Kagi Changelog 2/13: Faster and more accurate instant answers and Wikipedia page
kagi.com
kagi.com
I expected to like lenses and favoring/blocking specific domains. What I didn’t expect was how much their “Quick Answer” would change how I search.
I’ve been “AI hesitant”, in general the chance that an LLM will hallucinate makes these kinds of tools more trouble than they’re worth for me personally. In Kagi’s case, though, the individual facts it states in the quick answers have citations linking to the site it drew that information from.
Here’s what I’ve found:
- it’s been accurate most of the time, but not 100% (as expected)
- citations are pretty accurate most of the time
- every so often the citation links to a page that seemingly doesn’t back the claim in the quick answer
Unsurprisingly, I don’t trust the AI generated quick answer in isolation, what it does do is let me scan a few paragraphs, find the one that answers my question most specifically, and visit the sites it links to as citations for that piece of the answer. This saves me the time of clicking through the top $N results and scanning each page to find the one that seems to answer my query most directly. It’s like a layer on top of the page rank.
I remember using Google the first time and being impressed how the top answers were so much more relevant than Yahoo, it was a huge time saver. Now I find myself wondering if the “quick answer” citations will prove to be a similar jump in accelerating my ability to find the right web page.
It also makes me wonder if their own page rank algorithm could incorporate the quick answer output as an input to a site’s rank? That would be an interesting experiment!
On a side note, as a software dev I was curious about how it would "answer" queries that weren't questions, so I've tried feeding it queries like "linux distro" or other things that are most certainly not questions. About half the time it does a pretty good job at answering! For the really hard ones, it seems to just morph it into "what is <query>?" so nothing real complicated, but I did find it interesting.
Example: If you search for a specific movie title on Netflix but they don't have it, then they will give you a list of movies that they think are similar to the one you searched for. That is because their database actually knows about the movie and therefore can find links to other vaguely related stuff, e.g. movies made by the same director, with a similar theme, etc. But if I search for a specific title, then none of this is what I want, and I don't want to spend the extra 10-20 seconds scrolling through the list to realize that they actually don't have what I want. This is clearly a search experience which is optimized for maximizing engagement rather than user experience because a small minority will end up watching something from the garbage results while the majority will waste their time and be burdened by extra cognitive load. Shareholders are happy, users suffer.
With Netflix I assume they use data from IMDb for finding similar movies.
But one platform having particularly surprising ability to find “similar” things is AliExpress.
On AliExpress if you search for a brand and model of something without saying what it is, AliExpress is still sometimes able to know what kind of thing you are looking for and show similar products from other brands. And I’ve been wondering how they do that.
Maybe AliExpress has a big database of products that they scrape from the internet and classify, even for brands and models that have never been on AliExpress.
Or they could be able to do it based on similar queries that people made in the past where someone for example included extra keywords about what they were looking for. Or those people first having searched for a brand name and model and then made subsequent searches for more generic descriptions of what they looked for.
Or sellers could be including names of brands and models for products that are similar in the description or other input fields for metadata for their listings.
I absolutely hated that when I was a subscriber. That 1/4 of seconds of believing the search will succeed, just to give me the subpar copycat of the movie I was looking for.
This has never been the case. If it can’t find your title, it’ll display “titles similar to”, right at the head part of your search. No 20 seconds of confirmation needed.
I actually prefer Netflix’ way because if I search for “Demolition Man” and they don’t have it, it might be that I’m in the mood for any <2000s action schlok, and who says I already know about “Escape From New York”?
I searched for "Ted Lasso".
It has grey text "More to explore:", white text "Ted Lasso" and then thumbnail list of different shows, and it's literally just thumbnails, you can't even Ctrl + F and you have to read all the titles in different colored and stylised fonts.
It's as if it is intentionally built in such a way to make it hard to understand that it's really not there.
Edit:
And in TV it says nothing, just gives you the thumbnails and since it takes longer to type you must check after each character whether one of the thumbnails happens to be what you are searching for.
