@agentor/dashscope 0.0.0 → 0.0.2

package/README.md

@@ -10,6 +10,12 @@
 
 - **Chat Completions API** - Standard `/chat/completions` with function calling, streaming, and reasoning
 - **Responses API** - `/responses` endpoint with built-in tools support
+- **Embedding** - Text vectorization via OpenAI-compatible `/embeddings` endpoint
+- **Reranking** - Document reranking via `/reranks` endpoint
+- **Image Generation** - Text-to-image via multimodal generation endpoint
+- **Video Generation** - Text-to-video and image-to-video with async polling
+- **Speech Synthesis** - Text-to-speech for CosyVoice and Qwen-TTS models
+- **Transcription** - Speech-to-text for short and long audio
 - **Built-in Tools** - Web search, code interpreter, web extractor, file search, image search, MCP integration
 - **Thinking Mode** - Enable reasoning/thinking with configurable budget
 - **Multi-region** - Beijing, Singapore, US, Germany regions
@@ -245,20 +251,277 @@ const first = await generateText({
 });
 ```
 
+## Embedding
+
+```typescript
+import { embed, embedMany } from "ai";
+
+// Single text embedding
+const { embedding, usage } = await embed({
+  model: dashscope.embeddingModel("text-embedding-v4"),
+  value: "The clothes quality is excellent",
+});
+
+console.log(embedding.length); // 1024 (default dimensions)
+
+// Batch embedding
+const { embeddings } = await embedMany({
+  model: dashscope.embeddingModel("text-embedding-v4"),
+  values: ["Hello world", "Machine learning is fascinating"],
+});
+```
+
+### Custom Dimensions
+
+```typescript
+const { embedding } = await embed({
+  model: dashscope.embeddingModel("text-embedding-v4"),
+  value: "Custom dimension embedding",
+  providerOptions: {
+    openaiCompatible: {
+      dimensions: 256,
+    },
+  },
+});
+
+console.log(embedding.length); // 256
+```
+
+## Reranking
+
+```typescript
+import { rerank } from "ai";
+
+const { ranking } = await rerank({
+  model: dashscope.rerankingModel("qwen3-rerank"),
+  query: "What is a reranking model?",
+  documents: [
+    "Reranking models sort candidate texts by relevance",
+    "Quantum computing is a frontier field",
+    "Pre-trained models brought advances to reranking",
+  ],
+});
+
+for (const item of ranking) {
+  console.log(`Index: ${item.originalIndex}, Score: ${item.score}`);
+}
+```
+
+### Top N Results
+
+```typescript
+const { ranking } = await rerank({
+  model: dashscope.rerankingModel("qwen3-rerank"),
+  query: "How to reset password?",
+  documents: [
+    "Go to Settings > Security > Change Password",
+    "Forgot your password?",
+    "Two-factor authentication is supported",
+  ],
+  topN: 2,
+});
+```
+
+## Image Generation
+
+```typescript
+import { generateImage } from "ai";
+
+const { images } = await generateImage({
+  model: dashscope.imageModel("qwen-image-plus"),
+  prompt: "A cute cat sitting on a windowsill with sunlight streaming in",
+  providerOptions: {
+    dashscope: {
+      size: "1024*1024",
+    },
+  },
+});
+
+// images[0].uint8Array — raw image data
+// images[0].base64 — base64 encoded image
+```
+
+## Video Generation
+
+```typescript
+import { experimental_generateVideo as generateVideo } from "ai";
+
+// Text-to-video
+const { videos } = await generateVideo({
+  model: dashscope.videoModel("wan2.6-t2v"),
+  prompt: "A golden retriever running through a field of sunflowers",
+  providerOptions: {
+    dashscope: {
+      size: "1280*720",
+      duration: 5,
+    },
+  },
+});
+```
+
+### Image-to-Video
+
+Use a model ID containing `-i2v` for image-to-video mode:
+
+```typescript
+const { videos } = await generateVideo({
+  model: dashscope.videoModel("wan2.6-i2v-turbo"),
+  prompt: "The cat stretches and walks away",
+  providerOptions: {
+    dashscope: {
+      resolution: "720P",
+    },
+  },
+  image: "data:image/png;base64,...", // or a URL string
+});
+```
+
+## Speech Synthesis (TTS)
+
+```typescript
+import { experimental_generateSpeech as generateSpeech } from "ai";
+import { writeFileSync } from "fs";
+
+const { audio } = await generateSpeech({
+  model: dashscope.speechModel("cosyvoice-v3-flash"),
