Support for Pre-Embedded MongoDB Atlas Vector Search in LangFlow

[ealss101]

Description
I am working on integrating LangFlow with MongoDB Atlas Vector Search, but running into an issue where LangChain’s MongoDBAtlasVectorSearch requires an embedding parameter, even though my backend already handles embedding before storing vectors in MongoDB.

My Setup

• Documents go through an ingestion pipeline where they are chunked, embedded, and stored in MongoDB Atlas.
• The vector search should only retrieve precomputed embeddings, not create new ones.
• LangFlow needs to take a search query and return relevant document chunks.

Issue

• The LangFlow component for MongoDB Atlas requires embedding, but this is not needed since vectors already exist.
• Passing a dummy embedding function did not work.
• Looking for a way to initialize MongoDBAtlasVectorSearch in LangFlow without needing an embedding function.

Expected Behavior

• The LangFlow component should allow pure vector search without requiring an embedding input.
• It should connect directly to the stored vector database and return search results based on existing embeddings.

Steps to Reproduce

• Set up MongoDB Atlas Vector Search with precomputed embeddings.
• Create a LangFlow component using MongoDBAtlasVectorSearch.
• Try to perform a search without providing an embedding function.
• Observe the error related to missing embedding.

Proposed Solution

• Allow an option to disable embedding when using MongoDBAtlasVectorSearch in LangFlow.
• Provide a direct query mode for retrieving stored vectors.
• Has anyone faced this issue before? Would appreciate any input on how to structure LangFlow to handle this scenario.