elevenlabs 0.4.4 → 0.6.0

package/dist/esm/adapters/voice.js

@@ -0,0 +1,154 @@
+import { arrayBufferToBase64, createElevenLabsClient, generateId, parseOutputFormat } from "../utils/client.js";
+import { BaseVoiceAdapter } from "@tanstack/ai/adapters";
+//#region src/adapters/voice.ts
+/**
+* ElevenLabs voice-design adapter built on the official
+* `@elevenlabs/elevenlabs-js` SDK.
+*
+* ElevenLabs designs voices in two steps — generate previews, then promote one
+* into a real voice — but that is an implementation detail. Pass `name` to get
+* a saved voice back; leave it off to audition previews first.
+*
+* @example Audition previews
+* ```ts
+* const result = await generateVoice({
+*   adapter: elevenlabsVoiceDesign('eleven_ttv_v3'),
+*   prompt: 'A warm, gravelly narrator in his sixties with a slight Irish lilt',
+* })
+* ```
+*
+* @example Save the voice
+* ```ts
+* const result = await generateVoice({
+*   adapter: elevenlabsVoiceDesign('eleven_ttv_v3'),
+*   prompt: 'A bright, upbeat product demo host',
+*   name: 'Demo Host',
+* })
+* ```
+*/
+var ElevenLabsVoiceAdapter = class extends BaseVoiceAdapter {
+	name = "elevenlabs";
+	client;
+	constructor(model, config) {
+		super(model, config ?? {});
+		this.client = createElevenLabsClient(config);
+	}
+	async generateVoice(options) {
+		const { logger } = options;
+		logger.request(`activity=generateVoice provider=elevenlabs model=${this.model}`, {
+			provider: "elevenlabs",
+			model: this.model
+		});
+		try {
+			if (!options.prompt) throw new Error("ElevenLabs voice design requires a `prompt` describing the voice. Reference audio alone is not supported; pass both to guide the design with a real speaker.");
+			const opts = options.modelOptions ?? {};
+			const referenceAudioBase64 = await toReferenceAudioBase64(options.referenceAudio);
+			if (referenceAudioBase64 && this.model !== "eleven_ttv_v3") throw new Error(`ElevenLabs only accepts reference audio on eleven_ttv_v3, but this adapter is using "${this.model}".`);
+			const requestOptions = options.abortSignal ? { abortSignal: options.abortSignal } : {};
+			const previewResponse = await this.client.textToVoice.design({
+				voiceDescription: options.prompt,
+				modelId: this.model,
+				...opts.outputFormat ? { outputFormat: opts.outputFormat } : {},
+				...opts.text ? { text: opts.text } : {},
+				...opts.autoGenerateText != null ? { autoGenerateText: opts.autoGenerateText } : {},
+				...opts.loudness != null ? { loudness: opts.loudness } : {},
+				...opts.seed != null ? { seed: opts.seed } : {},
+				...opts.guidanceScale != null ? { guidanceScale: opts.guidanceScale } : {},
+				...opts.quality != null ? { quality: opts.quality } : {},
+				...opts.shouldEnhance != null ? { shouldEnhance: opts.shouldEnhance } : {},
+				...opts.promptStrength != null ? { promptStrength: opts.promptStrength } : {},
+				...referenceAudioBase64 ? { referenceAudioBase64 } : {},
+				...opts.remixingSessionId ? { remixingSessionId: opts.remixingSessionId } : {},
+				...opts.remixingSessionIterationId ? { remixingSessionIterationId: opts.remixingSessionIterationId } : {}
+			}, requestOptions);
+			const { format, contentType } = parseOutputFormat(opts.outputFormat);
+			const voices = previewResponse.previews.map((preview) => ({
+				voiceId: preview.generatedVoiceId,
+				audio: preview.audioBase64,
+				format,
+				contentType: preview.mediaType || contentType,
+				duration: preview.durationSecs,
+				...preview.language ? { language: preview.language } : {},
+				saved: false,
+				status: "ready"
+			}));
+			if (voices.length === 0) throw new Error("ElevenLabs returned no voice previews for this description. Try a longer, more specific prompt.");
+			if (options.name) voices[0] = await this.promoteFirstPreview(voices, options.name, options.description ?? options.prompt, opts.labels, requestOptions);
+			return {
+				id: generateId(this.name),
+				model: this.model,
+				voices,
+				previewText: previewResponse.text
+			};
+		} catch (error) {
+			logger.errors("elevenlabs.generateVoice fatal", {
+				error,
+				source: "elevenlabs.generateVoice"
+			});
+			throw error;
+		}
+	}
+	/**
+	* Promote the best preview into a real library voice. The remaining
+	* generated ids go along as `playedNotSelectedVoiceIds` — ElevenLabs uses
+	* them as RLHF signal for future designs.
+	*/
+	async promoteFirstPreview(voices, voiceName, voiceDescription, labels, requestOptions) {
+		const [best, ...rest] = voices;
+		if (!best) throw new Error("No preview to promote into a library voice.");
+		const saved = await this.client.textToVoice.create({
+			voiceName,
+			voiceDescription,
+			generatedVoiceId: best.voiceId,
+			...labels ? { labels } : {},
+			...rest.length > 0 ? { playedNotSelectedVoiceIds: rest.map((voice) => voice.voiceId) } : {}
+		}, requestOptions);
+		return {
+			...best,
+			voiceId: saved.voiceId,
+			saved: true
+		};
+	}
+	generateId() {
+		return generateId(this.name);
+	}
+};
+/**
+* Normalize reference audio to the bare base64 the design endpoint wants.