It seems to me "everyone" think it is always about privacy or features or something.
But the main thing that keeps me on Kagi is the results. They seem to have most relevant results and few irrelevant results and if I decide to be specific using doublequotes I get no irrelevant results wrt that word. (And if you find one it is a bug and will be dealt with.)
I have lost enough hours of my life clicking through Google or Bing results that maybe has something relevant to my search.
Edit: I have been beating this drum since matt_cutts was in Google and used to frequent HN and so I think it is relatively clear that Google does not care about the quality of the search results.
A decent search engine should do the same, be able to tell that you're doing something wrong, and do better if you want some answers.
If we balk at AI when it hallucinates, we should balk at search engines when they hallucinate, too.
Kagi does the correct thing, IMHO.
Then it should suggest a better one and then evaluate the query anyway.
> This is what any decent search engine should do -- return nothing
WHY?! That's the opposite of it's job!
I have an account with a username that is spelled very similarly to a real word. Google will suggest searching for the real word instead. If you do that though, you'll never find the username!
I'm tired of people saying the computer should not do what I tell it to. It's like children who won't even attempt a multiple choice test because they aren't 100% sure
Hacker News in a nutshell.
I'm saying I don't like high cutoffs of similarity scores. I have no idea what you're talking about.
Very very few queries should have _literally zero results_. Surely you have at least a few words in common with something
I'm saying if you have a query that returns a similarity score, I don't want only results with > 0.1. I want all the results returned
> Then it should suggest a better one and then evaluate the query anyway.
Google does this, and they suck at it, unless you just spelled a word wrong. Do a niche or very specific query, for which Google has no answer and it will, without fail, remove the most relevant keyword and give you a bunch of junk results.
Like search for things you did not search for...?
No the hell not. It should do what I tell it to. For a search engine that is to show me what it has about the query I input. If that is nothing, that's what it should show. It should not show me entirely unrelated results, ads, or what it "think" I meant. Not its job.
That's what I said. "then evaluate the query anyway." I should have added "original" to that statement
> If that is nothing, that's what it should show. It should not show me entirely unrelated results, ads, or what it "think" I meant. Not its job.
I'm saying I don't want similarity cutoffs. Most FTS methods involve a similarity score, I'm saying I don't want only results with > 0.1 similarity. I want all of them that were returned.
I'm NOT saying it should somehow inject results that didn't originate from the original FTS query.
It’s terrible but far better than getting 100’s of irrelevant results because Google decided two words out of 10 in your query were the only ones that matter.
I've had queries of copy+pasted errors with zero results, but playing around with it a bit just to find a github result that was only like two words off.
> it’s terrible but far better than getting 100’s of irrelevant results because
Surely the similarity to the one on github would still have it ranked on the first page?
I have never asked for or even wanted "personalized" results, because on Kagi and everywhere else, personalized is shorthand for "very very poor guesses". It's very frustrating.
disclaimer: I work for Kagi
Yes but there's a >0% chance that you'll click on a potentially sponsored link (or a non-sponsored link to a page that itself contains ads) when you instead see a bunch of unrelated results. It makes financial sense to show random results vs not showing anything.
It's great for technical lookup, and answers, but for general search usage it's not there yet. I'd love to give actionable feedback on how they could improve, but can't figure out exactly what's missing on that department. It could just as well be a small crawled index.
I'll note that to get local news, I do have to switch the region selector from "International" to "Portugal". Kagi doesn't have Google's behaviour of using my IP location. Which is good. Getting international results from Google is a struggle.
Showing the score of a football match, and doing online shopping, are the two remaining use cases I have for google. My usage of Google has now gone down. 95%. Kagi is simply better.
This seems to be recent, so I guess that's why I didn't experience it.
This is the type of attention to detail that makes me a happy user. Killer feature for me, no more having to type !quick (and yes I set up quick bangs but always seem to forget).
