@ai-sdk/google 3.0.79 → 3.0.81

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

@@ -248,8 +248,14 @@ The following optional provider options are available for Google Generative AI m
 - **serviceTier** _'standard' | 'flex' | 'priority'_
 
   Optional. The service tier to use for the request.
-  Set to 'flex' for 50% cheaper processing at the cost of increased latency.
-  Set to 'priority' for ultra-low latency at a 75-100% price premium over 'standard'.
+  Set to `'flex'` for 50% cheaper processing at the cost of increased latency.
+  Set to `'priority'` for ultra-low latency at a 75-100% price premium over `'standard'`.
+
+  Because Priority can be gracefully downgraded to Standard under load, the
+  tier the request actually ran on is surfaced on
+  `result.providerMetadata.google.serviceTier`. See
+  [Priority inference](https://ai.google.dev/gemini-api/docs/priority-inference)
+  and [Flex inference](https://ai.google.dev/gemini-api/docs/flex-inference).
 
 - **threshold** _string_
 
@@ -1090,8 +1096,9 @@ The following Zod features are known to not work with Google Generative AI:
 
 The [Gemini Interactions API](https://ai.google.dev/gemini-api/docs/interactions)
 (`POST /v1beta/interactions`) is a separate Google endpoint with server-side
-state, unified content blocks, first-class built-in tools, agent presets, and
-native multimodal image output. It is reached via the `google.interactions(...)`
+state, unified content blocks, first-class built-in tools, agent presets,
+managed agents that run in a sandboxed Linux environment, and native
+multimodal image output. It is reached via the `google.interactions(...)`
 factory:
 
 ```ts
@@ -1104,10 +1111,12 @@ const { text } = await generateText({
 });
 ```
 
-`google.interactions(...)` accepts either a model ID string (e.g.
-`'gemini-2.5-flash'`, `'gemini-3-pro-preview'`) or `{ agent: <name> }` to use
-a Gemini [agent preset](#agent-presets). The returned model can be passed to
-`generateText` and `streamText` like any other AI SDK language model.
+`google.interactions(...)` accepts a model ID string (e.g.
+`'gemini-2.5-flash'`, `'gemini-3-pro-preview'`), `{ agent: <name> }` to use
+a Gemini [agent preset](#agent-presets), or `{ managedAgent: <name> }` to
+invoke a [managed agent](#managed-agents) you created on Google's side.
+The returned model can be passed to `generateText` and `streamText` like
+any other AI SDK language model.
 
 <Note>
   Use `google(...)` for the standard `:generateContent` /
@@ -1213,6 +1222,22 @@ The following optional provider options are available:
   Alternative to the AI SDK `system` message. If both are set, the AI SDK
   `system` message wins and a warning is emitted.
 
+- **background** _boolean_
+
+  Run the interaction in the background. Required for agents whose
+  server-side workflow cannot complete within a single request/response;
+  rejected by agents that only support synchronous calls. When `true`,
+  the POST returns a non-terminal status and the SDK polls
+  `GET /interactions/{id}` until the work completes.
+
+- **environment** _string \| object_
+
+  Sandbox environment configuration for [managed agents](#managed-agents).
+  Pass `'remote'` to provision a fresh sandbox, an `environment_id`
+  string to reuse an existing one, or an object of the form
+  `{ type: 'remote', sources?, network? }` to preload files and/or
+  constrain outbound traffic. Only applies to agent calls.
+
 - **pollingTimeoutMs** _number_
 
   Maximum time, in milliseconds, to poll a background interaction (agent
@@ -1466,7 +1491,10 @@ Pass `{ agent: <name> }` to target a Gemini agent preset. The factory
 type-checks the agent name against the supported set:
 
 ```ts
-import { google } from '@ai-sdk/google';
+import {
+  google,
+  type GoogleLanguageModelInteractionsOptions,
+} from '@ai-sdk/google';
 import { generateText } from 'ai';
 
 const result = await generateText({
@@ -1475,28 +1503,143 @@ const result = await generateText({
   }),
   prompt:
     'Briefly summarize the most-cited papers on retrieval-augmented generation since 2024 (2-3 sentences).',
+  providerOptions: {
+    google: {
+      background: true,
+    } satisfies GoogleLanguageModelInteractionsOptions,
+  },
 });
 ```
 
