@ai-sdk/google 4.0.0-canary.70 → 4.0.0-canary.72

package/CHANGELOG.md

@@ -1,5 +1,23 @@
 # @ai-sdk/google
 
+## 4.0.0-canary.72
+
+### Patch Changes
+
+- b71c0d7: feat(provider/google): support Google search grounding when using `generateImage` with Gemini
+
+## 4.0.0-canary.71
+
+### Patch Changes
+
+- 045d2e8: fix(google): read `serviceTier` from `usageMetadata.serviceTier` in both generate and stream paths
+
+  The previous implementation read `serviceTier` from the `x-gemini-service-tier`
+  response header, which is only populated on non-streaming responses. Gemini
+  streaming includes the value in `usageMetadata.serviceTier` on every chunk, so
+  `providerMetadata.google.serviceTier` was always `null` for streams. Read from
+  `usageMetadata` for both paths instead.
+
 ## 4.0.0-canary.70
 
 ### Patch Changes

package/dist/index.d.ts

@@ -171,6 +171,7 @@ declare const responseSchema: _ai_sdk_provider_utils.LazySchema<{
         candidatesTokenCount?: number | null | undefined;
         totalTokenCount?: number | null | undefined;
         trafficType?: string | null | undefined;
+        serviceTier?: string | null | undefined;
         promptTokensDetails?: {
             modality: string;
             tokenCount: number;
@@ -216,6 +217,17 @@ interface GoogleProviderMetadata {
 declare const googleImageModelOptionsSchema: _ai_sdk_provider_utils.LazySchema<{
     personGeneration?: "dont_allow" | "allow_adult" | "allow_all" | null | undefined;
     aspectRatio?: "1:1" | "3:4" | "4:3" | "9:16" | "16:9" | null | undefined;
+    googleSearch?: {
+        [x: string]: unknown;
+        searchTypes?: {
+            webSearch?: Record<string, never> | undefined;
+            imageSearch?: Record<string, never> | undefined;
+        } | undefined;
+        timeRangeFilter?: {
+            startTime: string;
+            endTime: string;
+        } | undefined;
+    } | undefined;
 }>;
 type GoogleImageModelOptions = InferSchema<typeof googleImageModelOptionsSchema>;
 

package/dist/index.js

@@ -7,7 +7,7 @@ import {
 } from "@ai-sdk/provider-utils";
 
 // src/version.ts
-var VERSION = true ? "4.0.0-canary.70" : "0.0.0-test";
+var VERSION = true ? "4.0.0-canary.72" : "0.0.0-test";
 
