jpbreda teaches Mike to run on a model you host yourself

Until now, Mike has run on two outside AI services, Claude and Gemini, both reached over the internet. jpbreda adds a third path: pointing Mike at a language model running on your own servers (using vLLM, a tool for serving open-source models on your own hardware). Because it's configured once at the server level rather than per user, everyone on the deployment gets the local option without managing their own keys.

There's a rough edge worth knowing before anyone leans on this: the local model shows up in the picker whether or not an operator has actually set it up, so an unconfigured deployment would offer a choice that fails when selected. The work was opened as a pull request but closed without being merged into the main project, so for now it lives as an experiment in this fork.