@ai-sdk/google-vertex 4.0.146 → 4.0.148

package/dist/index.mjs

@@ -1,5 +1,5 @@
 // src/google-vertex-provider-node.ts
-import { loadOptionalSetting as loadOptionalSetting2, resolve as resolve5 } from "@ai-sdk/provider-utils";
+import { loadOptionalSetting as loadOptionalSetting2, resolve as resolve6 } from "@ai-sdk/provider-utils";
 
 // src/google-vertex-auth-google-auth-library.ts
 import { GoogleAuth } from "google-auth-library";
@@ -23,13 +23,13 @@ import {
   loadOptionalSetting,
   loadSetting,
   normalizeHeaders,
-  resolve as resolve4,
+  resolve as resolve5,
   withoutTrailingSlash,
   withUserAgentSuffix
 } from "@ai-sdk/provider-utils";
 
 // src/version.ts
-var VERSION = true ? "4.0.146" : "0.0.0-test";
+var VERSION = true ? "4.0.148" : "0.0.0-test";
 
 // src/google-vertex-embedding-model.ts
 import {
@@ -108,7 +108,6 @@ var googleVertexEmbeddingModelOptions = z2.object({
 var GoogleVertexEmbeddingModel = class {
   constructor(modelId, config) {
     this.specificationVersion = "v3";
-    this.maxEmbeddingsPerCall = 2048;
     this.supportsParallelCalls = true;
     this.modelId = modelId;
     this.config = config;
@@ -116,12 +115,18 @@ var GoogleVertexEmbeddingModel = class {
   get provider() {
     return this.config.provider;
   }
+  // gemini-embedding-2 models only support :embedContent (one value per call),
+  // not the :predict batch endpoint. https://github.com/vercel/ai/issues/15853
+  get maxEmbeddingsPerCall() {
+    return usesEmbedContentEndpoint(this.modelId) ? 1 : 2048;
+  }
   async doEmbed({
     values,
     headers,
     abortSignal,
     providerOptions
   }) {
+    var _a;
     let googleOptions = await parseProviderOptions({
       provider: "vertex",
       providerOptions,
@@ -147,6 +152,37 @@ var GoogleVertexEmbeddingModel = class {
       await resolve(this.config.headers),
       headers
     );
+    if (usesEmbedContentEndpoint(this.modelId)) {
+      const {
+        responseHeaders: responseHeaders2,
+        value: response2,
+        rawValue: rawValue2
+      } = await postJsonToApi({
+        url: `${this.config.baseURL}/models/${this.modelId}:embedContent`,
+        headers: mergedHeaders,
+        body: {
+          content: { parts: [{ text: values[0] }] },
+          embedContentConfig: {
+            outputDimensionality: googleOptions.outputDimensionality,
+            taskType: googleOptions.taskType,
+            title: googleOptions.title,
+            autoTruncate: googleOptions.autoTruncate
+          }
+        },
+        failedResponseHandler: googleVertexFailedResponseHandler,
+        successfulResponseHandler: createJsonResponseHandler(
+          googleVertexEmbedContentResponseSchema
+        ),
+        abortSignal,
+        fetch: this.config.fetch
+      });
+      return {
+        warnings: [],
+        embeddings: [response2.embedding.values],
+        usage: ((_a = response2.usageMetadata) == null ? void 0 : _a.promptTokenCount) == null ? void 0 : { tokens: response2.usageMetadata.promptTokenCount },
+        response: { headers: responseHeaders2, body: rawValue2 }
+      };
+    }
     const url = `${this.config.baseURL}/models/${this.modelId}:predict`;
     const {
       responseHeaders,
@@ -200,6 +236,17 @@ var googleVertexTextEmbeddingResponseSchema = z3.object({
     })
   )
 });
+var googleVertexEmbedContentResponseSchema = z3.object({
+  embedding: z3.object({
+    values: z3.array(z3.number())
+  }),
+  usageMetadata: z3.object({
+    promptTokenCount: z3.number().nullish()
+  }).nullish()
+});
+function usesEmbedContentEndpoint(modelId) {
+  return modelId === "gemini-embedding-2" || modelId === "gemini-embedding-2-preview";
+}
 
 // src/google-vertex-image-model.ts
