We added a pronoun resolution pass for follow-up handling. This work targets references like "that", "it", "he", and "she" when prior conversation state exists.
The implementation goal is narrow: improve retrieval targeting when a referent can be grounded. If the referent cannot be resolved, the system should surface uncertainty instead of guessing.
Conversation context handling was added so follow-up questions can reuse prior state instead of being treated as isolated prompts.
We clarified the boundary between conversation memory, retrieved evidence, and answer synthesis in the Agentic RAG runtime.
Memory consolidation work was added to reduce duplicate memory state and keep useful context retrievable over time.