Impressive work on decision models for agentic search. Drex Search looks like a big upgrade for agents working over large file collections. It decides where to look, reads only the pages that matter, and stops when it can cite the source. This is a great fit for Claude Code, Codex, and Hermes Agent, where poor retrieval often leads to wrong answers. Nace reports 62% fewer agent steps and 56% fewer tokens per question, with +4 points of accuracy on average.