Overview

Catsu supports 11 major embedding providers with a unified, consistent API.

Supported Providers

Provider	Models	Key Features	Environment Variable
Voyage AI	11	Domain-specific, multimodal, quantization	`VOYAGE_API_KEY`
OpenAI	3	Industry standard, Matryoshka	`OPENAI_API_KEY`
Cohere	5	Multilingual, truncation	`COHERE_API_KEY`
Gemini	1	Long context, Matryoshka	`GEMINI_API_KEY`
Jina AI	6	Multimodal, code-specific, long context	`JINA_API_KEY`
Mistral AI	2	Code-optimized, quantization	`MISTRAL_API_KEY`
Nomic	2	Long text handling, Matryoshka	`NOMIC_API_KEY`
Cloudflare	7	Edge inference, BGE models	`CLOUDFLARE_API_KEY`
DeepInfra	16	Open-source models, Qwen3	`DEEPINFRA_API_KEY`
Mixed Bread	4	Multilingual, quantization	`MIXEDBREAD_API_KEY`
Together AI	7	Open-source, long context	`TOGETHERAI_API_KEY`

For detailed model information including pricing and benchmarks, visit the Models Catalog.

Feature Comparison

Input Type Support

Providers that support input_type="query" or input_type="document":

✅ Voyage AI, Cohere, Gemini, Jina AI, Mistral AI, Nomic, Mixed Bread
❌ OpenAI, Cloudflare, DeepInfra, Together AI (parameter ignored)

Matryoshka Embeddings (dimensions)

Providers that support custom embedding dimensions:

✅ Voyage AI, Gemini, Jina AI, Mistral AI (select models), Nomic, OpenAI (text-embedding-3), DeepInfra (Qwen3), Mixed Bread
❌ Cohere, Cloudflare, Together AI

Quantization Support

Providers that support binary or int8 embeddings:

✅ Voyage AI (voyage-3.5), Mistral AI (codestral-embed-2505), Mixed Bread
Partial: Jina AI, OpenAI
❌ Most other providers (float only)

Quick Examples

Using Different Providers

import catsu

client = catsu.Client()

# Voyage AI
voyage_response = client.embed(model="voyage-3", input="Text")

# OpenAI
openai_response = client.embed(model="text-embedding-3-small", input="Text")

# Cohere
cohere_response = client.embed(model="embed-v4.0", input="Text")

Provider-Specific Features

# Voyage AI with domain-specific model
finance = client.embed(model="voyage-finance-2", input="Financial text")

# Jina AI with multimodal
multimodal = client.embed(model="jina-embeddings-v4", input="Text with image support")

# Gemini with long context (2048 tokens)
long_text = client.embed(model="gemini-embedding-001", input="Very long document...")

# Mistral with code optimization
code = client.embed(model="codestral-embed-2505", input="def hello(): pass")