Long-Term Memory (Agent Memory)

Long-term memory is a persistent store that gives an agent access to information across sessions — user preferences, prior decisions, past tool results worth remembering, or accumulated background about a project. It is distinct from the context window, which is per-request and reset each call. Implementations typically combine a vector store (for semantic recall by similarity) with a key-value or structured store (for exact lookups like "what is this user's time zone"). Memory is often split into three kinds borrowed from cognitive science: episodic (specific past events), semantic (facts and preferences), and procedural (how-to patterns the agent has learned). Writing to memory needs policy — not every turn is worth remembering, and over-eager writes pollute future retrievals. Read policy also matters: surfacing too much memory bloats the context window and causes context rot.