jellz77 takes Mike fully on-prem

jellz77 has rebuilt the fork around Ollama - software that runs open-source AI models locally, on a server you control, rather than calling out to a hosted API. The model picker now reads whatever models are installed on that local server and offers them alongside the usual options, with per-user key overrides where needed. The owner's own description is modest: "an attempt at fully on-prem."

A couple of rough edges are worth knowing before you lean on it. Right now everyone shares a single local server reachable from the back end - there's no per-user endpoint - and the interface lists local models whether or not a server is actually wired up, so a missing connection only shows itself when you send a message.