@llumiverse/drivers 1.3.0 → 1.4.0

package/lib/{esm/vertexai → vertexai}/index.js

@@ -1,20 +1,30 @@
-import { AnthropicVertex } from "@anthropic-ai/vertex-sdk";
-import { PredictionServiceClient, v1beta1 } from "@google-cloud/aiplatform";
-import { GoogleGenAI } from "@google/genai";
-import { AbstractDriver, getConversationMeta, getModelCapabilities, incrementConversationTurn, modelModalitiesToArray, stripBase64ImagesFromConversation, stripHeartbeatsFromConversation, truncateLargeTextInConversation, } from "@llumiverse/core";
-import { FetchClient } from "@vertesia/api-fetch-client";
-import { GoogleAuth } from "google-auth-library";
-import { getEmbeddingsForImages } from "./embeddings/embeddings-image.js";
-import { getEmbeddingsForText } from "./embeddings/embeddings-text.js";
-import { getModelDefinition } from "./models.js";
-import { ANTHROPIC_REGIONS, NON_GLOBAL_ANTHROPIC_MODELS } from "./models/claude.js";
-import { ImagenModelDefinition } from "./models/imagen.js";
-export function trimModelName(model) {
-    const i = model.lastIndexOf("@");
-    return i > -1 ? model.substring(0, i) : model;
+import { AnthropicVertex } from '@anthropic-ai/vertex-sdk';
+import { GoogleGenAI } from '@google/genai';
+import { PredictionServiceClient, v1beta1 } from '@google-cloud/aiplatform';
+import { getConversationMeta, getModelCapabilities, incrementConversationTurn, modelModalitiesToArray, stripBase64ImagesFromConversation, stripHeartbeatsFromConversation, truncateLargeTextInConversation, } from '@llumiverse/core';
+import { AbstractDriver } from '@llumiverse/core/driver';
+import { mergeDriverHttpTimeoutOptions, resolveDriverHttpTimeouts } from '@llumiverse/core/http-agent';
+import { FetchClient } from '@vertesia/api-fetch-client';
+import { GoogleAuth } from 'google-auth-library';
+import { formatClaudeDebugPrompt } from '../shared/claude-messages.js';
+import { generateVertexAiEmbeddings } from './embeddings/embed.js';
+import { ANTHROPIC_REGIONS, NON_GLOBAL_ANTHROPIC_MODELS } from './models/claude.js';
+import { formatGeminiDebugPrompt } from './models/gemini.js';
+import { formatImagenDebugPrompt, ImagenModelDefinition } from './models/imagen.js';
+import { getModelDefinition, trimModelName } from './models.js';
+function isClaudePrompt(prompt) {
+    return ('messages' in prompt &&
+        ('system' in prompt || prompt.messages.some((message) => Array.isArray(message.content))));
 }
+function isClaudeStreamingPrompt(prompt) {
+    return (prompt !== null &&
+        typeof prompt === 'object' &&
+        'messages' in prompt &&
+        Array.isArray(prompt.messages));
+}
+export { trimModelName };
 export class VertexAIDriver extends AbstractDriver {
-    static PROVIDER = "vertexai";
+    static PROVIDER = 'vertexai';
     provider = VertexAIDriver.PROVIDER;
     aiplatform;
     anthropicClient;
@@ -25,6 +35,7 @@ export class VertexAIDriver extends AbstractDriver {
     llamaClient;
     modelGarden;
     imagenClient;
+    predictionClient;
     googleAuth;
     authClientPromise;
     constructor(options) {
@@ -38,19 +49,59 @@ export class VertexAIDriver extends AbstractDriver {
         this.modelGarden = undefined;
         this.llamaClient = undefined;
         this.imagenClient = undefined;
+        this.predictionClient = undefined;
         this.googleAuth = new GoogleAuth(options.googleAuthOptions);
         this.authClientPromise = undefined;
     }
+    /**
+     * Cleanup Google Cloud clients when the driver is evicted from the cache.
