Vector Search

Embedding text into a vector space and ranking by geometric closeness, usually cosine similarity.

Why it matters in a memory layer

It is a genuinely good tool for the read path and a genuinely bad proxy for relevance in a memory layer. Similarity has no opinion about recency, authority, or whether a fact has been retired, so it will confidently rank a dead belief above a live one. It answers "what looks like this", never "what is true".

Connections

Taught in: Embedding Recall

Used in: Retrieval Is Not Enough · Watching It Fail · Your First Memory Layer · Deduplication · Hybrid Ranking · Vector Stores for Data That Changes