I dumped Google for Kagi
arstechnica.com
arstechnica.com
Just kidding, I love Kagi and I get a ton of value from it! I always like the saying that "the best ideas are obvious in hindsight", and that is absolutely how I feel about Kagi. Being able to uprank, downrank, and/or pinning, blacklisting domains is a killer feature, and something that would have been so easy for Google to have done at any time. Their approach to AI has also been perfect IMHO. From the beginning they didn't have the models provide answers, but (basically, as far as I can tell) RAGged the results and summarized (and included sources). I find myself using that feature all the time. Also the ! support and generally clean interface make for a great UX. I also very much appreciate their pro-rating and transparent billing, and the prices being pretty reasonable.
The search results themselves are also better than Google (in no small part to my own fine-tuning of domains, i.e. downranking/upranking/blacklisting, but even just the vanilla results are better most of the time). It has been so long since I've had to !g something that I don't even remember the last time. I do routinely use !gm to open something in google maps though because Apple maps has not been a good experience for me, but I don't fault Kagi for that.
If you haven't tried Kagi, you should. Also play around with the lenses feature while you're at it, cause you can find some cool stuff that way that otherwise doesn't surface in the results.
Initial set up to make Kagi the default search engine is a little involved because you have to put your token in the query string, but the instructions are solid and once it's done, it's done. Very happy customer.
I use !g once a week or so in Kagi, and 100% of the time, google’s results are worse. It’s usually something bogus I’m searching for, and there really isn’t anything to be said about it, but when I search google for it, I get a ton more junk that is totally unrelated to my query.
Kagi’s results are better than google’s 100% of the time that I compare the two. It’s no exaggeration.
I made the switch to kagi because there was so much ai slop and stack overflow copy sites in my Google results. Sometimes I would hit three or four cheap clone sites of the same stack overflow answer trying to get my answer. Now I'm super happy with Kagi.
I think you can import filter lists similar to how it works with uBlock Origin and import some AI websites list.
Kagi is awesome, and their default search is just way, way better than google.
That being said, their upranking, downranking, and finetuning just straight up does not work. I spent like an hour pinning sites that I wanted to show up in my search results, tested extensively, and it turns out search just doesn't seem to pay attention to your pinned sites at all. It's one of the features that I was most excited about, and I'm super bummed it doesn't work. I've had an open issue since March about it on their feedback forum, and it seems like they have no plans to fix it.
https://kagifeedback.org/d/6670-pinned-domains-dont-appear-a...
While I'm usually good at avoiding such things, this one somehow worked on me and was insanly frustrating: you click the wrong button, the App Store pops up, you switch back to the web page, go back, and then (via a redirect) the same thing happens again. (Whoever implemented that deserves punishment.)
Anyway, between this and also that for many a technical topic Google search results are just so full of nonsense sites that asking the question of an LLM is actually the rational approach despite the risks of hallucinations, maybe it's time to give Kagi a go...
Hopefully ublock origin lite can keep the popups away for a while.
Ah yes, they do exactly the same thing with Maps.
I tried selecting it in the distraction picker but it seems to hide the entire page.
Early attempts at search-emabled AIs (e.g. ChatGPT) were nowhere near as good at first, so I stopped testing them against Perplexity/Kagi
It's well behind Kagi (which uses a bunch of sources for results and new models for quick answer or assistant). Perplexity is even well behind other competitors like chatgpt
Preference doesn't have to be based on anything, but providing misinformation that is easily verified is not that common on HN. And Perplexity Pro is different than vanilla Perplexity.
You're right though it seems that they don't just use bing + their own crawler, but who else they use seems to be community speculation.
+ I love Youtube Summaries https://chromewebstore.google.com/detail/chatgpt-summarize/c... (should be open source tough)
What's the difference compared to that setup?
Ad supported search is not incentivized to find you information as fast as possible; the goal is for you to click on ads.
I hate paying for anything. I am cheap as hell, but it saves me enough time that I can justify it. I felt google search was getting worse while I feel Kagi search gets better. Plus I can customize Kagi to ban entire websites or show websites, such as this one, more often.
