@cloudflare/tanstack-ai 0.1.6 → 0.1.7

package/dist/adapters/workers-ai-transcription.mjs

@@ -0,0 +1,199 @@
+import { a as validateWorkersAiConfig, i as isDirectCredentialsConfig, r as isDirectBindingConfig, t as createGatewayFetch } from "../create-fetcher-DY7wfYYy.mjs";
+import { t as _defineProperty } from "../defineProperty-CbyrzcbA.mjs";
+import { n as workersAiRestFetchBinary, t as workersAiRestFetch } from "../workers-ai-rest-GKy2r7eG.mjs";
+import { n as uint8ArrayToBase64 } from "../binary-p4H_N_3M.mjs";
+import { BaseTranscriptionAdapter } from "@tanstack/ai/adapters";
+//#region src/adapters/workers-ai-transcription.ts
+var WorkersAiTranscriptionAdapter = class extends BaseTranscriptionAdapter {
+	constructor(config, model) {
+		super({}, model);
+		_defineProperty(this, "name", "workers-ai-transcription");
+		_defineProperty(this, "adapterConfig", void 0);
+		validateWorkersAiConfig(config);
+		this.adapterConfig = config;
+	}
+	async transcribe(options) {
+		const { audio, language, prompt, modelOptions } = options;
+		const audioBytes = await normalizeAudioToBytes(audio);
+		const extra = { ...modelOptions };
+		if (language) extra.language = language;
+		if (prompt) extra.initial_prompt = prompt;
+		const audioPayload = this.buildAudioPayload(audioBytes, audio);
+		if (isDirectBindingConfig(this.adapterConfig)) return this.transcribeViaBinding(audioPayload, extra);
+		if (isDirectCredentialsConfig(this.adapterConfig)) {
+			if (this.model === "@cf/deepgram/nova-3") return this.transcribeViaRestBinary(audioBytes, audio, extra);
+			return this.transcribeViaRest(audioPayload, extra);
+		}
+		return this.transcribeViaGateway(audioPayload, extra);
+	}
+	/**
+	* Build the audio field for the request payload, handling model-specific formats.
+	*
+	* - `@cf/deepgram/nova-3` requires `{ body: base64, contentType: "audio/..." }`
+	* - `@cf/openai/whisper-large-v3-turbo` REST/gateway accepts a base64 string
+	* - Other Whisper models accept `number[]` (binding) or base64 (REST)
+	*/
+	buildAudioPayload(audioBytes, originalAudio) {
+		if (this.model === "@cf/deepgram/nova-3") return { audio: {
+			body: uint8ArrayToBase64(new Uint8Array(audioBytes)),
+			contentType: detectAudioContentType(originalAudio)
+		} };
+		if (this.model === "@cf/openai/whisper-large-v3-turbo") return { audio: uint8ArrayToBase64(new Uint8Array(audioBytes)) };
+		return { audio: audioBytes };
+	}
+	async transcribeViaBinding(audioPayload, options) {
+		const result = await this.adapterConfig.binding.run(this.model, {
+			...audioPayload,
+			...options
+		});
+		return this.normalizeResult(result);
+	}
+	async transcribeViaRest(audioPayload, options) {
+		const config = this.adapterConfig;
+		const data = await (await workersAiRestFetch(config, this.model, {
+			...audioPayload,
+			...options
+		}, {
+			label: "Workers AI transcription",
+			signal: options.signal
+		})).json();
+		return this.normalizeResult(data.result ?? data);
+	}
+	/**
+	* Transcribe via REST using raw binary audio.
+	* Required for models like Deepgram Nova-3 that expect raw audio bytes
+	* with a Content-Type header (e.g. "audio/wav") instead of JSON.
+	*/
+	async transcribeViaRestBinary(audioBytes, originalAudio, options) {
+		const config = this.adapterConfig;
+		const contentType = detectAudioContentType(originalAudio);
+		const data = await (await workersAiRestFetchBinary(config, this.model, new Uint8Array(audioBytes), contentType, {
+			label: "Workers AI transcription",
+			signal: options.signal
+		})).json();
+		return this.normalizeResult(data.result ?? data);
+	}
+	async transcribeViaGateway(audioPayload, options) {
+		const gatewayConfig = this.adapterConfig;
+		const response = await createGatewayFetch("workers-ai", gatewayConfig)("https://api.cloudflare.com/v1/audio/transcriptions", {
+			method: "POST",
+			body: JSON.stringify({
+				model: this.model,
+				...audioPayload,
+				...options
+			})
+		});
+		if (!response.ok) {
+			const errorText = await response.text();
+			throw new Error(`Workers AI transcription gateway request failed (${response.status}): ${errorText}`);
+		}
+		const data = await response.json();
+		return this.normalizeResult(data);
+	}
+	/**
+	* Normalize Workers AI transcription results into the standard
+	* TanStack AI TranscriptionResult shape.
+	*
+	* Handles three response formats:
+	* - Whisper: `{ text, words?, vtt? }`
+	* - Whisper v3-turbo: `{ text, segments?, transcription_info? }`
+	* - Deepgram Nova-3: `{ results: { channels: [{ alternatives: [{ transcript, words }] }] } }`
+	*/
+	normalizeResult(raw) {
+		const results = raw.results;
+		if (results?.channels) {
+			const alt = results.channels?.[0]?.alternatives?.[0];
+			const text = alt?.transcript ?? "";
+			const result = {
+				id: this.generateId(),
+				model: this.model,
+				text
+			};
+			if (alt?.words && Array.isArray(alt.words)) result.words = alt.words.map((w) => ({
+				word: w.word ?? "",
+				start: w.start ?? 0,
+				end: w.end ?? 0
+			}));
+			return result;
+		}
+		const result = {
+			id: this.generateId(),
+			model: this.model,
+			text: raw.text ?? ""
+		};
+		const transcriptionInfo = raw.transcription_info;
+		if (transcriptionInfo?.language) result.language = transcriptionInfo.language;
+		if (transcriptionInfo?.duration != null) result.duration = transcriptionInfo.duration;
+		if (raw.segments && Array.isArray(raw.segments)) result.segments = raw.segments.map((seg, idx) => ({
+			id: idx,
+			text: seg.text ?? "",
+			start: seg.start ?? 0,
+			end: seg.end ?? 0
+		}));
+		if (raw.words && Array.isArray(raw.words)) result.words = raw.words.map((w) => ({
+			word: w.word ?? "",
+			start: w.start ?? 0,
+			end: w.end ?? 0
+		}));
+		return result;
+	}
+};
+/**
+* Creates a Workers AI transcription adapter for speech-to-text.
+*
+* Works with TanStack AI's `generateTranscription()` activity function:
+* ```ts
+* import { generateTranscription } from "@tanstack/ai";
+* import { createWorkersAiTranscription } from "@cloudflare/tanstack-ai";
+*
+* const adapter = createWorkersAiTranscription(
+*   "@cf/openai/whisper-large-v3-turbo",
+*   { binding: env.AI },
+* );
+*
+* const result = await generateTranscription({ adapter, audio: audioData });
+* // result.text — the transcribed text
+* ```
+*
+* Note: Factory takes `(model, config)` for ergonomics — the class constructor
+* uses `(config, model)` to match TanStack AI's upstream convention.
+*/
+function createWorkersAiTranscription(model, config) {
+	return new WorkersAiTranscriptionAdapter(config, model);
+}
+/**
+* Normalize various audio input formats into a number[] (raw bytes).
+*
+* Note: `File extends Blob`, so `File` instances are handled by the
+* `instanceof Blob` branch. `Blob.arrayBuffer()` always reads the full
+* contents regardless of any prior reads — there's no cursor to worry about.
+*/
+async function normalizeAudioToBytes(audio) {
+	if (audio instanceof ArrayBuffer) return Array.from(new Uint8Array(audio));
+	if (audio instanceof Blob) {
+		const buffer = await audio.arrayBuffer();
+		return Array.from(new Uint8Array(buffer));
+	}
+	if (typeof audio === "string") {
+		const binary = atob(audio);
+		const bytes = new Uint8Array(binary.length);
+		for (let i = 0; i < binary.length; i++) bytes[i] = binary.charCodeAt(i);
+		return Array.from(bytes);
+	}
+	throw new Error("Unsupported audio format. Expected string, File, Blob, or ArrayBuffer.");
+}
+/**
+* Detect the MIME type of the audio input for models that require it
+* (e.g., Deepgram Nova-3).
+*
+* - `File` / `Blob`: use the `.type` property (e.g., "audio/wav")
+* - `ArrayBuffer` / `string`: defaults to "audio/wav"
+*/
+function detectAudioContentType(audio) {
+	if (audio instanceof Blob && audio.type) return audio.type;
+	return "audio/wav";
+}
+//#endregion
+export { WorkersAiTranscriptionAdapter, createWorkersAiTranscription };
+
+//# sourceMappingURL=workers-ai-transcription.mjs.map

