Passing metadata filter at time of input

[XariZaru]

Checked other resources

• I added a very descriptive title to this question.
• I searched the LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.

Commit to Help

• I commit to help with one of those options 👆

Example Code

from typing import List

from langchain.chains.conversational_retrieval.base import ConversationalRetrievalChain
from langchain.retrievers import ParentDocumentRetriever
from langchain.storage import create_kv_docstore, LocalFileStore
from langchain_core.callbacks import CallbackManagerForRetrieverRun
from langchain_core.documents import Document
from langchain_openai import OpenAIEmbeddings, ChatOpenAI
from langchain_pinecone import PineconeVectorStore
from langchain_text_splitters import RecursiveCharacterTextSplitter


class CustomParentDocumentRetriever(ParentDocumentRetriever):

    # See https://github.com/langchain-ai/langchain/blob/61dd92f8215daef3d9cf1734b0d1f8c70c1571c3/libs/langchain/langchain/vectorstores/base.py#L500
    def _get_relevant_documents(
            self, query: str, *, run_manager: CallbackManagerForRetrieverRun
    ) -> List[Document]:
        docs_and_similarities = (
            self.vectorstore.similarity_search_with_relevance_scores(
                query, **self.search_kwargs
            )
        )

        # Make the score part of the document metadata
        for doc, similarity in docs_and_similarities:
            doc.metadata["score"] = similarity

        docs = [doc for doc, _ in docs_and_similarities]
        return docs
    
vectorstore = PineconeVectorStore(index_name="MYTEST_VECTORSTORE",
                                  embedding=OpenAIEmbeddings(model="text-embedding-3-small"))

fs = LocalFileStore("MY KV DIRECTORY")
store = create_kv_docstore(fs)
parent_splitter = RecursiveCharacterTextSplitter(chunk_size=4000)
child_splitter = RecursiveCharacterTextSplitter(chunk_size=800)
parent_retriever = CustomParentDocumentRetriever(
        vectorstore=vectorstore,  # vectorstore,
        docstore=store,
        child_splitter=child_splitter,
        parent_splitter=parent_splitter,
        # top_k=3,
        search_kwargs={"k": 6}  # , "score_threshold": .80}
    )

llm = ChatOpenAI(
    temperature=0,
    verbose=False,
    # openai_api_key=key,
    model_name="gpt-4o-mini"
)

chain = ConversationalRetrievalChain.from_llm(
    llm,
    chain_type="stuff",
    retriever=parent_retriever,
    combine_docs_chain_kwargs={"prompt": "You're an assistant that helps me infer business names given a transaction description. Return only the name of the inferred business and nothing else. ### Input: 213124 In-N-Out Garden Grove, CA #### Output: In-N-Out"},
    return_source_documents=True,
    # verbose=True,
    rephrase_question=False,
    max_tokens_limit=128000,
    response_if_no_docs_found="""I'm sorry, but I was not able to find the answer to your question based on the information I know. You may have to reach out to the respective internal department for more details regarding your inquiry."""
)

rv = chain(
    inputs={
        'question': "5341 Marshalls 2132 21st St. NY",
        'chat_history': [],
        'filter': {'origin': 'engageware'}
    },
)

Description

I have the following ConversationalRetrievalChain set up with the following objects:

1. Pinecone Vector Store
2. Local KV Directory

My vector store has a metadata attribute called "origin".

I want to pass this metadata filter at input runtime instead of at compile time.

Something like this:

results = chain(
    inputs={
        'question': "5341 Marshalls 2132 21st St. NY",
        'chat_history': [],
        'filter': {'origin': 'engageware'}
    },
)

Should give me only documents back whose origin is "engageware"

System Info

langchain==0.3.20
langchain_pinecone==0.2.3
langchain_openai==0.3.7
datarobot_drum
datarobot
pandas
langchain_community==0.3.19

[XariZaru]

Any ideas? @dosubot