I am not sure what the user experience of your setup is but with Kagi I can tell it to summarize like 8 sites and quickly read through all of them on the page.
Also, I didn't know how to use AI well with search prior to Kagi; the ease of use really helped me maximize its use.
My home is ad-blocked up the wazoo. I love signal and hate noise. Kagi gives me that.
I basically pay only to have that. The rest are lots of niceties on top that I can pick if I want them or need them.
It respects your keywords, if it edits anything of the query it tells you why and how, and I think I can count on one hand the number of times I had to scroll to hit the "More results" button since I started using it.
Before I tried Kagi (I tried it out when it was invite-only for testing and then converted to paying user) Google was basically frustrating me with every single query. I had to scroll past the ads, ignore the stuff that looked like websites, ignore the first 5 results because they were useless, ignore the "other results from X" because nothing of value was ever found, then go to the second page, and still find maybe 1 result that was applicable to my query.
Kagi felt like a breath of fresh air. I have a few Googler friends that tried it out on my recommendation and prefer it and pay for it over Google Search, I think that speaks for itself.
Sure, it will live on as a zombie of sorts. AM radio still exists. OTA television still exists. But their key demographics have long left (and in the case of CBS, are being forced to leave even faster). They won't be back. Yahoo still exists but it's so dead that its last act of relevance was to have an ex-Google exec execute what was essentially a pump-and-dump for the board. Similarly, few people say Xerox any more and just say copy instead. I don't even know how to call a taxi. Uber/Lyft is orders of magnitude better than taxis ever were.
The author is just a key demographic leaving Google. I left for Perplexity. My brother sent me a ChatGPT conversation which I asked some follow up questions to. AI is really good, now.
So I rarely search anymore. Search was always just a component in me trying to find an answer. Today, it's just a noisy inaccurate distraction.
The author won't be going back. I won't be going back. But Google has plenty of money. Like AM, OTA and Yahoo, it will continue to exist and you shouldn't feel too sorry for their ex-McKinsey CEO.
Search is still useful. Claude is not good at citing sources. Search is great for doing my own research.
I’m hearing a lot of people just copy and paste LLM output as truth. From students to law clerks. LLMs are designed to output convincing text (not correct text) and for lazy people that’s good enough.
There will always be a need for searching raw data and making conclusions without an LLM
AI models can't possibly contain everything to the depth one might require on a niche topic. Additionally, the less training data available, the more egregious the hallucinations become.
I've worked on some pretty niche things where the only way to get actual info is painstaking manual search queries based on a tree of keyword combinations. Those barely come up because search is (practically) dead, but when I'd ask AI the same questions or to find the same queries, it would simply make up confident answers.
AI is only truly helpful for the common denominator of very well documented and often discussed topics.
… it would simply make up confident answers. Dunno if you've used Google recently, but it will very confidently provide simply useless search results, pages and pages and pages of useless search results.
I look at using LLMs as like using another programming language. My C++ skills are not going to be directly useful in Rust or OCaml. I need to learn a new paradigm.
I'd ask AI the same questions. This is the crux. You really have to learn how to write (engineer) good prompts and you have to check what the LLM wrote. You use 'confidently' as a disparaging word. I think LLMs clear writing is quite useful in the reduction phase.
LLMs are perfect? No, just way better than Google.
To organize the world's information and make it universally accessible and useful has become To serve an ad and call it a day.
I had assumed that this degredation of search with useless results was an attempt to drive more people towards its even more useless AI products/results.
There will no longer be a reason to interact with a search engine directly for most people, AI will decide by itself if it retrieves from memory or performs a search.
It will never be as intuitive or efficient, not even mentioning the reliability.
A picture is worth a thousand words, and no LLM is going to change that.
What a colossally stupid waste of energy. We could have had our answers, with a decent search engine.
I actually look down on LLM hopefuls. Just so irredeemably up their own ass. When the bubble collapses they should be sent off to their own island where they can sell the same coconut back and forth between themselves.