package/dist/adapters/workers-ai-transcription.mjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"workers-ai-transcription.mjs","names":[],"sources":["../../src/adapters/workers-ai-transcription.ts"],"sourcesContent":["import { BaseTranscriptionAdapter } from \"@tanstack/ai/adapters\";\nimport type { TranscriptionOptions, TranscriptionResult } from \"@tanstack/ai\";\nimport {\n\ttype WorkersAiAdapterConfig,\n\ttype WorkersAiDirectBindingConfig,\n\ttype WorkersAiDirectCredentialsConfig,\n\ttype AiGatewayAdapterConfig,\n\tcreateGatewayFetch,\n\tisDirectBindingConfig,\n\tisDirectCredentialsConfig,\n\tvalidateWorkersAiConfig,\n} from \"../utils/create-fetcher\";\nimport { workersAiRestFetch, workersAiRestFetchBinary } from \"../utils/workers-ai-rest\";\nimport { uint8ArrayToBase64 } from \"../utils/binary\";\n\n// ---------------------------------------------------------------------------\n// Model types\n// ---------------------------------------------------------------------------\n\n/**\n * Workers AI models that support speech-to-text transcription.\n *\n * Note: the typed `AiModels` interface in `@cloudflare/workers-types` may lag\n * behind what's deployed. We use a string union here that matches the known\n * models including Deepgram partner models.\n *\n * **Nova-3 note:** `@cf/deepgram/nova-3` uses a different input format than the\n * Whisper models. Via binding it accepts `{ audio: { body: base64, contentType } }`.\n * Via REST it requires multipart form data (not JSON). The adapter handles both.\n */\nexport type WorkersAiTranscriptionModel =\n\t| \"@cf/openai/whisper\"\n\t| \"@cf/openai/whisper-tiny-en\"\n\t| \"@cf/openai/whisper-large-v3-turbo\"\n\t| \"@cf/deepgram/nova-3\"\n\t| (string & {});\n\n// ---------------------------------------------------------------------------\n// WorkersAiTranscriptionAdapter\n// ---------------------------------------------------------------------------\n\nexport class WorkersAiTranscriptionAdapter extends BaseTranscriptionAdapter<WorkersAiTranscriptionModel> {\n\treadonly name = \"workers-ai-transcription\" as const;\n\tprivate adapterConfig: WorkersAiAdapterConfig;\n\n\tconstructor(config: WorkersAiAdapterConfig, model: WorkersAiTranscriptionModel) {\n\t\tsuper({}, model);\n\t\tvalidateWorkersAiConfig(config);\n\t\tthis.adapterConfig = config;\n\t}\n\n\tasync transcribe(options: TranscriptionOptions): Promise<TranscriptionResult> {\n\t\tconst { audio, language, prompt, modelOptions } = options;\n\n\t\t// Normalize audio to raw bytes\n\t\tconst audioBytes = await normalizeAudioToBytes(audio);\n\n\t\tconst extra: Record<string, unknown> = { ...modelOptions };\n\t\tif (language) extra.language = language;\n\t\tif (prompt) extra.initial_prompt = prompt;\n\n\t\t// Build the model-specific audio payload:\n\t\t// - Deepgram Nova-3 (binding): { audio: { body: base64, contentType: \"audio/...\" } }\n\t\t// - Deepgram Nova-3 (REST): multipart FormData (handled separately)\n\t\t// - Whisper Large v3 Turbo (REST/gateway): { audio: base64string }\n\t\t// - Other Whisper models (binding): { audio: number[] }\n\t\tconst audioPayload = this.buildAudioPayload(audioBytes, audio);\n\n\t\tif (isDirectBindingConfig(this.adapterConfig)) {\n\t\t\treturn this.transcribeViaBinding(audioPayload, extra);\n\t\t}\n\n\t\tif (isDirectCredentialsConfig(this.adapterConfig)) {\n\t\t\t// Nova-3 REST requires raw binary audio, not JSON\n\t\t\tif (this.model === \"@cf/deepgram/nova-3\") {\n\t\t\t\treturn this.transcribeViaRestBinary(audioBytes, audio, extra);\n\t\t\t}\n\t\t\treturn this.transcribeViaRest(audioPayload, extra);\n\t\t}\n\n\t\treturn this.transcribeViaGateway(audioPayload, extra);\n\t}\n\n\t/**\n\t * Build the audio field for the request payload, handling model-specific formats.\n\t *\n\t * - `@cf/deepgram/nova-3` requires `{ body: base64, contentType: \"audio/...\" }`\n\t * - `@cf/openai/whisper-large-v3-turbo` REST/gateway accepts a base64 string\n\t * - Other Whisper models accept `number[]` (binding) or base64 (REST)\n\t */\n\tprivate buildAudioPayload(\n\t\taudioBytes: number[],\n\t\toriginalAudio: string | File | Blob | ArrayBuffer,\n\t): Record<string, unknown> {\n\t\tif (this.model === \"@cf/deepgram/nova-3\") {\n\t\t\tconst b64 = uint8ArrayToBase64(new Uint8Array(audioBytes));\n\t\t\tconst contentType = detectAudioContentType(originalAudio);\n\t\t\treturn { audio: { body: b64, contentType } };\n\t\t}\n\n\t\tif (this.model === \"@cf/openai/whisper-large-v3-turbo\") {\n\t\t\treturn { audio: uint8ArrayToBase64(new Uint8Array(audioBytes)) };\n\t\t}\n\n\t\treturn { audio: audioBytes };\n\t}\n\n\tprivate async transcribeViaBinding(\n\t\taudioPayload: Record<string, unknown>,\n\t\toptions: Record<string, unknown>,\n\t): Promise<TranscriptionResult> {\n\t\tconst ai = (this.adapterConfig as WorkersAiDirectBindingConfig).binding;\n\t\tconst result = (await ai.run(this.model, {\n\t\t\t...audioPayload,\n\t\t\t...options,\n\t\t})) as Record<string, unknown>;\n\t\treturn this.normalizeResult(result);\n\t}\n\n\tprivate async transcribeViaRest(\n\t\taudioPayload: Record<string, unknown>,\n\t\toptions: Record<string, unknown>,\n\t): Promise<TranscriptionResult> {\n\t\tconst config = this.adapterConfig as WorkersAiDirectCredentialsConfig;\n\n\t\tconst response = await workersAiRestFetch(\n\t\t\tconfig,\n\t\t\tthis.model,\n\t\t\t{ ...audioPayload, ...options },\n\t\t\t{\n\t\t\t\tlabel: \"Workers AI transcription\",\n\t\t\t\tsignal: (options as { signal?: AbortSignal }).signal,\n\t\t\t},\n\t\t);\n\n\t\tconst data = (await response.json()) as {\n\t\t\tresult?: Record<string, unknown>;\n\t\t} & Record<string, unknown>;\n\n\t\t// Cloudflare REST API wraps responses in { success, result: {...} }.\n\t\t// Use `data.result` when present, fall back to `data` for direct responses.\n\t\treturn this.normalizeResult(data.result ?? data);\n\t}\n\n\t/**\n\t * Transcribe via REST using raw binary audio.\n\t * Required for models like Deepgram Nova-3 that expect raw audio bytes\n\t * with a Content-Type header (e.g. \"audio/wav\") instead of JSON.\n\t */\n\tprivate async transcribeViaRestBinary(\n\t\taudioBytes: number[],\n\t\toriginalAudio: string | File | Blob | ArrayBuffer,\n\t\toptions: Record<string, unknown>,\n\t): Promise<TranscriptionResult> {\n\t\tconst config = this.adapterConfig as WorkersAiDirectCredentialsConfig;\n\t\tconst contentType = detectAudioContentType(originalAudio);\n\n\t\tconst response = await workersAiRestFetchBinary(\n\t\t\tconfig,\n\t\t\tthis.model,\n\t\t\tnew Uint8Array(audioBytes),\n\t\t\tcontentType,\n\t\t\t{\n\t\t\t\tlabel: \"Workers AI transcription\",\n\t\t\t\tsignal: (options as { signal?: AbortSignal }).signal,\n\t\t\t},\n\t\t);\n\n\t\tconst data = (await response.json()) as {\n\t\t\tresult?: Record<string, unknown>;\n\t\t} & Record<string, unknown>;\n\n\t\treturn this.normalizeResult(data.result ?? data);\n\t}\n\n\tprivate async transcribeViaGateway(\n\t\taudioPayload: Record<string, unknown>,\n\t\toptions: Record<string, unknown>,\n\t): Promise<TranscriptionResult> {\n\t\tconst gatewayConfig = this.adapterConfig as AiGatewayAdapterConfig;\n\t\tconst gatewayFetch = createGatewayFetch(\"workers-ai\", gatewayConfig);\n\n\t\t// The URL here is a placeholder — createGatewayFetch for \"workers-ai\" extracts\n\t\t// the model from the body, sets it as the endpoint, and routes through the gateway.\n\t\t// The actual URL path is not used.\n\t\tconst response = await gatewayFetch(\"https://api.cloudflare.com/v1/audio/transcriptions\", {\n\t\t\tmethod: \"POST\",\n\t\t\tbody: JSON.stringify({\n\t\t\t\tmodel: this.model,\n\t\t\t\t...audioPayload,\n\t\t\t\t...options,\n\t\t\t}),\n\t\t});\n\n\t\tif (!response.ok) {\n\t\t\tconst errorText = await response.text();\n\t\t\tthrow new Error(\n\t\t\t\t`Workers AI transcription gateway request failed (${response.status}): ${errorText}`,\n\t\t\t);\n\t\t}\n\n\t\tconst data = (await response.json()) as Record<string, unknown>;\n\t\treturn this.normalizeResult(data);\n\t}\n\n\t/**\n\t * Normalize Workers AI transcription results into the standard\n\t * TanStack AI TranscriptionResult shape.\n\t *\n\t * Handles three response formats:\n\t * - Whisper: `{ text, words?, vtt? }`\n\t * - Whisper v3-turbo: `{ text, segments?, transcription_info? }`\n\t * - Deepgram Nova-3: `{ results: { channels: [{ alternatives: [{ transcript, words }] }] } }`\n\t */\n\tprivate normalizeResult(raw: Record<string, unknown>): TranscriptionResult {\n\t\t// Deepgram Nova-3 format: { results: { channels: [{ alternatives: [{ transcript, words }] }] } }\n\t\tconst results = raw.results as Record<string, unknown> | undefined;\n\t\tif (results?.channels) {\n\t\t\tconst channels = results.channels as Array<{\n\t\t\t\talternatives?: Array<{\n\t\t\t\t\ttranscript?: string;\n\t\t\t\t\tconfidence?: number;\n\t\t\t\t\twords?: Array<{ word: string; start: number; end: number; confidence: number }>;\n\t\t\t\t}>;\n\t\t\t}>;\n\t\t\tconst alt = channels?.