Choice of LLM

Users should be given the option to choose an LLM provider (e.g, OpenAI, Ollama) and to choose a specific model from the provider. For now, OpenAI is the only provider and gpt-4o-mini is the only model.
As a first step, the user should be able to choose the model from OpenAI, and should be able to use Ollama to run LLM locally.

Start by locating the current OpenAI integration and the user-facing entry point for choosing the LLM model. Trace how the selected provider and model reach the LLM call, then determine the coverage needed for choosing OpenAI models and using Ollama locally. Done means both provider and model choices work for users.

Written by the indexing model from the issue text.