+  text: "Hello, welcome to Agentor.",
+  providerOptions: {
+    dashscope: {
+      voice: "longanyang",
+      format: "wav",
+      sampleRate: 24000,
+    },
+  },
+});
+
+writeFileSync("output.wav", audio.uint8Array);
+```
+
+## Transcription (Speech-to-Text)
+
+### Short Audio (Sync)
+
+```typescript
+import { experimental_transcribe as transcribe } from "ai";
+
+const { text } = await transcribe({
+  model: dashscope.transcriptionModel("qwen3-asr-flash"),
+  audio: new URL("https://example.com/audio.mp3"),
+});
+
+console.log(text);
+```
+
+### Long Audio (Async)
+
+For async models, provide the audio URL via `providerOptions`:
+
+```typescript
+const { text, segments } = await transcribe({
+  model: dashscope.transcriptionModel("qwen3-asr-flash-filetrans"),
+  audio: new Uint8Array(0), // placeholder
+  providerOptions: {
+    dashscope: {
+      fileUrl: "https://example.com/long-audio.mp3",
+      enableWords: true,
+    },
+  },
+});
+```
+
 ## Provider Configuration
 
 ```typescript
 import { createDashScope } from "@agentor/dashscope";
 
 const dashscope = createDashScope({
-  apiKey: "sk-xxx",                       // or set DASHSCOPE_API_KEY env var
-  region: "beijing",                       // beijing | singapore | us | germany
-  workspaceId: "ws-xxx",                   // required for germany region
-  baseURL: "https://custom-endpoint.com",  // override default base URL
+  apiKey: "sk-xxx", // or set DASHSCOPE_API_KEY env var
+  region: "beijing", // beijing | singapore | us | germany
+  workspaceId: "ws-xxx", // required for germany region
+  baseURL: "https://custom-endpoint.com", // override default base URL
   headers: { "X-Custom-Header": "value" }, // custom headers
 });
 ```
 
+## Available Models
+
+> For the complete and up-to-date model list, see [Alibaba Cloud Model Studio](https://help.aliyun.com/zh/model-studio/models).
+
+### Language Models (Chat)
+
+| Model                 | Description                               |
+| --------------------- | ----------------------------------------- |
+| `qwen3.6-max-preview` | Flagship model with strongest reasoning   |
+| `qwen3.6-plus`        | Recommended, balanced capability and cost |
+| `qwen3.6-flash`       | Fastest, ultra-low cost                   |
+| `qwen3.5-plus`        | Enhanced reasoning model                  |
+| `qwen3.5-flash`       | Fast and efficient model                  |
+| `qwen3-coder-plus`    | Code-optimized model                      |
+| `qwen3-coder-flash`   | Fast code model                           |
+| `qwq-plus`            | Dedicated reasoning model                 |
+| `deepseek-v4-pro`     | DeepSeek V4 Pro                           |
+| `deepseek-v4-flash`   | DeepSeek V4 Flash                         |
+| `kimi-k2.6`           | Moonshot Kimi K2.6                        |
+| `glm-5.1`             | Zhipu GLM 5.1                             |
+
+### Embedding Models
+
+| Model                          | Dimensions              | Description                         |
+| ------------------------------ | ----------------------- | ----------------------------------- |
+| `text-embedding-v4`            | 64-2048 (default 1024)  | Text embedding for search/RAG       |
+| `text-embedding-v3`            | 512-1024 (default 1024) | Legacy text embedding               |
+| `qwen3-vl-embedding`           | 256-2560 (default 2560) | Multimodal (text + image) embedding |
+| `tongyi-embedding-vision-plus` | 64-1152 (default 1152)  | Cross-modal search embedding        |
+
+### Reranking Models
+
+| Model             | Description                             |
+| ----------------- | --------------------------------------- |
+| `qwen3-rerank`    | Text reranking, 100+ languages          |
+| `qwen3-vl-rerank` | Multimodal reranking (text/image/video) |
+| `gte-rerank-v2`   | Semantic text reranking                 |
+
+### Image Models
+
+| Model                | Description                                  |
+| -------------------- | -------------------------------------------- |
+| `wan2.7-image-pro`   | Latest Wan image generation, up to 4096x4096 |
