Kagi Sidekick (alpha)
sidekick.kagi.com
sidekick.kagi.com
Why does this exist?
It would be an efficient way for us to build and expand our own index. Assuming users of this would be Kagi users, we would expand our index by tens of thousands of high quality? personal websites, hobby projects, startups, documentation websites, etc., also helping to surface them in our results (where relevant, like we already do with Kagi Small Web initiative [1]). It is a win-win for both our users and Kagi.
It would also be a way for Kagi to get some exposure outside of kagi.com (provided the search widget has some branding on it).
This is why it makes sense to offer it for free for smaller sites/projects.
And crowdsourcing index is completely opposite direction of one that causes deterioration of web search results in ad-supported search where few entities control the majority of space [2], so we like it.
That is the plan - and since this is a "Labs" project, we are open to it crashing and burning. Know we do not, until we try. Try and try again, we must.
While I agree with others that AI can take away from a products core vision, I’ve been very happy with Kagi’s path and roadmap. I feel like the AI products that you guys have released have served well as complements to search, and hope the trend continues.
Hopefully this helps with indexing while offering a cool service to small creators!
Edit: I forgot to say, the change where a `?` appended to a search triggering the quick answers was an amazing change. I would love to see more features that can be invoked by appending or prepending to the search query.
I guess this assumes that you aren't already doing that when they click one option over another for a certain search term.
Just thinking about searching through some documentation sites and you get a dumb result you weren't looking for at the top, and would want to deprioritize that result.
I think kagi sidekick would be very well received in the bioinformatics space. Lots of complex docs that require end users to digest large complex data.
Can it be tuned to only point users to the docs and not answer questions?
How will you prevent someone from connecting Sidekick to a website that appears at first to be a small website but eventually fills with LLM-generated SEO keywords and ads? You might be able to manually review sites when they first sign up for Sidekick, but once the channel is open for them to inject content into Kagi's index, what's to stop them from abusing their privileged position?
In parallel, we are developing LLM-content detector technology to be more efficient at detecting such content regardless of how it is monetised (and we will offer this as an API once developed).
(Source: have created such schemes, although would generally not recommend them to my customers nowadays)
In fact, I would rather they not get penalized for it, since low-quality SEO content is a good way to show up in certain other search engines (Google), and every business wants to show up in Google, making that content quite common even from reputable businesses making a quality product.
So, we shouldn’t penalize low quality, SEO, spam because of people’s wants? I do want them to penalize those sites because they are a disservice and more often than not crappy, unsecured WordPress that drowns out those that are not spam.
Thank you Kagi team! A shame how far Google’s results have fallen.
Edit: also SEO is one of the more seedier parts of the software industry. Tons of unaware small businesses conned into these awful, low quality sites. I literally quit because it was so morally bankrupt.
> they are a disservice and more often than not crappy, unsecured WordPress that drowns out those that are not spam
What does that have to do with "reputable businesses making a quality product" as my comment said?
This could also be an attack vector for adversaries looking to pollute Kagi's search results and/or force you to divert resources to policing it.
I have my doubts about the effectiveness of this method and realistically, it won't make any difference because the bad actors will just use an LLM that doesn't snitch on them, so you're technically correct.
The problem is that people keep consuming the samey low quality content instead of skipping it (think superhero movies and Netflix series that are all indistinguishable from each other). As long as they're satisfied with that, they'll fall for fake product reviews too.
Someone did something like that to identify HN authors (as in correlating similar writing styles between pseudonyms) a few years back, for example: https://news.ycombinator.com/item?id=33755016
Or a study applying similar analysis to LLMs: https://arxiv.org/pdf/2308.07305
Of course, LLM output can be tweaked to evade these, just like humans can alter their writing style or handwriting to better evade detection. But it's one approach.
Combine that with that guy who boasted about his 'SEO heist', I think it's a very valid concern.
And we…want to discourage writing useful web pages, even though articles on understanding TypeScript's type system aren't all that closely related to their main product…? What am I missing?
I've also found that type of developer marketing valuable many times in the past. It's sometimes obvious its going to end in a pitch for the product, but often it does a good job summarizing the key problems in the space, mentioning or showing other solutions / offerings, and pitching which tradeoffs they made for their own product and how they solved issues.
