For RAG (vectorstore), can files subset be dynamically selected

[vodkaslime]

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

N/A

Description

Currently as documented in https://js.langchain.com/docs/use_cases/question_answering/, it looks like we have the basic functionality to:

1. build vector store based on ALL FILES we stuff into a vector database.
2. then given a query, do an embedding and then send to vector database to retrieve most relevant k results.

This works fine in general cases, but what if I want to dynamically select files SUBSET for each chat session?

Let's say I have an app. Not multi-tenant, just single tenant, to put it simple. But here's the thing: I have a single vector database under the hood (let's say lancedb), where I stored 100 files. But in the app on top of it, for different chat sessions, I want to have the ability to dynamically select files that should be considered in that specific session. Say I want to select 3 out of the 100 files for chat session 1, and then select another 5 for chat session 2, since different chat sessions may stand for different kinds of tasks, therefore have different file sets.

I know this request is kinda like https://js.langchain.com/docs/use_cases/question_answering/per_user but not quite the same. The per user feature depends on namespace features in vector databases but the dynamic files feature requires more trivial operations to the vector databases.

So just wondering if we have this feature already, or is it on our roadmap?

Thanks

System Info

N/A