Provide a plan without AI functionality

Pleonasm

@Kaguy, your prior post is the most clear and succinct statement of the request in this thread. We all have our own personal preferences, and I certainly respect yours and others in the Kagi community who may share that perspective.

There is one aspect of this request, however, that I have struggled to understand. Relative to a Kagi customer who simply chooses to ignore and not use AI functionality in the product, what benefit does an “AI kill switch” provide? It does not change the fact that a subscription fee paid to Kagi partially funds the development of AI tools in the product, it does not impact the trajectory of AI adoption in the culture, it does not reduce the environmental impact of AI data centers, and it does not impact the displacement of jobs due to AI. Are the advantages of an “AI kill switch” only that it allows a customer to pretend that AI features do not exist in the Kagi product by visually hiding their existence, and thereby preventing the accidental usage one of the AI features? Perhaps I am failing to see the obvious and there exists a more substantive benefit?

As Edward R. Murrow noted, “The obscure we see eventually. The completely obvious, it seems, takes longer.”

Gredharm

Pleonasm There doesn't exist a real world benefit. It's a religious thing, like avoiding certain foods and thus not wanting them to be cooked in the same kitchen.

Yet I'd be willing to bet most of the people complaining are still willing to use Windows and Office 365. I'm sure there are some who are at least not hypocritical and only use FOSS rather than software made by malicious companies like Microsoft and Apple, but I suspect they are a small minority.

Kaguy

Pleonasm Are the advantages of an “AI kill switch” only that it allows a customer to pretend that AI features do not exist in the Kagi product by visually hiding their existence, and thereby preventing the accidental usage one of the AI features?

Correct. Like I said before, if one does not desire to have their money go toward AI development at all, they just won't use Kagi once they learn of the LLM features and their unwillingness to part with them (no, I don't think they should be here and I fundamentally disagree on their usefulness for search and, God forbid, news summarization). It's about peace of mind for customers who wish to enjoy the core product without the bloat. There's several users here saying, for example, that they came to Kagi specifically to get away from LLM features, so having them around is bothersome.
You may judge this as being nothing, and that's fine, but to me, and several others here, it's not nothing. Peace of mind is important.

Gredharm Yet I'd be willing to bet most of the people complaining are still willing to use Windows and Office 365. I'm sure there are some who are at least not hypocritical and only use FOSS rather than software made by malicious companies like Microsoft and Apple, but I suspect they are a small minority.

Come now, is it really necessary to explain to you how the companies you mentioned have created effective monopolies over the technology that has become necessary for our lives and we don't exactly have a choice not to use these products? Which is precisely the reason these companies can get away with being malicious in the first place? You know, the very problem Kagi alleges to wish to solve? Or are you going to try to guilt trip people for using Google's services next?

Gredharm

Kaguy If you're talking about enterprise, you have a point. It isn't easy to escape Office 365, Google Apps and so on if you're an employee of a company that uses it, or even in your own small business if you need to interact a lot with people using those services.

But as an individual user, most technology isn't necessary at all. It's just toys people want to use. And the stuff that is necessary, is easy to find non evil alternatives for.

As others have said, and I think I did earlier too, if Kagi forced people to use AI, I wouldn't want to use it. But they don't, it's optional. And it makes no difference if there were an AI free plan or not, as it's still the same company getting paid. That's how money works, I think someone also mentioned about how money is fungible by its nature. So what features you pay for makes no difference.

If what you really want is for Kagi to purge all AI from their services, I don't see that happening. Personally I find AI frequently useful for work and I'm sure many people are using it or they wouldn't bother with it.

And finally, AI isn't evil by nature, it's just a technology. Hate the companies using it to spy on people and the governments pushing it for that same purpose. But there's no reason to hate the technology.

Kaguy

Gredharm And the stuff that is necessary, is easy to find non evil alternatives for.