+     * `super.destroy()` releases the HTTP agent socket pool created by
+     * {@link AbstractDriver.getHttpAgent} / {@link AbstractDriver.getDriverFetch}.
+     */
+    destroy() {
+        this.aiplatform?.close();
+        this.modelGarden?.close();
+        this.imagenClient?.close();
+        this.predictionClient?.close();
+        super.destroy();
+    }
     async getAuthClient() {
         if (!this.authClientPromise) {
             this.authClientPromise = this.googleAuth.getClient();
         }
         return this.authClientPromise;
     }
-    getGoogleGenAIClient(region = this.options.region, flex = false) {
-        if (this.googleGenAI &&
-            this.googleGenAIRegion === region &&
-            this.googleGenAIFlex === flex) {
+    getSdkRequestTimeoutMs(httpTimeout) {
+        const timeouts = resolveDriverHttpTimeouts(mergeDriverHttpTimeoutOptions(this.options.httpTimeout, httpTimeout));
+        return Math.max(timeouts.headersTimeout, timeouts.bodyTimeout);
+    }
+    getGoogleGenAIHttpOptions(flex, httpTimeout) {
+        return {
+            timeout: this.getSdkRequestTimeoutMs(httpTimeout),
+            ...(flex
+                ? {
+                    headers: {
+                        'X-Vertex-AI-LLM-Request-Type': 'shared',
+                        'X-Vertex-AI-LLM-Shared-Request-Type': 'flex',
+                    },
+                }
+                : {}),
+        };
+    }
+    getAnthropicVertexClientOptions(region, authClient, httpTimeout) {
+        return {
+            timeout: this.getSdkRequestTimeoutMs(httpTimeout),
+            region,
+            projectId: this.options.project,
+            authClient: authClient,
+            fetch: this.getDriverFetch(),
+        };
+    }
+    getGoogleGenAIClient(region = this.options.region, flex = false, httpTimeout) {
+        if (httpTimeout) {
+            return this.buildGoogleGenAIClient(region, flex, httpTimeout);
+        }
+        if (this.googleGenAI && this.googleGenAIRegion === region && this.googleGenAIFlex === flex) {
             // Return existing client if region and flex settings match
             return this.googleGenAI;
         }
@@ -59,30 +110,24 @@ export class VertexAIDriver extends AbstractDriver {
         this.googleGenAIFlex = flex;
         return this.googleGenAI;
     }
-    buildGoogleGenAIClient(region, flex) {
+    buildGoogleGenAIClient(region, flex, httpTimeout) {
         return new GoogleGenAI({
             project: this.options.project,
             location: region,
             vertexai: true,
             googleAuthOptions: this.options.googleAuthOptions || {
-                scopes: ["https://www.googleapis.com/auth/cloud-platform"],
+                scopes: ['https://www.googleapis.com/auth/cloud-platform'],
             },
-            ...(flex ? {
-                httpOptions: {
-                    headers: {
-                        "X-Vertex-AI-LLM-Request-Type": "shared",
-                        "X-Vertex-AI-LLM-Shared-Request-Type": "flex",
-                    }
-                }
-            } : {}),
+            httpOptions: this.getGoogleGenAIHttpOptions(flex, httpTimeout),
         });
     }
-    getFetchClient() {
+    getFetchClient(region = this.options.region) {
         //Lazy initialization
         if (!this.fetchClient) {
             this.fetchClient = createFetchClient({
-                region: this.options.region,
+                region: region,
                 project: this.options.project,
+                fetchImpl: this.getDriverFetch(),
             }).withAuthCallback(async () => {
                 const token = await this.googleAuth.getAccessToken();
                 return `Bearer ${token}`;
@@ -90,23 +135,24 @@ export class VertexAIDriver extends AbstractDriver {
         }
         return this.fetchClient;
     }
-    getLLamaClient(region = "us-central1") {
+    getLLamaClient(region = 'us-central1') {
         //Lazy initialization
-        if (!this.llamaClient || this.llamaClient["region"] !== region) {
+        if (!this.llamaClient || this.llamaClient.region !== region) {
             this.llamaClient = createFetchClient({
                 region: region,
                 project: this.options.project,
-                apiVersion: "v1beta1",
+                apiVersion: 'v1beta1',
+                fetchImpl: this.getDriverFetch(),
             }).withAuthCallback(async () => {
                 const token = await this.googleAuth.getAccessToken();
                 return `Bearer ${token}`;
             });
             // Store the region for potential client reuse
-            this.llamaClient["region"] = region;
+            this.llamaClient.region = region;
         }
         return this.llamaClient;
