What's the difference between agent memory and RAG?

Agent memory is a record the agent writes about its own work and reads back later; RAG is a lookup the agent runs fresh each time against a separate corpus someone else built and keeps current, never something the agent itself writes to.

Anthropic’s memory tool docs call this just-in-time context retrieval: rather than loading everything up front, the agent records what it learns in memory files and reads them back on demand, so what’s in there came from its own sessions, not a document store someone else assembled. RAG’s chunks sit in a vector index built from documents nobody asked the agent to write, and a query only ever reads that index; it never adds to it.

That cuts the other way too: a RAG corpus gets fixed the moment someone edits the source document, while a wrong memory stays wrong until something notices and rewrites the file, since nothing re-checks it against the world the way a fresh retrieval would.