would be good if we could use a text-to-text chat to test the LLM component of our agents, and if this were exposed via the API so this could be automated