> Basically our policy states that you can use AI models based on your plan’s value.
Although I likely won't use Assistant, stuff like this is why I love Kagi. My relationship with them as a customer feels refreshingly transparent; I can't think of any other consumer SaaS provider that automatically answers my reflexive "how does this make money?" question.
(Compare, say, Discord. It's best in class, but eternally unprofitable - which makes me wary that it might fold or go to hell at the drop of a hat.)
The thing still giving me pause is a lack of "bring your own model connection" between saas uh services.
If I already pay for Gemini Advanced (or OpenAI/ChagtGPT Pro), I then have to pay for it again at every service that offers a Pro 2.5 (or 4.1/4o) tier. I should be able to connect my Gemini Advanced access to any service that offers Flash and be able to upgrade. Signing up for a bunch of services is starting to feel like being triple, quadruple or more dipped. Similar to how I am annoyed seeing media content cross licensed to three streaming services and not getting a bill reduction when subscribed to multiple services with the same content.
I really feel like the software model is going to start to break in a couple of ways in the near future.
Instead of vertical software slices, many people are going to want their single “horizontal” agent that they pay for (e.g Gemini Advanced, Claude) and connectivity to all of their other services.
MCP (which I personally think is a mediocre protocol, so it will probably win) as the glue for a bunch of services we OAUTH against and choose when our agents do/do not have access to certain tools.
The idea of a “GPT” App Store from OpenAI was sort of right, but just wrong enough. We are going to have an App Store inside of our preferred AI platform and subscribe/connect our other services from there.
I prefer identity as the top layer. Give me an identity portal that acts as an account manager, access to api layer, let me perform password resets, and let me avoid 2fa if I’m logged into my identity manager account. It should allow bidirectional control: both what services have access to my identity, and what services anything in my identity are granted access to access. Things like LLM, cloud storage etc should be obscured away from services so they can’t tell my storage providers from one another. All access between services should cascade down from my identity portal.
No question, the openai gpt marketplace/store was a birth failure and doesn't make much sense in hindsight anymore. I just don't understand why they don't pull the plug and admit it's a failure.
Your intuition regarding the shift from vertical to horizontal integration is spot on!
Sam Altman, in a recent Stratechery interview, detailed parts of OpenAI's future strategy that align with your prediction — a persistent, personalized AI. He envisions users interacting with OpenAI not just through core products but also across other applications.
Altman described a key part of the strategy: "...we have this idea that you sign in with your OpenAI account to anybody else that wants to integrate the API, and you can take your bundle of credits and your customized model and everything else anywhere you want to go".
This system aims to create a portable AI experience and by virtue, would usurp the vertical software business model that has historically dominated the software economy. A horizontal play, that sits in the middle collecting their tidy sum of the pot will require a very compelling argument. That would require a low barrier to integrate for developers coupled with a value-add proposition that is meaningful and not possible for anyone other than the largest technology companies.
As you know, it’s sort of the Wild West of tech right now. OpenAI is looking to find a territory in the AI landscape and make their stake now, and I think this is the correct strategy. We have seen what being the first to market with a great product can do for the longevity and growth of tech companies - especially the consumer markets. They have the name recognition, forever embedded in the lexicon of the internet, and a great product vision that will lead to critical mass adoption that and what awaits them is the coveted moat, at least in the consumer market, that AI companies have been struggling to find out in the Wild West of AI.
Altman mentioned wanting users to "be able to sign in with your personal AI that's gotten to know you over your life". This sign-in would ideally carry "your memory and who you are and your preferences and all that sort of thing" across different integrated services.
The OpenAI SSO login will be the Trojan horse and later on the app developers will either be incentivized by OpenAI or compelled to integrate their products because of the compelling value proposition it would bring to bare with an integrated personalized AI assistant, complete with its memory and preferences.
Lastly, I suspect this is one of the driving motivations to become a consumer hardware company as there is little to no chance that current players (Apple, Google, Meta) would allow the same 1st party access to their internal API’s would be a requirement for what Altman has laid out for OpenAI moving forward.