+| `wan2.7-image`       | Wan image generation, up to 2048x2048        |
+| `qwen-image-2.0-pro` | Qwen image generation and editing            |
+| `qwen-image-max`     | High quality image generation                |
+| `qwen-image-plus`    | Enhanced image generation                    |
+| `z-image-turbo`      | Fast image generation                        |
+
+### Video Models
+
+| Model              | Mode | Description                           |
+| ------------------ | ---- | ------------------------------------- |
+| `wan2.7-t2v`       | T2V  | Recommended text-to-video with audio  |
+| `wan2.6-t2v`       | T2V  | Text-to-video with audio              |
+| `wan2.2-t2v-plus`  | T2V  | Text-to-video (silent)                |
+| `wan2.7-i2v`       | I2V  | Recommended image-to-video with audio |
+| `wan2.6-i2v`       | I2V  | Image-to-video with audio             |
+| `wan2.6-i2v-flash` | I2V  | Fast image-to-video                   |
+
+### Speech Models (TTS)
+
+| Model                      | Description                        |
+| -------------------------- | ---------------------------------- |
+| `cosyvoice-v3.5-plus`      | Latest flagship, best quality      |
+| `cosyvoice-v3.5-flash`     | Latest lightweight                 |
+| `cosyvoice-v3-plus`        | V3 enhanced                        |
+| `cosyvoice-v3-flash`       | V3 fast synthesis                  |
+| `qwen3-tts-flash-realtime` | Qwen TTS with 17 human-like voices |
+
+### Transcription Models (STT)
+
+| Model                       | Mode  | Description                    |
+| --------------------------- | ----- | ------------------------------ |
+| `qwen3-asr-flash`           | Sync  | Short audio (up to 5 min)      |
+| `qwen3-asr-flash-filetrans` | Async | Long audio (up to 12 hours)    |
+| `fun-asr`                   | Async | Speaker diarization, hot words |
+| `paraformer-v2`             | Async | Legacy async transcription     |
+
 ## License
 
 MIT © [Demo Macro](https://www.demomacro.com/)

package/dist/index.d.mts

@@ -1,6 +1,7 @@
+import { OpenAICompatibleEmbeddingModel } from "@ai-sdk/openai-compatible";
 import * as _$_ai_sdk_provider_utils0 from "@ai-sdk/provider-utils";
 import { FetchFunction } from "@ai-sdk/provider-utils";
-import { LanguageModelV3 } from "@ai-sdk/provider";
+import { EmbeddingModelV3, Experimental_VideoModelV3, Experimental_VideoModelV3CallOptions, ImageModelV3, ImageModelV3CallOptions, LanguageModelV3, RerankingModelV3, RerankingModelV3CallOptions, SharedV3Warning, SpeechModelV3, SpeechModelV3CallOptions, TranscriptionModelV3, TranscriptionModelV3CallOptions } from "@ai-sdk/provider";
 
 //#region src/tools.d.ts
 declare const webSearchToolFactory: _$_ai_sdk_provider_utils0.ProviderToolFactoryWithOutputSchema<Record<string, never>, {
@@ -131,16 +132,12 @@ type DashScopeResponsesTools = typeof responsesTools;
 //#endregion
 //#region src/types.d.ts
 type DashScopeRegion = "beijing" | "singapore" | "us" | "germany";
-declare const DASHSCOPE_REGION_BASE_URLS: Record<DashScopeRegion, {
-  baseURL: string;
-  videoBaseURL: string;
-}>;
+declare const DASHSCOPE_REGION_URLS: Record<DashScopeRegion, string>;
 interface DashScopeProviderSettings {
   apiKey?: string;
   region?: DashScopeRegion;
   workspaceId?: string;
   baseURL?: string;
-  videoBaseURL?: string;
   headers?: Record<string, string>;
   fetch?: FetchFunction;
   includeUsage?: boolean;
@@ -164,13 +161,6 @@ interface DashScopeChatOptions {
   /** Enable code interpreter (requires enableThinking). */
   enableCodeInterpreter?: boolean;
 }
-interface DashScopeChatConfig {
-  provider: string;
-  baseURL: string;
-  headers: () => Record<string, string>;
-  fetch?: FetchFunction;
-  includeUsage?: boolean;
-}
 interface DashScopeResponsesOptions {
   enableThinking?: boolean;
   reasoning?: {
@@ -188,6 +178,12 @@ interface DashScopeResponsesNamespace {
 interface DashScopeProvider {
   (modelId: string): LanguageModelV3;
   languageModel(modelId: string): LanguageModelV3;
+  embeddingModel(modelId: string): EmbeddingModelV3;
+  rerankingModel(modelId: string): RerankingModelV3;