     }
-    async getAnthropicClient(region = this.options.region) {
+    async getAnthropicClient(region = this.options.region, httpTimeout) {
         // Extract region prefix and map if it exists in ANTHROPIC_REGIONS, otherwise use as-is
         const getRegionPrefix = (r) => r.split('-')[0];
         const regionPrefix = getRegionPrefix(region);
@@ -116,22 +162,12 @@ export class VertexAIDriver extends AbstractDriver {
         // Get auth client to avoid version mismatch with GoogleAuth generic types
         const authClient = await this.getAuthClient();
         // If mapped region is different from default mapped region, create one-off client
-        if (mappedRegion !== defaultMappedRegion) {
-            return new AnthropicVertex({
-                timeout: 20 * 60 * 10000, // Set to 20 minutes, 10 minute default, setting this disables long request error: https://github.com/anthropics/anthropic-sdk-typescript?#long-requests
-                region: mappedRegion,
-                projectId: this.options.project,
-                authClient: authClient,
-            });
+        if (httpTimeout || mappedRegion !== defaultMappedRegion) {
+            return new AnthropicVertex(this.getAnthropicVertexClientOptions(mappedRegion, authClient, httpTimeout));
         }
         //Lazy initialization for default region
         if (!this.anthropicClient) {
-            this.anthropicClient = new AnthropicVertex({
-                timeout: 20 * 60 * 10000, // Set to 20 minutes, 10 minute default, setting this disables long request error: https://github.com/anthropics/anthropic-sdk-typescript?#long-requests
-                region: mappedRegion,
-                projectId: this.options.project,
-                authClient: authClient,
-            });
+            this.anthropicClient = new AnthropicVertex(this.getAnthropicVertexClientOptions(mappedRegion, authClient));
         }
         return this.anthropicClient;
     }
@@ -172,10 +208,22 @@ export class VertexAIDriver extends AbstractDriver {
         }
         return this.imagenClient;
     }
+    async getPredictionServiceClient() {
+        //Lazy initialization using the driver's configured region
+        if (!this.predictionClient) {
+            const authClient = await this.getAuthClient();
+            this.predictionClient = new PredictionServiceClient({
+                projectId: this.options.project,
+                apiEndpoint: `${this.options.region}-${API_BASE_PATH}`,
+                authClient,
+            });
+        }
+        return this.predictionClient;
+    }
     validateResult(result, options) {
         // Optionally preprocess the result before validation
         const modelDef = getModelDefinition(options.model);
-        if (typeof modelDef.preValidationProcessing === "function") {
+        if (typeof modelDef.preValidationProcessing === 'function') {
             const processed = modelDef.preValidationProcessing(result, options);
             result = processed.result;
             options = processed.options;
@@ -189,7 +237,7 @@ export class VertexAIDriver extends AbstractDriver {
         return Promise.resolve(getModelDefinition(options.model).model.can_stream === true);
     }
     isImageModel(model) {
-        return model.includes("imagen");
+        return model.includes('imagen');
     }
     createPrompt(segments, options) {
         if (this.isImageModel(options.model)) {
@@ -197,6 +245,18 @@ export class VertexAIDriver extends AbstractDriver {
         }
         return getModelDefinition(options.model).createPrompt(this, segments, options);
     }
+    formatDebugPrompt(prompt) {
+        if (isClaudePrompt(prompt)) {
+            return formatClaudeDebugPrompt(prompt);
+        }
+        if ('contents' in prompt) {
+            return formatGeminiDebugPrompt(prompt);
+        }
+        if ('messages' in prompt) {
+            return prompt;
+        }
+        return formatImagenDebugPrompt(prompt);
+    }
     async requestTextCompletion(prompt, options) {
         return getModelDefinition(options.model).requestTextCompletion(this, prompt, options);
     }
@@ -210,7 +270,7 @@ export class VertexAIDriver extends AbstractDriver {
      */
     buildStreamingConversation(prompt, result, toolUse, options) {
         // Handle Claude-style prompts (has 'messages' array)
-        if ('messages' in prompt && Array.isArray(prompt.messages)) {
+        if (isClaudeStreamingPrompt(prompt)) {
             return this.buildClaudeStreamingConversation(prompt, result, toolUse, options);
         }
         // Only handle Gemini-style prompts with contents array
@@ -220,7 +280,7 @@ export class VertexAIDriver extends AbstractDriver {
         const completionResults = result;
