@core-ai/google-genai 0.18.0 → 0.19.0

package/dist/index.d.ts

@@ -1,7 +1,18 @@
 import { GoogleGenAI } from '@google/genai';
-import { ChatModel, EmbeddingModel, ImageModel, ModelCapabilities } from '@core-ai/core-ai';
+import { ModelCapabilities, ChatModel, EmbeddingModel, ImageModel } from '@core-ai/core-ai';
 import { z } from 'zod';
 
+type GoogleModelCapabilities = Omit<ModelCapabilities, 'reasoning'> & {
+    reasoning: ModelCapabilities['reasoning'] & {
+        thinkingParam: 'thinkingLevel' | 'thinkingBudget';
+    };
+};
+declare function getGoogleModelCapabilities(modelId: string): GoogleModelCapabilities;
+
+type GoogleReasoningMetadata = {
+    thoughtSignature?: string;
+};
+
 type GoogleGenAIClient = {
     models: GoogleGenAI['models'];
 };
@@ -23,17 +34,6 @@ type GoogleGenAIProvider = {
 declare function createGoogleGenAIProvider(options?: GoogleGenAIProviderBaseOptions, factoryOptions?: GoogleGenAIProviderFactoryOptions): GoogleGenAIProvider;
 declare function createGoogleGenAI(options?: GoogleGenAIProviderOptions): GoogleGenAIProvider;
 
-type GoogleReasoningMetadata = {
-    thoughtSignature?: string;
-};
-
-type GoogleModelCapabilities = {
-    reasoning: ModelCapabilities['reasoning'] & {
-        thinkingParam: 'thinkingLevel' | 'thinkingBudget';
-    };
-};
-declare function getGoogleModelCapabilities(modelId: string): GoogleModelCapabilities;
-
 declare const googleGenerateProviderOptionsSchema: z.ZodObject<{
     stopSequences: z.ZodOptional<z.ZodArray<z.ZodString>>;
     frequencyPenalty: z.ZodOptional<z.ZodNumber>;

package/dist/index.js

@@ -1,20 +1,15 @@
 // src/provider.ts
 import { GoogleGenAI } from "@google/genai";
 
-// src/chat-model.ts
-import {
-  StructuredOutputNoObjectGeneratedError,
-  StructuredOutputParseError,
-  StructuredOutputValidationError,
-  createObjectStream,
-  createChatStream
-} from "@core-ai/core-ai";
-
 // src/chat-adapter.ts
 import {
   FunctionCallingConfigMode
 } from "@google/genai";
-import { getProviderMetadata, zodSchemaToJsonSchema } from "@core-ai/core-ai";
+import {
+  getProviderMetadata,
+  validateInputModalities,
+  zodSchemaToJsonSchema
+} from "@core-ai/core-ai";
 
 // src/model-capabilities.ts
 import {
@@ -27,6 +22,10 @@ var ALL_EFFORTS = [
   "high",
   "max"
 ];
+var GOOGLE_INPUT_MODALITIES = {
+  input: ["text", "image", "file", "audio"],
+  output: ["text"]
+};
 function createCapabilities(config) {
   return {
     reasoning: {
@@ -35,7 +34,8 @@ function createCapabilities(config) {
       restrictsSamplingParams: false,
       supportedToolChoices: ["auto", "none", "required", "tool"],
       thinkingParam: config.thinkingParam
-    }
+    },
+    modalities: GOOGLE_INPUT_MODALITIES
   };
 }
 var DEFAULT_CAPABILITIES = createCapabilities({
@@ -151,6 +151,7 @@ function parseGoogleImageProviderOptions(providerOptions) {
 var googleProviderOptionsSchema = googleGenerateProviderOptionsSchema;
 