+  imageModel(modelId: string): ImageModelV3;
+  videoModel(modelId: string): Experimental_VideoModelV3;
+  speechModel(modelId: string): SpeechModelV3;
+  transcriptionModel(modelId: string): TranscriptionModelV3;
   chatOptions: (options: DashScopeChatOptions) => {
     providerOptions: {
       dashscope: DashScopeChatOptions;
@@ -201,10 +197,208 @@ interface DashScopeProvider {
   responses: DashScopeResponsesNamespace;
 }
 //#endregion
+//#region src/utils.d.ts
+interface DashScopeConfig {
+  provider: string;
+  baseURL: string;
+  headers: () => Record<string, string>;
+  fetch?: FetchFunction;
+  includeUsage?: boolean;
+}
+//#endregion
+//#region src/embedding.d.ts
+interface DashScopeEmbeddingOptions {
+  /** Output embedding dimensions. Supported by text-embedding-v4, text-embedding-v3, etc. */
+  dimensions?: number;
+}
+declare class DashScopeEmbeddingModel extends OpenAICompatibleEmbeddingModel {
+  constructor(modelId: string, config: DashScopeConfig);
+}
+//#endregion
+//#region src/image.d.ts
+interface DashScopeImageOptions {
+  /** Output image size, e.g. "2048*2048", "1024*1024", "1K", "2K". */
+  size?: string;
+  /** Negative prompt describing what to avoid. */
+  negativePrompt?: string;
+  /** Enable prompt extension/rewriting. Default depends on model. */
+  promptExtend?: boolean;
+  /** Add watermark. Default false. */
+  watermark?: boolean;
+  /** Number of images to generate. Default 1. */
+  n?: number;
+}
+declare class DashScopeImageModel implements ImageModelV3 {
+  readonly specificationVersion: "v3";
+  readonly modelId: string;
+  private readonly config;
+  constructor(modelId: string, config: DashScopeConfig);
+  get provider(): string;
+  get maxImagesPerCall(): number | undefined;
+  doGenerate(options: ImageModelV3CallOptions): Promise<{
+    images: string[];
+    warnings: SharedV3Warning[];
+    response: {
+      timestamp: Date;
+      modelId: string;
+      headers: Record<string, string> | undefined;
+    };
+  }>;
+}
+//#endregion
+//#region src/rerank.d.ts
+interface DashScopeRerankOptions {
+  /** English instruction to guide the reranking strategy. */
+  instruct?: string;
+}
+declare class DashScopeRerankingModel implements RerankingModelV3 {
+  readonly specificationVersion: "v3";
+  readonly modelId: string;
+  private readonly config;
+  constructor(modelId: string, config: DashScopeConfig);
+  get provider(): string;
+  doRerank(options: RerankingModelV3CallOptions): Promise<{
+    ranking: {
+      index: number;
+      relevanceScore: number;
+    }[];
+    warnings: SharedV3Warning[];
+    response: {
+      id: string | undefined;
+      modelId: string | undefined;
+      headers: Record<string, string> | undefined;
+    };
+  }>;
+}
+//#endregion
+//#region src/speech.d.ts
+interface DashScopeSpeechOptions {
+  /** Voice name. Model-specific, e.g. "longanyang" for CosyVoice, "Cherry" for Qwen-TTS. */
+  voice?: string;
+  /** Output audio format: "wav", "mp3", "pcm". Default depends on model. */
+  format?: string;
+  /** Sample rate. Default depends on model. */
+  sampleRate?: number;
+  /** Language type for Qwen-TTS: "Chinese" | "English" | "Japanese" | etc. */
+  languageType?: string;
+  /** Speaking speed. 0.5-2.0, default 1.0. */
+  speed?: number;
+  /** Volume. 0.5-2.0, default 1.0. */
+  volume?: number;
+  /** Pitch. -12 to 12, default 0. */
+  pitch?: number;
+}
+declare class DashScopeSpeechModel implements SpeechModelV3 {
+  readonly specificationVersion: "v3";
+  readonly modelId: string;
+  private readonly config;
+  constructor(modelId: string, config: DashScopeConfig);
+  get provider(): string;
+  doGenerate(options: SpeechModelV3CallOptions): Promise<{
+    audio: Uint8Array<ArrayBuffer>;
+    warnings: SharedV3Warning[];
+    request: {
+      body: Record<string, unknown>;
+    };
+    response: {
+      timestamp: Date;
+      modelId: string;
+      headers: Record<string, string> | undefined;
+    };
+  }>;
+}
+//#endregion
+//#region src/transcription.d.ts
+interface DashScopeTranscriptionOptions {
+  /**
+   * Publicly accessible audio file URL for async transcription.
