Detailed Description

I'm proposing to use an llm api to give Naomi more conversational skills.

Context

This makes Naomi more conversational and able to handle topics that are not directly addressed by plugins.

Possible Implementation

My implementation would be to use the current text to intent system to decide which plugin to activate, then activate that plugin and pass the output into an LLM as a system prompt. This does not require multiple calls to the LLM, so should run pretty quickly. There are several possible issues with this system, one being that it does not handle expect() or confirm() calls well (it would get stuck waiting for user input without providing the prompt) and it might end up activating devices or making changes without realizing it or reporting it to the user. I'm just going to play with it and see what happens.