Developer Chooses SQLite FTS5 Over Vector Search for Coding Agent Memory

A developer building a coding agent needed a way to search through tool output, git logs, and fetched documents without overloading the model's context window. Instead of the commonly recommended vector database approach, they opted for SQLite's built-in FTS5 full-text search engine. The decision was driven by the nature of the queries involved — searching for exact terms like error codes, stack traces, and log lines — where semantic similarity offers little advantage over precise keyword matching. FTS5 provides BM25-style relevance ranking and snippet highlighting out of the box, requiring no embedding model or external database. The developer acknowledges that vector search remains the better choice for prose-heavy knowledge bases where the same concept may be expressed in varied language.