It's not perfect but I know that I'm more likely to find a result that serves me, rather than a result that's optimizing for an algorithm. And if I do come across something spammy, I know that telling Kagi to adjust a site lower (or remove a site from results) is just 2 clicks away.
Even though it's a small thing, it's one I encountered with every Google search, manyamnymany times per day. Relieving that is worth more than $10/mo to me. I think that pricing is right, though—if it were more than $10/mo and I likely never would have tried it.
The only thing I use Google Search for now is hyperlocal stuff, mostly I expect to interact with Maps result. Kagi Maps isn't even close.
I can just search something, glance at the results and pick the one that seems the closest one to whatever I'm looking for.
With Google, I have to first go "is this an ad pretending to be a search result?", then I need to spend more time looking that the domain and preview to see if it's some listicle crap or AI slop. Only then I can click the result.
Yes they are very helpful. But what is the incentive to create more blogs for them LLM companies to scrape for novel tech? They are great for answering questions about things that are very well documented and understood but the incentive structure that aligns technical bloggers with search is being undermined.
In tech, knowledge scales non-linearly. No amount of trivial search can amount to finding excellent technical writing. Most of the stuff LLMs are great at answering are things that an individual can typically figure out already just much more slowly by RTFM (eg: react component, MVC code etc.) However, the LLMs fail at deeply technical or highly novel subjects.
I worry LLM usage overtime will create a gap between research papers and engineering as nobody is incentivizeded to write about their implementations/explorations.
I am hearing more and more tech-people jumping to Kagi regardless of role (SWE,SRE, PD,DS) which is encouraging.
Please delete your comment before a Google product manager reads it and puts it on their idea board.
Still, I think paying for search remains a taboo and this is unlikely to change in the near future. So Kagi seems poised to remain a service for the tech-literate — which is precisely the kind of audience that already knows how to use ad-blocking, avoid Google’s AI snippets and so forth.
I get it won't ever become as big as Google if everyone has to pay buy I don't think everyone has to become that big anyways. I'd rather have multiple search engines with varying strengths and weaknesses than another monopoly surrounded by smaller companies keeping Microsoft Bing on life support.
The Translator outperforms Google Translate and DeepL in my experience, too, and provides very nice context and such for translations. Kagi Maps might become the premiere OpenStreetMap interface, too, with amazing integration of other resources. Just one anecdote about it: I just checked in Kagi Maps a local restaurant I added a few weeks ago in OpenStreetMap, and on Kagi they somehow have some interior photos that circulated in a local magazine. Amazing stuff
There will always be users who refuse (not going to convince my parents ever), but for many power users, or semi-power users, it's becoming more acceptable to just pay the $20/mo and get a better product.
The limited/free functionality seems to more than suffice if your use case is just replacing Google.
Damn, you just gave me a faint glimmer of hope for a search renaissance somewhere down the line, when it becomes so niche that it won't even be worth it to SEO-spam.
OTOH, the existence of Kagi Assistant made Kagi Teams a very easy sell to my employer.
I happen to be an Ars subscriber and that's about as much as I'd be willing to pay for Kagi too.
I'm hoping enough people agree with me for Kagi to be viable.
!ai sends the query directly to kagi.com/assistant with you default model. If you pick a solid and fast default, like kimi-k2, it's a great experience
Wholeheartedly agree. Millionshort was based on this idea, of effectively skipping the "first 10k" sites that most commonly come up in results, skipping to the good stuff. But they've implemented forced logins and made other design choices I haven't loved.
The article "The Men Who Killed Google" by Ed Zitron argues that Google's search engine has declined due to a shift from a user-centric approach to a revenue-focused one. The author pinpoints Prabhakar Raghavan, with the support of CEO Sundar Pichai, as the key individuals behind this change.
The article details a "code yellow" in February 2019, declared by the ads team, due to underperforming search revenue. This led to a conflict between the ads and finance teams, who wanted to boost revenue at any cost, and the search team, which prioritized user experience.