+*
+* https URLs are rejected rather than fetched: ElevenLabs has no URL field
+* here, and downloading caller-supplied media into memory to inline it is a
+* footgun on large files.
+*/
+async function toReferenceAudioBase64(audio) {
+	if (audio == null) return void 0;
+	if (audio instanceof ArrayBuffer) return arrayBufferToBase64(audio);
+	if (typeof audio !== "string") return arrayBufferToBase64(await audio.arrayBuffer());
+	if (audio.startsWith("data:")) {
+		const commaIndex = audio.indexOf(",");
+		const header = commaIndex === -1 ? "" : audio.slice(5, commaIndex);
+		if (commaIndex === -1 || !/;base64$/i.test(header)) throw new Error("ElevenLabs voice design needs base64 reference audio. Pass a base64 data URL, a base64 string, a Blob, or an ArrayBuffer.");
+		return audio.slice(commaIndex + 1);
+	}
+	if (/^https?:\/\//i.test(audio)) throw new Error("ElevenLabs voice design does not accept reference audio URLs. Read the file yourself and pass a Blob, ArrayBuffer, or base64 string.");
+	return audio;
+}
+/**
+* Create an ElevenLabs voice-design adapter using `ELEVENLABS_API_KEY` from env.
+*/
+function elevenlabsVoiceDesign(model, config) {
+	return new ElevenLabsVoiceAdapter(model, config);
+}
+/**
+* Create an ElevenLabs voice-design adapter with an explicit API key.
+*/
+function createElevenLabsVoiceDesign(model, apiKey, config) {
+	return new ElevenLabsVoiceAdapter(model, {
+		apiKey,
+		...config
+	});
+}
+//#endregion
+export { ElevenLabsVoiceAdapter, createElevenLabsVoiceDesign, elevenlabsVoiceDesign };
+
+//# sourceMappingURL=voice.js.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"voice.js","names":[],"sources":["../../../src/adapters/voice.ts"],"sourcesContent":["import { BaseVoiceAdapter } from '@tanstack/ai/adapters'\nimport {\n  arrayBufferToBase64,\n  createElevenLabsClient,\n  generateId,\n  parseOutputFormat,\n} from '../utils/client'\nimport type { ElevenLabsClient } from '@elevenlabs/elevenlabs-js'\nimport type {\n  GeneratedVoice,\n  VoiceGenerationOptions,\n  VoiceResult,\n} from '@tanstack/ai'\nimport type { ElevenLabsClientConfig } from '../utils/client'\nimport type {\n  ElevenLabsOutputFormat,\n  ElevenLabsVoiceModel,\n} from '../model-meta'\n\n/**\n * Provider-specific voice-design options. Fields map 1:1 onto the SDK's\n * `VoiceDesignRequestModel` — mirroring the names so ElevenLabs'\n * documentation stays useful here.\n * @see https://elevenlabs.io/docs/api-reference/text-to-voice/design\n */\nexport interface ElevenLabsVoiceProviderOptions {\n  /** Output audio format for the previews, `codec_samplerate[_bitrate]`. */\n  outputFormat?: ElevenLabsOutputFormat\n  /** Line the previews speak, 100..1000 characters. */\n  text?: string\n  /** Let ElevenLabs write the preview line from the description. */\n  autoGenerateText?: boolean\n  /** Preview loudness, -1 (quietest) to 1 (loudest). 0 is roughly -24 LUFS. */\n  loudness?: number\n  /** Deterministic sampling seed — same seed and inputs produce the same voice. */\n  seed?: number\n  /** How closely to follow the description. High values can sound robotic. */\n  guidanceScale?: number\n  /** Higher quality trades variety for fidelity. */\n  quality?: number\n  /** Let ElevenLabs expand a short description into a detailed one. */\n  shouldEnhance?: boolean\n  /**\n   * Balance of description against reference audio, 0 (almost all reference)\n   * to 1 (almost all description). `eleven_ttv_v3` only.\n   */\n  promptStrength?: number\n  /** Metadata stored on the voice. Only used when the voice is saved. */\n  labels?: Record<string, string>\n  /** Remixing session to attach these generations to. */\n  remixingSessionId?: string\n  /** Remixing session iteration to attach these generations to. */\n  remixingSessionIterationId?: string\n}\n\n/**\n * ElevenLabs voice-design adapter built on the official\n * `@elevenlabs/elevenlabs-js` SDK.\n *\n * ElevenLabs designs voices in two steps — generate previews, then promote one\n * into a real voice — but that is an implementation detail. Pass `name` to get\n * a saved voice back; leave it off to audition previews first.\n *\n * @example Audition previews\n * ```ts\n * const result = await generateVoice({\n *   adapter: elevenlabsVoiceDesign('eleven_ttv_v3'),\n *   prompt: 'A warm, gravelly narrator in his sixties with a slight Irish lilt',\n * })\n * ```\n *\n * @example Save the voice\n * ```ts\n * const result = await generateVoice({\n *   adapter: elevenlabsVoiceDesign('eleven_ttv_v3'),\n *   prompt: 'A bright, upbeat product demo host',\n *   name: 'Demo Host',\n * })\n * ```\n */\nexport class ElevenLabsVoiceAdapter<\n  TModel extends ElevenLabsVoiceModel,\n> extends BaseVoiceAdapter<TModel, ElevenLabsVoiceProviderOptions> {\n  readonly name = 'elevenlabs' as const\n\n  private readonly client: ElevenLabsClient\n\n  constructor(model: TModel, config?: ElevenLabsClientConfig) {\n    super(model, config ?? {})\n    this.client = createElevenLabsClient(config)\n  }\n\n  async generateVoice(\n    options: VoiceGenerationOptions<ElevenLabsVoiceProviderOptions>,\n  ): Promise<VoiceResult> {\n    const { logger } = options\n    logger.request(\n      `activity=generateVoice provider=elevenlabs model=${this.model}`,\n      { provider: 'elevenlabs', model: this.model },\n    )\n    try {\n      // `voiceDescription` is required by the design endpoint, so reference\n      // audio alone is not enough here — unlike clone-only providers.