I have been really pleased with Kagi, though I do live in perpetual fear that they'll be acquired. Please no. And God help us if Apple acquires them, which seems like a risk given Kagi's great work on search and AI, two things Apple really wants. Don't do it Kagi!
But they're being cautious with it, so I'm optimistic. I'm still a happy supporter and user.
Of course, that's the only answer any successful founder would give you...so I still live in fear :(
Just this morning I realized that when I look up a classic Disney movie Kagi turns up the classic and not the 2010s remake, where Google invariably turns up the newer one. Little user-centric things like that are why I love Kagi so much!
The latest Firefox ESR I can run on my old laptop doesn't like ??= and some other newer JS syntax. I see this sort of problem in my own daily development and I know with tools like esbuild you can simply specify --target=firefox78 and end up JS that works fine on older yet perfectly working hardware.
So what does this have to do with Kagi? Well after a bit of looking around I found https://kagifeedback.org and reported the issue [1]. A day or so later Vlad triaged it and it was fixed within the next few days.
A month later the problem reappeared in another place[2], Vlad triaged it again and it was promptly fixed.
Needless to say I'm very impressed with attention Vlad and people at Kagi give to problems and the effort they spend improving their product.
Makes me want to learn Crystal :-)
1- https://kagifeedback.org/d/2766-kagi-search-interface-no-lon...
2- https://kagifeedback.org/d/3055-kagi-search-on-firefox-esr-g...
1. Location based or aware searching - it keeps giving me the UK versions of storefronts or the identically-named steakhouse a thousand miles away and not the one I can walk to. Kind of lame, particularly giving me random countries’ versions of sites. 2. The map — maybe this is a setting, but the map always gives me nearly useless results. This is probably tied to geolocation or location aware results.
Google is still extremely competitive in the maps/location space, and I’ve heard Apple Maps is very good these days. Secondly, these features are not trivial. Maps are much more interactive - giant teams or even orgs are managing these products at faang.
On the other hand, having language-awareness is perhaps more important. Google frequently gives me results based on location, although I don’t speak the language (well). Getting results in English is hard where I am, in the typical infantilizing fashion of megacorps (we think you want X, and we’re not gonna offer you a setting).
Edit: apparently Google added Spanish as a language automatically based on some searches, and then search results for say API docs came up in Spanish first instead, despite English being my “preferred language”.
I don't know if there's a toggle I just haven't found yet for just automatically using Google Maps for location based searches until their feature gets a little better.
Beyond that, Kagi has been an incredible breath of fresh air.
[0]: https://kagifeedback.org/d/3022-english-results-given-too-mu...
Now I can just keep it on my local language, even while doing English searches. Previously I had to switch between national and international all the time, that was a pain.
> Now I can just keep it on my local language, even while doing English searches.
That’s exactly the issue. I search for a name, so it can’t know what language it’s in. But I want the German results, that’s why I switched it to German.
> Previously I had to switch between national and international all the time, that was a pain.
That wasn’t an issue at all for me, I use !de for German searches, and nothing for automatic international searches.
My language, Norwegian, is also a much "smaller" one than German, so search results without any English are outright worse. Personally, even if I spoke German, I'd prefer to get Wikipedia on top for a search like "Olympus E-P2".
It seems like they should implement a "verbatim language" search option, so both use cases can be satisfied.
And never the US Amazon or eBay over the German version.
To be clear, I'm not saying that "this is already good enough so what are you complaining about", I'm merely suggesting something that might go most of the way to fixing it until Kagi eventually implements a better search. They are a small company and I'd imagine that non-english language (and especially multi-lingual) behavior is not going to be the top-most priority, as understandably frustrating as that is for the people who need it.
Kagi is not broken in general, 95% of the time it's still fine because almost all my searches are in English, so without !de in the German region. It's only the regional search that is not working properly anymore because they went further into "we know what you actually want to search for" territory. Not doing that was a huge part of what I loved about kagi.
In addition, this worked fine all the time. I've been using Kagi since the earliest betas, they only broke it very recently.