[0]?.alternatives?.[0];\n\t\t\tconst text = alt?.transcript ?? \"\";\n\t\t\tconst result: TranscriptionResult = {\n\t\t\t\tid: this.generateId(),\n\t\t\t\tmodel: this.model,\n\t\t\t\ttext,\n\t\t\t};\n\t\t\tif (alt?.words && Array.isArray(alt.words)) {\n\t\t\t\tresult.words = alt.words.map((w) => ({\n\t\t\t\t\tword: w.word ?? \"\",\n\t\t\t\t\tstart: w.start ?? 0,\n\t\t\t\t\tend: w.end ?? 0,\n\t\t\t\t}));\n\t\t\t}\n\t\t\treturn result;\n\t\t}\n\n\t\t// Whisper format: { text, words?, vtt? }\n\t\t// Whisper v3-turbo format: { text, segments?, transcription_info? }\n\t\tconst result: TranscriptionResult = {\n\t\t\tid: this.generateId(),\n\t\t\tmodel: this.model,\n\t\t\ttext: (raw.text as string) ?? \"\",\n\t\t};\n\n\t\t// Language from transcription_info (whisper-large-v3-turbo)\n\t\tconst transcriptionInfo = raw.transcription_info as Record<string, unknown> | undefined;\n\t\tif (transcriptionInfo?.language) {\n\t\t\tresult.language = transcriptionInfo.language as string;\n\t\t}\n\n\t\t// Duration\n\t\tif (transcriptionInfo?.duration != null) {\n\t\t\tresult.duration = transcriptionInfo.duration as number;\n\t\t}\n\n\t\t// Segments (whisper-large-v3-turbo returns these)\n\t\tif (raw.segments && Array.isArray(raw.segments)) {\n\t\t\tresult.segments = raw.segments.map((seg: Record<string, unknown>, idx: number) => ({\n\t\t\t\tid: idx,\n\t\t\t\ttext: (seg.text as string) ?? \"\",\n\t\t\t\tstart: (seg.start as number) ?? 0,\n\t\t\t\tend: (seg.end as number) ?? 0,\n\t\t\t}));\n\t\t}\n\n\t\t// Words — basic whisper returns top-level words[], v3-turbo nests them in segments\n\t\tif (raw.words && Array.isArray(raw.words)) {\n\t\t\tresult.words = raw.words.map((w: Record<string, unknown>) => ({\n\t\t\t\tword: (w.word as string) ?? \"\",\n\t\t\t\tstart: (w.start as number) ?? 0,\n\t\t\t\tend: (w.end as number) ?? 0,\n\t\t\t}));\n\t\t}\n\n\t\treturn result;\n\t}\n}\n\n// ---------------------------------------------------------------------------\n// Factory function\n// ---------------------------------------------------------------------------\n\n/**\n * Creates a Workers AI transcription adapter for speech-to-text.\n *\n * Works with TanStack AI's `generateTranscription()` activity function:\n * ```ts\n * import { generateTranscription } from \"@tanstack/ai\";\n * import { createWorkersAiTranscription } from \"@cloudflare/tanstack-ai\";\n *\n * const adapter = createWorkersAiTranscription(\n *   \"@cf/openai/whisper-large-v3-turbo\",\n *   { binding: env.AI },\n * );\n *\n * const result = await generateTranscription({ adapter, audio: audioData });\n * // result.text — the transcribed text\n * ```\n *\n * Note: Factory takes `(model, config)` for ergonomics — the class constructor\n * uses `(config, model)` to match TanStack AI's upstream convention.\n */\nexport function createWorkersAiTranscription(\n\tmodel: WorkersAiTranscriptionModel,\n\tconfig: WorkersAiAdapterConfig,\n) {\n\treturn new WorkersAiTranscriptionAdapter(config, model);\n}\n\n// ---------------------------------------------------------------------------\n// Utilities\n// ---------------------------------------------------------------------------\n\n/**\n * Normalize various audio input formats into a number[] (raw bytes).\n *\n * Note: `File extends Blob`, so `File` instances are handled by the\n * `instanceof Blob` branch. `Blob.arrayBuffer()` always reads the full\n * contents regardless of any prior reads — there's no cursor to worry about.\n */\nasync function normalizeAudioToBytes(audio: string | File | Blob | ArrayBuffer): Promise<number[]> {\n\tif (audio instanceof ArrayBuffer) {\n\t\treturn Array.from(new Uint8Array(audio));\n\t}\n\n\tif (audio instanceof Blob) {\n\t\t// This also handles `File` (which extends Blob)\n\t\tconst buffer = await audio.arrayBuffer();\n\t\treturn Array.from(new Uint8Array(buffer));\n\t}\n\n\tif (typeof audio === \"string\") {\n\t\t// Assume base64 string — decode to bytes\n\t\tconst binary = atob(audio);\n\t\tconst bytes = new Uint8Array(binary.length);\n\t\tfor (let i = 0; i < binary.length; i++) {\n\t\t\tbytes[i] = binary.charCodeAt(i);\n\t\t}\n\t\treturn Array.from(bytes);\n\t}\n\n\tthrow new Error(\"Unsupported audio format. Expected string, File, Blob, or ArrayBuffer.\");\n}\n\n/**\n * Detect the MIME type of the audio input for models that require it\n * (e.g., Deepgram Nova-3).\n *\n * - `File` / `Blob`: use the `.type` property (e.g., \"audio/wav\")\n * - `ArrayBuffer` / `string`: defaults to \"audio/wav\"\n */\nfunction detectAudioContentType(audio: string | File | Blob | ArrayBuffer): string {\n\t// File and Blob carry their own MIME type\n\tif (audio instanceof Blob && audio.type) {\n\t\treturn audio.type;\n\t}\n\n\t// For raw bytes, default to audio/wav — this is the most common\n\t// format for transcription inputs and what the E2E tests use.\n\treturn \"audio/wav\";\n}\n"],"mappings":";;;;;;AAyCA,IAAa,gCAAb,cAAmD,yBAAsD;CAIxG,YAAY,QAAgC,OAAoC;AAC/E,QAAM,EAAE,EAAE,MAAM;wBAJR,QAAO,2BAAoC;wBAC5C,iBAAA,KAAA,EAAsC;AAI7C,0BAAwB,OAAO;AAC/B,OAAK,gBAAgB;;CAGtB,MAAM,WAAW,SAA6D;EAC7E,MAAM,EAAE,OAAO,UAAU,QAAQ,iBAAiB;EAGlD,MAAM,aAAa,MAAM,sBAAsB,MAAM;EAErD,MAAM,QAAiC,EAAE,GAAG,cAAc;AAC1D,MAAI,SAAU,OAAM,WAAW;AAC/B,MAAI,OAAQ,OAAM,iBAAiB;EAOnC,MAAM,eAAe,KAAK,kBAAkB,YAAY,MAAM;AAE9D,MAAI,sBAAsB,KAAK,cAAc,CAC5C,QAAO,KAAK,qBAAqB,cAAc,MAAM;AAGtD,MAAI,0BAA0B,KAAK,cAAc,EAAE;AAElD,OAAI,KAAK,UAAU,sBAClB,QAAO,KAAK,wBAAwB,YAAY,OAAO,MAAM;AAE9D,UAAO,KAAK,kBAAkB,cAAc,MAAM;;AAGnD,SAAO,KAAK,qBAAqB,cAAc,MAAM;;;;;;;;;CAUtD,kBACC,YACA,eAC0B;AAC1B,MAAI,KAAK,UAAU,sBAGlB,QAAO,EAAE,OAAO;GAAE,MAFN,mBAAmB,IAAI,WAAW,WAAW,CAAC;GAE7B,aADT,uBAAuB,cAAc;GACf,EAAE;AAG7C,MAAI,KAAK,UAAU,oCAClB,QAAO,EAAE,OAAO,mBAAmB,IAAI,WAAW,WAAW,CAAC,EAAE;AAGjE,SAAO,EAAE,OAAO,YAAY;;CAG7B,MAAc,qBACb,cACA,SAC+B;EAE/B,MAAM,SAAU,MADJ,KAAK,cAA+C,QACvC,IAAI,KAAK,OAAO;GACxC,GAAG;GACH,GAAG;GACH,CAAC;AACF,SAAO,KAAK,gBAAgB,OAAO;;CAGpC,MAAc,kBACb,cACA,SAC+B;EAC/B,MAAM,SAAS,KAAK;EAYpB,MAAM,OAAQ,OAVG,MAAM,mBACtB,QACA,KAAK,OACL;GAAE,GAAG;GAAc,GAAG;GAAS,EAC/B;GACC,OAAO;GACP,QAAS,QAAqC;GAC9C,CACD,EAE4B,MAAM;AAMnC,SAAO,KAAK,gBAAgB,KAAK,UAAU,KAAK;;;;;;;CAQjD,MAAc,wBACb,YACA,eACA,SAC+B;EAC/B,MAAM,SAAS,KAAK;EACpB,MAAM,cAAc,uBAAuB,cAAc;EAazD,MAAM,OAAQ,OAXG,MAAM,yBACtB,QACA,KAAK,OACL,IAAI,WAAW,WAAW,EAC1B,aACA;GACC,OAAO;GACP,QAAS,QAAqC;GAC9C,CACD,EAE4B,MAAM;AAInC,SAAO,KAAK,gBAAgB,KAAK,UAAU,KAAK;;CAGjD,MAAc,qBACb,cACA,SAC+B;EAC/B,MAAM,gBAAgB,KAAK;EAM3B,MAAM,WAAW,MALI,mBAAmB,cAAc,cAAc,CAKhC,sDAAsD;GACzF,QAAQ;GACR,MAAM,KAAK,UAAU;IACpB,OAAO,KAAK;IACZ,GAAG;IACH,GAAG;IACH,CAAC;GACF,CAAC;AAEF,MAAI,CAAC,SAAS,IAAI;GACjB,MAAM,YAAY,MAAM,SAAS,MAAM;AACvC,SAAM,IAAI,MACT,oDAAoD,SAAS,OAAO,KAAK,YACzE;;EAGF,MAAM,OAAQ,MAAM,SAAS,MAAM;AACnC,SAAO,KAAK,gBAAgB,KAAK;;;;;;;;;;;CAYlC,gBAAwB,KAAmD;EAE1E,MAAM,UAAU,IAAI;AACpB,MAAI,SAAS,UAAU;GAQtB,MAAM,MAPW,QAAQ,WAOF,IAAI,eAAe;GAC1C,MAAM,OAAO,KAAK,cAAc;GAChC,MAAM,SAA8B;IACnC,IAAI,KAAK,YAAY;IACrB,OAAO,KAAK;IACZ;IACA;AACD,OAAI,KAAK,SAAS,MAAM,QAAQ,IAAI,MAAM,CACzC,QAAO,QAAQ,IAAI,MAAM,KAAK,OAAO;IACpC,MAAM,EAAE,QAAQ;IAChB,OAAO,EAAE,SAAS;IAClB,KAAK,EAAE,OAAO;IACd,EAAE;AAEJ,UAAO;;EAKR,MAAM,SAA8B;GACnC,IAAI,KAAK,YAAY;GACrB,OAAO,KAAK;GACZ,MAAO,IAAI,QAAmB;GAC9B;EAGD,MAAM,oBAAoB,IAAI;AAC9B,MAAI,mBAAmB,SACtB,QAAO,WAAW,kBAAkB;AAIrC,MAAI,mBAAmB,YAAY,KAClC,QAAO,WAAW,kBAAkB;AAIrC,MAAI,IAAI,YAAY,MAAM,QAAQ,IAAI,SAAS,CAC9C,QAAO,WAAW,IAAI,SAAS,KAAK,KAA8B,SAAiB;GAClF,IAAI;GACJ,MAAO,IAAI,QAAmB;GAC9B,OAAQ,IAAI,SAAoB;GAChC,KAAM,IAAI,OAAkB;GAC5B,EAAE;AAIJ,MAAI,IAAI,SAAS,MAAM,QAAQ,IAAI,MAAM,CACxC,QAAO,QAAQ,IAAI,MAAM,KAAK,OAAgC;GAC7D,MAAO,EAAE,QAAmB;GAC5B,OAAQ,EAAE,SAAoB;GAC9B,KAAM,EAAE,OAAkB;GAC1B,EAAE;AAGJ,SAAO;;;;;;;;;;;;;;;;;;;;;;;AA4BT,SAAgB,6BACf,OACA,QACC;AACD,QAAO,IAAI,8BAA8B,QAAQ,MAAM;;;;;;;;;AAcxD,eAAe,sBAAsB,OAA8D;AAClG,KAAI,iBAAiB,YACpB,QAAO,MAAM,KAAK,IAAI,WAAW,MAAM,CAAC;AAGzC,KAAI,iBAAiB,MAAM;EAE1B,MAAM,SAAS,MAAM,MAAM,aAAa;AACxC,SAAO,MAAM,KAAK,IAAI,WAAW,OAAO,CAAC;;AAG1C,KAAI,OAAO,UAAU,UAAU;EAE9B,MAAM,SAAS,KAAK,MAAM;EAC1B,MAAM,QAAQ,IAAI,WAAW,OAAO,OAAO;AAC3C,OAAK,IAAI,IAAI,GAAG,IAAI,OAAO,QAAQ,IAClC,OAAM,KAAK,OAAO,WAAW,EAAE;AAEhC,SAAO,MAAM,KAAK,MAAM;;AAGzB,OAAM,IAAI,MAAM,yEAAyE;;;;;;;;;AAU1F,SAAS,uBAAuB,OAAmD;AAElF,KAAI,iBAAiB,QAAQ,MAAM,KAClC,QAAO,MAAM;AAKd,QAAO"}

