AI overview

Cratebase’s AI surface is four primitives on top of the same collections engine, not a separate service: a vector field type with nearestTo search, auto-embedding on write, an LLM chat gateway that streams over the realtime connection you already have, and an MCP server that exposes every collection as agent tools automatically. What it explicitly is not: an approximate-nearest-neighbor index at massive scale — ranking is application-side cosine similarity over a capped candidate set (see Vector search), which is honest and fast enough for most app-scoped corpora, not a dedicated vector database.