\n      if (!options.prompt) {\n        throw new Error(\n          'ElevenLabs voice design requires a `prompt` describing the voice. Reference audio alone is not supported; pass both to guide the design with a real speaker.',\n        )\n      }\n\n      const opts = options.modelOptions ?? {}\n      const referenceAudioBase64 = await toReferenceAudioBase64(\n        options.referenceAudio,\n      )\n      if (referenceAudioBase64 && this.model !== 'eleven_ttv_v3') {\n        throw new Error(\n          `ElevenLabs only accepts reference audio on eleven_ttv_v3, but this adapter is using \"${this.model}\".`,\n        )\n      }\n\n      const requestOptions = options.abortSignal\n        ? { abortSignal: options.abortSignal }\n        : {}\n\n      const previewResponse = await this.client.textToVoice.design(\n        {\n          voiceDescription: options.prompt,\n          modelId: this.model,\n          ...(opts.outputFormat ? { outputFormat: opts.outputFormat } : {}),\n          ...(opts.text ? { text: opts.text } : {}),\n          ...(opts.autoGenerateText != null\n            ? { autoGenerateText: opts.autoGenerateText }\n            : {}),\n          ...(opts.loudness != null ? { loudness: opts.loudness } : {}),\n          ...(opts.seed != null ? { seed: opts.seed } : {}),\n          ...(opts.guidanceScale != null\n            ? { guidanceScale: opts.guidanceScale }\n            : {}),\n          ...(opts.quality != null ? { quality: opts.quality } : {}),\n          ...(opts.shouldEnhance != null\n            ? { shouldEnhance: opts.shouldEnhance }\n            : {}),\n          ...(opts.promptStrength != null\n            ? { promptStrength: opts.promptStrength }\n            : {}),\n          ...(referenceAudioBase64 ? { referenceAudioBase64 } : {}),\n          ...(opts.remixingSessionId\n            ? { remixingSessionId: opts.remixingSessionId }\n            : {}),\n          ...(opts.remixingSessionIterationId\n            ? { remixingSessionIterationId: opts.remixingSessionIterationId }\n            : {}),\n        },\n        requestOptions,\n      )\n\n      const { format, contentType } = parseOutputFormat(opts.outputFormat)\n      const voices: Array<GeneratedVoice> = previewResponse.previews.map(\n        (preview) => ({\n          voiceId: preview.generatedVoiceId,\n          audio: preview.audioBase64,\n          format,\n          contentType: preview.mediaType || contentType,\n          duration: preview.durationSecs,\n          ...(preview.language ? { language: preview.language } : {}),\n          saved: false,\n          // The design endpoint returns a finished preview; nothing trains.\n          status: 'ready' as const,\n        }),\n      )\n\n      // A caller who passed `name` asked for a persisted voice. Returning an\n      // empty list would report success for a request that produced nothing.\n      if (voices.length === 0) {\n        throw new Error(\n          'ElevenLabs returned no voice previews for this description. Try a longer, more specific prompt.',\n        )\n      }\n\n      if (options.name) {\n        voices[0] = await this.promoteFirstPreview(\n          voices,\n          options.name,\n          options.description ?? options.prompt,\n          opts.labels,\n          requestOptions,\n        )\n      }\n\n      return {\n        id: generateId(this.name),\n        model: this.model,\n        voices,\n        previewText: previewResponse.text,\n      }\n    } catch (error) {\n      logger.errors('elevenlabs.generateVoice fatal', {\n        error,\n        source: 'elevenlabs.generateVoice',\n      })\n      throw error\n    }\n  }\n\n  /**\n   * Promote the best preview into a real library voice. The remaining\n   * generated ids go along as `playedNotSelectedVoiceIds` — ElevenLabs uses\n   * them as RLHF signal for future designs.\n   */\n  private async promoteFirstPreview(\n    voices: ReadonlyArray<GeneratedVoice>,\n    voiceName: string,\n    voiceDescription: string,\n    labels: Record<string, string> | undefined,\n    requestOptions: { abortSignal?: AbortSignal },\n  ): Promise<GeneratedVoice> {\n    const [best, ...rest] = voices\n    if (!best) {\n      throw new Error('No preview to promote into a library voice.')\n    }\n\n    const saved = await this.client.textToVoice.create(\n      {\n        voiceName,\n        voiceDescription,\n        generatedVoiceId: best.voiceId,\n        ...(labels ? { labels } : {}),\n        ...