Understanding the Agent and LLM interaction

[ygicp]

In the documentation, following image depicts the flow of Runner, Agent, LLM through events. How can LLMAgent determine the need for calling the tool, is not the brains required to determine the need for tool use with the LLM only?
Page link in the documentation

[boyangsvl]

LlmAgent uses model to determine whether to call a tool, as well as what parameters to use.
I will convert this to a discussion since we use issues for bugs and feature requests.