         // Convert accumulated results to text content for assistant message
         const textContent = completionResults
-            .map(r => {
+            .map((r) => {
             switch (r.type) {
                 case 'text':
                     return r.value;
@@ -229,8 +289,10 @@ export class VertexAIDriver extends AbstractDriver {
                 case 'image':
                     // Skip images in conversation - they're in the result
                     return '';
-                default:
-                    return String(r.value || '');
+                default: {
+                    const _exhaustive = r;
+                    return String(_exhaustive);
+                }
             }
         })
             .join('');
@@ -246,7 +308,7 @@ export class VertexAIDriver extends AbstractDriver {
                     functionCall: {
                         name: tool.tool_name,
                         args: tool.tool_input,
-                    }
+                    },
                 };
                 // Include thought_signature for Gemini thinking models (2.5+/3.0+)
                 // This must be preserved in the conversation for subsequent API calls
@@ -259,15 +321,13 @@ export class VertexAIDriver extends AbstractDriver {
         // prompt.contents already includes the conversation history
         // (merged in requestTextCompletionStream via updateConversation),
         // so we use it directly — do NOT prepend options.conversation again.
-        let conversation = [
-            ...prompt.contents,
-        ];
+        let conversation = [...prompt.contents];
         // Only add assistant message if there's actual content
         // (Empty text parts can cause API errors)
         if (parts.length > 0) {
             conversation.push({
                 role: 'model',
-                parts: parts
+                parts: parts,
             });
         }
         // Increment turn counter
@@ -277,7 +337,7 @@ export class VertexAIDriver extends AbstractDriver {
         const stripOptions = {
             keepForTurns: options.stripImagesAfterTurns ?? Infinity,
             currentTurn,
-            textMaxTokens: options.stripTextMaxTokens
+            textMaxTokens: options.stripTextMaxTokens,
         };
         let processedConversation = stripBase64ImagesFromConversation(conversation, stripOptions);
         processedConversation = truncateLargeTextInConversation(processedConversation, stripOptions);
@@ -291,7 +351,10 @@ export class VertexAIDriver extends AbstractDriver {
         const geminiPrompt = prompt;
         if (geminiPrompt.system) {
             if (typeof processedConversation === 'object' && processedConversation !== null) {
-                processedConversation = { ...processedConversation, _llumiverse_system: geminiPrompt.system };
+                processedConversation = {
+                    ...processedConversation,
+                    _llumiverse_system: geminiPrompt.system,
+                };
             }
         }
         return processedConversation;
@@ -304,7 +367,7 @@ export class VertexAIDriver extends AbstractDriver {
         const completionResults = result;
         // Convert accumulated results to text content
         const textContent = completionResults
-            .map(r => {
+            .map((r) => {
             switch (r.type) {
                 case 'text':
                     return r.value;
@@ -312,8 +375,10 @@ export class VertexAIDriver extends AbstractDriver {
                     return typeof r.value === 'string' ? r.value : JSON.stringify(r.value);
                 case 'image':
                     return '';
-                default:
-                    return String(r.value || '');
+                default: {
+                    const _exhaustive = r;
+                    return String(_exhaustive);
+                }
             }
         })
             .join('');
@@ -323,7 +388,7 @@ export class VertexAIDriver extends AbstractDriver {
         if (textContent) {
             content.push({
                 type: 'text',
-                text: textContent
+                text: textContent,
             });
         }
         // Add tool_use blocks in Claude format
@@ -333,31 +398,29 @@ export class VertexAIDriver extends AbstractDriver {
                     type: 'tool_use',
                     id: tool.id,
                     name: tool.tool_name,
-                    input: tool.tool_input ?? {}
+                    input: tool.tool_input ?? {},
                 });
             }
         }
         // Claude's requestTextCompletionStream does NOT mutate prompt.messages
         // to include history, so we must prepend options.conversation here.
