@ai-sdk/google 4.0.98 → 4.0.100

package/docs/15-google.mdx

@@ -395,6 +395,14 @@ 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": "4.0.98",
+  "version": "4.0.100",
   "type": "module",
   "license": "Apache-2.0",
   "sideEffects": false,
@@ -35,8 +35,8 @@
     }
   },
   "dependencies": {
-    "@ai-sdk/provider": "4.0.28",
-    "@ai-sdk/provider-utils": "5.0.61"
+    "@ai-sdk/provider": "4.0.30",
+    "@ai-sdk/provider-utils": "5.0.63"
   },
   "devDependencies": {
     "@ai-sdk/test-server": "2.0.4",

package/src/google-error.ts

@@ -6,15 +6,18 @@ import {
 } from '@ai-sdk/provider-utils';
 import { z } from 'zod/v4';
 
+export const getGoogleErrorSchema = () =>
+  z.object({
+    code: z.number().nullable(),
+    message: z.string(),
+    status: z.string(),
+    details: z.array(z.unknown()).nullish(),
+  });
+
 const googleErrorDataSchema = lazySchema(() =>
   zodSchema(
     z.object({
-      error: z.object({
-        code: z.number().nullable(),
-        message: z.string(),
-        status: z.string(),
-        details: z.array(z.unknown()).nullish(),
-      }),
+      error: getGoogleErrorSchema(),
     }),
   ),
 );

package/src/google-language-model.ts

@@ -18,6 +18,7 @@ import {
   createToolNameMapping,
   createEventSourceResponseHandler,
   createJsonResponseHandler,
+  createProviderStreamError,
   generateId,
   isCustomReasoning,
   lazySchema,
@@ -45,7 +46,10 @@ import {
 import { convertToGoogleMessages } from './convert-to-google-messages';
 import { downloadToolResultFiles } from './download-tool-result-files';
 import { getModelPath } from './get-model-path';
-import { googleFailedResponseHandler } from './google-error';
+import {
+  getGoogleErrorSchema,
+  googleFailedResponseHandler,
+} from './google-error';
 import { sanitizeResponseJsonSchema } from './sanitize-response-json-schema';
 import {
   googleLanguageModelOptions,
@@ -516,6 +520,7 @@ export class GoogleLanguageModel implements LanguageModelV4 {
     toolNameMapping,
     jsonResponseToolName,
     initialPromptTokenCount,
+    groundingMetadataPerRequest,
   }: {
     config: GoogleLanguageModelConfig;
     response: InferSchema<typeof responseSchema>;
@@ -524,6 +529,7 @@ export class GoogleLanguageModel implements LanguageModelV4 {
     toolNameMapping?: ReturnType<typeof createToolNameMapping>;
     jsonResponseToolName?: string;
     initialPromptTokenCount?: number | null;
+    groundingMetadataPerRequest?: Array<GroundingMetadataSchema | null>;
   }): LanguageModelV4GenerateResult {
     const wrapProviderMetadata = (payload: Record<string, unknown>) =>
       Object.fromEntries(
@@ -714,6 +720,9 @@ export class GoogleLanguageModel implements LanguageModelV4 {
       providerMetadata: wrapProviderMetadata({
         promptFeedback: response.promptFeedback ?? null,
         groundingMetadata: candidate?.groundingMetadata ?? null,
+        ...(groundingMetadataPerRequest != null
+          ? { groundingMetadataPerRequest }
+          : {}),
         urlContextMetadata: candidate?.urlContextMetadata ?? null,
         safetyRatings: candidate?.safetyRatings ?? null,
         usageMetadata: usageMetadata ?? null,
@@ -777,12 +786,15 @@ export class GoogleLanguageModel implements LanguageModelV4 {
     // 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 content: LanguageModelV4Content[] = [];
+    const groundingMetadataPerRequest: Array<GroundingMetadataSchema | null> =
+      [];
     let usageMetadata: GoogleUsageMetadata | undefined;
     const firstPromptTokenCount = response.usageMetadata?.promptTokenCount;
     let isContinued = false;
 
     while (true) {
       const candidate = response.candidates?.[0];
+      groundingMetadataPerRequest.push(candidate?.groundingMetadata ?? null);
       const continuationToken =
         candidate?.finishReason === 'CONTINUATION'
           ? candidate.continuationToken
@@ -824,6 +836,9 @@ export class GoogleLanguageModel implements LanguageModelV4 {
       toolNameMapping,
       jsonResponseToolName,
       initialPromptTokenCount: isContinued ? firstPromptTokenCount : undefined,
+      groundingMetadataPerRequest: isContinued
+        ? groundingMetadataPerRequest
+        : undefined,
     });
 
     const result = {
@@ -881,6 +896,10 @@ export class GoogleLanguageModel implements LanguageModelV4 {
     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
@@ -889,12 +908,22 @@ export class GoogleLanguageModel implements LanguageModelV4 {
       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: LanguageModelV4FinishReason = {
@@ -995,6 +1024,28 @@ export class GoogleLanguageModel implements LanguageModelV4 {
 
             const value = chunk.value;
 
+            if (value.error != null) {
+              finishReason = { unified: 'error', raw: undefined };
+              const error = value.error;
+              controller.enqueue({
+                type: 'error',
+                error: createProviderStreamError({
+                  message: error.message,
+                  type: error.status,
+                  code: error.code ?? undefined,
+                  statusCode: error.code ?? undefined,
+                  isRetryable:
+                    error.code != null &&
+                    (error.code === 408 ||
+                      error.code === 409 ||
+                      error.code === 429 ||
+                      error.code >= 500),
+                  data: error,
+                }),
+              });
+              return;
+            }
+
             if (!hasEmittedResponseMetadata && value.responseId != null) {
               hasEmittedResponseMetadata = true;
               controller.enqueue({
@@ -1463,6 +1514,14 @@ export class GoogleLanguageModel implements LanguageModelV4 {
               providerMetadata: wrapProviderMetadata({
                 promptFeedback,
                 groundingMetadata: lastGroundingMetadata,
+                ...(groundingMetadataPerRequest.length > 0
+                  ? {
+                      groundingMetadataPerRequest: [
+                        ...groundingMetadataPerRequest,
+                        requestGroundingMetadata,
+                      ],
+                    }
+                  : {}),
                 urlContextMetadata: lastUrlContextMetadata,
                 safetyRatings: lastSafetyRatings,
                 usageMetadata: usage ?? null,
@@ -2012,6 +2071,7 @@ export type UsageMetadataSchema = NonNullable<
 const chunkSchema = lazySchema(() =>
   zodSchema(
     z.object({
+      error: getGoogleErrorSchema().nullish(),
       responseId: z.string().nullish(),
       candidates: z
         .array(

package/src/google-prompt.ts

@@ -94,6 +94,8 @@ export type GoogleUsageMetadata = UsageMetadataSchema;
 export interface GoogleProviderMetadata {
   promptFeedback: GooglePromptFeedback | null;
   groundingMetadata: GoogleGroundingMetadata | null;
+  /** Grounding from each automatically continued request, in request order. */
+  groundingMetadataPerRequest?: Array<GoogleGroundingMetadata | null>;
   urlContextMetadata: GoogleUrlContextMetadata | null;
   safetyRatings: GoogleSafetyRating[] | null;
   usageMetadata: GoogleUsageMetadata | null;