These integrations with other services and specific data are remarkably useful. Glad to see more! One weak spot of Kagi for me is their local / maps search. It's Apple Maps which is still not as good as Google Maps. But moreso Google Local has become a very strong product with a unique database. Nothing stopping me from using Google for local business info when I need it though.
It doesn't help to rely on Apple Maps when everyone can contribute to OSM directly. It's also easy to make it accessible enough (but not to much to avoid trolls) just to increase OSM contributors base.
That's what Kagi does with their Wikipedia oneboxes and I regularly open the article to correct it when I notice an issue. I wish Kago (and others) would do that with places, so we can avoid opening the Maps tab to have quick place info.
Imagine being a business owner and ignoring the very basics...
The thing is: Submitting your info costs nothing in money and almost none of your time, but brings enormous benefits in bringing in customers. So if the business owner is too lazy to do it, why should anybody else do it?
If you submit correct info for your favourite local business to a map service, that might bring them thousands of new customers and the revenue from them. Then go to them and tell them you'd like a coffee since you put their info neatly into the maps. They'd probably spit in your face. Then turn around and pay thousands for Google Ads or for a local social media influencer to bring in 50 new customers.
Business owners being dumb about their internet presence is why for example Booking.com is a hundred billion dollar company.
I’d say toward the end of 2023 we passed the inflection point where Google is now generally worse. In fact these days Google is often worse or at least no better than Bing and DuckDuckGo. Kagi is better than all of them.
It seems like the Internet being winner take all is only part of the story. The reality is winner takes all, then turns to shit, then an unbundling occurs.
Still like Kagi better for the results (especially technical stuff), but I do still use Google for regional/news/shopping searches or things where Google Search behaves more like an assistant.
I think this is an area where Kagi will have some trouble competing because of Google's data moat and the fact they have been honing non-web integrations for a long time. This long tail must be 90% of the job.
Search is so much their golden goose and they've let it languish and now there are real dangers to that product from several flanks. They have Gmail after that but is that enough to keep the coffers full?
I had so much hope in Google Workspace as competition to Microsoft 365 but even that is in trouble if you ask me. I've been helping a company migrate from Workspace to M365 and one of their reasons was because they couldn't get any support from Google. Also, Workspace is hard to use. Not that M365 is easy, or that their support is any good but, Google has clear perception problems here.
I almost never hear of anyone talking about GCP anymore so I'm guessing they're losing mindshare in addition to market share.
Dart and Flutter are wonderful projects but it seems to me that since there's no way to stick ads on them, Google is content to give that department a little money and let them do their thing. But, of course, when some of the aforementioned products start losing more money, great products like Dart and Flutter will certainly be cut.
Maybe I'm looking at it the wrong way. Maybe Google's time has come and companies like Kagi can come in and fill the void Google seems hell-bent on leaving behind.
Gemini 1.5 was "announced" and you can't use it or see what it really does, it's just "better". (Admittedly, the context length sounds incredible, but we can't really test it and hype only works when there are no alternatives.)
Sora then dropped immediately afterward and was showing numerous really impressive demos and then taking user requests all day.
To maintain your level of privacy requires a drastically different lifestyle than most of us have or want.
That is for me one of the downsides of Kagi.
Also, you probably are aware of this, but an incognito window won't magically stop trackers.
They didn't state anything about their desired level of privacy other than not wanting a username attached to every search. That's not equivalent to needing to use TOR for every search.
>99% of people are probably signed into a google account when they search google
What is the relevance of this statement? Obviously this wouldn't apply to the parent poster, it doesn't refute anything the parent said, and at least anecdotally it's not true at all.
>If you want the sort of privacy you seem to want, stick to the dark web and similarly inconvenient tools.
Privacy is a spectrum, not some binary choice between having a username (and often real name) attached to every search vs. using the dark web for every search. You can land somewhere in the middle, for example: "I don't want a username attached to every search".
There is no technical reason Kagi needs usernames, but they choose to require them. For some people, that points to the company not being as privacy-friendly as other people seem to think/claim.
