The AIpiary is no longer live.
The AIpiary was a swarm-based "inverse cloud" for running machine learning models. Only a small server ran in the cloud: it hosted the website, kept track of users and chats, and handed out tasks. The models themselves ran on volunteers' own GPUs. Anyone with an account could start a worker bee on their machine, which then picked up chat requests from everyone in the swarm. The swarm served one model, a Llama 3.1 chat assistant called Beatrix Buzz. Workers bid for tasks through an auction queue on the server, which put a task back in the queue when the worker that took it went away. Each worker ran the model with llama.cpp in a scattermind graph and kept no chat history on the volunteer's machine: it read and wrote the conversation through the server's API. The entire project is now open source and available on GitHub.