In a recent conversation with Vlad, we discussed the possibility of providing a free API allotment for paid users of Kagi Search. This feature would allow users to receive a few hundred API calls for free per month, or a few dozen calls per day.
After the allotment has been exhausted, the extra calls would be charged to the user's API credits or search credits.
This would be useful for users to try API clients, like my Raycast extension for FastGPT.
What about just having a set number of free AI calls a month, which would include both the website and the API, and, after that, both would be pay per use
Grooty No. Kagi's goal was to have unlimited AI calls (it used to be limited for Ultimate users previously). However, since APIs are easier to abuse and have fewer restrictions, Vlad wanted a cap on those.
Would 100 calls a month be sufficient?
One of the problems with an easy solution like this is 100 calls to universal summariezer model is $100 in cost.
Vlad Really? That's so much more expensive than I thought.
You could add quotas to different services depending on cost, or just not provide Universal Summarizer credits at all.
100 free FastGPT calls would make a big difference.
Vlad I don't think the average user will need the Muriel model. By default, Kagi already restricts free access to the default model, so you can do the same with the free calls
Maybe include an asterisk (*) which says "not including Muriel", but make it pretty visible and obvious so it's not fine print?
You can create a simple table showing what's included and what's not
Perhaps just dollar value is simpler. $1 in Free API credit every month?
Sounds good
THe more I think about this it should not be every month (this means the API is useful and the user should really pay for it - otherwise someone else has to pay for this cost). So perhaps a one time allotment like $2-$3 in API credits to test everything as you create a Kagi acount (and become a paid member) makes more sense?
But the whole point of this was to allow somewhat free access to tools like the Raycast extension for the average user.
Vlad FastGPT and other AI features are already unlimited for paying users, so if you set a cap on API requests monthly, it doesn’t seem much different. APIs are easier to automate and abuse, but a small monthly cap would pretty much stop it immediately. Even if it’s not super generous, it may attract more API users by letting them dip their toes in the water. For existing API users, it will offset their costs marginally.
FastGPT and other AI features are already unlimited for paying users
FastGPT is an experiment and at some time will be deprecated (the API will remain). Also if it became very popular we would need to lock it down as again someone has to pay for it and we are not in the business of providing free search.
AI features in Kagi are unlimited but in order to use "quick answers" you first need to perform a (paid) search. So the user is paying for the underlying search. And the main driver of cost we are talking about is not AI but underlying search.
Adding just a $1/mo free API allotment, automatically adds 20% cost for $5/mo plans and 10% for $10/mo plans to our cost base, and we do not have such margin to absorb. So the safe thing to do would be to offer a trial of the API and if it is worth paying for, users will, if it is not, then we should probably work on improving it.
This concept came up for me recently when thinking about how I could more effectively integrate Kagi into my phone usage. I think there are at least two reasons for wanting to use the API in general:
I am only focusing on the personal usage perspective. As a user, I want to be able to use Kagi in the most effective manner possible and retrieve information in a fashion that suits me personally in the context I am in at the time.
This concept exists in another form today -- Custom CSS. Kagi cannot please everyone with the default web interface, so Custom CSS offloads this responsibility to the user.
The API is another form of this. It would be great if Kagi provided an Android app or other phone interface so I could use it much like Google Assistant. However, Kagi resources are scarce. This work could be offloaded to Kagi users if API usage were available for personal use, and someone could build a Tasker integration or other phone interface and make it available. This makes Kagi more valuable without having Kagi do the work.
It's another level above Custom CSS -- Custom Interface.
As for cost: my argument is that I am going to perform the search, or use Universal Summarizer, regardless of whether I use the API or not. I may not be as pleased with the user experience, but I do want the information.
The benefit of having the API available for subscribers is that Kagi may get more subscribers if they know they can use alternative interfaces to Kagi. People may not want to even try Kagi if they can't get the same kind of Android or Raycast or Alfred integration they get with Google today. An API with personal use limits (which are the same limits, if they exist, as for the web interface) would allow this to happen.
I think a MVP version of this would be to provide a $2 floor of API credits each month to Ultimate users. If credits are >=$2, nothing is added, if <=$2, set it to $2. This adds incentive to subscribe to Ultimate, which has few benefits at the moment, and allows users to experiment with the API and build interfaces that may draw in more users in the future. (As before, I think "personal usage of API" should just be included, but implementing that is more work than adding a relatively safe amount of credits to Ultimate users.)
