Important Update: Deprecation of RAG Module (Effective v2026.05.01)

[admonstrator]

Deprecation of RAG Module (Effective v2026.05.01)

RAG will be removed as of version 2026.05.01

I have been developing Paperless AI next for several months now. During this time, I have frequently struggled with the overhead of maintaining two distinct architectural versions. This creates a level of complexity that is no longer sustainable, especially since the original source was not necessarily designed with long-term maintainability in mind.

Why this change?

To ensure the stability and future growth of the project, I have decided to discontinue support for the RAG (Retrieval-Augmented Generation) module within the main codebase, starting with version 2026.05.01.

My primary vision for Paperless AI next is to make tagging and managing documents within Paperless-ngx as seamless as possible. While RAG is a powerful feature, it currently does not align with the core benefit of the tool and adds too much maintenance debt.

Future Outlook

This doesn't necessarily mean RAG is gone forever. I may re-introduce RAG functionality at a later date, but if I do, it will be as a completely separate, decoupled system to keep the core of Paperless AI next clean and maintainable.

What does this mean for you?

• Focus on Core Features: By removing this complexity, I can dedicate more time to improving auto-tagging, metadata extraction, and UI enhancements.
• Cleanup: Following the cutoff date, RAG-related code will be removed from the main branch to streamline the repository.

Thank you for your understanding and for supporting the continued development of Paperless AI next!

[Xyz00777]

Sad, do i understand it correct that we will than not able anymore to "talk" with the documents? And the nextcloud-ai-next is more getting like paperless-gpt?

[admonstrator]

Yes, you are right. "Talking" to documents will be removed from my fork. The main focus is tagging and working with documents instead of talking to them.

[Xyz00777]

sad :/

[pbinksma]

Your RAG implementation never really worked well for me (probably my fault!). Answers were quite unreliable and it often did not find the right documents.
I meanwhile setup my own instance of a RAG Chatbot using n8n and a workflow which crawls all paperless documents (via scheduled reconciliation crawls and webhook for new docs) into my quad VectorDB. The nice thing is that I embedded a bot account on my Matrix-server so that I can actually "chat" about my documents from my mobile on the fly. Meanwhile it produces very good results on 1200+ documents.
Thanks for your implementation. It was actually the thing which got the ball rolling!

[admonstrator]

Oh, I never implemented RAG. It was in the product before I forked it.

[pbinksma]

True! I haven't used the original paperless-ai long enough. I switched to your fork at day 1 :D