package/dist/adapters/workers-ai-tts.cjs

@@ -1,13 +1,114 @@
-"use strict";Object.defineProperty(exports, "__esModule", {value: true});
+Object.defineProperty(exports, Symbol.toStringTag, { value: "Module" });
+require("../workers-ai-Bm7Up4or.cjs");
+const require_create_fetcher = require("../create-fetcher-5iL34e6H.cjs");
+const require_defineProperty = require("../defineProperty-DQoAg20E.cjs");
+const require_workers_ai_rest = require("../workers-ai-rest-CkNCtBwv.cjs");
+const require_binary = require("../binary-C9FAYwZj.cjs");
+let _tanstack_ai_adapters = require("@tanstack/ai/adapters");
+//#region src/adapters/workers-ai-tts.ts
+var WorkersAiTTSAdapter = class extends _tanstack_ai_adapters.BaseTTSAdapter {
+	constructor(config, model) {
+		super({}, model);
+		require_defineProperty._defineProperty(this, "name", "workers-ai-tts");
+		require_defineProperty._defineProperty(this, "adapterConfig", void 0);
+		require_create_fetcher.validateWorkersAiConfig(config);
+		this.adapterConfig = config;
+	}
+	async generateSpeech(options) {
+		const { text, voice, format, speed, modelOptions } = options;
+		const extra = { ...modelOptions };
+		if (voice) extra.voice = voice;
+		if (speed != null) extra.speed = speed;
+		if (require_create_fetcher.isDirectBindingConfig(this.adapterConfig)) return this.generateViaBinding(text, format, extra);
+		if (require_create_fetcher.isDirectCredentialsConfig(this.adapterConfig)) return this.generateViaRest(text, format, extra);
+		return this.generateViaGateway(text, format, extra);
+	}
+	async generateViaBinding(text, format, options) {
+		const result = await this.adapterConfig.binding.run(this.model, {
+			text,
+			...options
+		});
+		return this.normalizeResult(result, format);
+	}
+	async generateViaRest(text, format, options) {
+		const config = this.adapterConfig;
+		const buffer = await (await require_workers_ai_rest.workersAiRestFetch(config, this.model, {
+			text,
+			...options
+		}, {
+			label: "Workers AI TTS",
+			signal: options.signal
+		})).arrayBuffer();
+		return this.wrapAudioResult(new Uint8Array(buffer), format);
+	}
+	async generateViaGateway(text, format, options) {
+		const gatewayConfig = this.adapterConfig;
+		const response = await require_create_fetcher.createGatewayFetch("workers-ai", gatewayConfig)("https://api.cloudflare.com/v1/audio/speech", {
+			method: "POST",
+			body: JSON.stringify({
+				model: this.model,
+				text,
+				...options
+			})
+		});
+		if (!response.ok) {
+			const errorText = await response.text();
+			throw new Error(`Workers AI TTS gateway request failed (${response.status}): ${errorText}`);
+		}
+		const buffer = await response.arrayBuffer();
+		return this.wrapAudioResult(new Uint8Array(buffer), format);
+	}
+	/**
+	* Normalize binding results. Workers AI TTS can return:
+	* - Uint8Array / ArrayBuffer (raw audio bytes)
+	* - ReadableStream<Uint8Array> (streamed audio bytes)
+	* - { audio: "base64..." } (JSON wrapper)
+	*/
+	async normalizeResult(result, format) {
+		const b64 = await require_binary.binaryToBase64(result, "audio");
+		return {
+			id: this.generateId(),
+			model: this.model,
+			audio: b64,
+			format: format ?? "mp3",
+			contentType: `audio/${format ?? "mp3"}`
+		};
+	}
+	wrapAudioResult(bytes, format) {
+		return {
+			id: this.generateId(),
+			model: this.model,
+			audio: require_binary.uint8ArrayToBase64(bytes),
+			format: format ?? "mp3",
+			contentType: `audio/${format ?? "mp3"}`
+		};
+	}
+};
+/**
+* Creates a Workers AI text-to-speech adapter.
+*
+* Works with TanStack AI's `generateSpeech()` activity function:
+* ```ts
+* import { generateSpeech } from "@tanstack/ai";
+* import { createWorkersAiTts } from "@cloudflare/tanstack-ai";
+*
+* const adapter = createWorkersAiTts(
+*   "@cf/deepgram/aura-1",
+*   { binding: env.AI },
+* );
+*
+* const result = await generateSpeech({ adapter, text: "Hello world" });
+* // result.audio — base64-encoded audio
+* ```
+*
+* Note: Factory takes `(model, config)` for ergonomics — the class constructor
+* uses `(config, model)` to match TanStack AI's upstream convention.
+*/
+function createWorkersAiTts(model, config) {
+	return new WorkersAiTTSAdapter(config, model);
+}
+//#endregion
+exports.WorkersAiTTSAdapter = WorkersAiTTSAdapter;
+exports.createWorkersAiTts = createWorkersAiTts;
 