+   * Required for async models (filetrans, fun-asr, paraformer) when using long audio.
+   */
+  fileUrl?: string;
+  /** Language hint(s), e.g. ["zh", "en"]. */
+  languageHints?: string[];
+  /** Enable inverse text normalization (convert spoken numbers/dates to written form). */
+  enableItn?: boolean;
+  /** Enable word-level timestamps. */
+  enableWords?: boolean;
+  /** Channel IDs to transcribe. Default [0]. */
+  channelId?: number[];
+  /** Polling interval in ms. Default 5000. (async mode only) */
+  pollIntervalMs?: number;
+  /** Polling timeout in ms. Default 600000. (async mode only) */
+  pollTimeoutMs?: number;
+}
+declare class DashScopeTranscriptionModel implements TranscriptionModelV3 {
+  readonly specificationVersion: "v3";
+  readonly modelId: string;
+  private readonly config;
+  constructor(modelId: string, config: DashScopeConfig);
+  get provider(): string;
+  doGenerate(options: TranscriptionModelV3CallOptions): Promise<{
+    text: string;
+    segments: {
+      text: string;
+      startSecond: number;
+      endSecond: number;
+    }[];
+    language: undefined;
+    durationInSeconds: undefined;
+    warnings: SharedV3Warning[];
+    response: {
+      timestamp: Date;
+      modelId: string;
+      headers: Record<string, string> | undefined;
+    };
+  }>;
+  private doSync;
+  private doAsync;
+}
+//#endregion
+//#region src/video.d.ts
+interface DashScopeVideoOptions {
+  /** Negative prompt. */
+  negativePrompt?: string;
+  /** Enable prompt extension. */
+  promptExtend?: boolean;
+  /** Add watermark. Default false. */
+  watermark?: boolean;
+  /** Resolution for I2V: "720P" | "1080P". For T2V: use size "WIDTH*HEIGHT". */
+  resolution?: string;
+  /** Size in "WIDTH*HEIGHT" format (T2V/R2V). */
+  size?: string;
+  /** Video duration in seconds. */
+  duration?: number;
+  /** Polling interval in ms. Default 5000. */
+  pollIntervalMs?: number;
+  /** Polling timeout in ms. Default 600000. */
+  pollTimeoutMs?: number;
+}
+declare class DashScopeVideoModel implements Experimental_VideoModelV3 {
+  readonly specificationVersion: "v3";
+  readonly modelId: string;
+  private readonly config;
+  constructor(modelId: string, config: DashScopeConfig);
+  get provider(): string;
+  get maxVideosPerCall(): number | undefined;
+  doGenerate(options: Experimental_VideoModelV3CallOptions): Promise<{
+    videos: {
+      type: "url";
+      url: string;
+      mediaType: string;
+    }[];
+    warnings: SharedV3Warning[];
+    response: {
+      timestamp: Date;
+      modelId: string;
+      headers: Record<string, string> | undefined;
+    };
+  }>;
+}
+//#endregion
 //#region src/provider.d.ts
 declare function createDashScope(options?: DashScopeProviderSettings): DashScopeProvider;
 //#endregion
 //#region src/index.d.ts
 declare const dashscope: DashScopeProvider;
 //#endregion
-export { DASHSCOPE_REGION_BASE_URLS, DashScopeChatConfig, DashScopeChatOptions, DashScopeProvider, DashScopeProviderSettings, DashScopeRegion, DashScopeResponsesNamespace, DashScopeResponsesOptions, DashScopeResponsesTools, createDashScope, dashscope, responsesTools };
+export { DASHSCOPE_REGION_URLS, DashScopeChatOptions, DashScopeEmbeddingModel, DashScopeEmbeddingOptions, DashScopeImageModel, DashScopeImageOptions, DashScopeProvider, DashScopeProviderSettings, DashScopeRegion, DashScopeRerankOptions, DashScopeRerankingModel, DashScopeResponsesNamespace, DashScopeResponsesOptions, DashScopeResponsesTools, DashScopeSpeechModel, DashScopeSpeechOptions, DashScopeTranscriptionModel, DashScopeTranscriptionOptions, DashScopeVideoModel, DashScopeVideoOptions, createDashScope, dashscope, responsesTools };