llama.cpp

LLM inference in C/C++ — runs models locally on CPU/GPU across platforms with minimal dependencies.