-        const existingMessages = options.conversation?.messages ?? [];
-        const existingSystem = options.conversation?.system ?? prompt.system;
+        const existingConversation = options.conversation;
+        const existingMessages = existingConversation?.messages ?? [];
+        const existingSystem = existingConversation?.system ?? prompt.system;
         // Build the new messages array
-        const newMessages = [
-            ...existingMessages,
-            ...prompt.messages,
-        ];
+        const newMessages = [...existingMessages, ...prompt.messages];
         // Only add assistant message if there's actual content
         // (Claude API rejects empty text content blocks)
         if (content.length > 0) {
             newMessages.push({
                 role: 'assistant',
-                content: content
+                content: content,
             });
         }
         // Build the new conversation in ClaudePrompt format
         const conversation = {
             messages: newMessages,
-            system: existingSystem
+            system: existingSystem,
         };
         // Increment turn counter
         const withTurn = incrementConversationTurn(conversation);
@@ -366,7 +429,7 @@ export class VertexAIDriver extends AbstractDriver {
         const stripOptions = {
             keepForTurns: options.stripImagesAfterTurns ?? Infinity,
             currentTurn,
-            textMaxTokens: options.stripTextMaxTokens
+            textMaxTokens: options.stripTextMaxTokens,
         };
         let processedConversation = stripBase64ImagesFromConversation(withTurn, stripOptions);
         processedConversation = truncateLargeTextInConversation(processedConversation, stripOptions);
@@ -377,7 +440,7 @@ export class VertexAIDriver extends AbstractDriver {
         return processedConversation;
     }
     async requestImageGeneration(_prompt, _options) {
-        const splits = _options.model.split("/");
+        const splits = _options.model.split('/');
         const modelName = trimModelName(splits[splits.length - 1]);
         return new ImagenModelDefinition(modelName).requestImageGeneration(this, _prompt, _options);
     }
@@ -393,33 +456,39 @@ export class VertexAIDriver extends AbstractDriver {
         // Get clients
         const modelGarden = await this.getModelGardenClient();
         const aiplatform = await this.getAIPlatformClient();
-        const globalGenAiClient = this.getGoogleGenAIClient("global");
+        const globalGenAiClient = this.getGoogleGenAIClient('global');
         let models = [];
         //Model Garden Publisher models - Pretrained models
-        const publishers = ["google", "anthropic", "meta"];
+        const publishers = ['google', 'anthropic', 'meta'];
         // Meta "maas" models are LLama Models-As-A-Service. Non-maas models are not pre-deployed.
