Providers Overview

barrel_embed supports multiple embedding providers, each with different characteristics and use cases.

Cloud Providers

Production-ready cloud embedding APIs with high availability and no infrastructure to manage.

Local Providers

Run embedding models locally without external API calls.

Specialized Providers

Advanced embedding types for specific use cases.

Choosing a Provider

For Production

Cloud providers are recommended:

• OpenAI - Best general-purpose quality
• Cohere - Input type optimization (search_query vs search_document)
• Voyage AI - Top retrieval performance, domain-specific models
• Jina AI - Longest context length (8K tokens)

For Enterprise

Hyperscaler providers with compliance features:

• Azure OpenAI - SOC 2, HIPAA, regional deployment
• AWS Bedrock - VPC integration, IAM, CloudWatch
• Google Vertex AI - VPC-SC, CMEK, BigQuery integration

For EU Data Residency

• Mistral - EU-based company
• Azure OpenAI - Deploy in EU regions
• Google Vertex AI - Deploy in EU regions

For Local Development

Ollama is recommended:

• No Python dependencies
• Easy model management
• Good performance

For Offline/Air-gapped

Local or FastEmbed:

• No external API calls
• Full data privacy
• FastEmbed is lighter (~100MB vs ~2GB)

For Specialized Use Cases

• Hybrid search: Use SPLADE for sparse + dense combination
• Passage retrieval: Use ColBERT for token-level matching
• Image search: Use CLIP for cross-modal embeddings

Provider Chain

Configure fallback providers for high availability:

#{embedder => [
    {voyage, #{api_key => <<"pa-...">>}},      % Primary: best retrieval
    {openai, #{api_key => <<"sk-...">>}},      % Fallback: OpenAI
    {ollama, #{url => <<"http://localhost:11434">>}}  % Offline fallback
]}

Providers are tried in order. If one fails, the next is attempted automatically.

Quick Comparison