Following this, Google implemented several changes, including a March 2019 update that rolled back quality improvements and a May 2019 ad redesign that made ads look like organic search results. In May 2020, Raghavan replaced Ben Gomes as the head of Google Search. The author notes Raghavan's history at Yahoo, where he was head of search from 2005 to 2012, a period during which Yahoo's search market share significantly dropped.
Zitron concludes that Raghavan and other "managerial types" are damaging technology by prioritizing shareholder value and growth over product quality.
We can't complain as it is exactly whats asked from management of public companies innit?
And that's why most products sucks nowadays
The founder started that he was not interested in serving users who want anonymity. I see they've since added Privacy Pass using VOPRF tokens, which does provide anonymity by decoupling searches from user accounts. But by now LLM search tools are good and don't require an account.
"Yes, this makes sense. This is possible because technically the extension does not care if you have an account or not. It just needs to be ‘loaded’ with valid tokens. And you can imagine a mechanism where you could also anonymously purchase them, eg. with monero, without ever creating an account at Kagi. Let us know here if you are excited about this, as it will help prioritize it." https://blog.kagi.com/kagi-privacy-pass
They also provide a Tor hidden service which can be used with Privacy Pass (although you'd have to install their extension on Tor Browser for it to work)
I don't want to be in a group of people, someone asks me to look up how old some celebrity is, and then information about that person gets pushed to me by algorithms from then on. It's a trite example, but a real one. I've found myself becoming aware of which topics the algorithms know I've looked into, and trying to groom that list.
In Europe, you can't presently use Gemini without being logged in - presumably having something to do with their recent tech laws. I don't know if there's a way to delete searches from Gemini either. I really don't like that.
I should be allowed to be curious about something without having that curiosity etched into my permanent record.
But it doesn't, at least not for me. But I think that's tied to poor implementation, design, and being a solution looking for a problem rather than a philosophical issue with using AI to improve search experience.
Google is…fine? Maybe I just learned how to use it for what I do
like, take existing bookmarks, make a crawler.
nice work, btw
The hardest part BY FAR is the crawler: initially I was using Apache Nutch but it got slower and slower as the index grew, so I replaced it with my own crawler that I wrote in PHP (comfortable for me) and made that multi-threaded using Supervisor.
The second hardest part was the amount of security I had to build in to prevent bots running spam searches and hogging my infra.
I'll try to write a blog soon and post it here.
1. Always identify your crawler via a consistent user-agent string, that explains its a web search crawler and not a generic web browser.
2. Always obey the directives in robots.txt.
3. Make sure your crawler is not too aggressive (low frequency of requests).
(updated for formatting)
- !w: results from last week only
- !y: results from yesterday and today only
- !io: Assistant, but offline
- !ig: Assistant, but for grammar correction
- !itr: Assistant, but for correcting and organizing transcript text
- !sh: Google shopping tab
Did you just shadow the Wikipedia bang?
If Kagi stopped using Yandex, or somehow allowed accounts to configure a way so that their funds do not go to Yandex (unsure if this is even possible to split accounts this way) then I would sign back up immediately.
What a waste of time this conversation is.
"Kagi has been using yandex search api since 2019, long before current geopolitical tensions"
The war in Ukraine has been ongoing since 2014.
"We understand these are deeply personal issues for many. As someone who was a refugee of two wars, I'm not indifferent to human suffering."
He is Serbian.
Regardless, the option to exclude Yandex (and other search engines) results has been suggested here: https://kagifeedback.org/d/4727-option-to-choose-or-exclude-...
> The job of a search engine is to produce the most relevant search results, period. Kagi excels at this precisely because we remain unimpressed by world politics. The moment 'politics' becomes a factor in search results is the moment I stop working on a search engine.
> We set out to fix search, not the world.
Kagi is awesome. I appreciate the Yandex results. I'm not a big fan of Russia.
Right, but there's lots of companies in the world. Lots more than there are tyrannical regimes. Lots more than there are total regimes! We're not individually obligated to support each of them, when they financially support things we find abhorrent.
But no, I don't like the Chinese government either.
Maybe they have to do it to survive, who knows, but you don't have to use their services or products that resell their services.
Holding and adhering to personal principles is the opposite of weak. You might disagree with that person's decision, but that does not make them (or their decision) weak.