(rest.length > 0\n          ? { playedNotSelectedVoiceIds: rest.map((voice) => voice.voiceId) }\n          : {}),\n      },\n      requestOptions,\n    )\n\n    return { ...best, voiceId: saved.voiceId, saved: true }\n  }\n\n  protected override generateId(): string {\n    return generateId(this.name)\n  }\n}\n\n/**\n * Normalize reference audio to the bare base64 the design endpoint wants.\n *\n * https URLs are rejected rather than fetched: ElevenLabs has no URL field\n * here, and downloading caller-supplied media into memory to inline it is a\n * footgun on large files.\n */\nasync function toReferenceAudioBase64(\n  audio: VoiceGenerationOptions['referenceAudio'],\n): Promise<string | undefined> {\n  if (audio == null) return undefined\n  if (audio instanceof ArrayBuffer) return arrayBufferToBase64(audio)\n  if (typeof audio !== 'string') {\n    return arrayBufferToBase64(await audio.arrayBuffer())\n  }\n\n  if (audio.startsWith('data:')) {\n    const commaIndex = audio.indexOf(',')\n    const header = commaIndex === -1 ? '' : audio.slice(5, commaIndex)\n    if (commaIndex === -1 || !/;base64$/i.test(header)) {\n      throw new Error(\n        'ElevenLabs voice design needs base64 reference audio. Pass a base64 data URL, a base64 string, a Blob, or an ArrayBuffer.',\n      )\n    }\n    return audio.slice(commaIndex + 1)\n  }\n\n  if (/^https?:\\/\\//i.test(audio)) {\n    throw new Error(\n      'ElevenLabs voice design does not accept reference audio URLs. Read the file yourself and pass a Blob, ArrayBuffer, or base64 string.',\n    )\n  }\n\n  return audio\n}\n\n/**\n * Create an ElevenLabs voice-design adapter using `ELEVENLABS_API_KEY` from env.\n */\nexport function elevenlabsVoiceDesign<TModel extends ElevenLabsVoiceModel>(\n  model: TModel,\n  config?: ElevenLabsClientConfig,\n): ElevenLabsVoiceAdapter<TModel> {\n  return new ElevenLabsVoiceAdapter(model, config)\n}\n\n/**\n * Create an ElevenLabs voice-design adapter with an explicit API key.\n */\nexport function createElevenLabsVoiceDesign<\n  TModel extends ElevenLabsVoiceModel,\n>(\n  model: TModel,\n  apiKey: string,\n  config?: Omit<ElevenLabsClientConfig, 'apiKey'>,\n): ElevenLabsVoiceAdapter<TModel> {\n  return new ElevenLabsVoiceAdapter(model, { apiKey, ...config })\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;AAgFA,IAAa,yBAAb,cAEU,iBAAyD;CACjE,OAAgB;CAEhB;CAEA,YAAY,OAAe,QAAiC;EAC1D,MAAM,OAAO,UAAU,CAAC,CAAC;EACzB,KAAK,SAAS,uBAAuB,MAAM;CAC7C;CAEA,MAAM,cACJ,SACsB;EACtB,MAAM,EAAE,WAAW;EACnB,OAAO,QACL,oDAAoD,KAAK,SACzD;GAAE,UAAU;GAAc,OAAO,KAAK;EAAM,CAC9C;EACA,IAAI;GAGF,IAAI,CAAC,QAAQ,QACX,MAAM,IAAI,MACR,8JACF;GAGF,MAAM,OAAO,QAAQ,gBAAgB,CAAC;GACtC,MAAM,uBAAuB,MAAM,uBACjC,QAAQ,cACV;GACA,IAAI,wBAAwB,KAAK,UAAU,iBACzC,MAAM,IAAI,MACR,wFAAwF,KAAK,MAAM,GACrG;GAGF,MAAM,iBAAiB,QAAQ,cAC3B,EAAE,aAAa,QAAQ,YAAY,IACnC,CAAC;GAEL,MAAM,kBAAkB,MAAM,KAAK,OAAO,YAAY,OACpD;IACE,kBAAkB,QAAQ;IAC1B,SAAS,KAAK;IACd,GAAI,KAAK,eAAe,EAAE,cAAc,KAAK,aAAa,IAAI,CAAC;IAC/D,GAAI,KAAK,OAAO,EAAE,MAAM,KAAK,KAAK,IAAI,CAAC;IACvC,GAAI,KAAK,oBAAoB,OACzB,EAAE,kBAAkB,KAAK,iBAAiB,IAC1C,CAAC;IACL,GAAI,KAAK,YAAY,OAAO,EAAE,UAAU,KAAK,SAAS,IAAI,CAAC;IAC3D,GAAI,KAAK,QAAQ,OAAO,EAAE,MAAM,KAAK,KAAK,IAAI,CAAC;IAC/C,GAAI,KAAK,iBAAiB,OACtB,EAAE,eAAe,KAAK,cAAc,IACpC,CAAC;IACL,GAAI,KAAK,WAAW,OAAO,EAAE,SAAS,KAAK,QAAQ,IAAI,CAAC;IACxD,GAAI,KAAK,iBAAiB,OACtB,EAAE,eAAe,KAAK,cAAc,IACpC,CAAC;IACL,GAAI,KAAK,kBAAkB,OACvB,EAAE,gBAAgB,KAAK,eAAe,IACtC,CAAC;IACL,GAAI,uBAAuB,EAAE,qBAAqB,IAAI,CAAC;IACvD,GAAI,KAAK,oBACL,EAAE,mBAAmB,KAAK,kBAAkB,IAC5C,CAAC;IACL,GAAI,KAAK,6BACL,EAAE,4BAA4B,KAAK,2BAA2B,IAC9D,CAAC;GACP,GACA,cACF;GAEA,MAAM,EAAE,QAAQ,gBAAgB,kBAAkB,KAAK,YAAY;GACnE,MAAM,SAAgC,gBAAgB,SAAS,KAC5D,aAAa;IACZ,SAAS,QAAQ;IACjB,OAAO,QAAQ;IACf;IACA,aAAa,QAAQ,aAAa;IAClC,UAAU,QAAQ;IAClB,GAAI,QAAQ,WAAW,EAAE,UAAU,QAAQ,SAAS,IAAI,CAAC;IACzD,OAAO;IAEP,QAAQ;GACV,EACF;GAIA,IAAI,OAAO,WAAW,GACpB,MAAM,IAAI,MACR,iGACF;GAGF,IAAI,QAAQ,MACV,OAAO,KAAK,MAAM,KAAK,oBACrB,QACA,QAAQ,MACR,QAAQ,eAAe,QAAQ,QAC/B,KAAK,QACL,cACF;GAGF,OAAO;IACL,IAAI,WAAW,KAAK,IAAI;IACxB,OAAO,KAAK;IACZ;IACA,aAAa,gBAAgB;GAC/B;EACF,SAAS,OAAO;GACd,OAAO,OAAO,kCAAkC;IAC9C;IACA,QAAQ;GACV,CAAC;GACD,MAAM;EACR;CACF;;;;;;CAOA,MAAc,oBACZ,QACA,WACA,kBACA,QACA,gBACyB;EACzB,MAAM,CAAC,MAAM,GAAG,QAAQ;EACxB,IAAI,CAAC,MACH,MAAM,IAAI,MAAM,6CAA6C;EAG/D,MAAM,QAAQ,MAAM,KAAK,OAAO,YAAY,OAC1C;GACE;GACA;GACA,kBAAkB,KAAK;GACvB,GAAI,SAAS,EAAE,OAAO,IAAI,CAAC;GAC3B,GAAI,KAAK,SAAS,IACd,EAAE,2BAA2B,KAAK,KAAK,UAAU,MAAM,OAAO,EAAE,IAChE,CAAC;EACP,GACA,cACF;EAEA,OAAO;GAAE,GAAG;GAAM,SAAS,MAAM;GAAS,OAAO;EAAK;CACxD;CAEA,aAAwC;EACtC,OAAO,WAAW,KAAK,IAAI;CAC7B;AACF;;;;;;;;AASA,eAAe,uBACb,OAC6B;CAC7B,IAAI,SAAS,MAAM,OAAO,KAAA;CAC1B,IAAI,iBAAiB,aAAa,OAAO,oBAAoB,KAAK;CAClE,IAAI,OAAO,UAAU,UACnB,OAAO,oBAAoB,MAAM,MAAM,YAAY,CAAC;CAGtD,IAAI,MAAM,WAAW,OAAO,GAAG;EAC7B,MAAM,aAAa,MAAM,QAAQ,GAAG;EACpC,MAAM,SAAS,eAAe,KAAK,KAAK,MAAM,MAAM,GAAG,UAAU;EACjE,IAAI,eAAe,MAAM,CAAC,YAAY,KAAK,MAAM,GAC/C,MAAM,IAAI,MACR,2HACF;EAEF,OAAO,MAAM,MAAM,aAAa,CAAC;CACnC;CAEA,IAAI,gBAAgB,KAAK,KAAK,GAC5B,MAAM,IAAI,MACR,sIACF;CAGF,OAAO;AACT;;;;AAKA,SAAgB,sBACd,OACA,QACgC;CAChC,OAAO,IAAI,uBAAuB,OAAO,MAAM;AACjD;;;;AAKA,SAAgB,4BAGd,OACA,QACA,QACgC;CAChC,OAAO,IAAI,uBAAuB,OAAO;EAAE;EAAQ,GAAG;CAAO,CAAC;AAChE"}

