The RAG Experiment Accelerator should work with local models using ollama. This should be configuration based.