A constrained generation library for llama.cpp that prevents malformed outputs. Features include XML/JSON tag completion, multi-step reasoning support, and an ultra-simple API. Built on llama.cpp's sampler architecture for reliable token filtering.