Cache search results to reduce API costs for repeat searches

nathanl

I find myself somewhat frequently doing a repeat search for the same term a day or more later, looking for the same site I found the first time. If Kagi were able to cache the search results for a given query for some period of time, a repeat search could use the cached results instead of having to be run through the paid APIs and incurring those costs.

I realize this may create more problems than it solves though. For example, the cache would have to be global (shared between all Kagi users) to prevent user-specific tracking, although this would actually improve the caches effectiveness. Even then, you'd still have a pool of data that clearly indicates that someone searched for X at time Y. You'd also have to determine when to use cached results vs requesting new results. For something that is breaking news, the results could change hourly or more I'd guess. When searching for a company name and I'm just looking for their website, it wouldn't matter if the results were a week old.

Another less involved suggestion would be to not do a full API search for certain searches. For example, using the search bar as a calculator, doing a currency conversion, or defining a word. Here you could just display the objective result, with an addition button for "Show more" that runs the full search if a user wants. There is a good chance you don't need to do a full API search if my query was "35*24". I don't know what percentage of API requests this would save you though.

Just some thoughts.

Vlad

nathanl 99.9999% of searches are unique and thus cache does not help. Even when it does (for example a query like 'facebook') you may want to invalidate it in case there are some major news about faceook that just happened.

NoGoogle

Vlad I'm unaware of the limitations of website cookie technology, but if you do ever decide to introduce a cache feature for repeat searches, it might be a neat idea if it's even possible to simply store the last 24 hours of search results in cookies/local data which would allow for an offline privacy respecting way of temporarily remembering search history locally. Disregard this if that's not how cookies can work.

z64

NoGoogle

[...] it might be a neat idea if it's even possible to simply store the last 24 hours of search results in cookies/local data which would allow for an offline privacy respecting way of temporarily remembering search history locally

We actually already do this, to a degree, with localstorage. If you have JS enabled, we store a cache of output from our server that I believe is valid for 10 minutes. It is mainly there to smooth over any issues with browser's back/forward cache, and to prevent things like refreshing the page from sending us traffic when there's no use. But this is about as much as we can do.

There are two problems with storing more data here, or storing it for longer:

You can't store a lot of data on the browser itself. I think most browsers will cap us at 10MB by default. A lot of searches could fit there, but this isn't really something to waste it on. (and other things we do might need that space)
Consider time-sensitive queries, like current news or searching about an important event unfolding. If we extend the lifetime of this local cache too long, then quality is impacted: You will get old results until it wears off, you clear your browser data, or switch devices, etc.

About cookies: We don't use cookies to do this, because cookies are transmitted to the server with every request, which we do not need to do for a local cache; it would waste bandwidth, inflating the size of every request and response by a lot.

FunPost

z64
another option could be to allow the user to choose if they want to cache the results (or add a bang for caching results), it could solve the problem with time-sensitive queries, I guess there's still the problem of storage...

snowytrees

I’m a big fan of locally caching the previous 24 hours of queries and having them show up in the autocomplete results (with a UI element indicating cached). That way I can choose to have fresh results or not waste an api search on something I’m fine with a cached result for. This should be a toggleable option, with the ability to opt out specific searches from it. Eg searching -nc xxx would not cache the result beyond what it does currently.