SentenceWindowRetriever

Retrieve documents adjacent to a given document in the document store to provide fuller context. When an initial retriever finds a relevant sentence, SentenceWindowRetriever fetches the surrounding sentences before and after it, giving the LLM more context to work with.

Key Features

• Expands retrieval context by fetching neighboring documents around each retrieved document.
• Configurable window size to control how many surrounding documents to include.
• Compatible with BM25 and embedding-based retrievers.
• Supports multiple document store backends (OpenSearch, Elasticsearch, Pgvector, Pinecone, Qdrant, Astra).
• Flexible source document identification using single or multiple metadata fields.

Configuration

1. Drag the SentenceWindowRetriever 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 document_store to retrieve neighboring documents from.
   • Set window_size to control how many documents to fetch before and after each retrieved document (default: 3).
4. Go to the Advanced tab to configure source_id_meta_field, split_id_meta_field, and raise_on_missing_meta_fields.

Connections

SentenceWindowRetriever receives a list of retrieved documents through its retrieved_documents input, typically from a BM25 or embedding retriever. It outputs two values: context_windows (a list of concatenated strings) and context_documents (a list of document objects including context). Connect context_documents to a prompt builder or LLM for downstream processing.

Source Code

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

Usage Examples

Basic Configuration

  SentenceWindowRetriever:
    type: components.retrievers.sentence_window_retriever.SentenceWindowRetriever
    init_parameters:
      document_store:
        type: haystack_integrations.document_stores.opensearch.document_store.OpenSearchDocumentStore
        init_parameters:
          index: Standard-Index-English
      window_size: 3
      source_id_meta_field: source_id

components:
  SentenceWindowRetriever:
    type: components.retrievers.sentence_window_retriever.SentenceWindowRetriever
    init_parameters:
      document_store:
        type: haystack_integrations.document_stores.opensearch.document_store.OpenSearchDocumentStore
        init_parameters:
          hosts:
          index: 'Standard-Index-English'
      window_size: 3
      source_id_meta_field: source_id  # Can also be a list: ["source_id", "file_id"]

Parameters

Inputs

Parameter	Type	Default	Description
retrieved_documents	List[Document]		List of retrieved documents from the previous retriever.
window_size	Optional[int]	None	The number of documents to retrieve before and after the relevant one. Overrides the window_size parameter set at initialization.

Outputs

Parameter	Type	Default	Description
context_windows	List[str]		A list of strings, where each string represents the concatenated text from the context window of the corresponding document in retrieved_documents.
context_documents	List[Document]		A list of Document objects, containing the retrieved documents plus the context documents surrounding them. The documents are sorted by the split_idx_start metadata field.

Init Parameters

These are the parameters you can configure in Pipeline Builder:

Parameter	Type	Default	Description
document_store	DocumentStore		The document store to retrieve the surrounding documents from.
window_size	int	3	The number of documents to retrieve before and after the relevant one. For example, window_size: 2 fetches two preceding and two following documents.
source_id_meta_field	Union[str, List[str]]	source_id	The metadata field containing the ID of the original document. Can be a single field name or a list of field names. When a list is provided, only documents matching all specified fields are retrieved.
split_id_meta_field	str	"split_id"	The metadata field that contains the split ID of the document.
raise_on_missing_meta_fields	bool	True	If True, raises an error if the documents do not contain the required metadata fields. If False, skips retrieving context for documents with missing metadata fields, but still includes the original document in the results.

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
retrieved_documents	List[Document]		List of retrieved documents from the previous retriever.
window_size	Optional[int]	None	The number of documents to retrieve before and after the relevant one. Overrides the window_size parameter set at initialization.