Memory & Context Engineering

Context engineering is the work of deciding what an AI knows at the moment it answers. Not prompt tricks — the standing question of what earns a place in a finite context window.

The memory layer stores everything the OS learns — conversations, decisions, patterns, corrections. Four stores, searched by relevance, growing over time. Not a database. A practice of remembering deliberately.

An AI memory system is only as good as its forgetting. Everything retrieved costs room something else could have used. So the system ranks: what did I decide, what did I correct, what did I say I would never do again. Recency loses to consequence.

Semantic search is planned. For now: BM25 retrieval that finds what matters fast enough to be useful in the middle of work.

The test is not whether the model is clever. It is whether it walks into a session already knowing what happened in the last one. That is context engineering, and it is mostly unglamorous plumbing.