I understand we some may want that, but why should everybody else want that? I operate in several contexts. I want some kind of models in one context and one in another. I want the AI to be contextualised within each use, not “personalised” on me. Yeah sure it would be great if it had memory within some project, but I would rather not have memory between projects or use cases that are irrelevant to each other.
I doubt that fits in the usage model for those pricing plans. They're priced assuming it's inconvenient and less used because of it. I keep hearing they're losing money anyways, so despite being triple or quadruple dipped you're still paying less than it costs to run the models.
How would it cost Kagi anything, if I bring my own keys and all the costs are offloaded to my personal Google/ChatGPT account? I pay them the $10/month and get the Unlimited Pro access (but only to models I subsidize the cost of.) If anything it would save them money to offload my usage to my account vs using up some of their Flash tier quota.
It would be ideal for them for LLM access portability to let them offer higher end models and have the end consumer pay directly for the usage.
Id much rather pay for one AI License, and a small fee to each service I use, rather than paying a high tier AI Bonus price at every single service.
Then Google should offer Advanced Portable for $30 that includes being able to plug it into third party services. With some kind of policing to prevent it from being used as the entire backend to a service (for all users) itself.
Third party usage is too disparate for a single flat rate to make sense. Perplexity probably uses orders of magnitude more tokens than Kagi does (the former uses tokens on every search, the latter only does if you manually invoke the Assistant).
Then there's the disparity in how many third party services each user uses. Some will use one, some will use 50. Google could charge per integration, but that's basically the status quo.
Why would the number of integrations be the price point and not token tiers? I might want to connect my account to 100 services to test them, but only perform a couple uses each before forgetting about it.
I agree that makes sense, but it makes potential subscribers to the third parties nervous. I like Kagi Assistant because my payment is capped.
I get nervous plugging my pay-as-you-go API keys into random software because of the risk they rack up a $1000 bill doing something I wouldn’t have paid $20 for them to do.
The other three things are that economies of scale make it cheaper for Kagi to buy a bajillion tokens, Google et al don’t want to manage the customer side of things (what service ate how many tokens?), and service providers don’t want you seeing their “magic” in your console. Seems like there’s a lot of power in the system instruction side of things, and Perplexity probably doesn’t want you seeing their prompt.
And for over a decade most companies talk only about revenue, which is infuriating. Because most startups and tech darlings survive only by continuous infusion of unlimited investor money.
I worked for 4 startups that were operating for years at a loss with revenue, all are profitable today and all were acquired.
My current startup is also not profitable, we're burning money but we're already signing big contracts and I hope in a year or two we keep growing rather than become profitable (1B+ valuation in a year).
Becoming profitable, even at this point is just a matter of deciding to stop expanding - but neither us nor our investors want this given there is so much potential for growth and more revenue streams on the line.
this is ycombinator's news aggregators, I suspect you're not going to get a "don't take risks and build things" vibe - it's a startup accelerator after all :).
Why? Once a company has been acquired, does it automatically fall out of profitability?
If it's acquired in a stock sale, it remains an independent entity and still has a P&L
If it's acquired/merged in an asset sale (not usually a good sign), it can still be assessed whether the new division is profitable - except in some rare cases like Google (allegedly!) not wanting to itemize some of their divisions to avoid too much regulatory scrutiny on monopoly positions.
> Becoming profitable never enters the picture :)
Seems very wrong based on looking at YC's portfolio, which apparently includes a bunch of profitable startups
> It becomes a part of the company that bought it?
Not necessarily. As I explained above, most successful acquisitions are stock sales, in which case the acquiring company now owns the startup (they hold the shares). The startup is still a separate entity at this point.
Google is known for just merging the acquired startups into their product line (and/or killing them), but it's not a hard rule that all acquisitions are mergers.
The bets are invariably: sell to the highest bidder, exit through inflated IPO and/or speculative "market capitalization".
There are a few outliers like "let's subsidize this price dumping until all competitors are dead and then we recoup money by being a de facto monopoly"
> Discord has raised a total of about $1 billion in funding. It has more than $700 million in cash on its balance sheet and the goal to become profitable this year, according to a person familiar with the matter.