package/dist/esm/index.d.ts

@@ -2,6 +2,7 @@ export { elevenlabsRealtimeToken, elevenlabsRealtime } from './realtime/index.js
 export type { ElevenLabsRealtimeTokenOptions, ElevenLabsRealtimeOptions, ElevenLabsConversationMode, ElevenLabsVADConfig, ElevenLabsClientTool, } from './realtime/index.js';
 export { ElevenLabsSpeechAdapter, createElevenLabsSpeech, elevenlabsSpeech, type ElevenLabsSpeechProviderOptions, type ElevenLabsVoiceSettings, } from './adapters/speech.js';
 export { ElevenLabsAudioAdapter, createElevenLabsAudio, elevenlabsAudio, type ElevenLabsAudioProviderOptions, type ElevenLabsMusicProviderOptions, type ElevenLabsSoundEffectsProviderOptions, type ElevenLabsMusicCompositionPlan, } from './adapters/audio.js';
+export { ElevenLabsVoiceAdapter, createElevenLabsVoiceDesign, elevenlabsVoiceDesign, type ElevenLabsVoiceProviderOptions, } from './adapters/voice.js';
 export { ElevenLabsTranscriptionAdapter, createElevenLabsTranscription, elevenlabsTranscription, type ElevenLabsTranscriptionProviderOptions, } from './adapters/transcription.js';
-export { ELEVENLABS_TTS_MODELS, ELEVENLABS_AUDIO_MODELS, ELEVENLABS_TRANSCRIPTION_MODELS, isElevenLabsMusicModel, isElevenLabsSoundEffectsModel, type ElevenLabsTTSModel, type ElevenLabsAudioModel, type ElevenLabsMusicModel, type ElevenLabsSoundEffectsModel, type ElevenLabsTranscriptionModel, type ElevenLabsOutputFormat, } from './model-meta.js';
+export { ELEVENLABS_TTS_MODELS, ELEVENLABS_AUDIO_MODELS, ELEVENLABS_TRANSCRIPTION_MODELS, ELEVENLABS_VOICE_MODELS, isElevenLabsMusicModel, isElevenLabsSoundEffectsModel, type ElevenLabsTTSModel, type ElevenLabsAudioModel, type ElevenLabsMusicModel, type ElevenLabsSoundEffectsModel, type ElevenLabsTranscriptionModel, type ElevenLabsVoiceModel, type ElevenLabsOutputFormat, } from './model-meta.js';
 export { getElevenLabsApiKeyFromEnv, type ElevenLabsClientConfig, } from './utils/index.js';

package/dist/esm/index.js

@@ -3,8 +3,9 @@ import { elevenlabsRealtimeToken } from "./realtime/token.js";
 import { elevenlabsRealtime } from "./realtime/adapter.js";
 import "./realtime/index.js";
 import { ElevenLabsSpeechAdapter, createElevenLabsSpeech, elevenlabsSpeech } from "./adapters/speech.js";
-import { ELEVENLABS_AUDIO_MODELS, ELEVENLABS_TRANSCRIPTION_MODELS, ELEVENLABS_TTS_MODELS, isElevenLabsMusicModel, isElevenLabsSoundEffectsModel } from "./model-meta.js";
+import { ELEVENLABS_AUDIO_MODELS, ELEVENLABS_TRANSCRIPTION_MODELS, ELEVENLABS_TTS_MODELS, ELEVENLABS_VOICE_MODELS, isElevenLabsMusicModel, isElevenLabsSoundEffectsModel } from "./model-meta.js";
 import { ElevenLabsAudioAdapter, createElevenLabsAudio, elevenlabsAudio } from "./adapters/audio.js";
+import { ElevenLabsVoiceAdapter, createElevenLabsVoiceDesign, elevenlabsVoiceDesign } from "./adapters/voice.js";
 import { ElevenLabsTranscriptionAdapter, createElevenLabsTranscription, elevenlabsTranscription } from "./adapters/transcription.js";
 import "./utils/index.js";
-export { ELEVENLABS_AUDIO_MODELS, ELEVENLABS_TRANSCRIPTION_MODELS, ELEVENLABS_TTS_MODELS, ElevenLabsAudioAdapter, ElevenLabsSpeechAdapter, ElevenLabsTranscriptionAdapter, createElevenLabsAudio, createElevenLabsSpeech, createElevenLabsTranscription, elevenlabsAudio, elevenlabsRealtime, elevenlabsRealtimeToken, elevenlabsSpeech, elevenlabsTranscription, getElevenLabsApiKeyFromEnv, isElevenLabsMusicModel, isElevenLabsSoundEffectsModel };
+export { ELEVENLABS_AUDIO_MODELS, ELEVENLABS_TRANSCRIPTION_MODELS, ELEVENLABS_TTS_MODELS, ELEVENLABS_VOICE_MODELS, ElevenLabsAudioAdapter, ElevenLabsSpeechAdapter, ElevenLabsTranscriptionAdapter, ElevenLabsVoiceAdapter, createElevenLabsAudio, createElevenLabsSpeech, createElevenLabsTranscription, createElevenLabsVoiceDesign, elevenlabsAudio, elevenlabsRealtime, elevenlabsRealtimeToken, elevenlabsSpeech, elevenlabsTranscription, elevenlabsVoiceDesign, getElevenLabsApiKeyFromEnv, isElevenLabsMusicModel, isElevenLabsSoundEffectsModel };

