Guide · Developers and operators · Applies to 5.0.x

RAG on demand

RAG on demand keeps the platform focused: retrieve when a question needs external context, retain what proved useful, avoid storing everything by default.

Operating model#

1. Search existing knowledge first.
2. Ask with retrieved context when relevant material exists.
3. Let users accept or reject whether the result was useful.
4. Ingest durable material intentionally.
5. Review retained knowledge as part of knowledge-base hygiene.

Why it matters#

This avoids oversized knowledge bases and makes retention a human-confirmed workflow instead of an accidental side effect.