We added graph-oriented retrieval modules to the RAG surface. This gives the runtime a path for relationship-heavy questions where plain chunks may not be enough.
Graph retrieval remains one retrieval option, not a replacement for vector or keyword search. The system still needs to route based on query shape and evidence needs.
We clarified the boundary between conversation memory, retrieved evidence, and answer synthesis in the Agentic RAG runtime.
Datasource routing work was added so retrieval can target relevant sources instead of searching every corpus the same way.
Keyword retrieval work gave the RAG layer a lexical path for exact terms, identifiers, and phrases.