-
-var _chunkN6U27RQWcjs = require('../chunk-N6U27RQW.cjs');
-require('../chunk-3VQDXJLW.cjs');
-require('../chunk-7HSUHP63.cjs');
-require('../chunk-WBTQKWFN.cjs');
-require('../chunk-4DE2IREA.cjs');
-
-
-
-exports.WorkersAiTTSAdapter = _chunkN6U27RQWcjs.WorkersAiTTSAdapter; exports.createWorkersAiTts = _chunkN6U27RQWcjs.createWorkersAiTts;
 //# sourceMappingURL=workers-ai-tts.cjs.map

package/dist/adapters/workers-ai-tts.cjs.map

@@ -1 +1 @@
-{"version":3,"sources":["/home/runner/work/ai/ai/packages/tanstack-ai/dist/adapters/workers-ai-tts.cjs"],"names":[],"mappings":"AAAA;AACE;AACA;AACF,yDAA8B;AAC9B,iCAA8B;AAC9B,iCAA8B;AAC9B,iCAA8B;AAC9B,iCAA8B;AAC9B;AACE;AACA;AACF,uIAAC","file":"/home/runner/work/ai/ai/packages/tanstack-ai/dist/adapters/workers-ai-tts.cjs"}
+{"version":3,"file":"workers-ai-tts.cjs","names":["BaseTTSAdapter","isDirectBindingConfig","isDirectCredentialsConfig","workersAiRestFetch","createGatewayFetch","binaryToBase64","uint8ArrayToBase64"],"sources":["../../src/adapters/workers-ai-tts.ts"],"sourcesContent":["import { BaseTTSAdapter } from \"@tanstack/ai/adapters\";\nimport type { TTSOptions, TTSResult } from \"@tanstack/ai\";\nimport {\n\ttype WorkersAiAdapterConfig,\n\ttype WorkersAiDirectBindingConfig,\n\ttype WorkersAiDirectCredentialsConfig,\n\ttype AiGatewayAdapterConfig,\n\tcreateGatewayFetch,\n\tisDirectBindingConfig,\n\tisDirectCredentialsConfig,\n\tvalidateWorkersAiConfig,\n} from \"../utils/create-fetcher\";\nimport { workersAiRestFetch } from \"../utils/workers-ai-rest\";\nimport { binaryToBase64, uint8ArrayToBase64 } from \"../utils/binary\";\n\n// ---------------------------------------------------------------------------\n// Model types\n// ---------------------------------------------------------------------------\n\n/**\n * Workers AI models that support text-to-speech generation.\n *\n * Note: the typed `AiModels` interface in `@cloudflare/workers-types` may lag\n * behind what's deployed. We use a string union here that matches the known\n * models including Deepgram partner models.\n */\nexport type WorkersAiTTSModel =\n\t| \"@cf/deepgram/aura-1\"\n\t| \"@cf/deepgram/aura-2-en\"\n\t| \"@cf/deepgram/aura-2-es\"\n\t| (string & {});\n\n// ---------------------------------------------------------------------------\n// WorkersAiTTSAdapter\n// ---------------------------------------------------------------------------\n\nexport class WorkersAiTTSAdapter extends BaseTTSAdapter<WorkersAiTTSModel> {\n\treadonly name = \"workers-ai-tts\" as const;\n\tprivate adapterConfig: WorkersAiAdapterConfig;\n\n\tconstructor(config: WorkersAiAdapterConfig, model: WorkersAiTTSModel) {\n\t\tsuper({}, model);\n\t\tvalidateWorkersAiConfig(config);\n\t\tthis.adapterConfig = config;\n\t}\n\n\tasync generateSpeech(options: TTSOptions): Promise<TTSResult> {\n\t\tconst { text, voice, format, speed, modelOptions } = options;\n\n\t\t// Workers AI TTS models (Deepgram aura-1) accept { text, lang? }\n\t\tconst extra: Record<string, unknown> = { ...modelOptions };\n\t\tif (voice) extra.voice = voice;\n\t\tif (speed != null) extra.speed = speed;\n\n\t\tif (isDirectBindingConfig(this.adapterConfig)) {\n\t\t\treturn this.generateViaBinding(text, format, extra);\n\t\t}\n\n\t\tif (isDirectCredentialsConfig(this.adapterConfig)) {\n\t\t\treturn this.generateViaRest(text, format, extra);\n\t\t}\n\n\t\treturn this.generateViaGateway(text, format, extra);\n\t}\n\n\tprivate async generateViaBinding(\n\t\ttext: string,\n\t\tformat: string | undefined,\n\t\toptions: Record<string, unknown>,\n\t): Promise<TTSResult> {\n\t\tconst ai = (this.adapterConfig as WorkersAiDirectBindingConfig).binding;\n\t\tconst result = await ai.run(this.model, { text, ...options });\n\n\t\treturn this.normalizeResult(result, format);\n\t}\n\n\tprivate async generateViaRest(\n\t\ttext: string,\n\t\tformat: string | undefined,\n\t\toptions: Record<string, unknown>,\n\t): Promise<TTSResult> {\n\t\tconst config = this.adapterConfig as WorkersAiDirectCredentialsConfig;\n\t\tconst response = await workersAiRestFetch(\n\t\t\tconfig,\n\t\t\tthis.model,\n\t\t\t{ text, ...options },\n\t\t\t{ label: \"Workers AI TTS\", signal: (options as { signal?: AbortSignal }).signal },\n\t\t);\n\n\t\t// Workers AI TTS returns audio bytes directly\n\t\tconst buffer = await response.arrayBuffer();\n\t\treturn this.wrapAudioResult(new Uint8Array(buffer), format);\n\t}\n\n\tprivate async generateViaGateway(\n\t\ttext: string,\n\t\tformat: string | undefined,\n\t\toptions: Record<string, unknown>,\n\t): Promise<TTSResult> {\n\t\tconst gatewayConfig = this.adapterConfig as AiGatewayAdapterConfig;\n\t\tconst gatewayFetch = createGatewayFetch(\"workers-ai\", gatewayConfig);\n\n\t\t// The URL here is a placeholder — createGatewayFetch for \"workers-ai\" extracts\n\t\t// the model from the body, sets it as the endpoint, and routes through the gateway.\n\t\t// The actual URL path is not used.\n\t\tconst response = await gatewayFetch(\"https://api.cloudflare.com/v1/audio/speech\", {\n\t\t\tmethod: \"POST\",\n\t\t\tbody: JSON.stringify({\n\t\t\t\tmodel: this.model,\n\t\t\t\ttext,\n\t\t\t\t...options,\n\t\t\t}),\n\t\t});\n\n\t\tif (!response.ok) {\n\t\t\tconst errorText = await response.text();\n\t\t\tthrow new Error(\n\t\t\t\t`Workers AI TTS gateway request failed (${response.status}): ${errorText}`,\n\t\t\t);\n\t\t}\n\n\t\tconst buffer = await response.arrayBuffer();\n\t\treturn this.wrapAudioResult(new Uint8Array(buffer), format);\n\t}\n\n\t/**\n\t * Normalize binding results. Workers AI TTS can return:\n\t * - Uint8Array / ArrayBuffer (raw audio bytes)\n\t * - ReadableStream<Uint8Array> (streamed audio bytes)\n\t * - { audio: \"base64...\" } (JSON wrapper)\n\t */\n\tprivate async normalizeResult(result: unknown, format: string | undefined): Promise<TTSResult> {\n\t\t// Use the shared binaryToBase64 helper for Uint8Array/ArrayBuffer/ReadableStream\n\t\t// and { audio: \"base64...\" } JSON wrapper\n\t\tconst b64 = await binaryToBase64(result, \"audio\");\n\t\treturn {\n\t\t\tid: this.generateId(),\n\t\t\tmodel: this.model,\n\t\t\taudio: b64,\n\t\t\tformat: format ?? \"mp3\",\n\t\t\tcontentType: `audio/${format ?? \"mp3\"}`,\n\t\t};\n\t}\n\n\tprivate wrapAudioResult(bytes: Uint8Array, format: string | undefined): TTSResult {\n\t\treturn {\n\t\t\tid: this.generateId(),\n\t\t\tmodel: this.model,\n\t\t\taudio: uint8ArrayToBase64(bytes),\n\t\t\tformat: format ?? \"mp3\",\n\t\t\tcontentType: `audio/${format ?? \"mp3\"}`,\n\t\t};\n\t}\n}\n\n// ---------------------------------------------------------------------------\n// Factory function\n// ---------------------------------------------------------------------------\n\n/**\n * Creates a Workers AI text-to-speech adapter.\n *\n * Works with TanStack AI's `generateSpeech()` activity function:\n * ```ts\n * import { generateSpeech } from \"@tanstack/ai\";\n * import { createWorkersAiTts } from \"@cloudflare/tanstack-ai\";\n *\n * const adapter = createWorkersAiTts(\n *   \"@cf/deepgram/aura-1\",\n *   { binding: env.AI },\n * );\n *\n * const result = await generateSpeech({ adapter, text: \"Hello world\" });\n * // result.audio — base64-encoded audio\n * ```\n *\n * Note: Factory takes `(model, config)` for ergonomics — the class constructor\n * uses `(config, model)` to match TanStack AI's upstream convention.\n */\nexport function createWorkersAiTts(model: WorkersAiTTSModel, config: WorkersAiAdapterConfig) {\n\treturn new WorkersAiTTSAdapter(config, model);\n}\n"],"mappings":";;;;;;;;AAoCA,IAAa,sBAAb,cAAyCA,sBAAAA,eAAkC;CAI1E,YAAY,QAAgC,OAA0B;AACrE,QAAM,EAAE,EAAE,MAAM;+CAJR,QAAO,iBAA0B;+CAClC,iBAAA,KAAA,EAAsC;AAI7C,yBAAA,wBAAwB,OAAO;AAC/B,OAAK,gBAAgB;;CAGtB,MAAM,eAAe,SAAyC;EAC7D,MAAM,EAAE,MAAM,OAAO,QAAQ,OAAO,iBAAiB;EAGrD,MAAM,QAAiC,EAAE,GAAG,cAAc;AAC1D,MAAI,MAAO,OAAM,QAAQ;AACzB,MAAI,SAAS,KAAM,OAAM,QAAQ;AAEjC,MAAIC,uBAAAA,sBAAsB,KAAK,cAAc,CAC5C,QAAO,KAAK,mBAAmB,MAAM,QAAQ,MAAM;AAGpD,MAAIC,uBAAAA,0BAA0B,KAAK,cAAc,CAChD,QAAO,KAAK,gBAAgB,MAAM,QAAQ,MAAM;AAGjD,SAAO,KAAK,mBAAmB,MAAM,QAAQ,MAAM;;CAGpD,MAAc,mBACb,MACA,QACA,SACqB;EAErB,MAAM,SAAS,MADH,KAAK,cAA+C,QACxC,IAAI,KAAK,OAAO;GAAE;GAAM,GAAG;GAAS,CAAC;AAE7D,SAAO,KAAK,gBAAgB,QAAQ,OAAO;;CAG5C,MAAc,gBACb,MACA,QACA,SACqB;EACrB,MAAM,SAAS,KAAK;EASpB,MAAM,SAAS,OARE,MAAMC,wBAAAA,mBACtB,QACA,KAAK,OACL;GAAE;GAAM,GAAG;GAAS,EACpB;GAAE,OAAO;GAAkB,QAAS,QAAqC;GAAQ,CACjF,EAG6B,aAAa;AAC3C,SAAO,KAAK,gBAAgB,IAAI,WAAW,OAAO,EAAE,OAAO;;CAG5D,MAAc,mBACb,MACA,QACA,SACqB;EACrB,MAAM,gBAAgB,KAAK;EAM3B,MAAM,WAAW,MALIC,uBAAAA,mBAAmB,cAAc,cAAc,CAKhC,8CAA8C;GACjF,QAAQ;GACR,MAAM,KAAK,UAAU;IACpB,OAAO,KAAK;IACZ;IACA,GAAG;IACH,CAAC;GACF,CAAC;AAEF,MAAI,CAAC,SAAS,IAAI;GACjB,MAAM,YAAY,MAAM,SAAS,MAAM;AACvC,SAAM,IAAI,MACT,0CAA0C,SAAS,OAAO,KAAK,YAC/D;;EAGF,MAAM,SAAS,MAAM,SAAS,aAAa;AAC3C,SAAO,KAAK,gBAAgB,IAAI,WAAW,OAAO,EAAE,OAAO;;;;;;;;CAS5D,MAAc,gBAAgB,QAAiB,QAAgD;EAG9F,MAAM,MAAM,MAAMC,eAAAA,eAAe,QAAQ,QAAQ;AACjD,SAAO;GACN,IAAI,KAAK,YAAY;GACrB,OAAO,KAAK;GACZ,OAAO;GACP,QAAQ,UAAU;GAClB,aAAa,SAAS,UAAU;GAChC;;CAGF,gBAAwB,OAAmB,QAAuC;AACjF,SAAO;GACN,IAAI,KAAK,YAAY;GACrB,OAAO,KAAK;GACZ,OAAOC,eAAAA,mBAAmB,MAAM;GAChC,QAAQ,UAAU;GAClB,aAAa,SAAS,UAAU;GAChC;;;;;;;;;;;;;;;;;;;;;;;AA4BH,SAAgB,mBAAmB,OAA0B,QAAgC;AAC5F,QAAO,IAAI,oBAAoB,QAAQ,MAAM"}