-Agent calls run with `background: true` on the wire and the SDK polls the
-`GET /interactions/{id}` endpoint internally until the interaction
-completes. The default polling timeout is 30 minutes; raise it via
+Whether an agent runs synchronously or in the background depends on the
+agent. Long-running presets (such as the `deep-research-*` family)
+require `background: true` — the POST returns a non-terminal status and
+the SDK polls `GET /interactions/{id}` internally until the interaction
+completes. Other agents accept synchronous calls only and will reject
+`background: true`. Set the flag explicitly via
+`providerOptions.google.background`.
+
+The default polling timeout is 30 minutes; raise it via
 `pollingTimeoutMs` for slower agents:
 
 ```ts
+import {
+  google,
+  type GoogleLanguageModelInteractionsOptions,
+} from '@ai-sdk/google';
+import { generateText } from 'ai';
+
 await generateText({
   model: google.interactions({ agent: 'deep-research-max-preview-04-2026' }),
   prompt: 'Produce a long-form research brief on ...',
   providerOptions: {
     google: {
+      background: true,
       pollingTimeoutMs: 60 * 60 * 1000, // 1 hour
-    },
+    } satisfies GoogleLanguageModelInteractionsOptions,
   },
 });
 ```
 
 Agents also chain through `previousInteractionId` like model-id calls.
 
+### Managed Agents
+
+[Managed agents](https://ai.google.dev/gemini-api/docs/agents) run inside a
+sandboxed Linux environment provisioned per interaction. Pass the `environment`
+provider option to control how the sandbox is set up; the option is only
+accepted on agent calls.
+
+The simplest form provisions a fresh sandbox:
+
+```ts
+import {
+  google,
+  type GoogleLanguageModelInteractionsOptions,
+} from '@ai-sdk/google';
+import { generateText } from 'ai';
+
+const result = await generateText({
+  model: google.interactions({ agent: 'antigravity-preview-05-2026' }),
+  prompt: 'What is 2 + 2?',
+  providerOptions: {
+    google: {
+      environment: 'remote',
+    } satisfies GoogleLanguageModelInteractionsOptions,
+  },
+});
+```
+
+`environment` accepts three shapes:
+
+- `'remote'` — provision a fresh sandbox for this call.
+- any other string — an `environment_id` to reuse, forking the previous
+  sandbox so its filesystem and installed packages persist.
+- an object — provision a fresh sandbox and optionally preload `sources`
+  and/or constrain outbound traffic via `network`:
+
+```ts
+import {
+  google,
+  type GoogleLanguageModelInteractionsOptions,
+} from '@ai-sdk/google';
+import { generateText } from 'ai';
+
+await generateText({
+  model: google.interactions({ agent: 'antigravity-preview-05-2026' }),
+  prompt:
+    'Read the file at /data/note.txt and tell me exactly what it contains.',
+  providerOptions: {
+    google: {
+      environment: {
+        type: 'remote',
+        sources: [
+          {
+            type: 'inline',
+            content: 'hello from the AI SDK example\n',
+            target: '/data/note.txt',
+          },
+        ],
+      },
+    } satisfies GoogleLanguageModelInteractionsOptions,
+  },
+});
+```
+
+Three source types are supported: `inline` (write a string into the
+sandbox at `target`), `repository` (clone a git repository — pass the
+URL as `source`), and `gcs` (mount a Google Cloud Storage prefix).
+
+The `network` field accepts the string `'disabled'` to block all
+outbound traffic, or an object with an `allowlist` array whose entries
+each carry a `domain` plus an optional `transform` array of header
+objects to inject into matching requests.
+
+#### Custom managed agents
+
+For user-defined agents that you created on Google's side via the
+Gemini API's `/v1beta/agents` endpoint, pass the agent's name through the dedicated
+`managedAgent` factory shape instead of `agent` (which only accepts
+known preset names):
+
+```ts
+import {
+  google,
+  type GoogleLanguageModelInteractionsOptions,
+} from '@ai-sdk/google';
+import { generateText } from 'ai';
+
+const result = await generateText({
+  model: google.interactions({ managedAgent: 'my-custom-agent' }),
+  prompt: 'Hello!',
+  providerOptions: {
+    google: {
+      environment: 'remote',
+    } satisfies GoogleLanguageModelInteractionsOptions,
+  },
+});
+```
+
 ### Streaming
 
