sean-a-ward opens Mike up to OpenRouter and friends

Out of the box, Mike talks to OpenAI directly. sean-a-ward's fork adds a clean way to send those same calls to OpenAI-compatible services like OpenRouter (a marketplace that brokers access to many models through one interface), LiteLLM, or a local LM Studio instance. Admins can wire it up once for the whole deployment, and individual users can override it from their own account settings - including remapping Mike's hard-coded model choices to, say, Anthropic's Claude or any other model the chosen provider carries.

There are trade-offs. Some of OpenAI's newer threading and reasoning features quietly drop away when you route elsewhere, and the migration is staged in a way that lets the code ship before the database is ready. But for any team that wants model choice, cost control, or to keep traffic away from OpenAI specifically, this is the most polished take on the problem so far.