@ai-sdk/google 2.0.61 → 2.0.63

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/CHANGELOG.md CHANGED
@@ -1,5 +1,22 @@
1
1
  # @ai-sdk/google
2
2
 
3
+ ## 2.0.63
4
+
5
+ ### Patch Changes
6
+
7
+ - 7655bcf: feat(provider/google): Add multimodal tool-result support for Google function responses.
8
+
9
+ Tool results with `output.type = 'content'` now map media parts into
10
+ `functionResponse.parts` for Google models, including `image-data`,
11
+ `file-data`, and base64 `data:` URLs in URL-style content parts.
12
+ Remote HTTP(S) URLs in URL-style tool-result parts are not supported.
13
+
14
+ ## 2.0.62
15
+
16
+ ### Patch Changes
17
+
18
+ - a608d48: feat(provider/google): add `gemini-embedding-2-preview` and fix multimodal embedding support with `embedMany`
19
+
3
20
  ## 2.0.61
4
21
 
5
22
  ### Patch Changes
package/dist/index.d.mts CHANGED
@@ -180,18 +180,18 @@ declare const googleImageProviderOptionsSchema: _ai_sdk_provider_utils.LazySchem
180
180
  }>;
181
181
  type GoogleGenerativeAIImageProviderOptions = InferValidator<typeof googleImageProviderOptionsSchema>;
182
182
 
183
- type GoogleGenerativeAIEmbeddingModelId = 'gemini-embedding-001' | 'text-embedding-004' | (string & {});
183
+ type GoogleGenerativeAIEmbeddingModelId = 'gemini-embedding-001' | 'gemini-embedding-2-preview' | (string & {});
184
184
  declare const googleGenerativeAIEmbeddingProviderOptions: _ai_sdk_provider_utils.LazySchema<{
185
185
  outputDimensionality?: number | undefined;
186
186
  taskType?: "SEMANTIC_SIMILARITY" | "CLASSIFICATION" | "CLUSTERING" | "RETRIEVAL_DOCUMENT" | "RETRIEVAL_QUERY" | "QUESTION_ANSWERING" | "FACT_VERIFICATION" | "CODE_RETRIEVAL_QUERY" | undefined;
187
- content?: ({
187
+ content?: (({
188
188
  text: string;
189
189
  } | {
190
190
  inlineData: {
191
191
  mimeType: string;
192
192
  data: string;
193
193
  };
194
- })[] | undefined;
194
+ })[] | null)[] | undefined;
195
195
  }>;
196
196
  type GoogleGenerativeAIEmbeddingProviderOptions = InferValidator<typeof googleGenerativeAIEmbeddingProviderOptions>;
197
197
 
package/dist/index.d.ts CHANGED
@@ -180,18 +180,18 @@ declare const googleImageProviderOptionsSchema: _ai_sdk_provider_utils.LazySchem
180
180
  }>;
181
181
  type GoogleGenerativeAIImageProviderOptions = InferValidator<typeof googleImageProviderOptionsSchema>;
182
182
 
183
- type GoogleGenerativeAIEmbeddingModelId = 'gemini-embedding-001' | 'text-embedding-004' | (string & {});
183
+ type GoogleGenerativeAIEmbeddingModelId = 'gemini-embedding-001' | 'gemini-embedding-2-preview' | (string & {});
184
184
  declare const googleGenerativeAIEmbeddingProviderOptions: _ai_sdk_provider_utils.LazySchema<{
185
185
  outputDimensionality?: number | undefined;
186
186
  taskType?: "SEMANTIC_SIMILARITY" | "CLASSIFICATION" | "CLUSTERING" | "RETRIEVAL_DOCUMENT" | "RETRIEVAL_QUERY" | "QUESTION_ANSWERING" | "FACT_VERIFICATION" | "CODE_RETRIEVAL_QUERY" | undefined;
187
- content?: ({
187
+ content?: (({
188
188
  text: string;
189
189
  } | {
190
190
  inlineData: {
191
191
  mimeType: string;
192
192
  data: string;
193
193
  };
194
- })[] | undefined;
194
+ })[] | null)[] | undefined;
195
195
  }>;
196
196
  type GoogleGenerativeAIEmbeddingProviderOptions = InferValidator<typeof googleGenerativeAIEmbeddingProviderOptions>;
197
197
 
package/dist/index.js CHANGED
@@ -18,19 +18,19 @@ var __copyProps = (to, from, except, desc) => {
18
18
  var __toCommonJS = (mod) => __copyProps(__defProp({}, "__esModule", { value: true }), mod);
19
19
 
20
20
  // src/index.ts
21
- var src_exports = {};
22
- __export(src_exports, {
21
+ var index_exports = {};
22
+ __export(index_exports, {
23
23
  VERSION: () => VERSION,
24
24
  createGoogleGenerativeAI: () => createGoogleGenerativeAI,
25
25
  google: () => google
26
26
  });
27
- module.exports = __toCommonJS(src_exports);
27
+ module.exports = __toCommonJS(index_exports);
28
28
 
