How does AI memory actually work?

Past conversations are stored externally, turned into vector embeddings, and the most relevant pieces are retrieved into the prompt when they would help the response. The model itself remains stateless.

What is recall gating?

A coordinator that decides whether to inject any long-term memory at all on a given turn — based on signals like explicit recall intent, repeated themes, and emotional cues.

Why don't most chatbots use this?

Cost, complexity, and safety surface area. Stateless chat is cheaper to build and operate, even though it makes companion-shaped use cases impossible.

How AI Memory Works – Embeddings, Retrieval & Recall Gating