I use DDG mostly but if I need to actually search the web I go to Yandex (I don't have Kagi... at least not yet). I don't know of another place that can do that.
Like, is there a viable alternative for Kagi?
There isn't really an escape from helping despotic, genocidal regimes.
I recommend https://dzygaspaw.com personally, but there are many good crowdfunding organizers.
It has brought genocide, concentration camps and forced starvation back into the Overton window for state actors pursuing security objectives. This should terrify everyone, whether inside Europe or not.
In the UK, simply expressing support for a protest group is now imprisonable under anti-terror legislation. Germany is dishing out fines and deportations for as little as holding up signs.
Post-war Europe is ostensibly built on rules and norms around human rights, democracy, and the rule of law. The behaviour of the state of Israel, and our governments' craven support for it, is causing rapid democratic backsliding and laying the groundwork for our next round of atrocities.
Whatever one's moral perspective, or position on the middle east, this concerns everyone.
I do hope Kagi eventually has enough revenue to offer a more generous free tier, though (and without logging in). I fear it will remain a niche thing otherwise.
Despite a tough market, Kagi succeeds the old way: with better products.
Kagi proves plain old commerce - making a good thing and selling it - can win a fair fight against capital. Keep it up folks!
easily worth 4 coffees a month just to not worry anymore about the direct consequences of my searches, and being able to shitcan garbage sites like pinboard.* while pranking wikipedia in my own search results is just gravy.
they also have a search API now in beta which you can request access to; 2.5c/search.
The only smaller search engine with its own real index is Brave.
The index was why I looked elsewhere.
The reason I switched to Kagi was because of their business model where I'm a customer not a product. So results are theoretically tailored for value to me not value to the company.
With Kagi, the results are better than Google, and the first couple times I couldn't find something with Kagi and went back to Google, the Google results were even worse.
This is because I don’t have consistent usage habits, and the price isn’t stopping me, the psychology of a monthly subscription I don’t use is.
[1]: https://help.kagi.com/kagi/faq/faq.html#fair-pricing
[2]: https://kagi.com/changelog#6155
[3]: https://help.kagi.com/kagi/faq/faq.html#if-i-upgrade-my-plan...
And it doesn't look like there's any sort of "you only used 20 searches so we'll Free Tier you rather than $10 tier you this month".
This is not true, you don’t pay if you don’t use it. I think you’re assuming “credit” means some sort of redeemable voucher, but it’s account balance that automatically covers your next bill. Then at the end of the second unused month you get a credit that covers the third month, and so on.
The thing I’m mainly on about is the UX of not having to babysit another subscription. I once had a newsgroups sub that was exactly like this and it was amazing. You’d say “I want to spend at most X” and it would just ramp up towards X tier as needed.
Not sure how long this scales because like I said, my discovery rate has pretty much been zero for years now of new domains worth visiting. I'm sure they are out there but modern internet doesn't really let you find them short of finding a reference to them in some forum comment someplace, thanks to SEO arms race.
And that you can set the model to answer those query as any of the grok models?
Kagi isn't behind anyone else on this AI-powered search, in fact the assistant is nearly a strict superset of all the other ones.
You can set the model to chatGPT, claude, gemini, grok, etc. And it'll use kagi search as a backend instead of, respectively, bing, brave, google, google+twitter.
Kimi-k2 for instance was a positive surprise recently and is being added to your standard plan tonight as well as the brand new GPT-OSS models.
Except the last few months. And, I can't stress this enough, its not DDG's fault. The results are relevant to my search, and at first glance seem to be what I'm looking for. But take more than 2 seconds to read, and you'll notice the websites themselves are trash heaps of AI slop, serving errors and mistakes.
The human web is quickly shrinking out of existence. The last content silos of human expression have walled themselves in and are striking deals with the evil megacorps; pushing their shitty, error prone, planet killing, word prediction engines down everyone's throats.