29
29
  // src/google-provider.ts
30
30
  var import_provider_utils15 = require("@ai-sdk/provider-utils");
31
31
 
32
32
  // src/version.ts
33
- var VERSION = true ? "2.0.61" : "0.0.0-test";
33
+ var VERSION = true ? "2.0.63" : "0.0.0-test";
34
34
 
35
35
  // src/google-generative-ai-embedding-model.ts
36
36
  var import_provider = require("@ai-sdk/provider");
@@ -99,11 +99,15 @@ var googleGenerativeAIEmbeddingProviderOptions = (0, import_provider_utils2.lazy
99
99
  "CODE_RETRIEVAL_QUERY"
100
100
  ]).optional(),
101
101
  /**
102
- * Optional. Multimodal content parts for embedding non-text content
103
- * (images, video, PDF, audio). When provided, these parts are merged
104
- * with the text values in the embedding request.
102
+ * Optional. Per-value multimodal content parts for embedding non-text
103
+ * content (images, video, PDF, audio). Each entry corresponds to the
104
+ * embedding value at the same index and its parts are merged with the
105
+ * text value in the request. Use `null` for entries that are text-only.
106
+ *
107
+ * The array length must match the number of values being embedded. In
108
+ * the case of a single embedding, the array length must be 1.
105
109
  */
106
- content: import_v42.z.array(googleEmbeddingContentPartSchema).min(1).optional()
110
+ content: import_v42.z.array(import_v42.z.array(googleEmbeddingContentPartSchema).min(1).nullable()).optional()
107
111
  })
108
112
  )
109
113
  );
@@ -126,7 +130,6 @@ var GoogleGenerativeAIEmbeddingModel = class {
126
130
  abortSignal,
127
131
  providerOptions
128
132
  }) {
129
- var _a;
130
133
  const googleOptions = await (0, import_provider_utils3.parseProviderOptions)({
131
134
  provider: "google",
132
135
  providerOptions,
@@ -144,10 +147,16 @@ var GoogleGenerativeAIEmbeddingModel = class {
144
147
  await (0, import_provider_utils3.resolve)(this.config.headers),
145
148
  headers
146
149
  );
147
- const multimodalContent = (_a = googleOptions == null ? void 0 : googleOptions.content) != null ? _a : [];
150
+ const multimodalContent = googleOptions == null ? void 0 : googleOptions.content;
151
+ if (multimodalContent != null && multimodalContent.length !== values.length) {
152
+ throw new Error(
153
+ `The number of multimodal content entries (${multimodalContent.length}) must match the number of values (${values.length}).`
154
+ );
155
+ }
148
156
  if (values.length === 1) {
157
+ const valueParts = multimodalContent == null ? void 0 : multimodalContent[0];
149
158
  const textPart = values[0] ? [{ text: values[0] }] : [];
150
- const parts = multimodalContent.length > 0 ? [...textPart, ...multimodalContent] : [{ text: values[0] }];
159
+ const parts = valueParts != null ? [...textPart, ...valueParts] : [{ text: values[0] }];
151
160
  const {
152
161
  responseHeaders: responseHeaders2,
153
162
  value: response2,
@@ -184,13 +193,14 @@ var GoogleGenerativeAIEmbeddingModel = class {
184
193
  url: `${this.config.baseURL}/models/${this.modelId}:batchEmbedContents`,
185
194
  headers: mergedHeaders,
186
195
  body: {
187
- requests: values.map((value) => {
196
+ requests: values.map((value, index) => {
197
+ const valueParts = multimodalContent == null ? void 0 : multimodalContent[index];
188
198
  const textPart = value ? [{ text: value }] : [];
189
199
  return {
190
200
  model: `models/${this.modelId}`,
191
201
  content: {
192
202
  role: "user",
193
- parts: multimodalContent.length > 0 ? [...textPart, ...multimodalContent] : [{ text: value }]
203
+ parts: valueParts != null ? [...textPart, ...valueParts] : [{ text: value }]
194
204
  },
195
205
  outputDimensionality: googleOptions == null ? void 0 : googleOptions.outputDimensionality,
196
206
  taskType: googleOptions == null ? void 0 : googleOptions.taskType
@@ -350,11 +360,12 @@ function isEmptyObjectSchema(jsonSchema) {
350
360
  var import_provider2 = require("@ai-sdk/provider");
351
361
  var import_provider_utils4 = require("@ai-sdk/provider-utils");
352
362
  function convertToGoogleGenerativeAIMessages(prompt, options) {
353
- var _a;
363
+ var _a, _b;
354
364
  const systemInstructionParts = [];
355
365
  const contents = [];
356
366
  let systemMessagesAllowed = true;
357
367
  const isGemmaModel = (_a = options == null ? void 0 : options.isGemmaModel) != null ? _a : false;
368
+ const supportsFunctionResponseParts = (_b = options == null ? void 0 : options.supportsFunctionResponseParts) != null ? _b : true;
