How search works (for the curious)

1. 1. Flat text the model reads

   Simple; blows context; slow/expensive at scale.

2. 2. SQL / full-text search

   Great for exact words and filters; weak on find the vibe of X.

3. 3. Vector store (what Fjorj uses)

   Chunks + embeddings; good semantic recall; still local on disk; files remain source of truth.

Fjorj ships (1) local folder pick to watch/index on Mac, Windows, and Linux, (2) local vectors, (3) MCP so other apps can search (writes gated), and (4) Briefcase — Bundles of Items you pack when you want exact context, not a library dump. No need to run a chat server.