 `streamText` is supported and consumes the seven Interactions SSE event
@@ -1522,22 +1665,6 @@ const googleMetadata = (await result.providerMetadata)?.google;
 console.log('Interaction id:', googleMetadata?.interactionId);
 ```
 
-### Runnable Examples
-
-Paired `generateText` + `streamText` examples live under:
-
-- `examples/ai-functions/src/generate-text/google/interactions-*.ts`
-- `examples/ai-functions/src/stream-text/google/interactions-*.ts`
-
-Notable examples: `interactions-basic`, `interactions-multi-turn-stateful`,
-`interactions-multi-turn-stateless`, `interactions-tool-call`,
-`interactions-google-search`, `interactions-image-output`,
-`interactions-image-output-modify`, `interactions-image-base64`,
-`interactions-image-reference`, `interactions-image-url`,
-`interactions-pdf`, `interactions-structured-output`,
-`interactions-service-tier`, `interactions-agent-single-turn`, and
-`interactions-agent-multi-turn`.
-
 ## Gemma Models
 
 You can use [Gemma models](https://deepmind.google/models/gemma/) with the Google Generative AI API.
@@ -1768,6 +1895,29 @@ const { image } = await generateImage({
   details.
 </Note>
 
+#### Google Search Grounding
+
+Gemini image models support [Google Search grounding](#google-search) through `providerOptions.google.googleSearch`. The value matches the args of `google.tools.googleSearch(...)`; pass `{}` to enable with defaults, or `{ searchTypes: { imageSearch: {} } }` to ground on reference photos.
+
+```ts
+import { google } from '@ai-sdk/google';
+import { generateImage } from 'ai';
+
+const result = await generateImage({
+  model: google.image('gemini-3.1-flash-image-preview'),
+  prompt:
+    'Search for live footage of the 2026 Super Bowl halftime show artist, then generate a close-up in space.',
+  providerOptions: {
+    google: {
+      googleSearch: { searchTypes: { imageSearch: {} } },
+    },
+  },
+});
+
+// Grounding metadata is forwarded onto the image result:
+console.log(result.providerMetadata?.google?.groundingMetadata);
+```
+
 #### Gemini Image Model Capabilities
 
 | Model                            | Image Generation    | Image Editing       | Aspect Ratios                                       |

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ai-sdk/google",
-  "version": "3.0.79",
+  "version": "3.0.81",
   "license": "Apache-2.0",
   "sideEffects": false,
   "main": "./dist/index.js",
@@ -36,8 +36,8 @@
     }
   },
   "dependencies": {
-    "@ai-sdk/provider": "3.0.10",
-    "@ai-sdk/provider-utils": "4.0.27"
+    "@ai-sdk/provider-utils": "4.0.28",
+    "@ai-sdk/provider": "3.0.10"
   },
   "devDependencies": {
     "@types/node": "20.17.24",

package/src/convert-to-google-generative-ai-messages.ts

@@ -1,6 +1,7 @@
 import {
   UnsupportedFunctionalityError,
   type LanguageModelV3Prompt,
+  type SharedV3Warning,
 } from '@ai-sdk/provider';
 import { convertToBase64 } from '@ai-sdk/provider-utils';
 import type {
@@ -10,6 +11,48 @@ import type {
   GoogleGenerativeAIPrompt,
 } from './google-generative-ai-prompt';
 
+/**
+ * Sentinel value Google documents for replaying functionCall parts whose
+ * original thoughtSignature is not available to the client.
+ *
+ * Gemini 3 models reject `functionCall` parts that lack a `thoughtSignature`
+ * with HTTP 400 "Function call is missing a thought_signature in functionCall
+ * parts." Sending this sentinel string in place of the missing signature
+ * makes Gemini skip the validator and continue the turn.
+ *
+ * See https://ai.google.dev/gemini-api/docs/thought-signatures.
+ */
+export const SKIP_THOUGHT_SIGNATURE_VALIDATOR =
+  'skip_thought_signature_validator';