I'll mull the decision for a while but I'm probably going to give Kagi another try.
If I search for `foo bar baz`, I want pages with `foo`, `bar` and `baz`, please. No, verbatim doesn't cut it because it somehow is far too strict and even common queries return half a dozen results at most. Google started sucking in the 2010s when they decided they know better than me, and the more Kagi improves, the more they follow down Google's path of trying to be outsmart me, and falling short.
I just want a 2003 Google experience. A computer is not smarter than me, stop trying to interpret and make assumptions.
---
Aside, is it so hard to make a search engine with a query interface like any trigram-based SQL full-text engine? With quotes, AND, OR and -word. I still believe any AI-powered solution of the past 20 years to be a mockery of this gold standard of precision and simplicity, except maybe Google 1.0.
Google in 2003 was good because the web wasn't quite so full of utter garbage. There must be 100x more content at least now.
But Kagi often being as good as Google used to be proves this isn't the only reason.
Lately I have felt Kagi also sometimes stray down this "knows better than me what I want" path.
I recommend using the forums to let them know. They actually read the forums and act on it.
I've reported both ux improvements and search quality issues and more and they have always taken it seriously.
Since Kagi became Google for my purposes, I started using Google again. I don’t have much motivation to curate the customization features personally.
I would imagine that it could depend on the general type of query. As a developer I definitely noticed a decline in quality.
I’m quite privacy focused but for whatever reason don’t care if Google knows that I want to know how to write a helm chart or eat pasta later
I noticed something similar with Ecosia, they apparently switched to using both Bing and Google, and the results, while still pretty good seems a little worse. That's all anecdotal and may be completely random, but I think one issue could be that Google is getting worse.
There's more blatantly obvious spam results seeping into Kagi as well, though at least I can block those domains when I come across them. A few of them were hacked websites, notably government websites in South Asia that had been thoroughly compromised.
(Just found what you meant by Verbatim search - nope, I would have always had Personalized turned on instead. Seeing those results would be the first time I even encountered those domains, so I didn't know to block the domain until then. Since the domain is still in my blocklist, I can confirm it was a government website in Mumbai. I don't have any reason to be visiting Indian government sites.)
I'm not saying every Kagi search is infected with hacked results, it's only common at times when I'm really researching something and need to get to a second page of results.
I also note that some Warez sites rank highly in Kagi too. Knowing that Kagi's customer base is largely from HN, that's less surprising to me, and I'm thankful that I can block those domains from my own personal results.
"train from Galway to Dublin"
"Ilac shopping centre opening hours"
Kagi gives you a bunch of links that might be absolutely trash, if the service provider (eg: rail company) has a shitty website. Google gives you precisely what you're looking for, eg: a table of departure/arrival times for the next 5 trains, with a date picker. It can even handle the atrocious TFI (transport for ireland) car-crash, providing me with the precise info I need to get a bus from one part of Dublin to another. You can't even get that with the official TFI app, and they notoriously block access to their data API (breaking the only decent Play Store app for buses in the city)
I'm not sure if this is because Google have a budget for curated content like this, or if they have some magic way to extract that data without making a balls of it.
They have a system where public transport operators can upload their routes in a specific and quite complicated file format. Google does not integrate any public transit unless the authorised organisation uploads the data to them.
(Decent results, sure, but that is subjective.)
So, a big reason I was using DDG over Google, despite often subpar results, was bangs. Now I pay for Kagi, and guess what convinced me to subscribe? Bangs indicate caring about power users. How cool would it be if I could set Kagi as default search in any browser and get powerful shortcuts at my fingertips across all devices?
Well, for some reason Kagi seems to expect the bang to be added at the end, separated by a space, with exclamation mark after (not before). If you don’t do it just right, you’ll just be on Kagi, losing 5+ seconds and your train of thought.
