flashgrep is not just another search wrapper. It is meant to solve a very specific problem: agent search is too slow in real codebases.
The core is two-stage: first recall candidate files through a character-level inverted index, then run exact regex verification. That moves regex from a full-repo scan to a much smaller candidate set.
To keep that useful over time, it also has to deal with base snapshots, dirty catalogs, published overlays, daemon protocols, MCP integration, and workspace-state recovery.
When I work on flashgrep, I am really trying to turn search into infrastructure that agents can depend on.