Link Search Menu Expand Document Documentation Menu
Documentation

OpenSearch-provided connector blueprints

OpenSearch provides two types of connector blueprints:

  • Standard blueprints (recommended for OpenSearch 2.14+):

    • Designed for connectors that pass input directly to the model and return its raw output without requiring additional transformations.
    • Use a registered model without any pre-processing or post-processing functions.
    • Compatible with machine learning (ML) inference processors: ingest processors, search request processors, and search response processors.
    • Ideal for new implementations in OpenSearch 2.14 and later.

  • Legacy blueprints:

    • Suitable for specific input and output formatting requirements.
    • Include pre- and post-processing functions.
    • Recommended for existing implementations.
    • Compatible with neural search queries.

Find your model and platform

OpenSearch provides connector blueprints for the following platforms and models. Each blueprint link provides an example request for creating the connector, including all required parameters, and an example Predict API request. The Protocol column links to the credential fields and cluster prerequisites for that platform’s authentication method. For an overview of the available protocols, see Connector authentication.

Platform Model Protocol Legacy blueprint Standard blueprint
Aleph Alpha Luminous-Base Embedding Model http Blueprint N/A
Amazon Bedrock AI21 Labs Jurassic-2 Mid aws_sigv4 Blueprint Blueprint
Amazon Bedrock Anthropic Claude v2 aws_sigv4 Blueprint Blueprint
Amazon Bedrock Anthropic Claude v3 aws_sigv4 Blueprint Blueprint
Amazon Bedrock Anthropic Claude v3.7 aws_sigv4 Blueprint Blueprint
Amazon Bedrock Anthropic Claude v4 aws_sigv4 Blueprint Blueprint
Amazon Bedrock Cohere Embed Model v3 - English aws_sigv4 Blueprint Blueprint
Amazon Bedrock Cohere Embed Model v3 - Multilingual aws_sigv4 Blueprint Blueprint
Amazon Bedrock Titan Text Embeddings aws_sigv4 Blueprint Blueprint
Amazon Bedrock Titan Multimodal Embeddings aws_sigv4 Blueprint Blueprint
Amazon Bedrock Nova Multimodal Embeddings aws_sigv4 Blueprint Blueprint
Amazon Bedrock OpenAI GPT-OSS aws_sigv4 Blueprint Blueprint
Amazon Bedrock Converse Anthropic Claude 3 Sonnet aws_sigv4 Blueprint Blueprint
Amazon Comprehend Metadata Embedding Model aws_sigv4 Blueprint Blueprint
Amazon Textract Amazon Textract Model aws_sigv4 Blueprint Blueprint
Amazon SageMaker Text embedding models for batch inference aws_sigv4 Blueprint Blueprint
Amazon SageMaker Text embedding models aws_sigv4 Blueprint Blueprint
Amazon SageMaker Semantic highlighter models aws_sigv4 N/A Blueprint
Azure OpenAI Chat models (for example, gpt-4) http Blueprint Blueprint
Azure OpenAI Text embedding models (for example, text-embedding-ada-002) http Blueprint Blueprint
Cohere Chat models http Blueprint Blueprint
Cohere Multimodal embedding models (for example, embed-english-v3.0) http Blueprint Blueprint
Cohere Text embedding models http Blueprint Blueprint
DeepSeek Chat model http Blueprint Blueprint
Google Cloud Vertex AI Gemini google_cloud N/A Blueprint
Google Cloud Vertex AI text embeddings google_cloud N/A Blueprint
Google Cloud Vertex AI text embeddings (manually refreshed token) http N/A Blueprint
Google Cloud Vertex AI Gemini streaming google_cloud N/A Blueprint
Google Cloud Vertex AI batch inference google_cloud N/A Blueprint
Ollama Chat models http Blueprint Blueprint
OpenAI Chat models (for example, gpt-4o-mini) http Blueprint Blueprint
OpenAI Completion models (legacy; OpenAI retired the /v1/completions endpoint) http Blueprint Blueprint
OpenAI Text embedding models (for example, text-embedding-ada-002) http Blueprint Blueprint
OpenAI Text embedding models for batch inference http Blueprint Blueprint
Yandex Cloud Text embedding models http Blueprint Blueprint

LLM judgment blueprints for Search Relevance Workbench

Search Relevance Workbench uses a chat model to generate relevance ratings using an LLM-as-a-Judge. For an overview of these blueprints, see LLM judgment blueprints. The following table lists the available blueprints for connecting Search Relevance Workbench to a chat model.

Provider Protocol Blueprint
OpenAI http GPT-5, GPT-4.1, GPT-4o
Azure OpenAI http GPT-5, GPT-4.1, GPT-4o
DeepSeek http DeepSeek-V3, DeepSeek-R1
Ollama and other local OpenAI-compatible servers http Qwen 3, Llama 3.3, Mistral, Phi-4, Gemma 3
Google Gemini http Gemini 2.5 Pro, Gemini 2.5 Flash
Amazon Bedrock aws_sigv4 Anthropic Claude (native invoke API)
Amazon Bedrock Converse aws_sigv4 Anthropic Claude, Amazon Nova, Meta Llama, Mistral, AI21, Cohere

For more information about using these blueprints with Search Relevance Workbench, see Using different LLM providers.

Next steps

350 characters left

Have a question? .

Want to contribute? or .