Feat/agent frame mul llm

[Xm0onh]

This PR is following the issue #105

Anthropic and open ai work well. However, it is a different story for llama. The problem is: my running model is weak and too small, and the model cannot handle parsing jsons very well.
I started to write a script to optimize the parsing and help the model but it is time consuming.

[jfrank-summit]

This is a bit more complicated than I was thinking but should work for now, we can rethink structures and config ease later. I left a few comments on the models.

[Xm0onh]

Yea it is a bit complex, but I tried to make it type-safe and string-free.
Also the models were placeholders if someone wants to use them.

[jfrank-summit]

lgtm