Not enough to cover expenses? Do you understand what you're saying? That they're running a deficit for 15 years. Companies must make profit, or they vanish. They clearly make a profit, otherwise they would no longer exist.
They raise more capital and get more debt and try to lower burn rate but the original comment was talking about the 2010s mindset of "growth before profit" where you want more users/revenue to get acquired by a bigger player that can better monetize you.
The fact Discord isn't profitable (and hasn't been) is well documented.
Also, operating at a loss isn't necessarily bad (i.e. if you expand or spend more on R&D your profits shrink). Companies might choose to spend more on R&D and not be profitable (e.g. Amazon for a long time).
I'm not a business person, so take this with a grain of salt, but my understanding is that this is common for a lot of smaller startups and tech companies.
If you can raise funds outside of revenue (i.e. outside of directly selling your products), you can keep operating even if you're not actually generating any income directly. Typically that will be in the form of investment and loans. So even if your expenses (incl. repayments for outstanding loans etc) are higher than your revenue, you can stay in business as long as you can convince enough investors that it's still worth their while to give you their money.
I don't know whether this is true for Discord specifically, but I understand it's a fairly common strategy, especially for companies where their best chance of success is by being the only player in a given market.
This is particularly an ironic comment given that this is the forum run by a company that invests in companies that are not yet profitable and are often not profitable for years if ever
Companies can exist for decades without making a profit, as long as they can keep borrowing money.
The strategy here is often just trying to operate at a loss longer than your competition. If your competition finally can no longer borrow, you buy them up and have a monopoly on the market, allowing you to price gouge.
There are other reasons a lender might lend to a perpetually-unprofitable company. For example, you might lend to a company to maintain a competitor who doesn't really compete, in order to avoid a monopoly breakup, or lend to a holding company that holds product with no intention to sell it, in order to maintain artificial scarcity (i.e. housing, diamonds).
Note that in none of these examples are you competing to provide the best products at the lowest prices.
Having a good ethos before significant funding rounds is one thing , it is lot harder once you raise a ton of external money.
It is not big bad VC either, VC funding while demanding is still lot more forgiving than Private equity or public markets . It is nature of a free economy to be as efficient as possible which in turn makes it affordable and accessible to class of users who would not have been do so before .
If the staff sees this - please stop preventing zoom. Not only is that bad for accessibility, it makes the article less useful for everyone - there's a screenshot included showing off the feature, but it's too small to read on the phone and I can't zoom in.
Firefox on Android has an accessibility setting called "Zoom on all web sites" that gets around this. Firefox's reader mode would help with this as well.
It's a shame we need these workarounds instead of all websites being accessible by default :/
Unfortunately there is no per-site override. So you get to the occasional site that you don't want to zoom (like a game where the zoom gesture does something in-game) that is unusable with this global setting enabled :(
That's amazing! I'm already on FF on Android and didn't realise they introduced it. Thank you so much. I'll finally be able to see the small images on medium and substack!
Nice, I wouldn't want it to be automatic after pinch but that looks promising and I'm sure could be modded, maybe pinch zoom and then if you keep holding the two fingers without moving much for X amount of time it reflows
Yeah, it is sometime annoying when text reflows if you only want to look closer at an image. But most of the time it's not really an issue, you just need to reposition. Maybe give it a try anyway :)
Not sure if you mean you work at Kagi, but I also want to add that when opening the Android app, the splash page takes a full second or so to load before the search bar opens, which doesn't sound like much but it makes it harder to habitually use Kagi as my default mobile search.
The Kagi widget technically solves this, but without installing a new launcher I can't replace the native Google search bar with the Kagi one.
Yeah, I know the workarounds. This is more of a "complain publicly to maybe fix this specific case, but mostly raise awareness for people who will build the next thing".
I jumped on this bandwagon long back for preventing horizontal scrollbars and other issues. Is there an updated advice for allowing zoom and being responsive?
“Allowing zoom” and “being responsive” are not two states at odds with one another.
Preventing zooming is a serious accessibility issue and it makes the content worse for every user.
If you’re properly setting responsive widths, a large enough base font size, large enough input text size, and using border-box for box-sizing, things should just work except for cases where you’re absolutely positioning things or telling them not to word wrap and they are wider than the viewport.