I’m sure I’m in a minority but I swear, if Kagi is not going to make bangs work properly soon, I’m not renewing. If you do emulate DDG, go all the way, otherwise it’s just taunting your users…
/rant
!yt cat videos (works)
cat videos !yt (works)
cat !yt videos (not working)
cat videos!yt (not working)
I always prefix but I did that on ddg too, because chromium's tab search thingy has trained me to do it.
Basically you have your request and you add a criteria on it, not in it.
It's like ordering at a restaurant : you wouldn't say "I'll have a burger to take away, some fries and a coke". You either specify how you'll have your order before or after the actual order. And you don't create a new word like "a cokedinein".
Furthermore, I want Kagi to be able to reply to the request "how do the !g bang works?"; if I want something else to reply, I'll add the relevant bang around my request.
Or I'm missing something?
Right now it just falls back on Kagi to execute the bangs, but in the future:
- Bangs executed locally (when possible)
- Local "custom bangs" configuration (create new bangs, update existing bangs)
- Select fallback bang provider (queries with bangs: DDG, Kagi, Brave, etc)
- Select search engine separately (queries without bangs: Google, Kagi, DDG, Yandex, etc)
- Lots of other features, like executing multiple bangs at once.
Also note DuckDuckGoog supports DDB bangs with Google search: https://hw.leftium.com/#/item/4260964
What?.. bangs work both at the beginning and end of the query, and the exclamation mark has to be put before the bang.
However, I wonder if they [ninja-]fixed it, because it does work as of now. Kudos if so!
I always wanted to try Ultimate but I was in a Duo account with my wife so it wasn't really possible. Now I can, and trying out all those LLMs is fun!
E.g., XSS / HTML injection in summarizer or discuss document. Or their broken CSP which allows injecting forms to e.g., change settings.
They haven't fixed many reported issues in a while, and just to prove I'm not lying: https://kagi.com/discussdoc?url=https%3A%2F%2Fkagi.com%2Fcha...
Still, would have liked an official take on this. I was about to re-signup but now I'll hold off on that.
Admittedly, no full XSS anymore, but still dangerous and shows their lack of understanding and caring about security.
It's not the only place you can inject HTML and not every page has a CSP...
Is it? They say searches aren't tied to account: https://help.kagi.com/kagi/privacy/privacy-protection.html#e...
Anything specific about the privacy angle?
The alternatives demand your data and/or lets you pay an attention fee by showing you ads or irrelevant results.
Neither is perfect, but it's hard to do worse than "you have to be logged in with an account associated with your payment details".
At the moment I am aware of two search engines available to me that doesn't try to drive me crazy by wasting my time on irrelevant results:
- kagi.com
- and search.marginalia.nu
and the second one, while being honest and high quality, has a rather limited index.
A directed effort toward unfucking query understanding and index execution is up next on the list of tasks to tackle though. Hopefully it'll make a decent impact.
As far as I know, Google doesn't even have that feature and I like to be able to switch between international and local results.
Aside from that, I wish Safari had a setting to add custom search URLs, so that I can paste my Kagi auth token, instead of having to use xxSearch, which also has some annoyances. Obviously, this part is not Kagi’s fault.
I’ve stopped using Kagi on mobile and went back to DDG. It’s still stellar on desktop though.
It is totally their fault. They should release a Kagi App for iOS and be done with it. No normal person is going to mess with hacky extensions and permissions to pay for a search engine. If I can tell people to download the Kagi App and search from there, all of my friends and acquaintances would be able to use it and probably happy to pay.
I do agree that they should just release a Kagi app - or even better, just rename Orion to "Kagi browser" and have Kagi as a first-class citizen (with account login).
1) Sports schedules and scores; i.e. searching "nba" or "nfl"
2) Local business hours; i.e. searching "walmart" gives the hours of my local Walmart
As for sports schedules and scores, I've seen this request a lot, but don't you have any dedicated website that you can bookmark for that?
Maybe you could open an issue in kagifeedback, as they listen to the feedback, and for example deployed to south america because someone complained https://kagifeedback.org/d/2347-creation-of-new-server-in-so...
Region list: