T-Crypt / berth

Decides who gets your GPU. Chat, image and voice engines lease the card in turn instead of grabbing it, so two of them never load at once and crash each other. One Go binary, nothing else to install.

mostlygeek / llama-swap

Reliable model swapping for any local OpenAI/Anthropic compatible server - llama.cpp, vllm, etc