We continued cache work with context isolation for semantic reuse. Similar language is not enough; the surrounding conversation can change what a question refers to.
The cache path therefore needs to account for language, pronoun resolution, and conversation state. This favors correctness over aggressive reuse.
We clarified the boundary between conversation memory, retrieved evidence, and answer synthesis in the Agentic RAG runtime.
We tightened the Agentic RAG path around avoidable work: repeated decomposition, repeated embeddings, and unnecessary retrieval fan-out.
Memory consolidation work was added to reduce duplicate memory state and keep useful context retrievable over time.