Just for concrete confirmation that LLM(s) are being used, there's an open issue on the GitHub repository, on hallucinations with made up information, where a Kagi employee specifically mentions "an LLM hallucination problem":
There's also a line at the bottom of the about page at https://kite.kagi.com/about that says "Summaries may contain errors. Please verify important information."
https://github.com/kagisearch/kite-public/issues/97
There's also a line at the bottom of the about page at https://kite.kagi.com/about that says "Summaries may contain errors. Please verify important information."