Azxiana We don't do the correction "in house" - the reason for the current semantics are due to limitations and consistency issues with consuming multiple upstream indexes. Some indexes autocorrect, others don't, and some may or may not provide a way for us to control this behavior.
Say for example, you submit a query with a typo - if Kagi were to do "nothing", you might get some results based on your typo verbatim, and some results based on the "autocorrected" version of your query. To make sure the results simply make sense we had to pick the semantics that was the best quality for the lowest latency. And we built the "verbatim mode" feature so that you can always search for exact terms - I can suggest trying that in the meantime if you haven't yet, and seeing if that works better for you.
So TLDR - its like the way it is because we were between a rock & hard place compiling results from multiple upstream indexes with regards to balancing quality vs consitency vs latency. It's tricky.
But - the good news is that we are reviewing other internal changes that I expect to allow us to change the semantics to be more in line with what most expect, which is to show the results consistently for your terms as-written, and prompt for a "Did you mean?" instead of a "Search for <possibly typo'd terms> instead". So please stay tuned in the coming weeks for a potential update on this.