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

@@ -1,37 +1,62 @@
 import { ElevenLabs } from '@elevenlabs/elevenlabs-js';
 /**
  * ElevenLabs model identifiers. The lists below are the source of truth —
- * callers are blocked from passing unknown model IDs. Keep them in sync with
- * the ElevenLabs SDK via the automated update pipeline.
+ * callers are blocked from passing unknown model IDs.
+ *
+ * Maintained by hand: `scripts/sync-provider-models.ts` deliberately excludes
+ * elevenlabs, because these are media endpoint ids rather than OpenRouter
+ * models. Where the SDK publishes its own union, pin the list to it with
+ * `satisfies` so a removed id fails the build instead of a request.
+ *
+ * Each list is ordered newest-first, with models ElevenLabs has deprecated
+ * kept at the bottom so existing callers don't break on an upgrade. An id the
+ * SDK has dropped outright is removed rather than kept, because the adapter
+ * can no longer send it — see `eleven_text_to_sound_v1`.
+ *
+ * @see https://elevenlabs.io/docs/models
  */
 /**
  * Text-to-speech models.
  * @see https://elevenlabs.io/docs/models
  */
-export declare const ELEVENLABS_TTS_MODELS: readonly ["eleven_v3", "eleven_multilingual_v2", "eleven_flash_v2_5", "eleven_flash_v2", "eleven_turbo_v2_5", "eleven_turbo_v2", "eleven_monolingual_v1"];
+export declare const ELEVENLABS_TTS_MODELS: readonly ["eleven_v3", "eleven_v3_conversational", "eleven_multilingual_v2", "eleven_flash_v2_5", "eleven_flash_v2", "eleven_turbo_v2_5", "eleven_turbo_v2", "eleven_monolingual_v1"];
 export type ElevenLabsTTSModel = (typeof ELEVENLABS_TTS_MODELS)[number];
 /**
- * Audio generation models — music (`music_v1`) + sound effects
+ * Audio generation models — music (`music_v*`) + sound effects
  * (`eleven_text_to_sound_v*`) share one `generateAudio` adapter.
  * The adapter dispatches by model id so callers pick behavior via the model.
  *
  * @see https://elevenlabs.io/docs/overview/capabilities/music
  * @see https://elevenlabs.io/docs/overview/capabilities/sound-effects
  */
-export declare const ELEVENLABS_AUDIO_MODELS: readonly ["music_v1", "eleven_text_to_sound_v2", "eleven_text_to_sound_v1"];
+export declare const ELEVENLABS_AUDIO_MODELS: readonly ["music_v2_5", "music_v2", "eleven_text_to_sound_v2", "music_v1"];
 export type ElevenLabsAudioModel = (typeof ELEVENLABS_AUDIO_MODELS)[number];
-/** Music models within the audio family. */
-export type ElevenLabsMusicModel = 'music_v1';
-/** SFX models within the audio family. */
-export type ElevenLabsSoundEffectsModel = 'eleven_text_to_sound_v2' | 'eleven_text_to_sound_v1';
+/** Music models within the audio family, derived from the audio list. */
+export type ElevenLabsMusicModel = Extract<ElevenLabsAudioModel, `music_${string}`>;
+/** SFX models within the audio family, derived from the audio list. */
+export type ElevenLabsSoundEffectsModel = Extract<ElevenLabsAudioModel, `eleven_text_to_sound_${string}`>;
+/**
+ * Matches the `music_` prefix so a new music release needs no branch here,
+ * but still checks membership: the narrowed type is the closed set above, so
+ * a `music_*` id this package does not ship must not pass.
+ */
 export declare function isElevenLabsMusicModel(model: string): model is ElevenLabsMusicModel;
 export declare function isElevenLabsSoundEffectsModel(model: string): model is ElevenLabsSoundEffectsModel;
 /**
  * Speech-to-text (transcription) models — Scribe family.
  * @see https://elevenlabs.io/docs/overview/capabilities/speech-to-text
  */
-export declare const ELEVENLABS_TRANSCRIPTION_MODELS: readonly ["scribe_v2", "scribe_v1"];
+export declare const ELEVENLABS_TRANSCRIPTION_MODELS: readonly ["scribe_v2", "scribe_v2_medical", "scribe_v1"];
 export type ElevenLabsTranscriptionModel = (typeof ELEVENLABS_TRANSCRIPTION_MODELS)[number];
+/**
+ * Voice design (text-to-voice) models, used by the `generateVoice` adapter.
+ * `eleven_ttv_v3` is the only one that accepts reference audio, so guiding a
+ * design with a recording of a real speaker requires it.
+ *
+ * @see https://elevenlabs.io/docs/overview/capabilities/voice-design
+ */
+export declare const ELEVENLABS_VOICE_MODELS: readonly ["eleven_ttv_v3", "eleven_multilingual_ttv_v2"];
+export type ElevenLabsVoiceModel = (typeof ELEVENLABS_VOICE_MODELS)[number];
 /**
  * Supported `output_format` strings, encoded as `codec_samplerate[_bitrate]`.
  * Aliased to the SDK's `AllowedOutputFormats` so the list stays in sync

package/dist/esm/model-meta.js

@@ -1,8 +1,19 @@
 //#region src/model-meta.ts
 /**
 * ElevenLabs model identifiers. The lists below are the source of truth —
-* callers are blocked from passing unknown model IDs. Keep them in sync with
-* the ElevenLabs SDK via the automated update pipeline.
+* callers are blocked from passing unknown model IDs.
+*
+* Maintained by hand: `scripts/sync-provider-models.ts` deliberately excludes
+* elevenlabs, because these are media endpoint ids rather than OpenRouter
+* models. Where the SDK publishes its own union, pin the list to it with
+* `satisfies` so a removed id fails the build instead of a request.
+*
+* Each list is ordered newest-first, with models ElevenLabs has deprecated
+* kept at the bottom so existing callers don't break on an upgrade. An id the
+* SDK has dropped outright is removed rather than kept, because the adapter
+* can no longer send it — see `eleven_text_to_sound_v1`.
+*
+* @see https://elevenlabs.io/docs/models
 */
 /**
 * Text-to-speech models.
@@ -10,6 +21,7 @@
 */
 var ELEVENLABS_TTS_MODELS = [
 	"eleven_v3",
+	"eleven_v3_conversational",
 	"eleven_multilingual_v2",
 	"eleven_flash_v2_5",
 	"eleven_flash_v2",
@@ -18,7 +30,7 @@ var ELEVENLABS_TTS_MODELS = [
 	"eleven_monolingual_v1"
 ];
 /**
-* Audio generation models — music (`music_v1`) + sound effects
+* Audio generation models — music (`music_v*`) + sound effects
 * (`eleven_text_to_sound_v*`) share one `generateAudio` adapter.
 * The adapter dispatches by model id so callers pick behavior via the model.
 *
@@ -26,12 +38,18 @@ var ELEVENLABS_TTS_MODELS = [
 * @see https://elevenlabs.io/docs/overview/capabilities/sound-effects
 */
 var ELEVENLABS_AUDIO_MODELS = [
-	"music_v1",
+	"music_v2_5",
+	"music_v2",
 	"eleven_text_to_sound_v2",
-	"eleven_text_to_sound_v1"
+	"music_v1"
 ];
+/**
+* Matches the `music_` prefix so a new music release needs no branch here,
+* but still checks membership: the narrowed type is the closed set above, so
+* a `music_*` id this package does not ship must not pass.
+*/
 function isElevenLabsMusicModel(model) {
-	return model === "music_v1";
+	return model.startsWith("music_") && ELEVENLABS_AUDIO_MODELS.includes(model);
 }
 function isElevenLabsSoundEffectsModel(model) {
 	return model.startsWith("eleven_text_to_sound_");
@@ -40,8 +58,20 @@ function isElevenLabsSoundEffectsModel(model) {
 * Speech-to-text (transcription) models — Scribe family.
 * @see https://elevenlabs.io/docs/overview/capabilities/speech-to-text
 */
-var ELEVENLABS_TRANSCRIPTION_MODELS = ["scribe_v2", "scribe_v1"];
+var ELEVENLABS_TRANSCRIPTION_MODELS = [
+	"scribe_v2",
+	"scribe_v2_medical",
+	"scribe_v1"
+];
+/**
+* Voice design (text-to-voice) models, used by the `generateVoice` adapter.
+* `eleven_ttv_v3` is the only one that accepts reference audio, so guiding a
+* design with a recording of a real speaker requires it.
+*
+* @see https://elevenlabs.io/docs/overview/capabilities/voice-design
+*/
+var ELEVENLABS_VOICE_MODELS = ["eleven_ttv_v3", "eleven_multilingual_ttv_v2"];
 //#endregion
-export { ELEVENLABS_AUDIO_MODELS, ELEVENLABS_TRANSCRIPTION_MODELS, ELEVENLABS_TTS_MODELS, isElevenLabsMusicModel, isElevenLabsSoundEffectsModel };
+export { ELEVENLABS_AUDIO_MODELS, ELEVENLABS_TRANSCRIPTION_MODELS, ELEVENLABS_TTS_MODELS, ELEVENLABS_VOICE_MODELS, isElevenLabsMusicModel, isElevenLabsSoundEffectsModel };
 