package/dist/adapters/workers-ai-tts.d.cts

@@ -1,3 +1,54 @@
-export { createWorkersAiTts_alias_1 as createWorkersAiTts } from '../_tsup-dts-rollup.cjs';
-export { WorkersAiTTSModel_alias_1 as WorkersAiTTSModel } from '../_tsup-dts-rollup.cjs';
-export { WorkersAiTTSAdapter } from '../_tsup-dts-rollup.cjs';
+import { i as WorkersAiAdapterConfig } from "../create-fetcher-vAQ8WW-p.cjs";
+import { TTSOptions, TTSResult } from "@tanstack/ai";
+import { BaseTTSAdapter } from "@tanstack/ai/adapters";
+
+//#region src/adapters/workers-ai-tts.d.ts
+/**
+ * Workers AI models that support text-to-speech generation.
+ *
+ * Note: the typed `AiModels` interface in `@cloudflare/workers-types` may lag
+ * behind what's deployed. We use a string union here that matches the known
+ * models including Deepgram partner models.
+ */
+type WorkersAiTTSModel = "@cf/deepgram/aura-1" | "@cf/deepgram/aura-2-en" | "@cf/deepgram/aura-2-es" | (string & {});
+declare class WorkersAiTTSAdapter extends BaseTTSAdapter<WorkersAiTTSModel> {
+  readonly name: "workers-ai-tts";
+  private adapterConfig;
+  constructor(config: WorkersAiAdapterConfig, model: WorkersAiTTSModel);
+  generateSpeech(options: TTSOptions): Promise<TTSResult>;
+  private generateViaBinding;
+  private generateViaRest;
+  private generateViaGateway;
+  /**
+   * Normalize binding results. Workers AI TTS can return:
+   * - Uint8Array / ArrayBuffer (raw audio bytes)
+   * - ReadableStream<Uint8Array> (streamed audio bytes)
+   * - { audio: "base64..." } (JSON wrapper)
+   */
+  private normalizeResult;
+  private wrapAudioResult;
+}
+/**
+ * Creates a Workers AI text-to-speech adapter.
+ *
+ * Works with TanStack AI's `generateSpeech()` activity function:
+ * ```ts
+ * import { generateSpeech } from "@tanstack/ai";
+ * import { createWorkersAiTts } from "@cloudflare/tanstack-ai";
+ *
+ * const adapter = createWorkersAiTts(
+ *   "@cf/deepgram/aura-1",
+ *   { binding: env.AI },
+ * );
+ *
+ * const result = await generateSpeech({ adapter, text: "Hello world" });
+ * // result.audio — base64-encoded audio
+ * ```
+ *
+ * Note: Factory takes `(model, config)` for ergonomics — the class constructor
+ * uses `(config, model)` to match TanStack AI's upstream convention.
+ */
+declare function createWorkersAiTts(model: WorkersAiTTSModel, config: WorkersAiAdapterConfig): WorkersAiTTSAdapter;
+//#endregion
+export { WorkersAiTTSAdapter, WorkersAiTTSModel, createWorkersAiTts };
+//# sourceMappingURL=workers-ai-tts.d.cts.map

package/dist/adapters/workers-ai-tts.d.mts

@@ -0,0 +1,54 @@
+import { i as WorkersAiAdapterConfig } from "../create-fetcher-6p6heb85.mjs";
+import { BaseTTSAdapter } from "@tanstack/ai/adapters";
+import { TTSOptions, TTSResult } from "@tanstack/ai";
+
+//#region src/adapters/workers-ai-tts.d.ts
+/**
+ * Workers AI models that support text-to-speech generation.
+ *
+ * Note: the typed `AiModels` interface in `@cloudflare/workers-types` may lag
+ * behind what's deployed. We use a string union here that matches the known
+ * models including Deepgram partner models.
+ */
+type WorkersAiTTSModel = "@cf/deepgram/aura-1" | "@cf/deepgram/aura-2-en" | "@cf/deepgram/aura-2-es" | (string & {});
+declare class WorkersAiTTSAdapter extends BaseTTSAdapter<WorkersAiTTSModel> {
+  readonly name: "workers-ai-tts";
+  private adapterConfig;
+  constructor(config: WorkersAiAdapterConfig, model: WorkersAiTTSModel);
+  generateSpeech(options: TTSOptions): Promise<TTSResult>;
+  private generateViaBinding;
+  private generateViaRest;
+  private generateViaGateway;
+  /**
+   * Normalize binding results. Workers AI TTS can return:
+   * - Uint8Array / ArrayBuffer (raw audio bytes)
+   * - ReadableStream<Uint8Array> (streamed audio bytes)
+   * - { audio: "base64..." } (JSON wrapper)
+   */
+  private normalizeResult;
+  private wrapAudioResult;
+}
+/**
+ * Creates a Workers AI text-to-speech adapter.
+ *
+ * Works with TanStack AI's `generateSpeech()` activity function:
+ * ```ts
+ * import { generateSpeech } from "@tanstack/ai";
+ * import { createWorkersAiTts } from "@cloudflare/tanstack-ai";
+ *
+ * const adapter = createWorkersAiTts(
+ *   "@cf/deepgram/aura-1",
+ *   { binding: env.AI },
+ * );
+ *
+ * const result = await generateSpeech({ adapter, text: "Hello world" });
+ * // result.audio — base64-encoded audio
+ * ```
+ *
+ * Note: Factory takes `(model, config)` for ergonomics — the class constructor
+ * uses `(config, model)` to match TanStack AI's upstream convention.
+ */
+declare function createWorkersAiTts(model: WorkersAiTTSModel, config: WorkersAiAdapterConfig): WorkersAiTTSAdapter;
+//#endregion
+export { WorkersAiTTSAdapter, WorkersAiTTSModel, createWorkersAiTts };
+//# sourceMappingURL=workers-ai-tts.d.mts.map

