How do I separate retrieval corpus data from durable workflow state?

Host retrieval corpora and vector stores entirely outside Catalyst’s agentic durable execution layer. Catalyst, built on Dapr, manages replayable workflow state, while retrieval corpora provide static or updated context for RAG workflows; your agent logic should explicitly call these external retrieval stores instead of embedding retrieval data into persistent workflow checkpoints. A key caveat is to avoid accidentally persisting retrieval results directly into workflow checkpoint storage or related workflow state.