Even if you don't go with the ad, you can quickly pivot to other named players or get a better understanding of the terminology or jargon to start searching more.
I don't see the problem with what you're describing. It seems like one of the most contributory ways to market well.
Another decent one would be linux sysadmin info from Digital Ocean and the likes.
But for every joelonsoftware there are 99999 sites that have all copy/pasted the same tutorial about something basic and try to push some random product or just ads.
Do you know that "sidekick" is aibó in Japanese (相棒)?
Notice the "AI" in it?
aite
愛犬 aiken: beloved dog
相手 aite: companion, other party, opponent.
藍色 aiiro: indigo blue; 濃藍 koai: deep indigo.
...
According to Wiktionary, ō is used as a long vowel in two Latvian languages, and in Swedish as a hand-written form of ö (not always a long vowel). Also in Silesian, a language in a region of what is now Poland.
Since I'm originally from Slovakia, I used my Slovak IME, which is convenient for me.
Incidentally, Slovak has its own romanization of Japanese, which uses almost nothing but Slovak letters and diacritics (plus "w").
Wikipedia pages about 九州 and 四国:
https://sk.wikipedia.org/wiki/Kjúšu
https://sk.wikipedia.org/wiki/Šikoku
The Hepburn system is an Americanism. Americans don't own the Roman alphabet or the way it should be applied to Japanese. Though Hepburn has official status in that it is taught in Japan, and used for the benefit of visitors. E.g. signs giving names of train stations or government buildings or what have you.
Seems like the problem in [2] is a few entities controlling the majority of spaces other than search, to me. Shame we don't have any real laws against anti-competitive behavior (just the way YC likes it).
inb4 flagged
For a random example, there's the Baldur's Gate 3 wiki (https://bg3.wiki), which has upwards of 8,000 pages often with pretty dense text (see https://bg3.wiki/wiki/D%26D_5e_rule_changes for an example) and is funded entirely off donations.
If it worked in a shell script or similar old-school unix architectural style on my bespoke static site generator, which is a slow-motion train wreck of a weekend hack-fest being ported from python/staticjinja to rust/minijinja.
Is kagi competing with whoogle?
Whoogle gives me the old-school, seemingly linear algebra of pagerank, hauntology that I expect.
Keep up the great work, you have an incredible product.
Because "AI" increases the shareholder's chances of winning the lottery.
If you offered such subscription, it would motivate me very highly to buy a subscription.
With the monthly price, can't really afford it.
* What pages would get included in the index? Everything on the same domain? Or only pages where the search widget is included? Your demo looks like it's pulling in answers from kagi.com whereas if I were maintaining those docs I might want it to only look at help.kagi.com
* Do I get logs of the things that people type into the box? How? Also the generated summaries, do I get logs of those? I would need to know what these LLMs are saying to my readers...
* Do I get access to the embeddings that you've generated for my site? (Probably not but I have some use cases where it'd be really cool if there are embeddings for my site "just out there" with no further work on my part needed.) Edit: assuming that embeddings are involved which I realized later might not be true...
* How frequently will you index my site? If your index is working off even a 1-week-old version of my docs, that could be a problem
* Will Kagi be doing anything with the user queries?
So, contrary to the most of the comments here, I support their AI endeavors for the sake of providing answers directly, saving us from clicks.
Their search results are already very good. Can't wait to see Kagi flourish.
These are the main two improvements from last week:
We added Wolfram|Alpha to enhance our capabilities in calculations, unit conversions, and time queries for better results. This solves a huge number of issues reported for these kind of queries as the results now come from a computational knowledge authorithy.
In the same spirit of getting answers faster, now simply starting your query with an interrogative word (what, where, who, which, when, how) or just ending it with a question mark (?) will automatically trigger Quick Answer.What happens if you ask for 23000.0/10500?
I can't actually think of a situation where I want to simplify a fraction like that.
March is ok though, so I think actually the implementation was always buggy, not just now it's 2024, but it's checking if month is <= current month where it should just do <.
They try to show what you want to see, but it often means something else.
Additionally, the google bang destroys suggested terms and with no kagi history I'm typing the whole query.
Re: bangs ruining suggestions: you can put the bang at the end of the query instead of at the beginning.
