Memory

Short-term memory is two layers — the session log you store and the smaller working set that actually fits in the context window.

Long-term memory stores distilled facts outside the model so a new session can retrieve them — the weights never learned the fact.

Long-term memory splits into facts (semantic), events (episodic), and skills (procedural) — each needs a different store and a different forgetting rule.

User memory is personal data about the human; agent memory is operational — skills and run state. Split the stores.

Memory is a pipeline — extract structured writes, store each type in the right engine, retrieve a small filtered set, and forget on purpose with TTLs and supersedes.

Memory fails by silent overwrite, poisoned writes, and cross-user retrieval — govern writes and filter every retrieve.