358
369
  for (const { role, content } of prompt) {
359
370
  switch (role) {
360
371
  case "system": {
@@ -402,8 +413,8 @@ function convertToGoogleGenerativeAIMessages(prompt, options) {
402
413
  contents.push({
403
414
  role: "model",
404
415
  parts: content.map((part) => {
405
- var _a2, _b, _c;
406
- const thoughtSignature = ((_b = (_a2 = part.providerOptions) == null ? void 0 : _a2.google) == null ? void 0 : _b.thoughtSignature) != null ? String((_c = part.providerOptions.google) == null ? void 0 : _c.thoughtSignature) : void 0;
416
+ var _a2, _b2, _c;
417
+ const thoughtSignature = ((_b2 = (_a2 = part.providerOptions) == null ? void 0 : _a2.google) == null ? void 0 : _b2.thoughtSignature) != null ? String((_c = part.providerOptions.google) == null ? void 0 : _c.thoughtSignature) : void 0;
407
418
  switch (part.type) {
408
419
  case "text": {
409
420
  return part.text.length === 0 ? void 0 : {
@@ -456,36 +467,10 @@ function convertToGoogleGenerativeAIMessages(prompt, options) {
456
467
  for (const part of content) {
457
468
  const output = part.output;
458
469
  if (output.type === "content") {
459
- for (const contentPart of output.value) {
460
- switch (contentPart.type) {
461
- case "text":
462
- parts.push({
463
- functionResponse: {
464
- name: part.toolName,
465
- response: {
466
- name: part.toolName,
467
- content: contentPart.text
468
- }
469
- }
470
- });
471
- break;
472
- case "media":
473
- parts.push(
474
- {
475
- inlineData: {
476
- mimeType: contentPart.mediaType,
477
- data: contentPart.data
478
- }
479
- },
480
- {
481
- text: "Tool executed successfully and returned this image as a response"
482
- }
483
- );
484
- break;
485
- default:
486
- parts.push({ text: JSON.stringify(contentPart) });
487
- break;
488
- }
470
+ if (supportsFunctionResponseParts) {
471
+ appendToolResultParts({ parts, part, output });
472
+ } else {
473
+ appendLegacyToolResultParts({ parts, part, output });
489
474
  }
490
475
  } else {
491
476
  parts.push({
@@ -516,6 +501,77 @@ function convertToGoogleGenerativeAIMessages(prompt, options) {
516
501
  contents
517
502
  };
518
503
  }
504
+ function appendToolResultParts({
505
+ parts,
506
+ part,
507
+ output
508
+ }) {
509
+ const responseTextParts = [];
510
+ const functionResponseParts = [];
511
+ for (const contentPart of output.value) {
512
+ switch (contentPart.type) {
513
+ case "text":
514
+ responseTextParts.push(contentPart.text);
515
+ break;
516
+ case "media":
517
+ functionResponseParts.push({
518
+ inlineData: {
519
+ mimeType: contentPart.mediaType,
520
+ data: contentPart.data
521
+ }
522
+ });
523
+ break;
524
+ }
525
+ }
526
+ const responseText = responseTextParts.length > 0 ? responseTextParts.join("\n") : "Tool executed successfully.";
527
+ parts.push({
528
+ functionResponse: {
529
+ name: part.toolName,
530
+ response: {
531
+ name: part.toolName,
532
+ content: responseText
533
+ },
534
+ ...functionResponseParts.length > 0 ? { parts: functionResponseParts } : {}
535
+ }
536
+ });
537
+ }
538
+ function appendLegacyToolResultParts({
539
+ parts,
540
+ part,
541
+ output
542
+ }) {
543
+ for (const contentPart of output.value) {
544
+ switch (contentPart.type) {
545
+ case "text":
546
+ parts.push({
547
+ functionResponse: {
548
+ name: part.toolName,
549
+ response: {
550
+ name: part.toolName,
551
+ content: contentPart.text
552
+ }
553
+ }
554
+ });
555
+ break;
556
+ case "media":
557
+ parts.push(
558
+ {
559
+ inlineData: {
560
+ mimeType: contentPart.mediaType,
561
+ data: contentPart.data
562
+ }
563
+ },
564
+ {
565
+ text: "Tool executed successfully and returned this image as a response"
566
+ }
567
+ );
568
+ break;
569
+ default:
570
+ parts.push({ text: JSON.stringify(contentPart) });
571
+ break;
572
+ }
573
+ }
574
+ }
519
575
 
520
576
  // src/get-model-path.ts
521
577
  function getModelPath(modelId) {
@@ -916,9 +972,10 @@ var GoogleGenerativeAILanguageModel = class {
916
972
  });
917
973
  }
918
974
  const isGemmaModel = this.modelId.toLowerCase().startsWith("gemma-");
975
+ const supportsFunctionResponseParts = this.modelId.startsWith("gemini-3");
919
976
  const { contents, systemInstruction } = convertToGoogleGenerativeAIMessages(
920
977
  prompt,
921
- { isGemmaModel }
978
+ { isGemmaModel, supportsFunctionResponseParts }
922
979
  );
923
980
  const {
924
981
  tools: googleTools2,