Uses kalosm for interfacing with the llm, just runs commands through sh, logs the command and output, and passes the history to the llm for prompting.
Maybe someone will have suggestions for making it better, or it could give someone a start for their own little project.