+
+type GoogleProviderOptions = {
+  thought?: unknown;
+  thoughtSignature?: unknown;
+  serverToolCallId?: unknown;
+  serverToolType?: unknown;
+};
+
+function getGoogleProviderOptions(
+  providerOptions: Record<string, GoogleProviderOptions> | undefined,
+  providerOptionsName: string,
+): GoogleProviderOptions | undefined {
+  const namespaces = [
+    providerOptionsName,
+    'google',
+    'googleVertex',
+    'vertex',
+  ].filter((namespace, index, allNamespaces) => {
+    return allNamespaces.indexOf(namespace) === index;
+  });
+
+  for (const namespace of namespaces) {
+    const options = providerOptions?.[namespace];
+    if (options != null) {
+      return options;
+    }
+  }
+}
+
 const dataUrlRegex = /^data:([^;,]+);base64,(.+)$/s;
 
 function parseBase64DataUrl(
@@ -168,17 +211,41 @@ export function convertToGoogleGenerativeAIMessages(
   prompt: LanguageModelV3Prompt,
   options?: {
     isGemmaModel?: boolean;
+    /**
+     * Whether the target model is in the Gemini 3 family. Gemini 3 enforces a
+     * `thoughtSignature` on every replayed `functionCall` part; when one is
+     * missing we inject the documented `skip_thought_signature_validator`
+     * sentinel and emit a warning via `onWarning` so the developer can find
+     * and fix the upstream serialization that lost the signature.
+     */
+    isGemini3Model?: boolean;
     providerOptionsName?: string;
     supportsFunctionResponseParts?: boolean;
+    /**
+     * Called once for the request when a Gemini 3 `functionCall` part is
+     * about to be sent without a `thoughtSignature` and the sentinel is
+     * injected.
+     */
+    onWarning?: (warning: SharedV3Warning) => void;
   },
 ): GoogleGenerativeAIPrompt {
   const systemInstructionParts: Array<{ text: string }> = [];
   const contents: Array<GoogleGenerativeAIContent> = [];
   let systemMessagesAllowed = true;
   const isGemmaModel = options?.isGemmaModel ?? false;
+  const isGemini3Model = options?.isGemini3Model ?? false;
   const providerOptionsName = options?.providerOptionsName ?? 'google';
   const supportsFunctionResponseParts =
     options?.supportsFunctionResponseParts ?? true;
+  const onWarning = options?.onWarning;
+
+  let sentinelInjected = false;
+  const missingSignatureToolNames: string[] = [];
+  const injectSkipSignature = (toolName: string) => {
+    missingSignatureToolNames.push(toolName);
+    sentinelInjected = true;
+    return SKIP_THOUGHT_SIGNATURE_VALIDATOR;
+  };
 
   for (const { role, content } of prompt) {
     switch (role) {
@@ -243,11 +310,10 @@ export function convertToGoogleGenerativeAIMessages(
           role: 'model',
           parts: content
             .map(part => {
-              const providerOpts =
-                part.providerOptions?.[providerOptionsName] ??
-                (providerOptionsName !== 'google'
-                  ? part.providerOptions?.google
-                  : part.providerOptions?.vertex);
+              const providerOpts = getGoogleProviderOptions(
+                part.providerOptions,
+                providerOptionsName,
+              );
               const thoughtSignature =
                 providerOpts?.thoughtSignature != null
                   ? String(providerOpts.thoughtSignature)
@@ -303,6 +369,16 @@ export function convertToGoogleGenerativeAIMessages(
                       ? String(providerOpts.serverToolType)
                       : undefined;
 
+                  // For Gemini 3, every replayed functionCall part must carry a
+                  // thoughtSignature or the API returns HTTP 400. If the upstream