 // src/google-embedding-model.ts
 import {
@@ -1855,7 +1855,7 @@ var GoogleLanguageModel = class _GoogleLanguageModel {
         safetyRatings: (_p = candidate.safetyRatings) != null ? _p : null,
         usageMetadata: usageMetadata != null ? usageMetadata : null,
         finishMessage: (_q = candidate.finishMessage) != null ? _q : null,
-        serviceTier: (_r = responseHeaders == null ? void 0 : responseHeaders["x-gemini-service-tier"]) != null ? _r : null
+        serviceTier: (_r = usageMetadata == null ? void 0 : usageMetadata.serviceTier) != null ? _r : null
       }),
       request: { body: args },
       response: {
@@ -1866,7 +1866,6 @@ var GoogleLanguageModel = class _GoogleLanguageModel {
     };
   }
   async doStream(options) {
-    var _a;
     const { args, warnings, providerOptionsNames, extraHeaders } = await this.getArgs(options, { isStreaming: true });
     const wrapProviderMetadata = (payload) => Object.fromEntries(
       providerOptionsNames.map((name) => [name, payload])
@@ -1895,7 +1894,6 @@ var GoogleLanguageModel = class _GoogleLanguageModel {
     let providerMetadata = void 0;
     let lastGroundingMetadata = null;
     let lastUrlContextMetadata = null;
-    const serviceTier = (_a = responseHeaders == null ? void 0 : responseHeaders["x-gemini-service-tier"]) != null ? _a : null;
     const generateId3 = this.config.generateId;
     let hasToolCalls = false;
     let currentTextBlockId = null;
@@ -1940,7 +1938,7 @@ var GoogleLanguageModel = class _GoogleLanguageModel {
             controller.enqueue({ type: "stream-start", warnings });
           },
           transform(chunk, controller) {
-            var _a2, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n, _o;
+            var _a, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n, _o, _p;
             if (options.includeRawChunks) {
               controller.enqueue({ type: "raw", rawValue: chunk.rawValue });
             }
@@ -1953,7 +1951,7 @@ var GoogleLanguageModel = class _GoogleLanguageModel {
             if (usageMetadata != null) {
               usage = usageMetadata;
             }
-            const candidate = (_a2 = value.candidates) == null ? void 0 : _a2[0];
+            const candidate = (_a = value.candidates) == null ? void 0 : _a[0];
             if (candidate == null) {
               return;
             }
@@ -2248,7 +2246,7 @@ var GoogleLanguageModel = class _GoogleLanguageModel {
                 safetyRatings: (_n = candidate.safetyRatings) != null ? _n : null,
                 usageMetadata: usageMetadata != null ? usageMetadata : null,
                 finishMessage: (_o = candidate.finishMessage) != null ? _o : null,
-                serviceTier
+                serviceTier: (_p = usage == null ? void 0 : usage.serviceTier) != null ? _p : null
               });
             }
           },
@@ -2575,6 +2573,7 @@ var usageSchema = z5.object({
   totalTokenCount: z5.number().nullish(),
   // https://cloud.google.com/vertex-ai/generative-ai/docs/reference/rest/v1/GenerateContentResponse#TrafficType
   trafficType: z5.string().nullish(),
+  serviceTier: z5.string().nullish(),
   // https://ai.google.dev/api/generate-content#Modality
   promptTokensDetails: tokenDetailsSchema,
   candidatesTokensDetails: tokenDetailsSchema
@@ -2836,7 +2835,17 @@ var googleImageModelOptionsSchema = lazySchema12(
   () => zodSchema12(
     z13.object({
       personGeneration: z13.enum(["dont_allow", "allow_adult", "allow_all"]).nullish(),
-      aspectRatio: z13.enum(["1:1", "3:4", "4:3", "9:16", "16:9"]).nullish()
+      aspectRatio: z13.enum(["1:1", "3:4", "4:3", "9:16", "16:9"]).nullish(),
+      /**
+       * Enable Google Search grounding for Gemini image models. The value is
+       * forwarded as the args of the `google.tools.googleSearch` provider
+       * tool on the underlying language-model call. Pass `{}` for defaults.
+       *
+       * `generateImage` does not accept a `tools` parameter, so this is the
+       * dedicated escape hatch for grounding image generation the same way
+       * `generateText` does.
+       */
+      googleSearch: googleSearchToolArgsBaseSchema.optional()
     })
   )
 );
@@ -2928,7 +2937,15 @@ var GoogleImageModel = class _GoogleImageModel {
       parameters.aspectRatio = aspectRatio;
     }
     if (googleOptions) {
-      Object.assign(parameters, googleOptions);
+      const { googleSearch: imagenGoogleSearch, ...imagenOptions } = googleOptions;
+      if (imagenGoogleSearch != null) {
+        warnings.push({
+          type: "unsupported",
+          feature: "googleSearch",
+          details: "Google Search grounding is only supported on Gemini image models."
+        });
+      }
+      Object.assign(parameters, imagenOptions);
     }
     const body = {
       instances: [{ prompt }],
@@ -2968,7 +2985,7 @@ var GoogleImageModel = class _GoogleImageModel {
     };
   }
   async doGenerateGemini(options) {
-    var _a, _b, _c, _d, _e, _f, _g, _h, _i;
+    var _a, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k;
     const {
       prompt,
       n,
@@ -3026,12 +3043,18 @@ var GoogleImageModel = class _GoogleImageModel {
     const languageModelPrompt = [
       { role: "user", content: userContent }
     ];
+    const googleImageOptions = await parseProviderOptions3({
+      provider: "google",
+      providerOptions,
+      schema: googleImageModelOptionsSchema
+    });
+    const { googleSearch: _strippedGoogleSearch, ...passthroughGoogleOptions } = (_a = providerOptions == null ? void 0 : providerOptions.google) != null ? _a : {};
     const languageModel = new GoogleLanguageModel(this.modelId, {
       provider: this.config.provider,
       baseURL: this.config.baseURL,
-      headers: (_a = this.config.headers) != null ? _a : {},
+      headers: (_b = this.config.headers) != null ? _b : {},
       fetch: this.config.fetch,
-      generateId: (_b = this.config.generateId) != null ? _b : defaultGenerateId
+      generateId: (_c = this.config.generateId) != null ? _c : defaultGenerateId
     });
     const result = await languageModel.doGenerate({
       prompt: languageModelPrompt,
@@ -3042,9 +3065,17 @@ var GoogleImageModel = class _GoogleImageModel {
           imageConfig: aspectRatio ? {
             aspectRatio
           } : void 0,
-          ...(_c = providerOptions == null ? void 0 : providerOptions.google) != null ? _c : {}
+          ...passthroughGoogleOptions
         }
       },
+      tools: (googleImageOptions == null ? void 0 : googleImageOptions.googleSearch) != null ? [
+        {
+          type: "provider",
+          id: "google.google_search",
+          name: "google_search",
+          args: googleImageOptions.googleSearch
+        }
+      ] : void 0,
       headers,
       abortSignal
     });
@@ -3055,23 +3086,25 @@ var GoogleImageModel = class _GoogleImageModel {
         images.push(convertToBase642(part.data.data));
       }
     }
+    const languageModelGoogleMetadata = (_h = (_g = result.providerMetadata) == null ? void 0 : _g.google) != null ? _h : {};
     return {
       images,
       warnings,
       providerMetadata: {
         google: {
+          ...languageModelGoogleMetadata,
           images: images.map(() => ({}))
         }
       },
       response: {
         timestamp: currentDate,
         modelId: this.modelId,
-        headers: (_g = result.response) == null ? void 0 : _g.headers
+        headers: (_i = result.response) == null ? void 0 : _i.headers
       },
       usage: result.usage ? {
         inputTokens: result.usage.inputTokens.total,
         outputTokens: result.usage.outputTokens.total,
-        totalTokens: ((_h = result.usage.inputTokens.total) != null ? _h : 0) + ((_i = result.usage.outputTokens.total) != null ? _i : 0)
+        totalTokens: ((_j = result.usage.inputTokens.total) != null ? _j : 0) + ((_k = result.usage.outputTokens.total) != null ? _k : 0)
       } : void 0
     };
   }