Skip to content

The agent engineering platform.

docs(langchain-classic): clarify MultiVectorRetriever usage (#35053)

Mohammed Yaseen Ahmed Khana50d86c353bd
diff --git a/libs/langchain/langchain_classic/retrievers/multi_vector.py b/libs/langchain/langchain_classic/retrievers/multi_vector.py
index dcf438bb7f..ff5f310390 100644
--- a/libs/langchain/langchain_classic/retrievers/multi_vector.py
+++ b/libs/langchain/langchain_classic/retrievers/multi_vector.py
@@ -27,7 +27,20 @@ class SearchType(str, Enum):
class MultiVectorRetriever(BaseRetriever):
- """Retrieve from a set of multiple embeddings for the same document."""
+ """Retriever that supports multiple embeddings per parent document.
+
+ This retriever is designed for scenarios where documents are split into
+ smaller chunks for embedding and vector search, but retrieval returns
+ the original parent documents rather than individual chunks.
+
+ It works by:
+ - Performing similarity (or MMR) search over embedded child chunks
+ - Collecting unique parent document IDs from chunk metadata
+ - Fetching and returning the corresponding parent documents from the docstore
+
+ This pattern is commonly used in RAG pipelines to improve answer grounding
+ while preserving full document context.
+ """
vectorstore: VectorStore
"""The underlying `VectorStore` to use to store small chunks