 // src/chat-adapter.ts
+var DEFAULT_PROVIDER_ID = "google";
 var DEFAULT_STRUCTURED_OUTPUT_TOOL_NAME = "core_ai_generate_object";
 var DEFAULT_STRUCTURED_OUTPUT_TOOL_DESCRIPTION = "Return a JSON object that matches the requested schema.";
 function convertMessages(messages) {
@@ -254,6 +255,14 @@ function convertUserContentPart(part) {
       }
     };
   }
+  if (part.type === "audio") {
+    return {
+      inlineData: {
+        data: part.source.data,
+        mimeType: part.source.mediaType
+      }
+    };
+  }
   return {
     inlineData: {
       data: part.data,
@@ -361,17 +370,24 @@ function inferMimeTypeFromUrl(url) {
   }
   return "application/octet-stream";
 }
-function createGenerateRequest(modelId, options) {
+function createGenerateRequest(modelId, options, provider = DEFAULT_PROVIDER_ID, adapterOptions = {}) {
   const googleOptions = parseGoogleGenerateProviderOptions(
     options.providerOptions
   );
+  const capabilities = adapterOptions.capabilities ?? getGoogleModelCapabilities(modelId);
+  validateInputModalities({
+    messages: options.messages,
+    capabilities,
+    modelId,
+    providerId: provider
+  });
   const convertedMessages = convertMessages(options.messages);
   const requestConfig = {
     ...convertedMessages.systemInstruction ? { systemInstruction: convertedMessages.systemInstruction } : {},
     ...options.tools && Object.keys(options.tools).length > 0 ? { tools: convertTools(options.tools) } : {},
     ...options.toolChoice ? { toolConfig: convertToolChoice(options.toolChoice) } : {},
     ...mapSamplingToConfig(options),
-    ...mapReasoningToConfig(modelId, options),
+    ...mapReasoningToConfig(options, capabilities),
     ...mapGoogleProviderOptionsToConfig(googleOptions),
     ...options.signal ? { abortSignal: options.signal } : {}
   };
@@ -572,11 +588,10 @@ async function* transformStream(stream) {
     usage
   };
 }
-function mapReasoningToConfig(modelId, options) {
+function mapReasoningToConfig(options, capabilities) {
   if (!options.reasoning) {
     return {};
   }
-  const capabilities = getGoogleModelCapabilities(modelId);
   if (capabilities.reasoning.thinkingParam === "thinkingLevel") {
     return {
       thinkingConfig: {
@@ -681,6 +696,15 @@ function mapUsage(response, fallback) {
   };
 }
 
+// src/chat-model.ts
+import {
+  StructuredOutputNoObjectGeneratedError,
+  StructuredOutputParseError,
+  StructuredOutputValidationError,
+  createObjectStream,
+  createChatStream
+} from "@core-ai/core-ai";
+
 // src/google-error.ts
 import { ApiError } from "@google/genai";
 import {
@@ -814,6 +838,8 @@ function toNumericHttpCode(body) {
 
 // src/chat-model.ts
 function createGoogleGenAIChatModel(client, modelId, provider = "google") {
+  const capabilities = getGoogleModelCapabilities(modelId);
+  const adapterOptions = { capabilities };
   async function callGenerateContentApi(request) {
     try {
       return await client.models.generateContent(request);
@@ -829,12 +855,22 @@ function createGoogleGenAIChatModel(client, modelId, provider = "google") {
     }
   }
   async function generateChat(options) {
-    const request = createGenerateRequest(modelId, options);
+    const request = createGenerateRequest(
+      modelId,
+      options,
+      provider,
+      adapterOptions
+    );
     const response = await callGenerateContentApi(request);
     return mapGenerateResponse(response);
   }
   async function streamChat(options) {
-    const request = createGenerateRequest(modelId, options);
+    const request = createGenerateRequest(
+      modelId,
+      options,
+      provider,
+      adapterOptions
+    );
     return createChatStream(
       async () => transformStream(await callGenerateContentStreamApi(request)),
       { signal: options.signal }
@@ -843,7 +879,7 @@ function createGoogleGenAIChatModel(client, modelId, provider = "google") {
   return {
     provider,
     modelId,
-    capabilities: getGoogleModelCapabilities(modelId),
+    capabilities,
     generate: generateChat,
     stream: streamChat,
     async generateObject(options) {
@@ -1247,7 +1283,6 @@ function greatestCommonDivisor(a, b) {
 }
 
 // src/provider.ts
-var DEFAULT_PROVIDER_ID = "google";
 function createGoogleGenAIProvider(options = {}, factoryOptions = {}) {
   const client = options.client ?? new GoogleGenAI({
     apiKey: options.apiKey,

package/package.json

@@ -1,6 +1,6 @@
 {
     "name": "@core-ai/google-genai",
-    "version": "0.18.0",
+    "version": "0.19.0",
     "description": "Google GenAI provider package for @core-ai/core-ai",
     "license": "MIT",
     "author": "Omnifact (https://omnifact.ai)",
@@ -45,7 +45,7 @@
         "test:watch": "vitest"
     },
     "dependencies": {
-        "@core-ai/core-ai": "^0.18.0",
+        "@core-ai/core-ai": "^0.19.0",
         "@google/genai": "^1.42.0"
     },
     "peerDependencies": {