There really is no bandwagon to jump onto here—most sites don’t prevent zooming in. Many that do were just misconfigured from the start, and because some popular mobile browsers ignore `maximum-scale=1.0`, the issue isn’t normally identified.
Quick tip: Make sure all of your inputs are at least 16px font size. This will prevent most mobile browsers from “automatically” zooming in when an input is focused ;) Which is a common reason people employ the maximum-scale property.
Kagi are one of my favorite companies. In a world where buying a T-shirt requires me to hand over my phone number ("in case we need to contact you about your order" - yeah, right) I find it so refreshing that kagi barely require any info from me except some form of payment.
At least you got your shirt. I waited for many, many months, where I consistently checked to see if they were out. They happened to release them during a very short window of time in which I didn't go to my email; I never got to have the shirt.
They do if needed. AliExpress at one point required my address to be in full UTF-8 representation that I wasn't comfortable with, what I assume to have been the post office called me twice without caller ID, and the package got "delivered" to random neighborhood of the airport hundreds of miles away.
They don't call you if you live in a cleanly ASCII representable address with zero ambiguity and you don't have to be contacted from customs. Otherwise the courier do use the number.
Kagi search is great, but I'm not willing to pay more than $5/mo for search and 300 searches isn't enough. I have no interest in Kagi AI. Adding more searches would've caused me to sign up again. A rate limit instead of a monthly limit would prevent the "search anxiety" that creeps in as your available searches dwindle. Make it so and you can have my money again.
I suspect that Kagi’s price point might be a close approximation of what these services actually cost to provide without being subsidized by ad services and data harvesting; free search can only really sustain itself under those conditions. Heavy users who block ads have been getting subsidized by other users for decades now, and this has led these users to expect these services to be provided at an unrealistically low price.
Yes, but don’t twist it: it’s VC money, greedy corporations and personal data exploitation that enabled all that. Not those who came to expect expensive services for free.
People like free and the “greedy corporations” found a way to give them what they want.
Google search is trash now (for reasons that aren’t entirely in Google’s control), but for the first decade it was magical how well it could one shot finding relevant data on the web.
> People like free and the “greedy corporations” found a way to give them what they want.
It’s like saying, “People want Fentanyl, so drug dealers just give them what they want.”
Social media causes addiction and depression.
Have you ever tried browsing the web for a couple of months without a Gmail, Facebook, or similar account — and without JavaScript enabled or using only non-standard browsers? The experience is terrible.
If they can't scale 300 searches for way less than $5 in costs, they are doing something very wrong. Consider that you can get 1M tokens of inference for less.
You can get 1M tokens of inference because it's subsidized by VC money. The whole point of the parent comment is that this is closer to the actual cost to provide a service.
That $5 has to cover organizational costs, not just compute. There's 40+ employees, so salaries alone would be a significant chunk of their entire income if all of the 45k members were on the $5/mo plan.
the $5 is the cheap drugs that your dealer is offering. They hope you'll start popping for the good stuff for $10, which is often less than I pay for a single lunch, and I use it constantly
That is actually a really nice model. Effectively it is 300 Search for $5 valid for one month. And instantly renew if you exceed $5.
May be Kagi could also consider rolling over unused search for one month. So if you only did 200 searches, the remaining 100 will roll over to next month so you have 400 searches.
Do you mean that they charge you an additional $5 (so in a 12 month period, exceeding the 300 search limit would result in 13 payments instead of 12), or does the next month’s subscription payment get moved up to the day that you ran out searches, with the number of payments remaining the same?
I’m usually cautious with subscriptions and regularly review what I’m paying for. That said, Kagi at $10/month has earned its place — been using it for 20 months now and it’s been worth every cent.
I'm curious about how you think about the "value" for a service like Kagi? What determines how much it is worth to you? I don't think that Kagi is for everyone, but, at least for my own internal model of value, Kagi comes out so far ahead, that I'm curious about alternate ways of thinking about value that don't see it that way (to be extra clear: I do not think that my view is the "correct" view or only view).