-        const supportedModels = { google: ["gemini", "imagen"], anthropic: ["claude"], meta: ["maas"] };
+        const supportedModels = { google: ['gemini', 'imagen'], anthropic: ['claude'], meta: ['maas'] };
         // Additional models not in the listings, but we want to include
         // TODO: Remove once the models are available in the listing API, or no longer needed
         const additionalModels = {
-            google: [
-                "imagen-3.0-fast-generate-001",
-            ],
+            google: ['imagen-3.0-fast-generate-001'],
             anthropic: [],
             meta: [
-                "llama-4-maverick-17b-128e-instruct-maas",
-                "llama-4-scout-17b-16e-instruct-maas",
-                "llama-3.3-70b-instruct-maas",
-                "llama-3.2-90b-vision-instruct-maas",
-                "llama-3.1-405b-instruct-maas",
-                "llama-3.1-70b-instruct-maas",
-                "llama-3.1-8b-instruct-maas",
+                'llama-4-maverick-17b-128e-instruct-maas',
+                'llama-4-scout-17b-16e-instruct-maas',
+                'llama-3.3-70b-instruct-maas',
+                'llama-3.2-90b-vision-instruct-maas',
+                'llama-3.1-405b-instruct-maas',
+                'llama-3.1-70b-instruct-maas',
+                'llama-3.1-8b-instruct-maas',
             ],
         };
         //Used to exclude retired models that are still in the listing API but not available for use.
         //Or models we do not support yet
         const unsupportedModelsByPublisher = {
-            google: ["gemini-pro", "gemini-ultra", "imagen-product-recontext-preview", "embedding", "gemini-live-2.5-flash-preview-native-audio", "computer-use-preview"],
+            google: [
+                'gemini-pro',
+                'gemini-ultra',
+                'imagen-product-recontext-preview',
+                'embedding',
+                'embed',
+                'gemini-live-2.5-flash-preview-native-audio',
+                'computer-use-preview',
+            ],
             anthropic: [],
             meta: [],
         };
@@ -430,7 +499,7 @@ export class VertexAIDriver extends AbstractDriver {
         const publisherPromises = publishers.map(async (publisher) => {
             const [response] = await modelGarden.listPublisherModels({
                 parent: `publishers/${publisher}`,
-                orderBy: "name",
+                orderBy: 'name',
                 listAllVersions: true,
             });
             return { publisher, response };
@@ -445,18 +514,22 @@ export class VertexAIDriver extends AbstractDriver {
         // Process aiplatform models, project specific models
         const [response] = aiplatformResult;
         models = models.concat(response.map((model) => ({
-            id: model.name?.split("/").pop() ?? "",
-            name: model.displayName ?? "",
-            provider: "vertexai"
+            id: model.name?.split('/').pop() ?? '',
+            name: model.displayName ?? '',
+            provider: 'vertexai',
         })));
         // Process global google models from GenAI
-        models = models.concat(globalGoogleResult.map((model) => {
-            const modelCapability = getModelCapabilities(model.name ?? '', "vertexai");
+        // Exclude embedding, retired and or unsupported models
+        const excludedModels = unsupportedModelsByPublisher.google;
+        models = models.concat(globalGoogleResult
+            .filter((model) => !excludedModels.includes(model.name ?? ''))
+            .map((model) => {
+            const modelCapability = getModelCapabilities(model.name ?? '', 'vertexai');
             return {
-                id: "locations/global/" + model.name,
-                name: "Global " + model.name?.split('/').pop(),
-                provider: "vertexai",
-                owner: "google",
+                id: `locations/global/${model.name}`,
+                name: `Global ${model.name?.split('/').pop()}`,
+                provider: 'vertexai',
+                owner: 'google',
                 input_modalities: modelModalitiesToArray(modelCapability.input),
                 output_modalities: modelModalitiesToArray(modelCapability.output),
                 tool_support: modelCapability.tool_support,
@@ -466,20 +539,22 @@ export class VertexAIDriver extends AbstractDriver {
         for (const result of publisherResults) {
             const { publisher, response } = result;
             const modelFamily = supportedModels[publisher];
-            const retiredModels = unsupportedModelsByPublisher[publisher];
-            models = models.concat(response.filter((model) => {
-                const modelName = model.name ?? "";
-                // Exclude retired models
-                if (retiredModels.some(retiredModel => modelName.includes(retiredModel))) {