 //# sourceMappingURL=model-meta.js.map

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

@@ -1 +1 @@
-{"version":3,"file":"model-meta.js","names":[],"sources":["../../src/model-meta.ts"],"sourcesContent":["import type { ElevenLabs } from '@elevenlabs/elevenlabs-js'\n\n/**\n * ElevenLabs model identifiers. The lists below are the source of truth —\n * callers are blocked from passing unknown model IDs. Keep them in sync with\n * the ElevenLabs SDK via the automated update pipeline.\n */\n\n/**\n * Text-to-speech models.\n * @see https://elevenlabs.io/docs/models\n */\nexport const ELEVENLABS_TTS_MODELS = [\n  'eleven_v3',\n  'eleven_multilingual_v2',\n  'eleven_flash_v2_5',\n  'eleven_flash_v2',\n  'eleven_turbo_v2_5',\n  'eleven_turbo_v2',\n  'eleven_monolingual_v1',\n] as const\n\nexport type ElevenLabsTTSModel = (typeof ELEVENLABS_TTS_MODELS)[number]\n\n/**\n * Audio generation models — music (`music_v1`) + sound effects\n * (`eleven_text_to_sound_v*`) share one `generateAudio` adapter.\n * The adapter dispatches by model id so callers pick behavior via the model.\n *\n * @see https://elevenlabs.io/docs/overview/capabilities/music\n * @see https://elevenlabs.io/docs/overview/capabilities/sound-effects\n */\nexport const ELEVENLABS_AUDIO_MODELS = [\n  'music_v1',\n  'eleven_text_to_sound_v2',\n  'eleven_text_to_sound_v1',\n] as const\n\nexport type ElevenLabsAudioModel = (typeof ELEVENLABS_AUDIO_MODELS)[number]\n\n/** Music models within the audio family. */\nexport type ElevenLabsMusicModel = 'music_v1'\n/** SFX models within the audio family. */\nexport type ElevenLabsSoundEffectsModel =\n  | 'eleven_text_to_sound_v2'\n  | 'eleven_text_to_sound_v1'\n\nexport function isElevenLabsMusicModel(\n  model: string,\n): model is ElevenLabsMusicModel {\n  return model === 'music_v1'\n}\n\nexport function isElevenLabsSoundEffectsModel(\n  model: string,\n): model is ElevenLabsSoundEffectsModel {\n  return model.startsWith('eleven_text_to_sound_')\n}\n\n/**\n * Speech-to-text (transcription) models — Scribe family.\n * @see https://elevenlabs.io/docs/overview/capabilities/speech-to-text\n */\nexport const ELEVENLABS_TRANSCRIPTION_MODELS = [\n  'scribe_v2',\n  'scribe_v1',\n] as const\n\nexport type ElevenLabsTranscriptionModel =\n  (typeof ELEVENLABS_TRANSCRIPTION_MODELS)[number]\n\n/**\n * Supported `output_format` strings, encoded as `codec_samplerate[_bitrate]`.\n * Aliased to the SDK's `AllowedOutputFormats` so the list stays in sync\n * automatically whenever the `@elevenlabs/elevenlabs-js` dependency is bumped.\n *\n * @see https://elevenlabs.io/docs/api-reference/text-to-speech/convert\n */\nexport type ElevenLabsOutputFormat = ElevenLabs.AllowedOutputFormats\n"],"mappings":";;;;;;;;;;AAYA,IAAa,wBAAwB;CACnC;CACA;CACA;CACA;CACA;CACA;CACA;AACF;;;;;;;;;AAYA,IAAa,0BAA0B;CACrC;CACA;CACA;AACF;AAWA,SAAgB,uBACd,OAC+B;CAC/B,OAAO,UAAU;AACnB;AAEA,SAAgB,8BACd,OACsC;CACtC,OAAO,MAAM,WAAW,uBAAuB;AACjD;;;;;AAMA,IAAa,kCAAkC,CAC7C,aACA,WACF"}
+{"version":3,"file":"model-meta.js","names":[],"sources":["../../src/model-meta.ts"],"sourcesContent":["import type { ElevenLabs } from '@elevenlabs/elevenlabs-js'\n\n/**\n * ElevenLabs model identifiers. The lists below are the source of truth —\n * callers are blocked from passing unknown model IDs.\n *\n * Maintained by hand: `scripts/sync-provider-models.ts` deliberately excludes\n * elevenlabs, because these are media endpoint ids rather than OpenRouter\n * models. Where the SDK publishes its own union, pin the list to it with\n * `satisfies` so a removed id fails the build instead of a request.\n *\n * Each list is ordered newest-first, with models ElevenLabs has deprecated\n * kept at the bottom so existing callers don't break on an upgrade. An id the\n * SDK has dropped outright is removed rather than kept, because the adapter\n * can no longer send it — see `eleven_text_to_sound_v1`.\n *\n * @see https://elevenlabs.io/docs/models\n */\n\n/**\n * Text-to-speech models.\n * @see https://elevenlabs.io/docs/models\n */\nexport const ELEVENLABS_TTS_MODELS = [\n  'eleven_v3',\n  'eleven_v3_conversational',\n  'eleven_multilingual_v2',\n  'eleven_flash_v2_5',\n  'eleven_flash_v2',\n  // Deprecated by ElevenLabs — use `eleven_flash_v2_5`.\n  'eleven_turbo_v2_5',\n  // Deprecated by ElevenLabs — use `eleven_flash_v2`.\n  'eleven_turbo_v2',\n  // Deprecated by ElevenLabs — use `eleven_multilingual_v2`.\n  'eleven_monolingual_v1',\n] as const\n\nexport type ElevenLabsTTSModel = (typeof ELEVENLABS_TTS_MODELS)[number]\n\n/**\n * Audio generation models — music (`music_v*`) + sound effects\n * (`eleven_text_to_sound_v*`) share one `generateAudio` adapter.\n * The adapter dispatches by model id so callers pick behavior via the model.\n *\n * @see https://elevenlabs.io/docs/overview/capabilities/music\n * @see https://elevenlabs.io/docs/overview/capabilities/sound-effects\n */\nexport const ELEVENLABS_AUDIO_MODELS = [\n  'music_v2_5',\n  'music_v2',\n  'eleven_text_to_sound_v2',\n  // Deprecated by ElevenLabs — use `music_v2_5`.\n  'music_v1',\n] as const satisfies ReadonlyArray<\n  ElevenLabs.MusicModelId | ElevenLabs.SfxModelId\n>\n\nexport type ElevenLabsAudioModel = (typeof ELEVENLABS_AUDIO_MODELS)[number]\n\n/** Music models within the audio family, derived from the audio list. */\nexport type ElevenLabsMusicModel = Extract<\n  ElevenLabsAudioModel,\n  `music_${string}`\n>\n/** SFX models within the audio family, derived from the audio list. */\nexport type ElevenLabsSoundEffectsModel = Extract<\n  ElevenLabsAudioModel,\n  `eleven_text_to_sound_${string}`\n>\n\n/**\n * Matches the `music_` prefix so a new music release needs no branch here,\n * but still checks membership: the narrowed type is the closed set above, so\n * a `music_*` id this package does not ship must not pass.\n */\nexport function isElevenLabsMusicModel(\n  model: string,\n): model is ElevenLabsMusicModel {\n  return (\n    model.startsWith('music_') &&\n    (ELEVENLABS_AUDIO_MODELS as ReadonlyArray<string>).includes(model)\n  )\n}\n\nexport function isElevenLabsSoundEffectsModel(\n  model: string,\n): model is ElevenLabsSoundEffectsModel {\n  return model.startsWith('eleven_text_to_sound_')\n}\n\n/**\n * Speech-to-text (transcription) models — Scribe family.\n * @see https://elevenlabs.io/docs/overview/capabilities/speech-to-text\n */\nexport const ELEVENLABS_TRANSCRIPTION_MODELS = [\n  'scribe_v2',\n  // Domain-tuned for clinical audio.\n  'scribe_v2_medical',\n  // Deprecated by ElevenLabs — use `scribe_v2`.\n  'scribe_v1',\n] as const\n\nexport type ElevenLabsTranscriptionModel =\n  (typeof ELEVENLABS_TRANSCRIPTION_MODELS)[number]\n\n/**\n * Voice design (text-to-voice) models, used by the `generateVoice` adapter.\n * `eleven_ttv_v3` is the only one that accepts reference audio, so guiding a\n * design with a recording of a real speaker requires it.\n *\n * @see https://elevenlabs.io/docs/overview/capabilities/voice-design\n */\nexport const ELEVENLABS_VOICE_MODELS = [\n  'eleven_ttv_v3',\n  'eleven_multilingual_ttv_v2',\n] as const satisfies ReadonlyArray<ElevenLabs.VoiceDesignRequestModelModelId>\n\nexport type ElevenLabsVoiceModel = (typeof ELEVENLABS_VOICE_MODELS)[number]\n\n/**\n * Supported `output_format` strings, encoded as `codec_samplerate[_bitrate]`.\n * Aliased to the SDK's `AllowedOutputFormats` so the list stays in sync\n * automatically whenever the `@elevenlabs/elevenlabs-js` dependency is bumped.\n *\n * @see https://elevenlabs.io/docs/api-reference/text-to-speech/convert\n */\nexport type ElevenLabsOutputFormat = ElevenLabs.AllowedOutputFormats\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;AAuBA,IAAa,wBAAwB;CACnC;CACA;CACA;CACA;CACA;CAEA;CAEA;CAEA;AACF;;;;;;;;;AAYA,IAAa,0BAA0B;CACrC;CACA;CACA;CAEA;AACF;;;;;;AAsBA,SAAgB,uBACd,OAC+B;CAC/B,OACE,MAAM,WAAW,QAAQ,KACxB,wBAAkD,SAAS,KAAK;AAErE;AAEA,SAAgB,8BACd,OACsC;CACtC,OAAO,MAAM,WAAW,uBAAuB;AACjD;;;;;AAMA,IAAa,kCAAkC;CAC7C;CAEA;CAEA;AACF;;;;;;;;AAYA,IAAa,0BAA0B,CACrC,iBACA,4BACF"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tanstack/ai-elevenlabs",
-  "version": "0.4.4",
+  "version": "0.6.0",
   "description": "ElevenLabs adapter for TanStack AI realtime voice, text-to-speech, transcription, music, and sound effects.",
   "author": "Tanner Linsley",
   "license": "MIT",
@@ -52,16 +52,16 @@
     "src"
   ],
   "dependencies": {
-    "@elevenlabs/client": "^1.3.1",
-    "@elevenlabs/elevenlabs-js": "^2.44.0",
+    "@elevenlabs/client": "^1.25.0",
+    "@elevenlabs/elevenlabs-js": "^2.68.0",
     "@tanstack/ai-utils": "^0.4.0"
   },
   "peerDependencies": {
-    "@tanstack/ai": "^0.55.0"
+    "@tanstack/ai": "^0.57.0"
   },
   "devDependencies": {
     "@vitest/coverage-v8": "4.1.10",
-    "@tanstack/ai": "0.55.0"
+    "@tanstack/ai": "0.57.0"
   },
   "scripts": {
     "build": "vite build",

