VoyageAI

Specialized embeddings optimized for search and retrieval tasks

Configuration

'voyageai' => [
    'api_key' => env('VOYAGEAI_API_KEY', ''),
    'url' => env('VOYAGEAI_URL', 'https://api.voyageai.com/v1'),
],

Basic Usage

VoyageAI specializes in generating high-quality embeddings for semantic search and retrieval:

use Prism\Prism\Enums\Provider;
use Prism\Prism\Facades\Prism;

$response = Prism::embeddings()
    ->using(Provider::VoyageAI, 'voyage-3-lite')
    ->fromInput('The food was delicious and the waiter was very friendly.')
    ->asEmbeddings();

// Access the embedding vector
$embedding = $response->embeddings->first()->embedding;

Provider-Specific Options

Input Type

By default, VoyageAI generates general purpose vectors. However, they tailor your vectors for the task they are intended for - for search (“query”) or for retrieval (“document”):

For Search / Querying

use Prism\Prism\Enums\Provider;
use Prism\Prism\Facades\Prism;

Prism::embeddings()
    ->using(Provider::VoyageAI, 'voyage-3-lite')
    ->fromInput('best restaurants in San Francisco')
    ->withProviderOptions(['inputType' => 'query'])
    ->asEmbeddings();

For Document Retrieval

Prism::embeddings()
    ->using(Provider::VoyageAI, 'voyage-3-lite')
    ->fromInput('The food was delicious and the waiter was very friendly.')
    ->withProviderOptions(['inputType' => 'document'])
    ->asEmbeddings();

Truncation

By default, VoyageAI truncates inputs that are over the context length. You can force it to throw an error instead by setting truncation to false:

Prism::embeddings()
    ->using(Provider::VoyageAI, 'voyage-3-lite')
    ->fromInput('Your very long text here...')
    ->withProviderOptions(['truncation' => false])
    ->asEmbeddings();

Use Cases

Semantic Search

VoyageAI embeddings excel at semantic search applications:

use Prism\Prism\Facades\Prism;

// Index documents
$documents = [
    'Laravel is a PHP web framework',
    'Vue.js is a JavaScript framework',
    'Python is a programming language',
];

$documentEmbeddings = [];
foreach ($documents as $doc) {
    $response = Prism::embeddings()
        ->using('voyageai', 'voyage-3')
        ->fromInput($doc)
        ->withProviderOptions(['inputType' => 'document'])
        ->asEmbeddings();
    
    $documentEmbeddings[] = [
        'text' => $doc,
        'embedding' => $response->embeddings->first()->embedding,
    ];
}

// Search with a query
$queryResponse = Prism::embeddings()
    ->using('voyageai', 'voyage-3')
    ->fromInput('PHP frameworks')
    ->withProviderOptions(['inputType' => 'query'])
    ->asEmbeddings();

$queryEmbedding = $queryResponse->embeddings->first()->embedding;

// Calculate similarity and find best match
// (use cosine similarity or your preferred metric)

Batch Processing

Process multiple texts efficiently:

$texts = [
    'First document to embed',
    'Second document to embed',
    'Third document to embed',
];

$response = Prism::embeddings()
    ->using('voyageai', 'voyage-3-lite')
    ->fromInput($texts)
    ->withProviderOptions(['inputType' => 'document'])
    ->asEmbeddings();

foreach ($response->embeddings as $index => $embedding) {
    echo "Embedding for: {$texts[$index]}\n";
    // Store $embedding->embedding in your database
}

RAG (Retrieval-Augmented Generation)

Use VoyageAI embeddings for RAG applications:

// 1. Embed your knowledge base
$knowledgeBase = [
    'Laravel provides elegant routing',
    'Laravel has built-in authentication',
    'Laravel supports multiple databases',
];

$embeddedKnowledge = [];
foreach ($knowledgeBase as $item) {
    $response = Prism::embeddings()
        ->using('voyageai', 'voyage-3')
        ->fromInput($item)
        ->withProviderOptions(['inputType' => 'document'])
        ->asEmbeddings();
    
    $embeddedKnowledge[] = [
        'text' => $item,
        'embedding' => $response->embeddings->first()->embedding,
    ];
}

// 2. Embed user query
$userQuery = 'How does authentication work?';
$queryResponse = Prism::embeddings()
    ->using('voyageai', 'voyage-3')
    ->fromInput($userQuery)
    ->withProviderOptions(['inputType' => 'query'])
    ->asEmbeddings();

// 3. Find relevant context (using similarity search)
// 4. Pass context to your LLM for generation

Available Models

Voyage 3 Series

• voyage-3 - Latest and most capable model
• voyage-3-lite - Faster and more cost-effective

Voyage 2 Series

• voyage-2 - Previous generation
• voyage-2-lite - Lighter version

Specialized Models

• voyage-code-2 - Optimized for code search
• voyage-law-2 - Optimized for legal documents
• voyage-finance-2 - Optimized for financial documents

Best Practices

Use Appropriate Input Types

// For indexing documents
$docResponse = Prism::embeddings()
    ->using('voyageai', 'voyage-3')
    ->fromInput($documentText)
    ->withProviderOptions(['inputType' => 'document'])
    ->asEmbeddings();

// For search queries
$queryResponse = Prism::embeddings()
    ->using('voyageai', 'voyage-3')
    ->fromInput($searchQuery)
    ->withProviderOptions(['inputType' => 'query'])
    ->asEmbeddings();

Batch When Possible

Process multiple texts in a single request for better performance:

$response = Prism::embeddings()
    ->using('voyageai', 'voyage-3-lite')
    ->fromInput([
        'Document 1',
        'Document 2',
        'Document 3',
    ])
    ->withProviderOptions(['inputType' => 'document'])
    ->asEmbeddings();

Choose the Right Model

• Use voyage-3 for highest quality
• Use voyage-3-lite for speed and cost efficiency
• Use specialized models (code, law, finance) for domain-specific tasks

Features

• ✅ High-quality embeddings
• ✅ Task-specific optimization (query vs document)
• ✅ Batch processing
• ✅ Multiple specialized models
• ✅ Customizable truncation behavior
• ❌ Text generation (not supported)
• ❌ Image processing (not supported)

Performance Characteristics

• voyage-3: Best quality, higher latency
• voyage-3-lite: Good quality, lower latency
• Specialized models: Optimized for specific domains

Integration Example

Here’s a complete example of building a semantic search system:

use Prism\Prism\Facades\Prism;
use Illuminate\Support\Facades\DB;

class DocumentSearchService
{
    public function indexDocument(string $content, int $documentId): void
    {
        $response = Prism::embeddings()
            ->using('voyageai', 'voyage-3')
            ->fromInput($content)
            ->withProviderOptions(['inputType' => 'document'])
            ->asEmbeddings();
        
        $embedding = $response->embeddings->first()->embedding;
        
        DB::table('document_embeddings')->insert([
            'document_id' => $documentId,
            'embedding' => json_encode($embedding),
            'created_at' => now(),
        ]);
    }
    
    public function search(string $query): array
    {
        $response = Prism::embeddings()
            ->using('voyageai', 'voyage-3')
            ->fromInput($query)
            ->withProviderOptions(['inputType' => 'query'])
            ->asEmbeddings();
        
        $queryEmbedding = $response->embeddings->first()->embedding;
        
        // Use your database's vector search capabilities
        // This is a simplified example
        return $this->findSimilarDocuments($queryEmbedding);
    }
}```