Reame — CPU-first LLM inference server on llama.cpp: disk KV cache, self-regulating speculation, generation archive, interleaved multi-user, the Conclave. Your hardware, your realm.