 import { GoogleGenerativeAILanguageModel } from "@ai-sdk/google/internal";
@@ -552,22 +599,201 @@ var googleVertexTools = {
   vertexRagStore: googleTools.vertexRagStore
 };
 
+// src/google-vertex-transcription-model.ts
+import {
+  combineHeaders as combineHeaders3,
+  convertUint8ArrayToBase64 as convertUint8ArrayToBase642,
+  createJsonResponseHandler as createJsonResponseHandler3,
+  parseProviderOptions as parseProviderOptions3,
+  postJsonToApi as postJsonToApi3,
+  resolve as resolve3
+} from "@ai-sdk/provider-utils";
+import { z as z6 } from "zod/v4";
+
+// src/google-vertex-transcription-model-options.ts
+import { z as z5 } from "zod/v4";
+var googleVertexTranscriptionProviderOptionsSchema = z5.object({
+  /**
+   * BCP-47 language codes to recognize (e.g. `['en-US']`), or `['auto']` to let
+   * Chirp auto-detect the spoken language. Defaults to `['auto']`. For
+   * `telephony`, pass a supported explicit language code.
+   */
+  languageCodes: z5.array(z5.string()).optional(),
+  /**
+   * Whether to add punctuation to the transcript. Defaults to `true`.
+   */
+  enableAutomaticPunctuation: z5.boolean().optional(),
+  /**
+   * Whether to include word-level timestamps. Defaults to `true` so the
+   * transcription result can include segments.
+   *
+   * Enabling word-level timestamps can reduce transcription quality and speed
+   * for Chirp models.
+   */
+  enableWordTimeOffsets: z5.boolean().optional(),
+  /**
+   * The Cloud Speech-to-Text region for the request (e.g. `'us'`, `'eu'`,
+   * `'us-central1'`). Defaults to the provider `location`.
+   *
+   * Note: Speech-to-Text regions differ from Vertex AI regions. Chirp is only
+   * available in specific Speech-to-Text regions and is not available in the
+   * `global` location.
+   */
+  region: z5.string().optional()
+});
+
+// src/google-vertex-transcription-model.ts
+function parseDurationSeconds(value) {
+  if (value == null) {
+    return void 0;
+  }
+  const seconds = Number.parseFloat(value);
+  return Number.isFinite(seconds) ? seconds : void 0;
+}
+function convertBcp47ToIso6391(value) {
+  if (value == null) {
+    return void 0;
+  }
+  try {
+    const language = new Intl.Locale(value).language;
+    return language.length === 2 ? language : void 0;
+  } catch (e) {
+    return void 0;
+  }
+}
+var GoogleVertexTranscriptionModel = class {
+  constructor(modelId, config) {
+    this.modelId = modelId;
+    this.config = config;
+    this.specificationVersion = "v3";
+  }
+  get provider() {
+    return this.config.provider;
+  }
+  async doGenerate(options) {
+    var _a, _b, _c, _d, _e, _f, _g, _h, _i, _j;
+    const currentDate = (_c = (_b = (_a = this.config._internal) == null ? void 0 : _a.currentDate) == null ? void 0 : _b.call(_a)) != null ? _c : /* @__PURE__ */ new Date();
+    const warnings = [];
+    let googleOptions;
+    for (const provider of ["googleVertex", "vertex", "google"]) {
+      googleOptions = await parseProviderOptions3({
+        provider,
+        providerOptions: options.providerOptions,
+        schema: googleVertexTranscriptionProviderOptionsSchema
+      });
+      if (googleOptions != null) {
+        break;
+      }
+    }
+    const region = (_d = googleOptions == null ? void 0 : googleOptions.region) != null ? _d : this.config.location;
+    const languageCodes = (_e = googleOptions == null ? void 0 : googleOptions.languageCodes) != null ? _e : ["auto"];
+    const content = typeof options.audio === "string" ? options.audio : convertUint8ArrayToBase642(options.audio);
+    const requestBody = {
+      config: {
+        model: this.modelId,
+        languageCodes,
+        // Let Speech-to-Text auto-detect the audio encoding (wav/mp3/flac/…).
