We added datasource routing work to make retrieval more selective. A question about a runbook and a question about a policy should not necessarily hit the same source set.
The trade-off is recall versus latency. Source targeting can reduce noise and cost, but the router must be explicit about what it skipped.
Language routing work was added so multilingual queries can be handled without treating language as a knowledge silo.
We clarified the boundary between conversation memory, retrieved evidence, and answer synthesis in the Agentic RAG runtime.
Keyword retrieval work gave the RAG layer a lexical path for exact terms, identifiers, and phrases.