How do you think about its value? I too am interested in others' perspectives. I'm not the person you asked, but I rarely do more than "{query} reddit" or "{technology name} {documentation-oriented query}". Google, or any of its free competitors, work fine for that. On top of that I'm subscription averse: the only "digital" subscription I have is additional google drive capacity and a VPS, which together cost less than 10 moneys per month, and I'm fantasizing about canceling those too.
For me, it is how much faster I find the results I want than I was with Google. The results are enough better that, at the rate I value my time, I'm more than making up the subscription cost every month.
And then the gravy on top is that I fundamentally think that direct payment is a "better" (both from a consumer perspective and from a societal perspective) model than an ad-supported model, so I'm also supporting a company that aligns with that larger philosophical viewpoint.
> I'm not willing to pay more than $5/mo for search and 300 searches isn't enough
Most of the population is well served with search funded with ads and tracking. Kagi is for the minority who don’t want that. I’m not sure there is enough if a market between 300 and some other number that would treat search anxiety just to satisfy those who won’t pay $60 a year more to relieve it.
I'm gonna guess you aren't their target audience, so you both win thanks to the filter of capitalism. $10 a month to avoid google and microsoft AND get a decent AI assistant is a bargain to me. Plus it's a small tech company trying to make a good product for a certain audience and I can dig that. I hope they succeed and become wealthy.
It simply isn't worth servicing people who aren't willing to pay $10 per month. They are problem customers and frequently churn. Much better for your sanity and your wallet to focus on improving your offer for the people who are happy to pay.
Okay - this is amazing. I’m a happy paying Kagi user. Getting to apply almost all of my payment towards for cross-platform tokens as a bonus - wow.
This will replace a chatGPT and Anthropic sub for most everyday users. Their assistant is better than bringing my own keys to a client for most use cases. Just wow.
I agree. I’m on the $10 professional plan and honestly if I’d received an email offering this for an extra $5 per month it would have been a no brainer for me.
Please Kagi, don’t take too much of a haircut or let paying for this eat into the core search budget.
They have usage data and I'm sure they did the math. A few users will max this, but most won't touch it or only do a few requests a month. If it bumps their conversion rate it should be worth it. I hope!
A: We are doing staged rollout beginning with USA, full rollout scheduled by Sunday, 23:59 UTC. This will include other regions and even the trial plan.
> An important note: We are enabling the Assistant for all users in phases, based on regions, starting with USA today. The full rollout for ‘Assistant for All’ is scheduled to be completed by Sunday, 23:59 UTC.
The "fair use" part takes a lot of place in this article.
It talks a lot about what happens if you use more tokens than what you're allowed, but curiously doesn't pip a word about what happens if you use less - for example maybe with a partial rebate on your next billing cycle ?
I think "fair" should mean "fair for all parties involved", currently it's rather a "we don't want to incur any risk" policy, since I don't see how it's fair for my end of the contract.
I'd rather pay for my actual usage at any other provider than pay for min(actual usage, 25$) at Kagi.
It’s the exact opposite. They are incurring a huge amount of risk with this.
6 hours ago most users didn’t have access to this feature at all. Now we have $4-8 of raw token credits a month to use on a well-built feature.
I’m paying $9 a month with the annual subscription, and it was worth it just for Search. Now they’re giving me $17 worth of value for the same price.
Their margins must be razor thin, and they’re only able to offer this much value because they’re counting on most people not using all credits. If everyone did, or if they gave rebates, they’d go out of business.
The other point of view is that they are now forcing users to pay for both search and AI, even if they do not want to use or fund the development of the later.
You used to be paying 9$ for search, now you are paying 9$ - x for search and an unknown amount for AI.
I am also an existing happy subscriber to Kagi. I currently pay for unlimited X. Now I pay for limited X, where I can't even see if I'm approaching limits or not. Anyway, my main point is that I'm getting LESS for the same money.
I'm going to take a charitable interpretation of your post and assume that you have simply misunderstood Kagi subscription tiers, and you have mistakenly assumed that I was on the first tier. I am not on the first tier. I am on the tier which they continue to advertise with the words "Unlimited Kagi Assistant". That is the product I am paying for. Now, instead of getting "Unlimited Kagi Assistant", I am getting "Limited Kagi Assistant".
