A language model is stateless: it generates a response and forgets everything the moment that response is sent, unless the entire history is stuffed back into its context window on every call. Persistent memory is the architecture that replaces that habit — extracting what an agent has learned, storing it outside the context window, and retrieving only what’s relevant, so an agent can carry knowledge across conversations, users, and sessions instead of starting over every time. This knowledge base covers that architecture from first principles, including how Weaviate Engram implements it on top of Weaviate.