Try living Brazil and not using Whatsapp, the message app bought and overseen by Meta. Literally every single person has it, so if I want to stay in touch with friends and family, I need to have it. It's also widely used for professional communication, especially in the freelance world, so all I'd need to do in order to go with a "non-evil alternative" would be to ask every potential client and/or employer, plus friends and family to use another message app specifically for me and just really hope they'll all say yes. How fun.
A non-evil alternative to Google? You mean Kagi, who is essentially forced to pay Google for its index if it wants to have any hope of delivering a good search product? Or perhaps you mean that Gecko-based browser whose lights are kept on by Google's bribe to keep it as its default search engine? Boy howdy, I sure hope no one I work with uses Google Drive, or Google docs, and it sure is easy to use Linux for all my personal needs despite it not working with basically anything out of the box and requiring time-consuming tinkering just so I can do a hobby like playing a game, which is important for my individual mental health.
I wonder what non-evil, user friendly and easily bought plug and play option I have for smartphones, btw? There's two OS's last time I checked, one from Google and another from Apple, and those are a no-no. But hey, those are just toys, right? It's not like I need a smartphone for essential stuff on the modern world! Like message apps, calling an Uber now that there are no taxis anymore, accessing my bank account without going to an ATM, looking up information away from my desktop. Nah, I just really like Candy Crush or whatever.
So many non-evil alternatives! And all I have to do is massively inconvenience myself and give up on a functional social and professional existence!

Gredharm Personally I find AI frequently useful for work and I'm sure many people are using it or they wouldn't bother with it.

I'm very skeptical AI is so on demand by users considering the corporations have yet to find a way to raise decent revenue from it compared to the massive investment. This also makes me particularly suspicious of Kagi's claims that the AI features are attracting customers who wouldn't otherwise consider Kagi (I keep being told they exist, but I'm not sure if I believe that). I honestly think Kagi's team just really enjoy LLMs as a concept. Which is fair. But certainly not part of the pitch that gets people here. Please don't link me their AI integration philosophy.

Gredharm And finally, AI isn't evil by nature, it's just a technology.

I know

Gredharm Hate the companies using it to spy on people and the governments pushing it for that same purpose.

I do.

Gredharm But there's no reason to hate the technology.

I don't. I hate that this interesting, however massively limited and fundamentally unrealiable technology is being sold as a revolution by the same companies you claim to hate when for the vast majority of cases it doesn't raise productivity, doesn't save you time, often hallucinates, makes shitty looking pictures and shitty sounding music, and creates several issues regarding intellectual property, data scraping and privacy, all for the low cost of environmental and economical disaster. But, uh, religion thing or whatever.

Gredharm

Kaguy Well to start with, you can move to GrapheneOS to escape the main evils, which admittedly still gives Google money in terms of the phone, but it importantly deprives them of all your data. There's also Fairphones with /e/ OS, which isn't as good, but at least Google don't get paid.

Banking apps work on it too most of the time, though I've never actually found a need for one. You can just go to their website in any browser usually. Unless Brazilian banks require mobile apps for authentication or something?

With WhatsApp, if you have GrapheneOS you can use contact scopes and storage scopes at least to minimize the data that is stolen from you. But of course it is a good idea to at least try to move people over to better alternatives for their own good. It's the complacency and ignorance of the masses that gives big tech their power.

And AI when used for what it's good at, is pretty useful. Of course big tech companies lie endlessly about it to up share prices. Probably soon there'll be a massive global collapse when everyone realises it's all just incestuous insider trader and lies, plus lots of data theft to try and get slightly better scores on worthless benchmarks. But when the dust settles, people will be using small, narrowly trained models a lot for things like translations (for example YT videos into a dozen languages rapidly, even dubbed), smart code completion (as opposed to full vibe coding), summaries of gathered information, and for NPCs in computer games probably.

Kaguy

Gredharm Well to start with, you can move to GrapheneOS to escape the main evils, which admittedly still gives Google money in terms of the phone

So, same difference, still reliant on compatibility with Google offerings, which means it's not a non-evil alternative, because it can't compete without piggybacking from the evil company. Which is precisely the problem you pretend is all a matter of individual choice.

Gredharm Unless Brazilian banks require mobile apps for authentication or something?

They do. We still can use the browser website, but it requires authentication through a token/QR Code that's necessarily sent to your phone app.

Gredharm With WhatsApp, if you have GrapheneOS you can use contact scopes and storage scopes at least to minimize the data that is stolen from you.

Data is still stolen, still have to use the product from Meta, your point is rendered moot.

Gredharm But of course it is a good idea to at least try to move people over to better alternatives for their own good.

Good luck with that when calling them all complacent and ignorant for daring to not be tech-savvy and wanting convenience out of the products they have no choice but to purchase. It's not like Big Tech lobbies governments to influence regulation or kill competitors off with anti-competitive practices like pressuring smaller business to sell or straight up operating at a huge loss to get them out of business. Stuff the general public has no blame for.

Gredharm And AI when used for what it's good at, is pretty useful.

It's pretty useful to do menial yet laborous work, data processing for example.
I'm not in the mood to discuss why it has no place generating art, translations and anything that requires output accuracy (like, say, search).
I agree with you that once the bubble bursts and we face bad economic downturn, whatever's left of LLM will actually be used for what it's good for though.

In any case, this has been a long tangent that's really not the point of this thread. I'll drop it from now on if you don't mind.

RoxyRoxyRoxy

Kaguy I would say that that's your definition and dividing line of what's acceptable AI. There are some who just don't want Quick Answer. There are others who don't want any "generation" (QA, Assistant, Summarize) but are fine with Translate. Do we turn off every single thing connected to an LLM? Then someone who likes Translate either cannot utilise the killswitch, or cannot utilise a product they like.
If everyone was on the same page as you, it would be very easy. But they're not, and so it is not.

Kaguy You may remember a while back we rolled out an update to the mobile experience that inadvertantly felt like we were pushing users to use Quick Answer. When we realised that, it was basically all hands on deck internally to find an acceptable solution and roll it out immediately.
A lot of our products like Summarize and Translate started out as experiments, then "treats" (little features as part of your Kagi sub), and now are big and mature enough to be their own product. They didn't start out as an executive saying "We need 5 AI-powered products to get another funding round", they started out as a way to give something more back to our users, and hopefully improve your lives.
We have a few internal experiments made by Vlad that, while including an optional AI integration, are not focused around adding AI for AI's sake. I've demo'd them and they are thoughtfully made and incredibly useful.

Kaguy

RoxyRoxyRoxy There are others who don't want any "generation" (QA, Assistant, Summarize) but are fine with Translate.

Are there? I personaly don't see them in this thread. Are there other threads or maybe you receive personal DMs with those specific demands about cherrypicking what to be blocked?

RoxyRoxyRoxy Do we turn off every single thing connected to an LLM?

Yes. That seems to be rather clear ask from the OP of this suggestion and others who sympathize.

RoxyRoxyRoxy Then someone who likes Translate either cannot utilise the killswitch, or cannot utilise a product they like.

I guess? Again, very specific problem to have that I'm not seeing being brought up by users. I'm not omniscient though, so I concede I may be missing these.

RoxyRoxyRoxy and now are big and mature enough to be their own product.

Cool! I'd suggest they're treated as their own product away from the one I'm paying for, which is specifically search. Though I don't presume to speak for others on that point specifically.

RoxyRoxyRoxy they didn't start out as an executive saying "We need 5 AI-powered products to get another funding round",

Never said it did. What I am saying is that there's been an undeniable focus on expanding these features, which your dev team seems very enthusiastic about. For whatever it's worth, I don't doubt your sincerity and belief on the potential of LLMs. I just fundamentally disagree with it.

RoxyRoxyRoxy I've demo'd them and they are thoughtfully made and incredibly useful.

