Vertex AI Provider

Vertex AI is Google Cloud’s enterprise AI platform, offering advanced features, broader model access, and deep integration with Google Cloud services. Use Vertex AI for production applications that require governance, IAM control, and access to Model Garden.

Vertex AI is part of the unified @genkit-ai/google-genai package. For simpler setup, see Google AI.

Installation

npm install @genkit-ai/google-genai

Setup

Vertex AI supports two authentication methods:

Method 1: Application Default Credentials (Recommended)

This is the standard authentication method for production applications. Prerequisites:

Google Cloud project with billing enabled
Vertex AI API enabled
Appropriate IAM permissions

Local Development:

# Authenticate with your Google account
gcloud auth application-default login

# Set your project
gcloud config set project YOUR_PROJECT_ID

Production (GCP): Use a service account with the required permissions:

roles/aiplatform.user
roles/storage.objectViewer (for some features)

Configure the Plugin:

import { genkit } from 'genkit';
import { vertexAI } from '@genkit-ai/google-genai';

const ai = genkit({
  plugins: [
    vertexAI({
      location: 'us-central1',  // Regional endpoint
      // projectId: 'your-project', // Optional, auto-detected from ADC
    }),
  ],
});

Method 2: Express Mode (API Key)

Vertex AI Express Mode provides a simplified way to try Vertex AI features using just an API key, without requiring full GCP project setup or billing. Get an API Key:

Visit Vertex AI Studio
Enable Express Mode
Generate an API key

Configure the Plugin:

import { genkit } from 'genkit';
import { vertexAI } from '@genkit-ai/google-genai';

const ai = genkit({
  plugins: [
    vertexAI({
      apiKey: process.env.VERTEX_EXPRESS_API_KEY,
    }),
  ],
});

With Express Mode, you don’t provide projectId or location in the config.

Available Models

Text Generation (Gemini)

All Gemini models available through Google AI are also available on Vertex AI:

gemini-2.5-flash - Balanced performance and speed
gemini-2.5-pro - Most powerful Gemini model
gemini-2.5-flash-lite - Fastest option

Image Generation (Imagen)

imagen-3.0-generate-002 - High-quality image generation
imagen-3.0-fast-generate-001 - Faster generation

Music Generation (Lyria)

Vertex AI exclusive:

lyria-002 - Generate music and audio from text descriptions

Video Generation (Veo)

veo-002 - Generate videos from text prompts

Embeddings

text-embedding-005 - Latest embedding model
text-embedding-004 - Previous generation
text-multilingual-embedding-002 - Multilingual support

Usage Examples

Basic Text Generation

import { genkit } from 'genkit';
import { vertexAI } from '@genkit-ai/google-genai';

const ai = genkit({
  plugins: [vertexAI({ location: 'us-central1' })],
});

const response = await ai.generate({
  model: vertexAI.model('gemini-2.5-pro'),
  prompt: 'Explain quantum entanglement to a 10-year-old.',
});

console.log(response.text());

Music Generation with Lyria

const response = await ai.generate({
  model: vertexAI.model('lyria-002'),
  prompt: 'A cheerful, upbeat instrumental piano melody with a jazzy feel',
  config: {
    duration: 30, // seconds
  },
});

const audioFile = response.media();
if (audioFile) {
  console.log('Generated audio:', audioFile.url);
}

Image Generation with Imagen

const response = await ai.generate({
  model: vertexAI.model('imagen-3.0-generate-002'),
  prompt: 'A futuristic city with flying cars at night, cyberpunk style, highly detailed',
  config: {
    numberOfImages: 4,
    aspectRatio: '16:9',
  },
});

const images = response.media();
images?.forEach((img, i) => {
  console.log(`Image ${i + 1}:`, img.url);
});

Multimodal Input

const response = await ai.generate({
  model: vertexAI.model('gemini-2.5-flash'),
  prompt: [
    { text: 'Analyze this image and describe what you see in detail.' },
    { media: { url: 'gs://my-bucket/image.jpg' } }, // Cloud Storage URL
  ],
});

console.log(response.text());

Context Caching

Reduce costs and latency for repeated prompts with large context:

const response = await ai.generate({
  model: vertexAI.model('gemini-2.5-pro'),
  system: {
    text: largeSystemPrompt, // e.g., 50K tokens
    metadata: {
      cache: true,
      cacheTTL: 3600, // 1 hour
    },
  },
  prompt: 'Based on the system instructions, what should I do?',
});