Soon, there will be nothing of value left for Kagi to index. Maybe they should pivot to web autopsy tools, not unlike Internet Archive. I wish the Kagi team well. I hope they succeed. And I really hope I'm wrong about the open web's grim future. But I'm probably not. And the AI arms race suggests the megacorps agree.
1. The Internet Archive
2. Anna's Archive
3. YouTube (if they're allowed to)
That's more than enough knowledge and information.
Am I too optimistic?
In my opinion, we have gone full circle and arrived back to human curation to provide quality search results.
We have something in the works to fight back on that front, do not be worried ;)
Sadly, Google has this dark pattern where you are trying to search for something and they'll show you a pop-up to install something, and they change the popup until you click the wrong button and all of the sudden instead of buying plywood you find yourself in the app store to install God knows what (but I bet my rear that it tracks me). So I have gone now through the hoops and installed Kagi in iOS as well.
I'm not a google user but - are you sure you don't have malware or something? I guess probably not as you said iOS, but I'm failing to imagine this being true in the way I'm reading it.
Like, the page loads very fast, but only with the search bar, and only then you get the results appearing after some loading. But it's acceptable.
Image search is rather slow, though.
Foe now I really like Qwant, they used to relay on bing, but now do their own indexing.
If you decide to stick with google, change your search to be web-only by default, by entering "https://www.google.com/search?q=%s&udm=14" as your default search engine.
This does cut out the AI summaries as well as most of the infobox cruft. On the other hand it also cuts out the convenient unit conversion and calculator stuff, and for local results you have to navigate to maps, etc. But the inconvenience here is worth it because the amount of spam you get on the main search page has grown to the point of absurdity.
The result quality has, alas, significantly decreased as Google has shifted its focus away from that, but with this change it is nice and snappy and mostly works.
"The obvious flaw here is that Kagi still records source IP addresses along with Privacy Pass searches, potentially de-anonymizing them"
Clearly such a thing is possible, and a company which insists it isn’t is fundamentally either uncreative or user hostile.
Just the nature of it? A subscription is always going to be a regular fee that pays for an aggregate of multiple things, some of which aren’t used by every subscriber. (And which some subscribers are going to object to paying for.) The only way to ensure that you only pay for what you use is a pay-per-use model.
And conceptually I really like pay-per-use models, but the public tends not to love them - in particular for things with low marginal costs of production/serving, like software services. (As opposed to medium marginal-cost goods, like say, home gas utility, where non-usage-based pricing is available, but tends to be poor value.)
and then i ran out of free searches, so I gave them $5.
and then i ran out of $5 worth of searches so i kept renewing, $5 more. And again, and again.
At this point I had all my defaults set to kagi because I loved it. Just the fact there are no ads or sponsored leads and I can more-or-less kill webpages completely generated and SEO'd with LLMs, I can actually do research again.
and then I became a full subscriber, with orion, blacklists, I use the quick answer (LLM) and small web searching fruitfully, and joined their discord.
It's been a long time since I fell head over heels for a technology product, and this being something so core to how I use the internet, it's a godsend.
Enjoying it at least until they decide money is nice and growth is imperative, they go public, and enshittifies.
Seriously, I haven't used Google in years. I stopped even relying on Google as a decent backup option because it's gotten so-o-o bad. DDG has gotten incrementally worse simply because they're unable to fend off the reams of AI slop taking over everything (and it's also based in large part on Bing's index).
I recently tried Kagi now that they have free and lower-cost tiers, and oh wow I can feel the difference. This is like 2010s-era search, and I mean that as a distinct compliment. I signed up and won't look back. The "Quick Peek" feature is truly fantastic…it feels like it's some sort of AI but it's simply just a good set of queries with good excerpts of REAL text from REAL links. Bravo!
Screw you guys! I'm creating my own search with open source.
... but seriously it just augments my search.
I think probably Kagi should just make this the default - doing good maps is a hard problem, and I dont think its one that is solvable on Kagi's scale.
We have new maps UI and working on improving search API accuracy so would love some feedback where we are currently falling short to guide our roadmap.
TL; DR: forced localization of YouTube video titles to the language of my location despite language being set to English ("Oh I'm sorry, that's just for the UI"?)