I'm glad you and potentially other users find them useful, but so long as they are based on LLMs inferring the statiscally most likely next piece of text, pixel, soundbite or whatever else, they are fundamentally unreliable and prone to error when used long enough, therefore I (and from what I can gather many others here) am not interested in it no matter how much
you assure me it's been well thought out. If it is an LLM trying to retrieve or summarize information back to me in any way, it will halucinate. If it is another form of the concept of "your personal assistant" (I really don't get why that idea is supposedly so appealable to people, I don't personaly desire an assistant for most of the things I do, let alone a robot one) I do not trust it to assist me correctly.

I want a search engine that provides good search results based on what I'm looking for. That's it. The more Kagi does it, especially if it grows its own index along the way, the more value I have to my subscription. On that I think a lot of the people asking for no AI here can agree.

Anonymous12

Kaguy I want a search engine that provides good search results based on what I'm looking for. That's it.

You’re in luck! It does that in spades.

Gely

Kaguy for what it’s worth just having everything LLM removed is what I want, you’re defo not alone in that. Anything I find that has it I avoid. I’d rather not have to remember what does and doesn’t use it and just trust a kill setting

Pleonasm

DuckDuckGo recently (16-26 JAN 2026) conducted a poll asking individuals to select their preference between two options: (a) "Search with AI" and (b) "Search without AI." Of the 175,354 responses, 90% were "Search without AI."

As is the case with many online polls, it is unknown whether the respondents are a representative sample of the population, and it is unknown what proportion may be bogus. Therefore, caution is required in interpreting the results.

Nonetheless, couldn't Kagi conduct analyses to discern the extent to which its current customers use AI in the product? For example, take a random sample of all customers and create and inspect a binned frequency distribution of the "AI Cost" (as reported in Settings | Billing | Usage Details) for the most recent full month. Alternatively, the Kagi Search Stats reports "Assistant Threads in the past day," and perhaps these data can be used to calculate and report the average number of threads per customer per day over the most recent 30 days? The objective is to generate intelligence about the size of the subset of Kagi customers who prefer "Search without AI."

Anonymous12

Pleonasm Therefore, caution is required in interpreting the results.

You could also quite easily vote multiple times in their broken poll. This was a marketing exercise and nothing more.

Kaguy

Pleonasm I mean, I'd assume Kagi already has all that data without the necessity to run analyses? Unless you're asking for them to disclose the data, which sure, why not.

Pleonasm

@Anonymous12, yes, I certainly agree: there are several tactics that could have been used to distort the results from the DuckDuckGo poll. The more important question, however, is: what are the corresponding results for Kagi customers?

Beyond my prior suggestions, perhaps Kagi could consider adding a one-question survey to the Kagi landing page for logged-in customers to query their preference for the "AI kill switch" concept? If it were limited to logged-in customers, then only those who have an active Kagi subscription could participate - and, I presume that Kagi would be able to suppress multiple responses from the same customer.

Although not without methodological concerns, I also noticed the existence of this Kagi survey: How Kagi Works for You. The data collected therein are relevant to AI usage in the product, and perhaps someone from Kagi can summarize and share those findings?

Finally, Kagi could consider sending a survey invitation to a random subset of its customers through email (since Kagi knows the email associated with each account) to collect relevant data.

P.S.: Personally, I suspect that very few Kagi customers are "no AI" users - but, it is an empirical question that should be answered using data.

Pleonasm

@Kaguy, yes, if Kagi has already conducted analyses to assess the proportion of customers who are "no AI" users, then I encourage the company to disclose the information to help guide the discussion in this thread. Additionally, the issue could be raised in the next AMA....

Guterrat247

...or at least the "Starter"-tier should have minimum 500 searches, this would be a real starter, 300 is ridiculously little.

tux0r

Guterrat247 300 is ridiculously little.

It depends. In my experience, 300 is "just right" for many casual users. Not everyone burns through them in a week, like I did. 😅

Kaguy

For whatever it's worth, 300 feels about right for me as well, I rarely burn through it all in a month, and they've disclosed the economics behind this decision.

fxgn

Honestly can't imagine how people use the starter plan. 300 searches is enough for maybe 2-3 days for me. But I guess some people just don't use search that much.

« Previous Page