Full-text search with ranking

Full-text search covers the cases where you know precisely what you want: a function name, an error string, a client, a product code. It matches literal terms across entry titles, source labels, and the full memory body, then orders results with BM25 ranking so the strongest matches rise to the top instead of arriving in raw date order.

Keyword and semantic search complement each other. Semantic recall is best for fuzzy, half-remembered ideas; ranked full-text search is best for auditing exact strings and confirming whether something specific was ever discussed. Because it ranks rather than merely filters, a query for a rare identifier lands its handful of real hits first, which keeps precise lookups fast even in a vault of thousands of entries.