package/dist/adapters/workers-ai-tts.mjs

@@ -0,0 +1,111 @@
+import { a as validateWorkersAiConfig, i as isDirectCredentialsConfig, r as isDirectBindingConfig, t as createGatewayFetch } from "../create-fetcher-DY7wfYYy.mjs";
+import { t as _defineProperty } from "../defineProperty-CbyrzcbA.mjs";
+import { t as workersAiRestFetch } from "../workers-ai-rest-GKy2r7eG.mjs";
+import { n as uint8ArrayToBase64, t as binaryToBase64 } from "../binary-p4H_N_3M.mjs";
+import { BaseTTSAdapter } from "@tanstack/ai/adapters";
+//#region src/adapters/workers-ai-tts.ts
+var WorkersAiTTSAdapter = class extends BaseTTSAdapter {
+	constructor(config, model) {
+		super({}, model);
+		_defineProperty(this, "name", "workers-ai-tts");
+		_defineProperty(this, "adapterConfig", void 0);
+		validateWorkersAiConfig(config);
+		this.adapterConfig = config;
+	}
+	async generateSpeech(options) {
+		const { text, voice, format, speed, modelOptions } = options;
+		const extra = { ...modelOptions };
+		if (voice) extra.voice = voice;
+		if (speed != null) extra.speed = speed;
+		if (isDirectBindingConfig(this.adapterConfig)) return this.generateViaBinding(text, format, extra);
+		if (isDirectCredentialsConfig(this.adapterConfig)) return this.generateViaRest(text, format, extra);
+		return this.generateViaGateway(text, format, extra);
+	}
+	async generateViaBinding(text, format, options) {
+		const result = await this.adapterConfig.binding.run(this.model, {
+			text,
+			...options
+		});
+		return this.normalizeResult(result, format);
+	}
+	async generateViaRest(text, format, options) {
+		const config = this.adapterConfig;
+		const buffer = await (await workersAiRestFetch(config, this.model, {
+			text,
+			...options
+		}, {
+			label: "Workers AI TTS",
+			signal: options.signal
+		})).arrayBuffer();
+		return this.wrapAudioResult(new Uint8Array(buffer), format);
+	}
+	async generateViaGateway(text, format, options) {
+		const gatewayConfig = this.adapterConfig;
+		const response = await createGatewayFetch("workers-ai", gatewayConfig)("https://api.cloudflare.com/v1/audio/speech", {
+			method: "POST",
+			body: JSON.stringify({
+				model: this.model,
+				text,
+				...options
+			})
+		});
+		if (!response.ok) {
+			const errorText = await response.text();
+			throw new Error(`Workers AI TTS gateway request failed (${response.status}): ${errorText}`);
+		}
+		const buffer = await response.arrayBuffer();
+		return this.wrapAudioResult(new Uint8Array(buffer), format);
+	}
+	/**
+	* Normalize binding results. Workers AI TTS can return:
+	* - Uint8Array / ArrayBuffer (raw audio bytes)
+	* - ReadableStream<Uint8Array> (streamed audio bytes)
+	* - { audio: "base64..." } (JSON wrapper)
+	*/
+	async normalizeResult(result, format) {
+		const b64 = await binaryToBase64(result, "audio");
+		return {
+			id: this.generateId(),
+			model: this.model,
+			audio: b64,
+			format: format ?? "mp3",
+			contentType: `audio/${format ?? "mp3"}`
+		};
+	}
+	wrapAudioResult(bytes, format) {
+		return {
+			id: this.generateId(),
+			model: this.model,
+			audio: uint8ArrayToBase64(bytes),
+			format: format ?? "mp3",
+			contentType: `audio/${format ?? "mp3"}`
+		};
+	}
+};
+/**
+* Creates a Workers AI text-to-speech adapter.
+*
+* Works with TanStack AI's `generateSpeech()` activity function:
+* ```ts
+* import { generateSpeech } from "@tanstack/ai";
+* import { createWorkersAiTts } from "@cloudflare/tanstack-ai";
+*
+* const adapter = createWorkersAiTts(
+*   "@cf/deepgram/aura-1",
+*   { binding: env.AI },
+* );
+*
+* const result = await generateSpeech({ adapter, text: "Hello world" });
+* // result.audio — base64-encoded audio
+* ```
+*
+* Note: Factory takes `(model, config)` for ergonomics — the class constructor
+* uses `(config, model)` to match TanStack AI's upstream convention.
+*/
+function createWorkersAiTts(model, config) {
+	return new WorkersAiTTSAdapter(config, model);
+}
+//#endregion
+export { WorkersAiTTSAdapter, createWorkersAiTts };
+
+//# sourceMappingURL=workers-ai-tts.mjs.map

