The LLM needs to understand your existing codebase if it's going to be useful building features that integrate with said codebase seamlessly without breaking things or assuming things that don't exist. That's not something you want to give away to a private AI company, so self-host an open source model.

I understood how that would help the LLM, I asked how that would help us. What decision would that inform for us that we're currently having trouble with?

What "decision" are you talking about? You said you don't code, but you administer and maintain code. It sounds like you integrate new features into an existing codebase?