Why I built gollama — a single Go binary that manages llama.cpp instances with full flag control, multi-GPU support, and a web UI. Honest account of the technical decisions behind building an open-source LLM inference manager for self-hosters.