[dklsf]

I'm really surprised how much a single search query is costing them in terms of compute/infrastructure (1.25 US cents). That buys you more than a minute of compute time on the top vps on hetzner. What's happening behind the scenes during a query?

Getting the compute costs down will probably be a major obstacle to growth, since they'll have to keep prices high (19 USD/month) to pay for their infrastructure.

[Melatonic]

It is not compute that is killing them - it is using the Google and Bing API's for many searches. They charge a lot for those. In your Kagi account you can see exactly how many searches you have done and how much you have cost them (and it could easily be higher than your 10$ a month if you do a lot of searching)

[wooque]

They wrote about it before. They use various paid third-party APIs to power their own search. I guess that is the majority of the cost.

[godtoldmetodoit]

Happy paying customer here, and agreed. Their cost per search feels very high to me, but I don't know enough about the search space to know what is normal.

[burkaman]

Some of that comes from using Bing and Google search APIs. They have their own index as well but can't fully rely on it yet for all queries.

[bongobingo1]

Is the end goal to ween onto their own index or will the search space always be dominated by some major upstream players?

[burkaman]

I'd be surprised if that isn't their eventual goal, but I don't think it's really possible to deliver the kind of general-purpose engine they want with a team of ~5-10. Hopefully if they're successful and grow they can expand their own indexes.

You can try them at https://www.teclis.com/ and https://tinygem.org/.

[eitland]

I guess it depends on how much traction they get, both by consumers and also with websites.

AFAIK some websites actively or passively block everyone except Google.

[maxnoe]

Much of that will be the crawling and indexing that has to be done independent of the number of users, right?

So costs per user will go down with the number of users as this fixed cost is shared between more requests/users.

[pleb_nz]

I'm guessing the query itself will be a very minor percentage of the total cost with just costs coming from crawling, data moving, storage, redundancy and resilience etc.