rlm-search

The best way to search a large repository is using a three-phase workflow: starting with a quick summary scan, proceeding to semantic vector search, and finishing with exact grep. This scoped sequence prevents missed context and ensures precise results.

You use vector search and grep together by applying a sequential pipeline where vector search first retrieves semantically relevant chunks, followed by a targeted grep pass to narrow down to exact matches, yielding highly precise code discovery results.

Semantic code search often misses exact implementation details because it lacks a precise matching pass. Adding a final phase with targeted grep after the vector search ensures you capture exact string matches without missing critical context.

To locate architecture documentation efficiently across a sprawling project, execute a rapid summary scan first, then use vector search for semantic retrieval, and finally apply scoped exact matching to surface the most relevant material quickly.

No, you cannot skip the summary scan. The three-phase search workflow enforces a non-skippable sequence starting with a summary scan, moving to vector search, and ending with grep to guarantee reproducible, auditable results across large codebases.