@ai-sdk/google 3.0.75 → 3.0.77

package/dist/internal/index.mjs

@@ -643,17 +643,32 @@ var googleLanguageModelOptions = lazySchema2(
        */
       streamFunctionCallArguments: z2.boolean().optional(),
       /**
-       * Optional. The service tier to use for the request.
+       * Optional. The service tier to use for the request. Sent as the
+       * `serviceTier` body field. Gemini API only.
        */
-      serviceTier: z2.enum(["standard", "flex", "priority"]).optional()
+      serviceTier: z2.enum(["standard", "flex", "priority"]).optional(),
+      /**
+       * Optional. Vertex AI only. Sent as the
+       * `X-Vertex-AI-LLM-Shared-Request-Type` request header to select a
+       * shared (PayGo) tier. With Provisioned Throughput allocated and
+       * `requestType` unset, the request falls back to this tier only if
+       * PT capacity is exhausted.
+       *
+       * https://docs.cloud.google.com/vertex-ai/generative-ai/docs/priority-paygo
+       * https://docs.cloud.google.com/vertex-ai/generative-ai/docs/flex-paygo
+       */
+      sharedRequestType: z2.enum(["priority", "flex", "standard"]).optional(),
+      /**
+       * Optional. Vertex AI only. Sent as the `X-Vertex-AI-LLM-Request-Type`
+       * request header. Set to `'shared'` together with `sharedRequestType`
+       * to bypass Provisioned Throughput entirely.
+       *
+       * https://docs.cloud.google.com/vertex-ai/generative-ai/docs/priority-paygo
+       */
+      requestType: z2.enum(["shared"]).optional()
     })
   )
 );
-var VertexServiceTierMap = {
-  standard: "SERVICE_TIER_STANDARD",
-  flex: "SERVICE_TIER_FLEX",
-  priority: "SERVICE_TIER_PRIORITY"
-};
 
 // src/google-prepare-tools.ts
 import {
@@ -1220,10 +1235,27 @@ var GoogleGenerativeAILanguageModel = class {
         message: `'streamFunctionCallArguments' is only supported on the Vertex AI API and will be ignored with the current Google provider (${this.config.provider}). See https://docs.cloud.google.com/vertex-ai/generative-ai/docs/multimodal/function-calling#streaming-fc`
       });
     }
-    let sanitizedServiceTier = googleOptions == null ? void 0 : googleOptions.serviceTier;
     if ((googleOptions == null ? void 0 : googleOptions.serviceTier) && isVertexProvider) {
-      sanitizedServiceTier = VertexServiceTierMap[googleOptions.serviceTier];
+      warnings.push({
+        type: "other",
+        message: "'serviceTier' is a Gemini API option and is not supported on Vertex AI. Use 'sharedRequestType' (and optionally 'requestType') instead. See https://docs.cloud.google.com/vertex-ai/generative-ai/docs/priority-paygo"
+      });
     }
+    if (((googleOptions == null ? void 0 : googleOptions.sharedRequestType) || (googleOptions == null ? void 0 : googleOptions.requestType)) && !isVertexProvider) {
+      warnings.push({
+        type: "other",
+        message: `'sharedRequestType' and 'requestType' are Vertex AI options and are ignored with the current Google provider (${this.config.provider}).`
+      });
+    }
+    const vertexPaygoHeaders = isVertexProvider && ((googleOptions == null ? void 0 : googleOptions.sharedRequestType) || (googleOptions == null ? void 0 : googleOptions.requestType)) ? {
+      ...googleOptions.sharedRequestType && {
+        "X-Vertex-AI-LLM-Shared-Request-Type": googleOptions.sharedRequestType
+      },
+      ...googleOptions.requestType && {
+        "X-Vertex-AI-LLM-Request-Type": googleOptions.requestType
+      }
+    } : void 0;
+    const bodyServiceTier = isVertexProvider ? void 0 : googleOptions == null ? void 0 : googleOptions.serviceTier;
     const isGemmaModel = this.modelId.toLowerCase().startsWith("gemma-");
     const supportsFunctionResponseParts = this.modelId.startsWith("gemini-3");
     const { contents, systemInstruction } = convertToGoogleGenerativeAIMessages(
@@ -1295,18 +1327,20 @@ var GoogleGenerativeAILanguageModel = class {
         toolConfig,
         cachedContent: googleOptions == null ? void 0 : googleOptions.cachedContent,
         labels: googleOptions == null ? void 0 : googleOptions.labels,
-        serviceTier: sanitizedServiceTier
+        serviceTier: bodyServiceTier
       },
       warnings: [...warnings, ...toolWarnings],
-      providerOptionsName
+      providerOptionsName,
+      extraHeaders: vertexPaygoHeaders
     };
   }
   async doGenerate(options) {
     var _a, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n, _o, _p, _q, _r;
-    const { args, warnings, providerOptionsName } = await this.getArgs(options);