+        autoDecodingConfig: {},
+        features: {
+          // Word timing populates `segments`.
+          enableWordTimeOffsets: (_f = googleOptions == null ? void 0 : googleOptions.enableWordTimeOffsets) != null ? _f : true,
+          enableAutomaticPunctuation: (_g = googleOptions == null ? void 0 : googleOptions.enableAutomaticPunctuation) != null ? _g : true
+        }
+      },
+      content
+    };
+    const host = region === "global" ? "speech.googleapis.com" : `${region}-speech.googleapis.com`;
+    const url = `https://${host}/v2/projects/${this.config.project}/locations/${region}/recognizers/_:recognize`;
+    const {
+      value: response,
+      responseHeaders,
+      rawValue: rawResponse
+    } = await postJsonToApi3({
+      url,
+      headers: combineHeaders3(
+        this.config.headers ? await resolve3(this.config.headers) : void 0,
+        options.headers
+      ),
+      body: requestBody,
+      failedResponseHandler: googleVertexFailedResponseHandler,
+      successfulResponseHandler: createJsonResponseHandler3(
+        googleVertexTranscriptionResponseSchema
+      ),
+      abortSignal: options.abortSignal,
+      fetch: this.config.fetch
+    });
+    const results = (_h = response.results) != null ? _h : [];
+    const text = results.map((result) => {
+      var _a2, _b2, _c2;
+      return (_c2 = (_b2 = (_a2 = result.alternatives) == null ? void 0 : _a2[0]) == null ? void 0 : _b2.transcript) != null ? _c2 : "";
+    }).join(" ").trim();
+    const segments = results.flatMap(
+      (result) => {
+        var _a2, _b2, _c2, _d2;
+        return (_d2 = (_c2 = (_b2 = (_a2 = result.alternatives) == null ? void 0 : _a2[0]) == null ? void 0 : _b2.words) == null ? void 0 : _c2.flatMap((word) => {
+          const startSecond = parseDurationSeconds(word.startOffset);
+          const endSecond = parseDurationSeconds(word.endOffset);
+          return word.word == null || startSecond == null || endSecond == null ? [] : [{ text: word.word, startSecond, endSecond }];
+        })) != null ? _d2 : [];
+      }
+    );
+    const language = convertBcp47ToIso6391((_i = results[0]) == null ? void 0 : _i.languageCode);
+    return {
+      text,
+      segments,
+      language,
+      durationInSeconds: parseDurationSeconds(
+        (_j = response.metadata) == null ? void 0 : _j.totalBilledDuration
+      ),
+      warnings,
+      response: {
+        timestamp: currentDate,
+        modelId: this.modelId,
+        headers: responseHeaders,
+        body: rawResponse
+      }
+    };
+  }
+};
+var googleVertexTranscriptionResponseSchema = z6.object({
+  results: z6.array(
+    z6.object({
+      alternatives: z6.array(
+        z6.object({
+          transcript: z6.string().nullish(),
+          words: z6.array(
+            z6.object({
+              word: z6.string().nullish(),
+              startOffset: z6.string().nullish(),
+              endOffset: z6.string().nullish()
+            })
+          ).nullish()
+        })
+      ).nullish(),
+      languageCode: z6.string().nullish()
+    })
+  ).nullish(),
+  metadata: z6.object({
+    totalBilledDuration: z6.string().nullish()
+  }).nullish()
+});
+
 // src/google-vertex-video-model.ts
 import {
   AISDKError
 } from "@ai-sdk/provider";
 import {
-  combineHeaders as combineHeaders3,
-  convertUint8ArrayToBase64 as convertUint8ArrayToBase642,
-  createJsonResponseHandler as createJsonResponseHandler3,
+  combineHeaders as combineHeaders4,
+  convertUint8ArrayToBase64 as convertUint8ArrayToBase643,
+  createJsonResponseHandler as createJsonResponseHandler4,
   delay,
   lazySchema,
