Complex questions can require the same decomposition more than once. We added a decomposition cache so the system can reuse intermediate planning work when the request and context are compatible.
This keeps sequential conversations cheaper without hiding the decomposition step. The system still needs to expose what sub-queries were used and why they were reused.
We tightened the Agentic RAG path around avoidable work: repeated decomposition, repeated embeddings, and unnecessary retrieval fan-out.
A public-safe engineering status note for Cortagent Agentic RAG: routing, retrieval, evidence selection, memory, caching, safety, and evaluation remain the active surface.
We shaped Agentic RAG evaluation cases around follow-ups, weak evidence, source targeting, and complex multi-part questions.