[Enhancement]: Would be Awesome to Have Global RAG Repository

[bjmaynard01]

What features would you like to see added?

I would love to be able to upload documents to a global document repository to be used as a RAG enhanced model. Where I could upload a collection of documents one time and just by default all users can now search against that document set.

More details

I have gotten as far as creating a document upload interface and using the rag_api to parse these documents into the vector db. I can see the docs in the vectordb, but cannot figure out how to get users profiles to be able to search/interact with these uploaded documents.

Which components are impacted by your request?

General

Pictures

No response

Code of Conduct

• I agree to follow this project's Code of Conduct

[MartyWind]

OpenWebUI has this and it is pretty nice. You can create knowledgebases (?) and dump all your documents in there. You'd need to host something like ragflow and integrate it via MCP to do this with LibreChat, which is already quite the effort. An integrated solution would be nice.

[onestardao]

Yeah, totally get the need here — once you go beyond single-user context, the whole RAG setup starts to feel like it's duct-taped to a whiteboard.

The moment you want shared context, versioned uploads, or multi-user search boundaries, things fall apart — vector scope collisions, outdated context bleeding in, or even worse: global queries hitting irrelevant personal docs. We ran into all of that trying to build persistent knowledge layers.

Ended up writing some tooling to formalize “semantic zones” per user/group/role and track query-to-document lineage. Still a work in progress, but the whole system’s open and MIT-licensed. Backed by the tesseract.js folks too.

If this ends up getting picked up as a native LibreChat feature, happy to contribute or share what we’ve built so far. (Not dropping links unless someone asks — just here to help.)

[NickIvanter]

Hey @onestardao, if you could share some links as to where to begin this work or perhaps what's already available out there, I would be super grateful. We are about to embark on this as well.

[bjmaynard01]

I'm in the very beginnings of trying to hack this into my build, but I am far from a dev. Just a sys admin type trying to get this working using some internal docs for my user base. Anything you have to get me going in the right direction would be awesome.

[onestardao]

Thanks for the follow-up — that context helps a lot.

From what you described, it sounds like you’re already hitting a few of the classic multi-user RAG pitfalls:

Global index queries returning personal or irrelevant data

No real way to formalize versioned uploads or semantic zones

Context leakage from unrelated sessions (especially in shared setups)

These aren’t easy to catch at first glance, which is why I’ve been mapping out a full set of RAG failure modes over the past few months — 16+ of them, all based on actual bugs I’ve helped debug in live deployments (LibreChat included).
Each one includes a precise diagnosis + tested fix path.

You can browse the whole problem map here:
🔗 https://github.com/onestardao/WFGY/tree/main/ProblemMap/README.md

MIT-licensed, and always happy to help if any of it overlaps with what you’re building.