@ai-sdk/google 2.0.59 → 2.0.61

package/CHANGELOG.md

@@ -1,5 +1,19 @@
 # @ai-sdk/google
 
+## 2.0.61
+
+### Patch Changes
+
+- 3399b61: feat(provider/google): support multimodal content parts in embedding provider options
+
+## 2.0.60
+
+### Patch Changes
+
+- 3f0102b: fix(provider/google): make `segment` optional in `groundingSupports` schema
+
+  Backport of #12002. When using image search grounding, the Google API returns `groundingSupports` entries without a `segment` field, causing schema validation to fail with "Invalid JSON response".
+
 ## 2.0.59
 
 ### Patch Changes

package/dist/index.d.mts

@@ -111,11 +111,11 @@ declare const responseSchema: _ai_sdk_provider_utils.LazySchema<{
                 } | null | undefined;
             }[] | null | undefined;
             groundingSupports?: {
-                segment: {
+                segment?: {
                     startIndex?: number | null | undefined;
                     endIndex?: number | null | undefined;
                     text?: string | null | undefined;
-                };
+                } | null | undefined;
                 segment_text?: string | null | undefined;
                 groundingChunkIndices?: number[] | null | undefined;
                 supportChunkIndices?: number[] | null | undefined;
@@ -184,6 +184,14 @@ type GoogleGenerativeAIEmbeddingModelId = 'gemini-embedding-001' | 'text-embeddi
 declare const googleGenerativeAIEmbeddingProviderOptions: _ai_sdk_provider_utils.LazySchema<{
     outputDimensionality?: number | undefined;
     taskType?: "SEMANTIC_SIMILARITY" | "CLASSIFICATION" | "CLUSTERING" | "RETRIEVAL_DOCUMENT" | "RETRIEVAL_QUERY" | "QUESTION_ANSWERING" | "FACT_VERIFICATION" | "CODE_RETRIEVAL_QUERY" | undefined;
+    content?: ({
+        text: string;
+    } | {
+        inlineData: {
+            mimeType: string;
+            data: string;
+        };
+    })[] | undefined;
 }>;
 type GoogleGenerativeAIEmbeddingProviderOptions = InferValidator<typeof googleGenerativeAIEmbeddingProviderOptions>;
 

package/dist/index.d.ts

@@ -111,11 +111,11 @@ declare const responseSchema: _ai_sdk_provider_utils.LazySchema<{
                 } | null | undefined;
             }[] | null | undefined;
             groundingSupports?: {
-                segment: {
+                segment?: {
                     startIndex?: number | null | undefined;
                     endIndex?: number | null | undefined;
                     text?: string | null | undefined;
-                };
+                } | null | undefined;
                 segment_text?: string | null | undefined;
                 groundingChunkIndices?: number[] | null | undefined;
                 supportChunkIndices?: number[] | null | undefined;
@@ -184,6 +184,14 @@ type GoogleGenerativeAIEmbeddingModelId = 'gemini-embedding-001' | 'text-embeddi
 declare const googleGenerativeAIEmbeddingProviderOptions: _ai_sdk_provider_utils.LazySchema<{
     outputDimensionality?: number | undefined;
     taskType?: "SEMANTIC_SIMILARITY" | "CLASSIFICATION" | "CLUSTERING" | "RETRIEVAL_DOCUMENT" | "RETRIEVAL_QUERY" | "QUESTION_ANSWERING" | "FACT_VERIFICATION" | "CODE_RETRIEVAL_QUERY" | undefined;
+    content?: ({
+        text: string;
+    } | {
+        inlineData: {
+            mimeType: string;
+            data: string;
+        };
+    })[] | undefined;
 }>;
 type GoogleGenerativeAIEmbeddingProviderOptions = InferValidator<typeof googleGenerativeAIEmbeddingProviderOptions>;
 

package/dist/index.js

@@ -30,7 +30,7 @@ module.exports = __toCommonJS(src_exports);
 var import_provider_utils15 = require("@ai-sdk/provider-utils");
 
 // src/version.ts
-var VERSION = true ? "2.0.59" : "0.0.0-test";
+var VERSION = true ? "2.0.61" : "0.0.0-test";
 
