Nace AI launches Drex for Search, a page-level retrieval layer that tells agents where to read
Nace AI made Drex for Search generally available. Instead of returning top-k chunks, its decision model narrows a workspace from folder to file to page (or table and column) and hands the LLM only those pages, returning answers with file and page citations. It ships as a CLI, Python and TypeScript SDKs, an MCP server and a REST API.
Why it matters
A drop-in alternative to chunk-based RAG for coding agents over large file sets; Nace reports 56% fewer input tokens and 62% fewer agent steps on its own tests.
Source: Read the full story on Nace AI