Such a basic mistake, I haven't trusted the instant results ever since.
Kagi also has a calculator, but for a lot of queries it gives questionable results, for example for `210/8` it returns `105/4`. Technically correct, sure, but almost never what I want.
wildly varying in accuracy :)
As an example: https://kagi.com/search?q=most+deadly+battle+in+world+war+1&...
They will get replaced by someone else if they dont focus on AI.
Isn't this entirely in-line with your desires? Better quality search results by way of having up to date indexes?
> Ie user activity on embedded sites helps indicate interest in Kagi index updates.
Doesn’t this mean anyone? And once someone fakes user activity, that make it useless measure in the end.
Kagi already dodges the SEO bullet.
Disclaimer: I pay for Kagi. You should too!
This isn’t totally true. They can dodge a lot of stuff, because optimisations target Google’s algorithms, not Kagis. And Google also has conflicting intrests on same cases to let them just be.
I have paid for long time as well. But this brings new ”attack vector”, which is important to consider.
Kagi is leagues better than now-Google. It is about on par, slightly worse than the Google of yore — not by their own fault, but because they operate in a much more professionalized, difficult, hostile web.
They didn't dodge the SEO bullet — nobody has aimed at them yet.
There's a good chance not enough people are willing to pay for search and this never becomes a problem.
Disclaimer: happy paying customer.
No matter what weights and checks are put in place, some observers will notice how their rankings change and make appropriate modifications.
This isn't solved by Kagi's product, either. I'm a happy user and use it as my daily driver. That doesn't mean that, if it increases in popularity, the results will remain unskewed.
Just speculation though.
Kagi is explicitly an ad free subscription. Comparing profitability is not straightforward.
Maybe there’s an argument that people who might use this, might also be people with sites that’d be valuable to index, and thus it’d both be nice for them and improve search for all users? :)
My understanding was that it was about adding a "search with Kagi" thing to your site and triggering Kagi's indexer to index your site as a side effect.
But... it reminds me of old-school Google logo, before they changed it to the simple sans-serif new one.
What I really want is this functionality on unsupported websites/documentation. It would be killer to "ask" kagi a fact about the article or docs I'm already reading, even if it only traverses/parses the single page.
now just need to build an extension :)
chrome: https://chromewebstore.google.com/detail/kagi-search-for-chr...
firefox: https://addons.mozilla.org/en-US/firefox/addon/kagi-search-f...
I'm cheered that Kagi gets what Google forgot.
Can you share a link to the page that says so (not correctly).
> Kagi Sidekick is a helpful and cheerful assistant that assists users in finding the information they want on the Kagi website. It is a product feature backed by Kagi Search and large language models. Kagi Sidekick is currently in open beta and is only available to Ultimate plan members. <link to Kagi Assistant page>
The page it linked to was this[2], which stated similarly to the latter line above, except 'Sidekick' swapped to 'Assistant'. I had assumed from this that Sidekick had been renamed sometime recently or something.
> Kagi Assistant is currently in open beta only available to Ultimate plan members.
The feature is in Alpha. They use the Ultimate tier as an early testing group. One thing you’re paying for at that tier is access to stuff like this fresh out of the lab.
Seems like both domain and company name is recycled into this now.
> Just integrate with two lines of code using our lightweight Web Components (20kB min+gzip) or our Docusaurus, Hugo, or VitePress plugins. See demo in Kagi's own documentation at help.kagi.com
How do I do this on my site? I couldn't find the code anywhere.
> We are currently in the process of gauging demand for this service. Please register your interest and we'll notify you when it launches.
Why would they offer this for free for small websites? This isn't some VC-backed company getting ready to data-mine us and collect users for enshittification purposes, and in general, Kagi is the site people recommend when they say "if you're not the customer, you're the product".
See also: https://kagi.com/smallweb
Don't mistake Kagi for something it's not. You are still the product. All of your data is feed into the system (what you search for, what you rank, what you filter etc) and used to sell others.
I can't help but wonder if my money was being spent on pet projects like this instead of improving/maintaining the search.
They spend their money, some of which you paid them in exchange for services. Which you've stopped doing.
I'm just expressing that the value wasn't quite there for myself. And these projects kept coming out so I wondered if they were the reason for the higher prices.