+            const excludedModels = unsupportedModelsByPublisher[publisher];
+            models = models.concat(response
+                .filter((model) => {
+                const modelName = model.name ?? '';
+                // Exclude embedding, retired and or unsupported models
+                if (excludedModels.some((retiredModel) => modelName.includes(retiredModel))) {
                     return false;
                 }
                 // Check if the model belongs to the supported model families
-                if (modelFamily.some(family => modelName.includes(family))) {
+                if (modelFamily.some((family) => modelName.includes(family))) {
                     return true;
                 }
                 return false;
-            }).map(model => {
-                const modelCapability = getModelCapabilities(model.name ?? '', "vertexai");
+            })
+                .map((model) => {
+                const modelCapability = getModelCapabilities(model.name ?? '', 'vertexai');
                 return {
                     id: model.name ?? '',
                     name: model.name?.split('/').pop() ?? '',
@@ -492,20 +567,21 @@ export class VertexAIDriver extends AbstractDriver {
             }));
             // Create global google gemini models for Gemini 2.5 and later, if missing from GenAI listing
             if (publisher === 'google') {
-                const globalGeminiModels = response.filter((model) => {
-                    const modelName = model.name ?? "";
-                    if (retiredModels.some(retiredModel => modelName.includes(retiredModel))) {
+                const globalGeminiModels = response
+                    .filter((model) => {
+                    const modelName = model.name ?? '';
+                    if (excludedModels.some((retiredModel) => modelName.includes(retiredModel))) {
                         return false;
                     }
-                    if (modelFamily.some(family => modelName.includes(family))) {
+                    if (modelFamily.some((family) => modelName.includes(family))) {
                         const versionMatch = modelName.match(/gemini-(\d+(?:\.\d+)?)/);
                         if (versionMatch) {
                             const version = parseFloat(versionMatch[1]);
                             if (version >= 2.5) {
                                 // Check if already present
                                 const shortName = modelName.split('/').pop();
-                                const globalName = "Global " + shortName;
-                                if (models.some(m => m.name === globalName)) {
+                                const globalName = `Global ${shortName}`;
+                                if (models.some((m) => m.name === globalName)) {
                                     return false;
                                 }
                                 return true;
@@ -514,11 +590,12 @@ export class VertexAIDriver extends AbstractDriver {
                         return false;
                     }
                     return false;
-                }).map(model => {
-                    const modelCapability = getModelCapabilities(model.name ?? '', "vertexai");
+                })
+                    .map((model) => {
+                    const modelCapability = getModelCapabilities(model.name ?? '', 'vertexai');
                     return {
-                        id: "locations/global/" + model.name,
-                        name: "Global " + model.name?.split('/').pop(),
+                        id: `locations/global/${model.name}`,
+                        name: `Global ${model.name?.split('/').pop()}`,
                         provider: 'vertexai',
                         owner: publisher,
                         input_modalities: modelModalitiesToArray(modelCapability.input),
@@ -530,23 +607,25 @@ export class VertexAIDriver extends AbstractDriver {
             }
             // Create global anthropic models for those not in NON_GLOBAL_ANTHROPIC_MODELS
             if (publisher === 'anthropic') {
-                const globalAnthropicModels = response.filter((model) => {
-                    const modelName = model.name ?? "";
-                    if (retiredModels.some(retiredModel => modelName.includes(retiredModel))) {
+                const globalAnthropicModels = response
+                    .filter((model) => {
+                    const modelName = model.name ?? '';
+                    if (excludedModels.some((retiredModel) => modelName.includes(retiredModel))) {
                         return false;
                     }
-                    if (modelFamily.some(family => modelName.includes(family))) {
-                        if (modelName.includes("claude-3-7")) {
