Pre-Fetch is the lifecycle stage that runs before the model starts reasoning. Its job is to search the shared memory vault, rank the most relevant historical context, and inject that material into the prompt or working state automatically. Without pre-fetch, the model starts cold and must first remember that memory exists, then decide to query it, then reason about the result — a fragile chain. Good pre-fetch systems use retrieval, reranking, and graph traversal so only the highest-value context lands in the active window.
clawmemlanggraphclaude-code