LangChain Retrievers

LangChain retrievers are Runnables that take a query string and return a list of relevant Documents. A vector store is the most common source, but any search system can be a retriever. Because retrievers are Runnables, they slot straight into LCEL chains.

From Vector Store to Retriever

retriever = store.as_retriever(search_kwargs={"k": 2})
docs = retriever.invoke("Do you deliver to Dubai?")
for d in docs:
    print(d.page_content)

Sample output:
We ship to Pakistan, UAE and the UK.
Shipping takes 3-5 business days.

Search Types

• similarity (default): the k closest chunks.
• mmr: Maximal Marginal Relevance, balances relevance with diversity to avoid near-duplicates.
• similarity_score_threshold: only return results above a relevance score.

MMR Example

mmr = store.as_retriever(
    search_type="mmr",
    search_kwargs={"k": 3, "fetch_k": 20, "lambda_mult": 0.5},
)
print(len(mmr.invoke("shipping")))

Sample output:
3

fetch_k candidates are retrieved first, then k diverse ones are chosen. lambda_mult closer to 1 favours relevance, closer to 0 favours diversity.

A Simple RAG Chain with a Retriever

from langchain_core.prompts import ChatPromptTemplate
from langchain_core.output_parsers import StrOutputParser
from langchain_core.runnables import RunnablePassthrough
from langchain_ollama import ChatOllama

def format_docs(docs):
    return "\n\n".join(d.page_content for d in docs)

prompt = ChatPromptTemplate.from_template(
    "Answer using only this context. If unsure, say you don't know.\n\n"
    "Context:\n{context}\n\nQuestion: {question}"
)
rag = (
    {"context": retriever | format_docs, "question": RunnablePassthrough()}
    | prompt | ChatOllama(model="llama3.2", temperature=0) | StrOutputParser()
)
print(rag.invoke("Can I return an item after two weeks?"))

Sample output:
Yes. Returns are free within 30 days, so two weeks is fine.

The dictionary runs two branches: one retrieves and formats context, the other passes the question through. This is the classic two-step RAG pattern. The RAG tutorial on this site goes much deeper.

Custom Retrievers

Any function from string to list of Documents can be wrapped with RunnableLambda and used as a retriever, for example a SQL full-text search or an API call.

Frequently asked questions

What is the difference between a vector store and a retriever?

A vector store stores and searches embeddings. A retriever is the simpler interface (query in, documents out) that chains and agents use; a vector store can produce one with as_retriever().

What does k mean in a LangChain retriever?

It is the number of documents to return. Common values are 3 to 6.

When should I use MMR?

Use MMR when your top results are often near-duplicates, such as overlapping chunks from the same page.