We should be able to set different LLMs for Ask and Agent

[id95613]

We can use the small model to ask normal questions, and use the large model to resolve complex questions. This can save our fast-requests.