Grounding with Google Search

Ground model responses with real-time search results:

const response = await ai.generate({
  model: vertexAI.model('gemini-2.5-flash'),
  prompt: 'What are the latest developments in quantum computing?',
  config: {
    googleSearchRetrieval: {
      dynamicRetrievalConfig: {
        mode: 'MODE_DYNAMIC',
        dynamicThreshold: 0.3,
      },
    },
  },
});

console.log(response.text());
// Response will include citations to search sources

Function Calling

import { z } from 'genkit';

const queryDatabase = ai.defineTool(
  {
    name: 'queryDatabase',
    description: 'Query the customer database',
    inputSchema: z.object({
      customerId: z.string(),
    }),
    outputSchema: z.object({
      name: z.string(),
      email: z.string(),
    }),
  },
  async ({ customerId }) => {
    // Query your database
    return { name: 'John Doe', email: 'john@example.com' };
  }
);

const response = await ai.generate({
  model: vertexAI.model('gemini-2.5-flash'),
  prompt: 'Get the information for customer ID 12345',
  tools: [queryDatabase],
});

console.log(response.text());

Embeddings

const embeddings = await ai.embed({
  embedder: vertexAI.embedder('text-embedding-005'),
  content: 'Vertex AI is a comprehensive ML platform',
});

console.log(embeddings); // 768-dimensional vector

Model Garden Access

Vertex AI Model Garden provides access to models from various providers:

Anthropic Claude via Model Garden

import { genkit } from 'genkit';
import { vertexAI } from '@genkit-ai/google-genai';

const ai = genkit({
  plugins: [
    vertexAI({
      location: 'us-east5',
      projectId: 'your-project',
    }),
  ],
});

const response = await ai.generate({
  model: 'vertexai/claude-3-5-sonnet-v2@20241022',
  prompt: 'Explain relativity',
});

Meta Llama Models

const response = await ai.generate({
  model: 'vertexai/llama-3-1-405b',
  prompt: 'Write a poem about AI',
});

Vector Search Integration

Vertex AI Vector Search enables high-performance similarity search for RAG applications.

Setup Vector Search Index

import { vertexAIVectorStore } from '@genkit-ai/google-genai';

const vectorStore = vertexAIVectorStore({
  projectId: 'your-project',
  location: 'us-central1',
  indexId: 'your-index-id',
  indexEndpointId: 'your-endpoint-id',
});

// Index documents
await vectorStore.index([
  { content: 'Document 1 content...' },
  { content: 'Document 2 content...' },
]);

Retrieval-Augmented Generation (RAG)

import { retrieve } from '@genkit-ai/rag';

const response = await ai.generate({
  model: vertexAI.model('gemini-2.5-flash'),
  prompt: 'What is our return policy?',
  context: await retrieve({
    retriever: vectorStore.retriever(),
    query: 'return policy',
    limit: 5,
  }),
});

Configuration Options

Regional Endpoints

Choose the region closest to your users:

vertexAI({ location: 'us-central1' })  // Iowa, USA
vertexAI({ location: 'us-west1' })     // Oregon, USA
vertexAI({ location: 'europe-west1' }) // Belgium
vertexAI({ location: 'asia-northeast1' }) // Tokyo

Global Endpoint

vertexAI({ location: 'global' })

Some features like Lyria are only available in specific regions.

Model Configuration

const response = await ai.generate({
  model: vertexAI.model('gemini-2.5-pro'),
  prompt: 'Generate a story',
  config: {
    temperature: 1.2,
    topK: 40,
    topP: 0.95,
    maxOutputTokens: 2048,
    stopSequences: ['THE END'],
  },
});

Advanced Features

Model Tuning

Vertex AI supports fine-tuning Gemini models on your data:

# Create a tuning job (via gcloud CLI)
gcloud ai models upload \
  --region=us-central1 \
  --display-name=my-tuned-model \
  --container-image-uri=gcr.io/cloud-aiplatform/training/...

