OpenSearchHybridRetriever

Retrieve documents from OpenSearch using a combination of BM25 keyword search and embedding-based semantic search. The component runs both retrieval methods in parallel, then combines the results using a configurable join strategy. This hybrid approach typically provides better retrieval quality than using either method alone.

Key Features

• Combines BM25 keyword search and embedding-based semantic search in a single component.
• Configurable join strategies: Reciprocal Rank Fusion (RRF), concatenate, merge, and distribution-based rank fusion.
• Separate filter policies for BM25 and embedding retrieval.
• Configurable top-k limits for each retrieval method and for the final combined result.
• Built-in text embedder for converting queries to embeddings.

Configuration

1. Drag the OpenSearchHybridRetriever component onto the canvas from the Component Library.
2. Click on the component to open the configuration panel.
3. On the General tab:
   • Configure the embedder (a text embedder for semantic search, such as SentenceTransformersTextEmbedder).
   • Set top_k_bm25 and top_k_embedding to control the number of documents retrieved by each method.
   • Set top_k to limit the final number of combined documents returned.
   • Optionally configure filters_bm25 and filters_embedding.
4. Go to the Advanced tab to configure fuzziness, scale_score, all_terms_must_match, filter_policy_bm25, filter_policy_embedding, custom_query_bm25, custom_query_embedding, join_mode, weights, and sort_by_score.

Connections

OpenSearchHybridRetriever receives a query string from the Input component. It outputs a documents list. Connect the documents output to a PromptBuilder, a Ranker, or directly to an LLM component.

Source Code

To check this component's source code, open open_search_hybrid_retriever.py in the Haystack Core Integrations repository.

Usage Examples

Basic Configuration

  hybrid_retriever:
    type: haystack_integrations.components.retrievers.opensearch.open_search_hybrid_retriever.OpenSearchHybridRetriever
    init_parameters:
      document_store:
        type: haystack_integrations.document_stores.opensearch.document_store.OpenSearchDocumentStore
        init_parameters:
          index: default
          max_chunk_bytes: 104857600
          embedding_dim: 768
          return_embedding: false
          create_index: true
      embedder:
        type: deepset_cloud_custom_nodes.embedders.nvidia.text_embedder.DeepsetNvidiaTextEmbedder
        init_parameters:
          normalize_embeddings: true
          model: intfloat/e5-base-v2
      fuzziness: AUTO
      top_k_bm25: 20
      scale_score: false
      all_terms_must_match: false
      filter_policy_bm25: replace
      top_k_embedding: 20
      filter_policy_embedding: replace
      join_mode: reciprocal_rank_fusion
      top_k: 10
      sort_by_score: true

This is an example RAG pipeline with OpenSearchHybridRetriever combining BM25 and embedding-based retrieval:

components:
  hybrid_retriever:
    type: haystack_integrations.components.retrievers.opensearch.open_search_hybrid_retriever.OpenSearchHybridRetriever
    init_parameters:
      document_store:
        type: haystack_integrations.document_stores.opensearch.document_store.OpenSearchDocumentStore
        init_parameters:
          hosts:
          index: 'default'
          max_chunk_bytes: 104857600
          embedding_dim: 768
          return_embedding: false
          method:
          mappings:
          settings:
          create_index: true
          http_auth:
          use_ssl:
          verify_certs:
          timeout:
      embedder:
        type: deepset_cloud_custom_nodes.embedders.nvidia.text_embedder.DeepsetNvidiaTextEmbedder
        init_parameters:
          normalize_embeddings: true
          model: intfloat/e5-base-v2
      filters_bm25:
      fuzziness: AUTO
      top_k_bm25: 20
      scale_score: false
      all_terms_must_match: false
      filter_policy_bm25: replace
      custom_query_bm25:
      filters_embedding:
      top_k_embedding: 20
      filter_policy_embedding: replace
      custom_query_embedding:
      join_mode: reciprocal_rank_fusion
      weights:
      top_k: 10
      sort_by_score: true

  ranker:
    type: deepset_cloud_custom_nodes.rankers.nvidia.ranker.DeepsetNvidiaRanker
    init_parameters:
      model: intfloat/simlm-msmarco-reranker
      top_k: 8

  meta_field_grouping_ranker:
    type: haystack.components.rankers.meta_field_grouping_ranker.MetaFieldGroupingRanker
    init_parameters:
      group_by: file_id
      subgroup_by:
      sort_docs_by: split_id

  answer_builder:
    type: deepset_cloud_custom_nodes.augmenters.deepset_answer_builder.DeepsetAnswerBuilder
    init_parameters:
      reference_pattern: acm