 // src/google-generative-ai-embedding-model.ts
 var import_provider = require("@ai-sdk/provider");
@@ -59,6 +59,15 @@ var googleFailedResponseHandler = (0, import_provider_utils.createJsonErrorRespo
 // src/google-generative-ai-embedding-options.ts
 var import_provider_utils2 = require("@ai-sdk/provider-utils");
 var import_v42 = require("zod/v4");
+var googleEmbeddingContentPartSchema = import_v42.z.union([
+  import_v42.z.object({ text: import_v42.z.string() }),
+  import_v42.z.object({
+    inlineData: import_v42.z.object({
+      mimeType: import_v42.z.string(),
+      data: import_v42.z.string()
+    })
+  })
+]);
 var googleGenerativeAIEmbeddingProviderOptions = (0, import_provider_utils2.lazySchema)(
   () => (0, import_provider_utils2.zodSchema)(
     import_v42.z.object({
@@ -88,7 +97,13 @@ var googleGenerativeAIEmbeddingProviderOptions = (0, import_provider_utils2.lazy
         "QUESTION_ANSWERING",
         "FACT_VERIFICATION",
         "CODE_RETRIEVAL_QUERY"
-      ]).optional()
+      ]).optional(),
+      /**
+       * Optional. Multimodal content parts for embedding non-text content
+       * (images, video, PDF, audio). When provided, these parts are merged
+       * with the text values in the embedding request.
+       */
+      content: import_v42.z.array(googleEmbeddingContentPartSchema).min(1).optional()
     })
   )
 );
@@ -111,6 +126,7 @@ var GoogleGenerativeAIEmbeddingModel = class {
     abortSignal,
     providerOptions
   }) {
+    var _a;
     const googleOptions = await (0, import_provider_utils3.parseProviderOptions)({
       provider: "google",
       providerOptions,
@@ -128,7 +144,10 @@ var GoogleGenerativeAIEmbeddingModel = class {
       await (0, import_provider_utils3.resolve)(this.config.headers),
       headers
     );
+    const multimodalContent = (_a = googleOptions == null ? void 0 : googleOptions.content) != null ? _a : [];
     if (values.length === 1) {
+      const textPart = values[0] ? [{ text: values[0] }] : [];
+      const parts = multimodalContent.length > 0 ? [...textPart, ...multimodalContent] : [{ text: values[0] }];
       const {
         responseHeaders: responseHeaders2,
         value: response2,
@@ -139,7 +158,7 @@ var GoogleGenerativeAIEmbeddingModel = class {
         body: {
           model: `models/${this.modelId}`,
           content: {
-            parts: [{ text: values[0] }]
+            parts
           },
           outputDimensionality: googleOptions == null ? void 0 : googleOptions.outputDimensionality,
           taskType: googleOptions == null ? void 0 : googleOptions.taskType
@@ -165,12 +184,18 @@ var GoogleGenerativeAIEmbeddingModel = class {
       url: `${this.config.baseURL}/models/${this.modelId}:batchEmbedContents`,
       headers: mergedHeaders,
       body: {
-        requests: values.map((value) => ({
-          model: `models/${this.modelId}`,
-          content: { role: "user", parts: [{ text: value }] },
-          outputDimensionality: googleOptions == null ? void 0 : googleOptions.outputDimensionality,
-          taskType: googleOptions == null ? void 0 : googleOptions.taskType
-        }))
+        requests: values.map((value) => {
+          const textPart = value ? [{ text: value }] : [];
+          return {
+            model: `models/${this.modelId}`,
+            content: {
+              role: "user",
+              parts: multimodalContent.length > 0 ? [...textPart, ...multimodalContent] : [{ text: value }]
+            },
+            outputDimensionality: googleOptions == null ? void 0 : googleOptions.outputDimensionality,
+            taskType: googleOptions == null ? void 0 : googleOptions.taskType
+          };
+        })
       },
       failedResponseHandler: googleFailedResponseHandler,
       successfulResponseHandler: (0, import_provider_utils3.createJsonResponseHandler)(
@@ -1471,7 +1496,7 @@ var getGroundingMetadataSchema = () => import_v45.z.object({
         startIndex: import_v45.z.number().nullish(),
         endIndex: import_v45.z.number().nullish(),
         text: import_v45.z.string().nullish()
-      }),
+      }).nullish(),
       segment_text: import_v45.z.string().nullish(),
       groundingChunkIndices: import_v45.z.array(import_v45.z.number()).nullish(),
       supportChunkIndices: import_v45.z.array(import_v45.z.number()).nullish(),