+                  // serialization layer dropped the signature, inject the
+                  // documented sentinel so the request still succeeds.
+                  const effectiveThoughtSignature =
+                    thoughtSignature ??
+                    (isGemini3Model
+                      ? injectSkipSignature(part.toolName)
+                      : undefined);
+
                   if (serverToolCallId && serverToolType) {
                     return {
                       toolCall: {
@@ -313,7 +389,7 @@ export function convertToGoogleGenerativeAIMessages(
                             : part.input,
                         id: serverToolCallId,
                       },
-                      thoughtSignature,
+                      thoughtSignature: effectiveThoughtSignature,
                     };
                   }
 
@@ -325,7 +401,7 @@ export function convertToGoogleGenerativeAIMessages(
                       name: part.toolName,
                       args: part.input,
                     },
-                    thoughtSignature,
+                    thoughtSignature: effectiveThoughtSignature,
                   };
                 }
 
@@ -371,11 +447,10 @@ export function convertToGoogleGenerativeAIMessages(
             continue;
           }
 
-          const partProviderOpts =
-            part.providerOptions?.[providerOptionsName] ??
-            (providerOptionsName !== 'google'
-              ? part.providerOptions?.google
-              : part.providerOptions?.vertex);
+          const partProviderOpts = getGoogleProviderOptions(
+            part.providerOptions,
+            providerOptionsName,
+          );
           const serverToolCallId =
             partProviderOpts?.serverToolCallId != null
               ? String(partProviderOpts.serverToolCallId)
@@ -465,6 +540,23 @@ export function convertToGoogleGenerativeAIMessages(
     contents[0].parts.unshift({ text: systemText + '\n\n' });
   }
 
+  if (sentinelInjected && onWarning != null) {
+    const uniqueToolNames = Array.from(new Set(missingSignatureToolNames));
+    onWarning({
+      type: 'other',
+      message:
+        `Replayed ${missingSignatureToolNames.length} \`functionCall\` part(s) ` +
+        `for a Gemini 3 model without a \`thoughtSignature\` ` +
+        `(tools: ${uniqueToolNames.map(name => `\`${name}\``).join(', ')}). ` +
+        `Injected the documented \`skip_thought_signature_validator\` sentinel ` +
+        `to keep the request from failing with HTTP 400. ` +
+        `The likely cause is application code that drops ` +
+        '`providerOptions.google.thoughtSignature` when persisting or ' +
+        'serializing assistant tool-call messages. ' +
+        'See https://ai.google.dev/gemini-api/docs/thought-signatures.',
+    });
+  }
+
   return {
     systemInstruction:
       systemInstructionParts.length > 0 && !isGemmaModel

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

@@ -193,14 +193,17 @@ export class GoogleGenerativeAILanguageModel implements LanguageModelV3 {
       : googleOptions?.serviceTier;
 
     const isGemmaModel = this.modelId.toLowerCase().startsWith('gemma-');
-    const supportsFunctionResponseParts = this.modelId.startsWith('gemini-3');
+    const isGemini3Model = /^gemini-3[.-]/.test(this.modelId);
+    const supportsFunctionResponseParts = isGemini3Model;
 
     const { contents, systemInstruction } = convertToGoogleGenerativeAIMessages(
       prompt,
       {
         isGemmaModel,
+        isGemini3Model,
         providerOptionsName,
         supportsFunctionResponseParts,
+        onWarning: warning => warnings.push(warning),
       },
     );