+                    if (modelFamily.some((family) => modelName.includes(family))) {
+                        if (modelName.includes('claude-3-7')) {
                             return true;
                         }
-                        return !NON_GLOBAL_ANTHROPIC_MODELS.some(nonGlobalModel => modelName.includes(nonGlobalModel));
+                        return !NON_GLOBAL_ANTHROPIC_MODELS.some((nonGlobalModel) => modelName.includes(nonGlobalModel));
                     }
                     return false;
-                }).map(model => {
-                    const modelCapability = getModelCapabilities(model.name ?? '', "vertexai");
+                })
+                    .map((model) => {
+                    const modelCapability = getModelCapabilities(model.name ?? '', 'vertexai');
                     return {
-                        id: "locations/global/" + model.name,
-                        name: "Global " + model.name?.split('/').pop(),
+                        id: `locations/global/${model.name}`,
+                        name: `Global ${model.name?.split('/').pop()}`,
                         provider: 'vertexai',
                         owner: publisher,
                         input_modalities: modelModalitiesToArray(modelCapability.input),
@@ -559,7 +638,7 @@ export class VertexAIDriver extends AbstractDriver {
             // Add additional models that are not in the listing
             for (const additionalModel of additionalModels[publisher]) {
                 const publisherModelName = `publishers/${publisher}/models/${additionalModel}`;
-                const modelCapability = getModelCapabilities(additionalModel, "vertexai");
+                const modelCapability = getModelCapabilities(additionalModel, 'vertexai');
                 models.push({
                     id: publisherModelName,
                     name: additionalModel,
@@ -572,35 +651,18 @@ export class VertexAIDriver extends AbstractDriver {
             }
         }
         //Remove duplicates
-        const uniqueModels = Array.from(new Set(models.map(a => a.id)))
-            .map(id => {
-            return models.find(a => a.id === id) ?? {};
-        }).sort((a, b) => a.id.localeCompare(b.id));
+        const uniqueModels = Array.from(new Set(models.map((a) => a.id)))
+            .map((id) => {
+            return models.find((a) => a.id === id) ?? {};
+        })
+            .sort((a, b) => a.id.localeCompare(b.id));
         return uniqueModels;
     }
     validateConnection() {
-        throw new Error("Method not implemented.");
+        throw new Error('Method not implemented.');
     }
     async generateEmbeddings(options) {
-        if (options.image || options.model?.includes("multimodal")) {
-            if (options.text && options.image) {
-                throw new Error("Text and Image simultaneous embedding not implemented. Submit separately");
-            }
-            return getEmbeddingsForImages(this, options);
-        }
-        const text_options = {
-            content: options.text ?? "",
-            model: options.model,
-        };
-        return getEmbeddingsForText(this, text_options);
-    }
-    /**
-     * Cleanup Google Cloud clients when the driver is evicted from the cache.
-     */
-    destroy() {
-        this.aiplatform?.close();
-        this.modelGarden?.close();
-        this.imagenClient?.close();
+        return generateVertexAiEmbeddings(this, options);
     }
     /**
      * Format VertexAI errors by routing to model-specific error handlers.
@@ -619,7 +681,7 @@ export class VertexAIDriver extends AbstractDriver {
             try {
                 return modelDef.formatLlumiverseError(this, error, context);
             }
-            catch (formattingError) {
+            catch {
                 // If model-specific handler throws, fall through to default handling
                 // This allows model handlers to explicitly opt out for certain errors
             }
@@ -629,11 +691,11 @@ export class VertexAIDriver extends AbstractDriver {
     }
 }
 //'us-central1-aiplatform.googleapis.com',
-const API_BASE_PATH = "aiplatform.googleapis.com";
-function createFetchClient({ region, project, apiEndpoint, apiVersion = "v1", }) {
+const API_BASE_PATH = 'aiplatform.googleapis.com';
+function createFetchClient({ region, project, apiEndpoint, apiVersion = 'v1', fetchImpl, }) {
     const vertexBaseEndpoint = apiEndpoint ?? `${region}-${API_BASE_PATH}`;
-    return new FetchClient(`https://${vertexBaseEndpoint}/${apiVersion}/projects/${project}/locations/${region}`).withHeaders({
-        "Content-Type": "application/json",
+    return new FetchClient(`https://${vertexBaseEndpoint}/${apiVersion}/projects/${project}/locations/${region}`, fetchImpl).withHeaders({
+        'Content-Type': 'application/json',
     });
 }
 //# sourceMappingURL=index.js.map