-  parseProviderOptions as parseProviderOptions3,
-  postJsonToApi as postJsonToApi3,
-  resolve as resolve3,
+  parseProviderOptions as parseProviderOptions4,
+  postJsonToApi as postJsonToApi4,
+  resolve as resolve4,
   zodSchema
 } from "@ai-sdk/provider-utils";
-import { z as z5 } from "zod/v4";
+import { z as z7 } from "zod/v4";
 var GoogleVertexVideoModel = class {
   constructor(modelId, config) {
     this.modelId = modelId;
@@ -584,7 +810,7 @@ var GoogleVertexVideoModel = class {
     var _a, _b, _c, _d, _e, _f;
     const currentDate = (_c = (_b = (_a = this.config._internal) == null ? void 0 : _a.currentDate) == null ? void 0 : _b.call(_a)) != null ? _c : /* @__PURE__ */ new Date();
     const warnings = [];
-    const vertexOptions = await parseProviderOptions3({
+    const vertexOptions = await parseProviderOptions4({
       provider: "vertex",
       providerOptions: options.providerOptions,
       schema: googleVertexVideoModelOptionsSchema
@@ -602,7 +828,7 @@ var GoogleVertexVideoModel = class {
           details: "Vertex AI video models require base64-encoded images or GCS URIs. URL will be ignored."
         });
       } else {
-        const base64Data = typeof options.image.data === "string" ? options.image.data : convertUint8ArrayToBase642(options.image.data);
+        const base64Data = typeof options.image.data === "string" ? options.image.data : convertUint8ArrayToBase643(options.image.data);
         instance.image = {
           bytesBase64Encoded: base64Data,
           mimeType: options.image.mediaType
@@ -660,17 +886,17 @@ var GoogleVertexVideoModel = class {
         }
       }
     }
-    const { value: operation } = await postJsonToApi3({
+    const { value: operation } = await postJsonToApi4({
       url: `${this.config.baseURL}/models/${this.modelId}:predictLongRunning`,
-      headers: combineHeaders3(
-        await resolve3(this.config.headers),
+      headers: combineHeaders4(
+        await resolve4(this.config.headers),
         options.headers
       ),
       body: {
         instances,
         parameters
       },
-      successfulResponseHandler: createJsonResponseHandler3(
+      successfulResponseHandler: createJsonResponseHandler4(
         vertexOperationSchema
       ),
       failedResponseHandler: googleVertexFailedResponseHandler,
@@ -703,16 +929,16 @@ var GoogleVertexVideoModel = class {
           message: "Video generation request was aborted"
         });
       }
-      const { value: statusOperation, responseHeaders: pollHeaders } = await postJsonToApi3({
+      const { value: statusOperation, responseHeaders: pollHeaders } = await postJsonToApi4({
         url: `${this.config.baseURL}/models/${this.modelId}:fetchPredictOperation`,
-        headers: combineHeaders3(
-          await resolve3(this.config.headers),
+        headers: combineHeaders4(
+          await resolve4(this.config.headers),
           options.headers
         ),
         body: {
           operationName
         },
-        successfulResponseHandler: createJsonResponseHandler3(
+        successfulResponseHandler: createJsonResponseHandler4(
           vertexOperationSchema
         ),
         failedResponseHandler: googleVertexFailedResponseHandler,
@@ -781,38 +1007,38 @@ var GoogleVertexVideoModel = class {
     };
   }
 };
-var vertexOperationSchema = z5.object({
-  name: z5.string().nullish(),
-  done: z5.boolean().nullish(),
-  error: z5.object({
-    code: z5.number().nullish(),
-    message: z5.string(),
-    status: z5.string().nullish()
+var vertexOperationSchema = z7.object({
+  name: z7.string().nullish(),
+  done: z7.boolean().nullish(),
