Guide · Developers and operators · Applies to 5.0.x

RAG architecture

Eldric RAG grounds answers in tenant-controlled documents by indexing source content and retrieving relevant passages at answer time.

Conceptual flow#

1. Upload source documents into a tenant knowledge base.
2. Extract text and metadata from supported file types.
3. Chunk content using a strategy appropriate to the source.
4. Embed chunks and store them in tenant-scoped retrieval storage.
5. Retrieve matching passages when the user asks a grounded question.
6. Return an answer with citations back to source passages.

Boundaries#

The public docs describe the customer-facing architecture only. Internal storage layouts, byte formats and retrieval formulas are intentionally not documented.