  PromptBuilder:
    type: haystack.components.builders.prompt_builder.PromptBuilder
    init_parameters:
      template: "          You are a technical expert.\n          You answer questions truthfully based on provided documents.\n          If the answer exists in several documents, summarize them.\n          Ignore documents that don't contain the answer to the question.\n          Only answer based on the documents provided. Don't make things up.\n          If no information related to the question can be found in the document, say so.\n          Always use references in the form [NUMBER OF DOCUMENT] when using information from a document, e.g. [3] for Document [3] .\n          Never name the documents, only enter a number in square brackets as a reference.\n          The reference must only refer to the number that comes in square brackets after the document.\n          Otherwise, do not use brackets in your answer and reference ONLY the number of the document without mentioning the word document.\n\n          These are the documents:\n          {%- if documents|length > 0 %}\n          {%- for document in documents %}\n          Document [{{ loop.index }}] :\n          Name of Source File: {{ document.meta.file_name }}\n          {{ document.content }}\n          {% endfor -%}\n          {%- else %}\n          No relevant documents found.\n          Respond with \"Sorry, no matching documents were found, please adjust the filters or try a different question.\"\n          {% endif %}\n\n          Question: {{ question }}\n          Answer:"

      required_variables:
      variables:
  OpenAIGenerator:
    type: haystack.components.generators.openai.OpenAIGenerator
    init_parameters:
      api_key:
        type: env_var
        env_vars:
        - OPENAI_API_KEY
        strict: false
      model: gpt-5-mini
      streaming_callback:
      api_base_url:
      organization:
      system_prompt:
      generation_kwargs:
      timeout:
      max_retries:
      http_client_kwargs:

connections:
- sender: hybrid_retriever.documents
  receiver: ranker.documents
- sender: ranker.documents
  receiver: meta_field_grouping_ranker.documents
- sender: meta_field_grouping_ranker.documents
  receiver: answer_builder.documents
- sender: meta_field_grouping_ranker.documents
  receiver: PromptBuilder.documents
- sender: PromptBuilder.prompt
  receiver: OpenAIGenerator.prompt
- sender: OpenAIGenerator.replies
  receiver: answer_builder.replies

inputs:
  query:
  - "hybrid_retriever.query"
  - "ranker.query"
  - "PromptBuilder.question"
  - "answer_builder.query"

outputs:
  documents: "meta_field_grouping_ranker.documents"
  answers: "answer_builder.answers"

max_runs_per_component: 100

metadata: {}

Parameters

Inputs

Parameter	Type	Default	Description
query	str		The query string to search for.
filters_bm25	Optional[Dict[str, Any]]	None	Filters to apply during BM25 retrieval.
filters_embedding	Optional[Dict[str, Any]]	None	Filters to apply during embedding retrieval.
top_k	Optional[int]	None	Maximum number of documents to return from the combined results.

Outputs

Parameter	Type	Default	Description
documents	List[Document]		Documents retrieved and ranked using hybrid search.

Init Parameters

These are the parameters you can configure in Pipeline Builder:

Parameter	Type	Default	Description
document_store	OpenSearchDocumentStore		An instance of OpenSearchDocumentStore to use with the retriever.
embedder	TextEmbedder		A text embedder to embed the query for semantic search.
filters_bm25	Optional[Dict[str, Any]]	None	Default filters for BM25 retrieval.
fuzziness	Union[int, str]	"AUTO"	The fuzziness setting for BM25 retrieval.
top_k_bm25	int	10	Number of documents to return from BM25 retrieval.
scale_score	bool	False	Whether to scale the BM25 scores.
all_terms_must_match	bool	False	Whether all query terms must match in BM25 retrieval.
filter_policy_bm25	Union[str, FilterPolicy]	"replace"	How to apply runtime filters for BM25. Options: replace, merge.
custom_query_bm25	Optional[Dict[str, Any]]	None	A custom OpenSearch query for BM25 retrieval.
filters_embedding	Optional[Dict[str, Any]]	None	Default filters for embedding retrieval.
top_k_embedding	int	10	Number of documents to return from embedding retrieval.
filter_policy_embedding	Union[str, FilterPolicy]	"replace"	How to apply runtime filters for embedding retrieval. Options: replace, merge.
custom_query_embedding	Optional[Dict[str, Any]]	None	A custom OpenSearch query for embedding retrieval.
join_mode	Union[str, JoinMode]	"reciprocal_rank_fusion"	How to combine results from both retrievers. Options: concatenate, merge, reciprocal_rank_fusion, distribution_based_rank_fusion.
weights	Optional[List[float]]	None	Weights for the joiner when combining results.
top_k	Optional[int]	None	Final number of documents to return after combining results.
sort_by_score	bool	True	Whether to sort the final results by score.

Run Method Parameters

These are the parameters you can configure for the component's run() method. This means you can pass these parameters at query time through the API, in Playground, or when running a job. For details, see Modify Pipeline Parameters at Query Time.

Parameter	Type	Default	Description
query	str		The query string to search for.
filters_bm25	Optional[Dict[str, Any]]	None	Filters to apply during BM25 retrieval. The way filters are applied depends on filter_policy_bm25.
filters_embedding	Optional[Dict[str, Any]]	None	Filters to apply during embedding retrieval. The way filters are applied depends on filter_policy_embedding.
top_k	Optional[int]	None	Maximum number of documents to return. Overrides the value set at initialization.

Related Information

• OpenSearchBM25Retriever
• OpenSearchEmbeddingRetriever