package/dist/adapters/workers-ai-tts.mjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"workers-ai-tts.mjs","names":[],"sources":["../../src/adapters/workers-ai-tts.ts"],"sourcesContent":["import { BaseTTSAdapter } from \"@tanstack/ai/adapters\";\nimport type { TTSOptions, TTSResult } from \"@tanstack/ai\";\nimport {\n\ttype WorkersAiAdapterConfig,\n\ttype WorkersAiDirectBindingConfig,\n\ttype WorkersAiDirectCredentialsConfig,\n\ttype AiGatewayAdapterConfig,\n\tcreateGatewayFetch,\n\tisDirectBindingConfig,\n\tisDirectCredentialsConfig,\n\tvalidateWorkersAiConfig,\n} from \"../utils/create-fetcher\";\nimport { workersAiRestFetch } from \"../utils/workers-ai-rest\";\nimport { binaryToBase64, uint8ArrayToBase64 } from \"../utils/binary\";\n\n// ---------------------------------------------------------------------------\n// Model types\n// ---------------------------------------------------------------------------\n\n/**\n * Workers AI models that support text-to-speech generation.\n *\n * Note: the typed `AiModels` interface in `@cloudflare/workers-types` may lag\n * behind what's deployed. We use a string union here that matches the known\n * models including Deepgram partner models.\n */\nexport type WorkersAiTTSModel =\n\t| \"@cf/deepgram/aura-1\"\n\t| \"@cf/deepgram/aura-2-en\"\n\t| \"@cf/deepgram/aura-2-es\"\n\t| (string & {});\n\n// ---------------------------------------------------------------------------\n// WorkersAiTTSAdapter\n// ---------------------------------------------------------------------------\n\nexport class WorkersAiTTSAdapter extends BaseTTSAdapter<WorkersAiTTSModel> {\n\treadonly name = \"workers-ai-tts\" as const;\n\tprivate adapterConfig: WorkersAiAdapterConfig;\n\n\tconstructor(config: WorkersAiAdapterConfig, model: WorkersAiTTSModel) {\n\t\tsuper({}, model);\n\t\tvalidateWorkersAiConfig(config);\n\t\tthis.adapterConfig = config;\n\t}\n\n\tasync generateSpeech(options: TTSOptions): Promise<TTSResult> {\n\t\tconst { text, voice, format, speed, modelOptions } = options;\n\n\t\t// Workers AI TTS models (Deepgram aura-1) accept { text, lang? }\n\t\tconst extra: Record<string, unknown> = { ...modelOptions };\n\t\tif (voice) extra.voice = voice;\n\t\tif (speed != null) extra.speed = speed;\n\n\t\tif (isDirectBindingConfig(this.adapterConfig)) {\n\t\t\treturn this.generateViaBinding(text, format, extra);\n\t\t}\n\n\t\tif (isDirectCredentialsConfig(this.adapterConfig)) {\n\t\t\treturn this.generateViaRest(text, format, extra);\n\t\t}\n\n\t\treturn this.generateViaGateway(text, format, extra);\n\t}\n\n\tprivate async generateViaBinding(\n\t\ttext: string,\n\t\tformat: string | undefined,\n\t\toptions: Record<string, unknown>,\n\t): Promise<TTSResult> {\n\t\tconst ai = (this.adapterConfig as WorkersAiDirectBindingConfig).binding;\n\t\tconst result = await ai.run(this.model, { text, ...options });\n\n\t\treturn this.normalizeResult(result, format);\n\t}\n\n\tprivate async generateViaRest(\n\t\ttext: string,\n\t\tformat: string | undefined,\n\t\toptions: Record<string, unknown>,\n\t): Promise<TTSResult> {\n\t\tconst config = this.adapterConfig as WorkersAiDirectCredentialsConfig;\n\t\tconst response = await workersAiRestFetch(\n\t\t\tconfig,\n\t\t\tthis.model,\n\t\t\t{ text, ...options },\n\t\t\t{ label: \"Workers AI TTS\", signal: (options as { signal?: AbortSignal }).signal },\n\t\t);\n\n\t\t// Workers AI TTS returns audio bytes directly\n\t\tconst buffer = await response.arrayBuffer();\n\t\treturn this.wrapAudioResult(new Uint8Array(buffer), format);\n\t}\n\n\tprivate async generateViaGateway(\n\t\ttext: string,\n\t\tformat: string | undefined,\n\t\toptions: Record<string, unknown>,\n\t): Promise<TTSResult> {\n\t\tconst gatewayConfig = this.adapterConfig as AiGatewayAdapterConfig;\n\t\tconst gatewayFetch = createGatewayFetch(\"workers-ai\", gatewayConfig);\n\n\t\t// The URL here is a placeholder — createGatewayFetch for \"workers-ai\" extracts\n\t\t// the model from the body, sets it as the endpoint, and routes through the gateway.\n\t\t// The actual URL path is not used.\n\t\tconst response = await gatewayFetch(\"https://api.cloudflare.com/v1/audio/speech\", {\n\t\t\tmethod: \"POST\",\n\t\t\tbody: JSON.stringify({\n\t\t\t\tmodel: this.model,\n\t\t\t\ttext,\n\t\t\t\t...options,\n\t\t\t}),\n\t\t});\n\n\t\tif (!response.ok) {\n\t\t\tconst errorText = await response.text();\n\t\t\tthrow new Error(\n\t\t\t\t`Workers AI TTS gateway request failed (${response.status}): ${errorText}`,\n\t\t\t);\n\t\t}\n\n\t\tconst buffer = await response.arrayBuffer();\n\t\treturn this.wrapAudioResult(new Uint8Array(buffer), format);\n\t}\n\n\t/**\n\t * Normalize binding results. Workers AI TTS can return:\n\t * - Uint8Array / ArrayBuffer (raw audio bytes)\n\t * - ReadableStream<Uint8Array> (streamed audio bytes)\n\t * - { audio: \"base64...\" } (JSON wrapper)\n\t */\n\tprivate async normalizeResult(result: unknown, format: string | undefined): Promise<TTSResult> {\n\t\t// Use the shared binaryToBase64 helper for Uint8Array/ArrayBuffer/ReadableStream\n\t\t// and { audio: \"base64...\" } JSON wrapper\n\t\tconst b64 = await binaryToBase64(result, \"audio\");\n\t\treturn {\n\t\t\tid: this.generateId(),\n\t\t\tmodel: this.model,\n\t\t\taudio: b64,\n\t\t\tformat: format ?? \"mp3\",\n\t\t\tcontentType: `audio/${format ?? \"mp3\"}`,\n\t\t};\n\t}\n\n\tprivate wrapAudioResult(bytes: Uint8Array, format: string | undefined): TTSResult {\n\t\treturn {\n\t\t\tid: this.generateId(),\n\t\t\tmodel: this.model,\n\t\t\taudio: uint8ArrayToBase64(bytes),\n\t\t\tformat: format ?? \"mp3\",\n\t\t\tcontentType: `audio/${format ?? \"mp3\"}`,\n\t\t};\n\t}\n}\n\n// ---------------------------------------------------------------------------\n// Factory function\n// ---------------------------------------------------------------------------\n\n/**\n * Creates a Workers AI text-to-speech adapter.\n *\n * Works with TanStack AI's `generateSpeech()` activity function:\n * ```ts\n * import { generateSpeech } from \"@tanstack/ai\";\n * import { createWorkersAiTts } from \"@cloudflare/tanstack-ai\";\n *\n * const adapter = createWorkersAiTts(\n *   \"@cf/deepgram/aura-1\",\n *   { binding: env.AI },\n * );\n *\n * const result = await generateSpeech({ adapter, text: \"Hello world\" });\n * // result.audio — base64-encoded audio\n * ```\n *\n * Note: Factory takes `(model, config)` for ergonomics — the class constructor\n * uses `(config, model)` to match TanStack AI's upstream convention.\n */\nexport function createWorkersAiTts(model: WorkersAiTTSModel, config: WorkersAiAdapterConfig) {\n\treturn new WorkersAiTTSAdapter(config, model);\n}\n"],"mappings":";;;;;;AAoCA,IAAa,sBAAb,cAAyC,eAAkC;CAI1E,YAAY,QAAgC,OAA0B;AACrE,QAAM,EAAE,EAAE,MAAM;wBAJR,QAAO,iBAA0B;wBAClC,iBAAA,KAAA,EAAsC;AAI7C,0BAAwB,OAAO;AAC/B,OAAK,gBAAgB;;CAGtB,MAAM,eAAe,SAAyC;EAC7D,MAAM,EAAE,MAAM,OAAO,QAAQ,OAAO,iBAAiB;EAGrD,MAAM,QAAiC,EAAE,GAAG,cAAc;AAC1D,MAAI,MAAO,OAAM,QAAQ;AACzB,MAAI,SAAS,KAAM,OAAM,QAAQ;AAEjC,MAAI,sBAAsB,KAAK,cAAc,CAC5C,QAAO,KAAK,mBAAmB,MAAM,QAAQ,MAAM;AAGpD,MAAI,0BAA0B,KAAK,cAAc,CAChD,QAAO,KAAK,gBAAgB,MAAM,QAAQ,MAAM;AAGjD,SAAO,KAAK,mBAAmB,MAAM,QAAQ,MAAM;;CAGpD,MAAc,mBACb,MACA,QACA,SACqB;EAErB,MAAM,SAAS,MADH,KAAK,cAA+C,QACxC,IAAI,KAAK,OAAO;GAAE;GAAM,GAAG;GAAS,CAAC;AAE7D,SAAO,KAAK,gBAAgB,QAAQ,OAAO;;CAG5C,MAAc,gBACb,MACA,QACA,SACqB;EACrB,MAAM,SAAS,KAAK;EASpB,MAAM,SAAS,OARE,MAAM,mBACtB,QACA,KAAK,OACL;GAAE;GAAM,GAAG;GAAS,EACpB;GAAE,OAAO;GAAkB,QAAS,QAAqC;GAAQ,CACjF,EAG6B,aAAa;AAC3C,SAAO,KAAK,gBAAgB,IAAI,WAAW,OAAO,EAAE,OAAO;;CAG5D,MAAc,mBACb,MACA,QACA,SACqB;EACrB,MAAM,gBAAgB,KAAK;EAM3B,MAAM,WAAW,MALI,mBAAmB,cAAc,cAAc,CAKhC,8CAA8C;GACjF,QAAQ;GACR,MAAM,KAAK,UAAU;IACpB,OAAO,KAAK;IACZ;IACA,GAAG;IACH,CAAC;GACF,CAAC;AAEF,MAAI,CAAC,SAAS,IAAI;GACjB,MAAM,YAAY,MAAM,SAAS,MAAM;AACvC,SAAM,IAAI,MACT,0CAA0C,SAAS,OAAO,KAAK,YAC/D;;EAGF,MAAM,SAAS,MAAM,SAAS,aAAa;AAC3C,SAAO,KAAK,gBAAgB,IAAI,WAAW,OAAO,EAAE,OAAO;;;;;;;;CAS5D,MAAc,gBAAgB,QAAiB,QAAgD;EAG9F,MAAM,MAAM,MAAM,eAAe,QAAQ,QAAQ;AACjD,SAAO;GACN,IAAI,KAAK,YAAY;GACrB,OAAO,KAAK;GACZ,OAAO;GACP,QAAQ,UAAU;GAClB,aAAa,SAAS,UAAU;GAChC;;CAGF,gBAAwB,OAAmB,QAAuC;AACjF,SAAO;GACN,IAAI,KAAK,YAAY;GACrB,OAAO,KAAK;GACZ,OAAO,mBAAmB,MAAM;GAChC,QAAQ,UAAU;GAClB,aAAa,SAAS,UAAU;GAChC;;;;;;;;;;;;;;;;;;;;;;;AA4BH,SAAgB,mBAAmB,OAA0B,QAAgC;AAC5F,QAAO,IAAI,oBAAoB,QAAQ,MAAM"}

package/dist/adapters/workers-ai.cjs

@@ -1,11 +1,4 @@
-"use strict";Object.defineProperty(exports, "__esModule", {value: true});
-
-
-var _chunkOUEAAP25cjs = require('../chunk-OUEAAP25.cjs');
-require('../chunk-WBTQKWFN.cjs');
-require('../chunk-4DE2IREA.cjs');
-
-
-
-exports.WorkersAiTextAdapter = _chunkOUEAAP25cjs.WorkersAiTextAdapter; exports.createWorkersAiChat = _chunkOUEAAP25cjs.createWorkersAiChat;
-//# sourceMappingURL=workers-ai.cjs.map
+Object.defineProperty(exports, Symbol.toStringTag, { value: "Module" });
+const require_workers_ai = require("../workers-ai-Bm7Up4or.cjs");
+exports.WorkersAiTextAdapter = require_workers_ai.WorkersAiTextAdapter;
+exports.createWorkersAiChat = require_workers_ai.createWorkersAiChat;