+  error: z7.object({
+    code: z7.number().nullish(),
+    message: z7.string(),
+    status: z7.string().nullish()
   }).nullish(),
-  response: z5.object({
-    videos: z5.array(
-      z5.object({
-        bytesBase64Encoded: z5.string().nullish(),
-        gcsUri: z5.string().nullish(),
-        mimeType: z5.string().nullish()
+  response: z7.object({
+    videos: z7.array(
+      z7.object({
+        bytesBase64Encoded: z7.string().nullish(),
+        gcsUri: z7.string().nullish(),
+        mimeType: z7.string().nullish()
       })
     ).nullish(),
-    raiMediaFilteredCount: z5.number().nullish()
+    raiMediaFilteredCount: z7.number().nullish()
   }).nullish()
 });
 var googleVertexVideoModelOptionsSchema = lazySchema(
   () => zodSchema(
-    z5.object({
-      pollIntervalMs: z5.number().positive().nullish(),
-      pollTimeoutMs: z5.number().positive().nullish(),
-      personGeneration: z5.enum(["dont_allow", "allow_adult", "allow_all"]).nullish(),
-      negativePrompt: z5.string().nullish(),
-      generateAudio: z5.boolean().nullish(),
-      gcsOutputDirectory: z5.string().nullish(),
-      referenceImages: z5.array(
-        z5.object({
-          bytesBase64Encoded: z5.string().nullish(),
-          gcsUri: z5.string().nullish()
+    z7.object({
+      pollIntervalMs: z7.number().positive().nullish(),
+      pollTimeoutMs: z7.number().positive().nullish(),
+      personGeneration: z7.enum(["dont_allow", "allow_adult", "allow_all"]).nullish(),
+      negativePrompt: z7.string().nullish(),
+      generateAudio: z7.boolean().nullish(),
+      gcsOutputDirectory: z7.string().nullish(),
+      referenceImages: z7.array(
+        z7.object({
+          bytesBase64Encoded: z7.string().nullish(),
+          gcsUri: z7.string().nullish()
         })
       ).nullish()
     }).passthrough()
@@ -871,7 +1097,7 @@ function createVertex(options = {}) {
   const createConfig = (name) => {
     const getHeaders = async () => {
       var _a;
-      const originalHeaders = await resolve4((_a = options.headers) != null ? _a : {});
+      const originalHeaders = await resolve5((_a = options.headers) != null ? _a : {});
       return withUserAgentSuffix(
         originalHeaders,
         `ai-sdk/google-vertex/${VERSION}`
@@ -914,6 +1140,21 @@ function createVertex(options = {}) {
       generateId: (_a = options.generateId) != null ? _a : generateId
     });
   };
+  const createTranscriptionModel = (modelId) => {
+    if (apiKey) {
+      throw new Error(
+        "Google Vertex transcription models do not support Express Mode API keys. Use standard Google Cloud credentials instead."
+      );
+    }
+    const config = createConfig("transcription");
+    return new GoogleVertexTranscriptionModel(modelId, {
+      provider: config.provider,
+      headers: config.headers,
+      fetch: config.fetch,
+      project: loadVertexProject(),
+      location: loadVertexLocation()
+    });
+  };
   const provider = function(modelId) {
     if (new.target) {
       throw new Error(
@@ -930,6 +1171,8 @@ function createVertex(options = {}) {
   provider.imageModel = createImageModel;
   provider.video = createVideoModel;
   provider.videoModel = createVideoModel;
+  provider.transcription = createTranscriptionModel;
+  provider.transcriptionModel = createTranscriptionModel;
   provider.tools = googleVertexTools;
   return provider;
 }
@@ -952,7 +1195,7 @@ function createVertex2(options = {}) {
     ...options,
     headers: async () => ({
       Authorization: `Bearer ${await generateAuthToken()}`,
-      ...await resolve5(options.headers)
+      ...await resolve6(options.headers)
     })
   });
 }