manueljpconde wires Mike up to run on your own servers

Until now, Mike could only talk to the big three commercial AI providers - OpenAI, Anthropic, and Google - which means client material has to leave the building to get an answer. manueljpconde's fork adds a fourth option that lets the firm plug in a model running on its own hardware, via tools like LM Studio, Ollama, or vLLM (open-source runtimes that host an AI model locally and expose it the same way OpenAI does).

The wiring runs all the way through: backend, account settings, the model picker, and the prompts that nag users when a key is missing. It's a server-wide switch for now - one local model for the whole deployment, set by an administrator rather than chosen per user. The fork name and the choice of move both point the same way: a build aimed at firms that can't or won't send matter data across borders to a US AI vendor.