Then use your tuned model:

const response = await ai.generate({
  model: 'vertexai/projects/YOUR_PROJECT/locations/us-central1/models/my-tuned-model',
  prompt: 'Test my tuned model',
});

Model Evaluation

Use Vertex AI’s evaluation tools to assess model performance:

import { evaluate } from '@genkit-ai/google-genai';

const results = await evaluate({
  model: vertexAI.model('gemini-2.5-flash'),
  testCases: [
    { input: 'What is 2+2?', expectedOutput: '4' },
    { input: 'Capital of France?', expectedOutput: 'Paris' },
  ],
  metrics: ['accuracy', 'latency'],
});

console.log(results);

Request Logging

Vertex AI automatically logs requests for monitoring and debugging:

// View logs in Cloud Console:
// https://console.cloud.google.com/logs

Vertex AI vs Google AI

Feature	Vertex AI	Google AI
Authentication	ADC or API Key	API Key only
Setup Complexity	Moderate (GCP project)	Simple
Production Ready	Yes	Limited
Model Access	All models + Model Garden	Gemini models only
IAM Integration	Full GCP IAM	None
Vector Search	Yes	No
Fine-tuning	Yes	No
Monitoring	Cloud Logging/Monitoring	None
SLA	Available	None
Pricing	Volume discounts available	Pay-per-use
Best For	Enterprise, Production	Prototyping

Best Practices

Use regional endpoints close to your users for lower latency
Enable context caching for prompts with repeated large context
Implement proper IAM controls with service accounts
Monitor usage with Cloud Monitoring
Use Vector Search for RAG applications instead of manual similarity
Enable request logging for debugging and auditing
Set up alerts for quota limits and errors
Use Model Garden to access diverse models without managing multiple APIs

Troubleshooting

Authentication Errors

Error: Could not load the default credentials

Solution:

gcloud auth application-default login

Or set GOOGLE_APPLICATION_CREDENTIALS to your service account key:

export GOOGLE_APPLICATION_CREDENTIALS=/path/to/service-account.json

API Not Enabled

Error: Vertex AI API has not been used in project...

Solution:

gcloud services enable aiplatform.googleapis.com

Permission Denied

Error: Permission denied on resource project

Solution: Ensure your account or service account has the required IAM roles:

gcloud projects add-iam-policy-binding YOUR_PROJECT \
  --member='serviceAccount:YOUR_SA@YOUR_PROJECT.iam.gserviceaccount.com' \
  --role='roles/aiplatform.user'

Region Not Supported

Some models are only available in specific regions. Check the Vertex AI documentation for model availability.

Pricing

Vertex AI pricing varies by:

Model type (Flash vs Pro)
Input/output token count
Additional features (context caching, grounding)
Region

See the Vertex AI Pricing page for details. Cost Optimization Tips:

Use gemini-2.5-flash for most tasks
Enable context caching for repeated prompts
Use batch predictions for high-volume processing
Set appropriate maxOutputTokens limits

Next Steps

Google AI Provider - Simpler alternative for prototyping
RAG with Vector Search - Build retrieval systems
Model Evaluation - Test model performance
Production Deployment - Deploy to production

​Installation

​Setup

​Method 1: Application Default Credentials (Recommended)

​Method 2: Express Mode (API Key)

​Available Models

​Text Generation (Gemini)

​Image Generation (Imagen)

​Music Generation (Lyria)

​Video Generation (Veo)

​Embeddings

​Usage Examples

​Basic Text Generation

​Music Generation with Lyria

​Image Generation with Imagen

​Multimodal Input

​Context Caching

​Grounding with Google Search

​Function Calling

​Embeddings

​Model Garden Access

​Anthropic Claude via Model Garden

​Meta Llama Models

​Vector Search Integration

​Setup Vector Search Index

​Retrieval-Augmented Generation (RAG)

​Configuration Options

​Regional Endpoints

​Global Endpoint

​Model Configuration

​Advanced Features

​Model Tuning

​Model Evaluation

​Request Logging

​Vertex AI vs Google AI

​Best Practices

​Troubleshooting

​Authentication Errors

​API Not Enabled

​Permission Denied

​Region Not Supported

​Pricing

​Next Steps

Installation

Setup

Method 1: Application Default Credentials (Recommended)

Method 2: Express Mode (API Key)

Available Models

Text Generation (Gemini)

Image Generation (Imagen)

Music Generation (Lyria)

Video Generation (Veo)

Embeddings

Usage Examples

Basic Text Generation

Music Generation with Lyria

Image Generation with Imagen

Multimodal Input

Context Caching

Grounding with Google Search

Function Calling

Embeddings

Model Garden Access

Anthropic Claude via Model Garden

Meta Llama Models

Vector Search Integration

Setup Vector Search Index

Retrieval-Augmented Generation (RAG)

Configuration Options

Regional Endpoints

Global Endpoint

Model Configuration

Advanced Features

Model Tuning

Model Evaluation

Request Logging

Vertex AI vs Google AI

Best Practices

Troubleshooting

Authentication Errors

API Not Enabled

Permission Denied

Region Not Supported

Pricing

Next Steps