I should probably add in that kagi has some secret modes.

If you precede the search query with `!expert` you get an LLM search response.

If you precede it with `!code` you get a GPT-4-ish tuned response. It’s… really good. And since you can do it from a browser bar if you make it your default browser, I find myself reaching for !expert at least 50% as often as ChatGPT these days. Also, it’s extremely good with citations. The !expert mode seems pretty darn up to date, no knowledge cutoff issues. I have no idea if it is search, doing a vector lookup, or they just bake the model every night. Totally magical.

Also, I’m following the LLM stuff pretty closely these days — their universal summarized and their search LLM — no one else is doing that right now. It’s genuinely amazing.