package/src/adapters/audio.ts

@@ -7,6 +7,7 @@ import {
   readStreamToArrayBuffer,
 } from '../utils/client'
 import {
+  ELEVENLABS_AUDIO_MODELS,
   isElevenLabsMusicModel,
   isElevenLabsSoundEffectsModel,
 } from '../model-meta'
@@ -55,7 +56,7 @@ interface CommonAudioOptions {
 }
 
 /**
- * Provider options for music generation (`music_v1`).
+ * Provider options for music generation (`music_v*`).
  */
 export interface ElevenLabsMusicProviderOptions extends CommonAudioOptions {
   /** Structured composition plan. Mutually exclusive with `prompt`/`duration`. */
@@ -94,7 +95,7 @@ export type ElevenLabsAudioProviderOptions =
  *
  * @example
  * ```ts
- * const music = elevenlabsAudio('music_v1')
+ * const music = elevenlabsAudio('music_v2_5')
  * await generateAudio({ adapter: music, prompt: 'lo-fi beat', duration: 15 })
  *
  * const sfx = elevenlabsAudio('eleven_text_to_sound_v2')
@@ -129,7 +130,7 @@ export class ElevenLabsAudioAdapter<
         return await this.runSoundEffects(options)
       }
       throw new Error(
-        `Unsupported ElevenLabs audio model "${this.model}". Expected one of: music_v1, eleven_text_to_sound_v2, eleven_text_to_sound_v1.`,
+        `Unsupported ElevenLabs audio model "${this.model}". Expected one of: ${ELEVENLABS_AUDIO_MODELS.join(', ')}.`,
       )
     } catch (error) {
       logger.errors('elevenlabs.generateAudio fatal', {