+    const { args, warnings, providerOptionsName, extraHeaders } = await this.getArgs(options);
     const mergedHeaders = combineHeaders(
       await resolve(this.config.headers),
-      options.headers
+      options.headers,
+      extraHeaders
     );
     const {
       responseHeaders,
@@ -1471,7 +1505,7 @@ var GoogleGenerativeAILanguageModel = class {
           safetyRatings: (_p = candidate.safetyRatings) != null ? _p : null,
           usageMetadata: usageMetadata != null ? usageMetadata : null,
           finishMessage: (_q = candidate.finishMessage) != null ? _q : null,
-          serviceTier: (_r = response.serviceTier) != null ? _r : null
+          serviceTier: (_r = responseHeaders == null ? void 0 : responseHeaders["x-gemini-service-tier"]) != null ? _r : null
         }
       },
       request: { body: args },
@@ -1483,13 +1517,12 @@ var GoogleGenerativeAILanguageModel = class {
     };
   }
   async doStream(options) {
-    const { args, warnings, providerOptionsName } = await this.getArgs(
-      options,
-      { isStreaming: true }
-    );
+    var _a;
+    const { args, warnings, providerOptionsName, extraHeaders } = await this.getArgs(options, { isStreaming: true });
     const headers = combineHeaders(
       await resolve(this.config.headers),
-      options.headers
+      options.headers,
+      extraHeaders
     );
     const { responseHeaders, value: response } = await postJsonToApi({
       url: `${this.config.baseURL}/${getModelPath(
@@ -1510,7 +1543,7 @@ var GoogleGenerativeAILanguageModel = class {
     let providerMetadata = void 0;
     let lastGroundingMetadata = null;
     let lastUrlContextMetadata = null;
-    let serviceTier = null;
+    const serviceTier = (_a = responseHeaders == null ? void 0 : responseHeaders["x-gemini-service-tier"]) != null ? _a : null;
     const generateId2 = this.config.generateId;
     let hasToolCalls = false;
     let currentTextBlockId = null;
@@ -1520,6 +1553,34 @@ var GoogleGenerativeAILanguageModel = class {
     let lastCodeExecutionToolCallId;
     let lastServerToolCallId;
     const activeStreamingToolCalls = [];
+    const finishActiveStreamingToolCall = (controller) => {
+      const active = activeStreamingToolCalls.pop();
+      if (active == null) {
+        return;
+      }
+      const { finalJSON, closingDelta } = active.accumulator.finalize();
+      if (closingDelta.length > 0) {
+        controller.enqueue({
+          type: "tool-input-delta",
+          id: active.toolCallId,
+          delta: closingDelta,
+          providerMetadata: active.providerMetadata
+        });
+      }
+      controller.enqueue({
+        type: "tool-input-end",
+        id: active.toolCallId,
+        providerMetadata: active.providerMetadata
+      });
+      controller.enqueue({
+        type: "tool-call",
+        toolCallId: active.toolCallId,
+        toolName: active.toolName,
+        input: finalJSON,
+        providerMetadata: active.providerMetadata
+      });
+      hasToolCalls = true;
+    };
     return {
       stream: response.pipeThrough(
         new TransformStream({
@@ -1527,7 +1588,7 @@ var GoogleGenerativeAILanguageModel = class {
             controller.enqueue({ type: "stream-start", warnings });
           },
           transform(chunk, controller) {
-            var _a, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n, _o;
+            var _a2, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n, _o;
             if (options.includeRawChunks) {
               controller.enqueue({ type: "raw", rawValue: chunk.rawValue });
             }
@@ -1540,10 +1601,7 @@ var GoogleGenerativeAILanguageModel = class {
             if (usageMetadata != null) {
               usage = usageMetadata;
             }
-            if (value.serviceTier != null) {
-              serviceTier = value.serviceTier;
-            }
-            const candidate = (_a = value.candidates) == null ? void 0 : _a[0];
+            const candidate = (_a2 = value.candidates) == null ? void 0 : _a2[0];
             if (candidate == null) {
               return;
             }
@@ -1732,7 +1790,7 @@ var GoogleGenerativeAILanguageModel = class {
                 const isCompleteCall = part.functionCall.name != null && part.functionCall.args != null && part.functionCall.partialArgs == null;
                 const isNoArgsCompleteCall = part.functionCall.name != null && part.functionCall.args == null && part.functionCall.partialArgs == null && part.functionCall.willContinue !== true;
                 if (isStreamingChunk) {
