Add support for using locally running Ollama models to provide more privacy and on-device processing.