Documents are split into passages and prepared so that similar content can be found. When a question arrives, the system picks out the matching passages and hands them to the language model with the task of forming an answer. This reduces invented statements but does not rule them out.
Typical applications
Questions on manuals, policies and procedures.
Search in offers, contracts or project documents.
Internal knowledge assistant for new employees.
Support with your own knowledge base.
What matters
Permissions: the AI may only show documents the user may see anyway.
Data quality: outdated or contradictory documents lead to poor answers.
Source references: answers should point to where they come from so they can be checked.
Operation: locally with Ollama the document stock stays in your environment; for cloud services the data protection check points apply.
Getting started
Start with a limited stock, for example one manual, and check the answers for quality. Storage locations such as Nextcloud can be connected as a source. The result of a pilot decides whether an expansion is worthwhile.
RAGretrieval-augmented generationAI own documentsknowledge base AIchat with documentsRAG Ollama
Frequently asked questions
What does RAG mean?
Retrieval-augmented generation: an AI answers questions based on passages previously retrieved from your documents.
Does the AI no longer invent things with RAG?
It happens less often but still occurs. Source references and spot checks help with verification.
Can I connect Nextcloud?
In principle yes. It is important that the permissions from the storage are respected.