RAG Temporary Storage

RAG means that elmtree chat prepares documents so the model can use relevant excerpts from them. To do this, documents must be extracted, split into sections, and prepared for fast lookup.

Why RAG data is cached

Without caching, a large document would have to be read, extracted, and prepared again for every follow-up question. That would be slow, error-prone, and waste compute.

When you use this feature, we temporarily store RAG preparation in the server's local database. This storage only serves follow-up questions for the same document.

What can be stored

Depending on the document and processing route, prepared text sections, search structures, document status, and expiration timestamps can be stored. This data is purpose-bound: it exists so the concrete document question and possible follow-up questions can work.

RAG caching is not permanent chat history and not a basis for analysis or training.

Deletion

Prepared RAG data is stored for a limited time and then deleted. Once RAG data has expired, the document must be prepared again for further document-grounded questions.