Externalizing Agent Memory

Use when the user asks how an agent should remember things beyond the context window — "should agent memory be files or a vector store?", "how do I persist agent state across sessions?", "what should the agent write to disk?", "do I need Mem0 or Zep?", "how do I make context compression restorable?". Walks through whether externalized memory is needed at all, choosing the substrate (filesystem + grep vs vector store vs managed memory product), what to write out (notes, plans, checkpoints, memory types), the restorable-only compression rule, and write policies for self-updating memory. Not for arranging what stays inside the window — prompt layout, tool-result placement, t…