-                  if (part.functionCall.name != null && part.functionCall.willContinue === true) {
+                  if (part.functionCall.name != null) {
                     const toolCallId = (_i = part.functionCall.id) != null ? _i : generateId2();
                     const accumulator = new GoogleJSONAccumulator();
                     activeStreamingToolCalls.push({
@@ -1748,9 +1806,8 @@ var GoogleGenerativeAILanguageModel = class {
                       providerMetadata: providerMeta
                     });
                     if (part.functionCall.partialArgs != null) {
-                      const { textDelta } = accumulator.processPartialArgs(
-                        part.functionCall.partialArgs
-                      );
+                      const partialArgs = part.functionCall.partialArgs;
+                      const { textDelta } = accumulator.processPartialArgs(partialArgs);
                       if (textDelta.length > 0) {
                         controller.enqueue({
                           type: "tool-input-delta",
@@ -1759,12 +1816,14 @@ var GoogleGenerativeAILanguageModel = class {
                           providerMetadata: providerMeta
                         });
                       }
+                      if (part.functionCall.willContinue !== true && partialArgs.every((arg) => arg.willContinue !== true)) {
+                        finishActiveStreamingToolCall(controller);
+                      }
                     }
                   } else if (part.functionCall.partialArgs != null && activeStreamingToolCalls.length > 0) {
                     const active = activeStreamingToolCalls[activeStreamingToolCalls.length - 1];
-                    const { textDelta } = active.accumulator.processPartialArgs(
-                      part.functionCall.partialArgs
-                    );
+                    const partialArgs = part.functionCall.partialArgs;
+                    const { textDelta } = active.accumulator.processPartialArgs(partialArgs);
                     if (textDelta.length > 0) {
                       controller.enqueue({
                         type: "tool-input-delta",
@@ -1773,31 +1832,12 @@ var GoogleGenerativeAILanguageModel = class {
                         providerMetadata: providerMeta
                       });
                     }
+                    if (part.functionCall.willContinue !== true && partialArgs.every((arg) => arg.willContinue !== true)) {
+                      finishActiveStreamingToolCall(controller);
+                    }
                   }
                 } else if (isTerminalChunk && activeStreamingToolCalls.length > 0) {
-                  const active = activeStreamingToolCalls.pop();
-                  const { finalJSON, closingDelta } = active.accumulator.finalize();
-                  if (closingDelta.length > 0) {
-                    controller.enqueue({
-                      type: "tool-input-delta",
-                      id: active.toolCallId,
-                      delta: closingDelta,
-                      providerMetadata: active.providerMetadata
-                    });
-                  }
-                  controller.enqueue({
-                    type: "tool-input-end",
-                    id: active.toolCallId,
-                    providerMetadata: active.providerMetadata
-                  });
-                  controller.enqueue({
-                    type: "tool-call",
-                    toolCallId: active.toolCallId,
-                    toolName: active.toolName,
-                    input: finalJSON,
-                    providerMetadata: active.providerMetadata
-                  });
-                  hasToolCalls = true;
+                  finishActiveStreamingToolCall(controller);
                 } else if (isCompleteCall) {
                   const toolCallId = (_j = part.functionCall.id) != null ? _j : generateId2();
                   const toolName = part.functionCall.name;
@@ -2156,8 +2196,7 @@ var responseSchema = lazySchema3(
       promptFeedback: z3.object({
         blockReason: z3.string().nullish(),
         safetyRatings: z3.array(getSafetyRatingSchema()).nullish()
-      }).nullish(),
-      serviceTier: z3.string().nullish()
+      }).nullish()
     })
   )
 );
@@ -2178,8 +2217,7 @@ var chunkSchema = lazySchema3(
       promptFeedback: z3.object({
         blockReason: z3.string().nullish(),
         safetyRatings: z3.array(getSafetyRatingSchema()).nullish()
-      }).nullish(),
-      serviceTier: z3.string().nullish()
+      }).nullish()
     })
   )
 );