Voyage AI

Voyage AI offers high-performance embedding models including domain-specific and multimodal variants.

Overview

Models: 11 models including specialized variants
Key Features: Domain-specific models (finance, law, code), multimodal support, quantization
API Docs: Voyage AI Documentation

Environment Variable

export VOYAGE_API_KEY="your-voyage-api-key"

Supported Parameters

Parameter	Type	Required	Description
`model`	str	Yes	Model identifier
`input`	str \| List[str]	Yes	Text(s) to embed
`input_type`	str	No	`"query"` or `"document"`
`dimensions`	int	No	Output dimensions: 256, 512, 1024, 2048
`api_key`	str	No	Override API key for this request

Examples

Basic Usage

import catsu

client = catsu.Client()

response = client.embed(
    model="voyage-3",
    input="Hello, world!"
)

print(f"Dimensions: {response.dimensions}")  # 1024
print(f"Cost: ${response.usage.cost:.8f}")

With input_type

# For queries
query_response = client.embed(
    model="voyage-3",
    input="What is machine learning?",
    input_type="query"
)

# For documents
doc_response = client.embed(
    model="voyage-3",
    input="Machine learning is a subset of artificial intelligence...",
    input_type="document"
)

With Custom Dimensions

# Smaller dimensions for faster similarity search
response = client.embed(
    model="voyage-3",
    input="Sample text",
    dimensions=256  # vs default 1024
)

print(f"Dimensions: {response.dimensions}")  # 256

Domain-Specific Models

# Finance-specific
finance_response = client.embed(
    model="voyage-finance-2",
    input="The company's Q4 earnings exceeded expectations..."
)

# Law-specific
law_response = client.embed(
    model="voyage-law-2",
    input="In accordance with Section 12(a) of the statute..."
)

# Code-specific
code_response = client.embed(
    model="voyage-code-3",
    input="def calculate_embeddings(texts): pass"
)

Multimodal Support

# Text and image embeddings
response = client.embed(
    model="voyage-multimodal-3",
    input="Description of an image"
)

With Quantization (voyage-3.5)

voyage-3.5 supports binary quantization for smaller storage:

# Note: Quantization is configured on the provider side
# Catsu returns float embeddings by default
response = client.embed(
    model="voyage-3.5",
    input="Text"
)

Async Usage

import asyncio

async def main():
    client = catsu.Client()

    response = await client.aembed(
        model="voyage-3",
        input="Async embedding generation",
        input_type="document"
    )

    print(response.embeddings)

asyncio.run(main())

Model Variants

Voyage AI offers several model variants:

voyage-3-large - Highest quality
voyage-3.5 - Latest version with quantization
voyage-3 - Balanced performance
voyage-3-lite, voyage-3.5-lite - Faster, lower cost
voyage-code-3 - Code search and understanding
voyage-finance-2 - Financial documents
voyage-law-2 - Legal documents
voyage-multilingual-2 - Multi-language support
voyage-multimodal-3 - Text and image

For complete model details, visit catsu.dev.

Special Notes

Supports both input_type="query" and input_type="document"
Dimensions can be 256, 512, 1024, or 2048
voyage-3.5 supports binary quantization for efficient storage
Domain-specific models optimized for finance, law, and code
Maximum input tokens vary by model (check via list_models())

Next Steps

Common Parameters - Learn about input_type and dimensions
Best Practices: Model Selection - Choose the right model

On this page