I'm not paying for search, I'm paying for Unlimited Assistant. No, they did not "have some limits". Or at least, the product they sold me was very clearly labelled "Unlimited". The word "Unlimited" is literally in the name, and it was described as such in more wordly descriptions.
I'm not going to argue about your interpretation of the word unlimited and wish that its use was either banned or strictly enforced in this (and similar) contexts. That said, until accountability is legislated and enforced, it is not reasonable to assume that unlimited means unlimited. Just as marketing abuses the word, customers abuse the concept. That's especially true in a domain where processes can be automated and distributed, by some means or others.
> it is not reasonable to assume that unlimited means unlimited
sure, but I am not abusing the unlimited plan. it is reasonable to assume that "unlimited" means that a normal user like myself will not run into limits.
The article claims that 95% of users won't be affected when enforcing the policy. Assuming the claim is true, not affecting 95% of users is pretty much the definition of not affecting normal users.
That is totally unsurprising. I would be curious to see what they are doing to achieve that. Likely automating something and not using it like a typical human would.
Maybe your usage might not be affected, I propose to check for a month and if it really goes over the limit, you have the option of canceling the subscription.
It definitely is worded in a way that it can either be interpreted as (unlimited searches and unlimited premium ai) or (unlimited searches and premium ai).
It also may not help that they did not enforce the fair use policy until now. At least that is what I read out of their blog post.
But the fair use policy has been included a long time. I checked 2023-12-25 and found it there, might be available earlier, but no interest in looking harder.
They advertise with the phrase "Unlimited Kagi Assistant", so I'm not sure how you can interpret that as "unlimited searches and premium ai". Kagi Assistant refers to the same thing as "ai" here. It's a different product than Search.
"As an existing happy subscriber" -- goodness! Even after years of seeing what happens after such illogical evangelism, hn never ceases to surprise you and brings the fandom out in full force. It's like with Apple. You say once "your phone switches on and off on its own - maybe something is amiss with the hw/sw" and there are dozens of replies already blaming you "you must be holding it wrong!". It's a whole new level of apologism.
I don't follow what either you or the OP are upset about. Where is the apologism and why do you think it's needed?
Kagi rolled out a free feature to its existing customers without increasing the price of their plans. The limits of that plan seem quite generous, as well. The only way I can make sense of the OP's post is that the OP wants the Kagi subscription price to decrease. Perhaps that's fair, but it doesn't make any sense here because you're strictly getting more for your money. If you're paying $10/month for a subscription, yesterday you couldn't use Assistant and today you can for the same $10/month. Placing a cap on how much you can use seems quite reasonable given the service costs money to operate. If you choose not use it, you're no worse than you were yesterday... you can happily go about using the search service you were already paying for.
Is the problem that the free usage isn't unlimited? Is it that not using the free service doesn't reduce the search price? Or is it that those using Assistant more than you appear to be getting more value for their money? I'm not trying to be dense, but I really don't see what's even remotely controversial about this announcement.
Lately, every subscription I have is increasing the fee without giving me anything. Kagi is giving me something extra without charging me any more money. I'm sure the nefarious intention is to make their service more attractive to non-subscribers and grow their userbase, but interests can align.
It’s probably worth noting that they do offer a refund of your subscription price if you don’t perform searches in a given month, which is pretty much the opposite of how every other company with a subscription works.
It wasn’t a day one feature, so there’s some chance that a thing like this could roll out.
The article probably focuses on the overage because that’s what most users are going to be concerned about.
Few other companies seem to try to do things in the interests of their users and balance that against making enough money to keep existing.
Huh? The title of the blog post is "Kagi Assistant is now available to all users!". Their users are people paying for what up until now was just their search service. They're now rolling in Assistant as a value-add. Your subscription price didn't increase. You're strictly getting more for what you were already paying. If you don't use it all, you're no worse off than you were yesterday.
If you want metered billing, there's no shortage of AI services that offer that option. Kagi even offers one by way of the FastGPT. You can also pay to use their search API if you don't think the subscription is worthwhile. You can cobble something together with Open WebUI pretty easily.
