@ai-sdk/google 3.0.134 → 3.0.136

package/docs/15-google-generative-ai.mdx

@@ -360,6 +360,14 @@ and long coding or agentic tasks.
 - Complex requests can run for several minutes. Make sure your HTTP client
   allows read timeouts of at least 15 to 20 minutes.
 
+**Grounding across requests.** When a response is continued automatically,
+`providerMetadata.google.groundingMetadataPerRequest` contains the grounding
+metadata from each request, in order. An entry is `null` if that request did
+not return grounding metadata. This preserves earlier Search results even if
+the final request returns none, and keeps citation indices scoped to their
+original request. The existing `groundingMetadata` field keeps its current
+behavior; use the per-request array when inspecting a continued response.
+
 **Usage across requests.** Each continued request re-sends the prompt plus all
 output so far as input, so its prompt token count includes everything
 generated before it. `usage` is the sum of all requests, which matches what

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ai-sdk/google",
-  "version": "3.0.134",
+  "version": "3.0.136",
   "license": "Apache-2.0",
   "sideEffects": false,
   "main": "./dist/index.js",
@@ -37,7 +37,7 @@
   },
   "dependencies": {
     "@ai-sdk/provider": "3.0.19",
-    "@ai-sdk/provider-utils": "4.0.60"
+    "@ai-sdk/provider-utils": "4.0.61"
   },
   "devDependencies": {
     "@types/node": "20.17.24",

package/src/google-generative-ai-language-model.ts

@@ -428,11 +428,14 @@ export class GoogleGenerativeAILanguageModel implements LanguageModelV3 {
     // the prompt plus all output so far as input, so the first request's
     // prompt token count is kept separately for context window checks.
     const continuedParts: ContentPartSchema[] = [];
+    const groundingMetadataPerRequest: Array<GroundingMetadataSchema | null> =
+      [];
     let continuedUsageMetadata: GoogleGenerativeAIUsageMetadata | undefined;
     const firstPromptTokenCount = response.usageMetadata?.promptTokenCount;
 
     while (true) {
       const candidate = response.candidates?.[0];
+      groundingMetadataPerRequest.push(candidate?.groundingMetadata ?? null);
       const continuationToken =
         candidate?.finishReason === 'CONTINUATION'
           ? candidate.continuationToken
@@ -642,6 +645,9 @@ export class GoogleGenerativeAILanguageModel implements LanguageModelV3 {
         [providerOptionsName]: {
           promptFeedback: response.promptFeedback ?? null,
           groundingMetadata: candidate?.groundingMetadata ?? null,
+          ...(groundingMetadataPerRequest.length > 1
+            ? { groundingMetadataPerRequest }
+            : {}),
           urlContextMetadata: candidate?.urlContextMetadata ?? null,
           safetyRatings: candidate?.safetyRatings ?? null,
           usageMetadata: usageMetadata ?? null,
@@ -695,6 +701,10 @@ export class GoogleGenerativeAILanguageModel implements LanguageModelV3 {
     const { responseHeaders, value: firstResponse } =
       await streamGenerateContent(args);
 
+    const groundingMetadataPerRequest: Array<GroundingMetadataSchema | null> =
+      [];
+    let requestGroundingMetadata: GroundingMetadataSchema | null = null;
+
     // Gemini 4 long decode: when a stream reaches the per-request cap while
     // `maxOutputTokens` budget remains, its last chunk finishes with
     // `CONTINUATION`. Resend the same request with the continuation token and
@@ -703,12 +713,22 @@ export class GoogleGenerativeAILanguageModel implements LanguageModelV3 {
       stream: firstResponse,
       getContinuationToken: chunk => {
         const candidate = chunk.candidates?.[0];
+        if (candidate?.groundingMetadata != null) {
+          requestGroundingMetadata = candidate.groundingMetadata;
+        }
         return candidate?.finishReason === 'CONTINUATION'
           ? (candidate.continuationToken ?? undefined)
           : undefined;
       },
-      continueStream: async continuationToken =>
-        (await streamGenerateContent({ ...args, continuationToken })).value,
+      continueStream: async continuationToken => {
+        const { value } = await streamGenerateContent({
+          ...args,
+          continuationToken,
+        });
+        groundingMetadataPerRequest.push(requestGroundingMetadata);
+        requestGroundingMetadata = null;
+        return value;
+      },
     });
 
     let finishReason: LanguageModelV3FinishReason = {
@@ -1293,6 +1313,14 @@ export class GoogleGenerativeAILanguageModel implements LanguageModelV3 {
                 [providerOptionsName]: {
                   promptFeedback,
                   groundingMetadata: lastGroundingMetadata,
+                  ...(groundingMetadataPerRequest.length > 0
+                    ? {
+                        groundingMetadataPerRequest: [
+                          ...groundingMetadataPerRequest,
+                          requestGroundingMetadata,
+                        ],
+                      }
+                    : {}),
                   urlContextMetadata: lastUrlContextMetadata,
                   safetyRatings: lastSafetyRatings,
                   usageMetadata: usage ?? null,

package/src/google-generative-ai-prompt.ts

@@ -74,6 +74,8 @@ export type GoogleGenerativeAIUsageMetadata = UsageMetadataSchema;
 export interface GoogleGenerativeAIProviderMetadata {
   promptFeedback: GoogleGenerativeAIPromptFeedback | null;
   groundingMetadata: GoogleGenerativeAIGroundingMetadata | null;
+  /** Grounding from each automatically continued request, in request order. */
+  groundingMetadataPerRequest?: Array<GoogleGenerativeAIGroundingMetadata | null>;
   urlContextMetadata: GoogleGenerativeAIUrlContextMetadata | null;
   safetyRatings: GoogleGenerativeAISafetyRating[] | null;
   usageMetadata: GoogleGenerativeAIUsageMetadata | null;