BitTorrent-style open-source system for running 100B+ LLMs collaboratively — share your GPU, run models 10x faster than offloading. Free distributed inference.