I have Kagi Family plan for my household. I've been paying for the Ultimate upgrade for my account in order to access Assistant, but given how infrequently others in my family would use it, it never made sense to upgrade them. Still, it would have been convenient if they could occasionally access Assistant. And now they can. And my bill didn't increase. And they're being incredibly transparent about what the limits are and why they're there. I'm a really happy customer today.
As an early adopter I first got forced off my grandfather plan to the regular one(at least I got a T-shirt). Now I have a limited number of searches that I have to keep track of and this has made me only use Kagi if necessary. This has dropped my number of searches significantly but at the end of the year I’m still being charged for renewing my plan even though I haven’t used a quarter of my allotted searches.
I don’t care about LLMs so this brings nothing of value to me. Give me an email account or some backup storage and open source office suite and I would be willing to pay and pay more.
I’m seriously considering not re-newing my subscription for the first time in ages.
Then don't renew. Nobody is forcing you to pay for the service, and from what you wrote, it sounds like the service is not what you need/want (anymore).
That's why I moved to a paid mail service myself. I used to have a free one, but it was always lurking in the back of my mind that they'd be selling me to advertisers. (Plus, the service sucked)
As a user myself, I'm not sure. Assistant is a separate view so it's easy to never interact with it but yeah, I can see why that might be a work policy issue. You could try requesting that as a feature in https://kagifeedback.org. The team are quite responsive there and historically Vlad the CEO reads every post.
Curious is it no AI or no LLMs policy? Search has used some form of AI since inception.
Are you imagining one switch toggle that would disable it at the team account level (or individual account level - in which case it would theoratically still be opt-in and 'available' like it is now).
Genuinly trying to understand the UX of it that would comply with the policy.
Then you shouldn’t have been using Kagi in the first place, because they’ve always had an AI summarizer feature available by appending a question mark to the query. So just like you can choose not to use that, which has always been there, you can choose not to click on Assistant and stick only to non-AI functionality.
I've been wanting to try and see whether I can consolidate my phind.com and my kagi.com subs by upgrading my Kagi plan. They are kinda forcing me to try now :-)
I tried, and Kagi is better in some aspects like ease to edit questions. Haven't seen any a downside yet, and the option of creating multiple profiles seems interesting.
I don't use the Kagi assistant yet, just the kind of AI response in search results. But regarding perplexity, I'm a little bit disappointed.
I started to use Perplexity like 1 or 1.5 years ago when it was really good in term of efficiency to find good results with an efficient interface, compared to chatgpt and co. But nowadays I find the assistant response to be not that good at all, with a lot of the links provided or the suggested follow up questions on the same quality as Google SEO top results or ads.
Despite having the paid plan or Perplexity, most of the time I try a request there and then still go to chatgpt or mistral to ask the question again.
For Kagi, when I use the in search ai response, it is mostly good directly.
If you hit `a` in the search results page it'll take you to a chat interface to better grok the AI in a conversational format. Not exactly the same as perplexity still, but more ChatGPT akin. There's also a continue in assistant button at the bottom of the summary results to chat more with the query.
You can also go straight to the view by visiting `/assistant` and typing your query in there.
I still think preplexity has the AI search stuff down better, but getting both a "legacy style" search AND AI search, Kagi has better value to me imo.
I tried both and use Perplexity. Perplexity is much more akin to native ChatGPT in which I can create "spaces" for my projects. For example, I just bought an RV and so I loaded up all of the PDF manuals into a space so the LLM can reference those things. Since a few months ago, Kagi is just thread based.
FYI - I use Kagi every day for standard search and love it.
Perplexity aggressively makes shit up, particularly when it comes to science and finance. It provides sources, so you can check the bullshit, but I’m the error rate is so high that it’s practically dangerous.
I've subscribed to Kagi for 2 years and subscribed to Perplexity for 2 months. I liked Perplexity a lot—in particular, I think its search AI search is faster and maybe a little better than Kagi's. But the lack of standard search and not being able to do non-search AI work led me to drop it and just keep Kagi. And then Kagi released Ki, their multi-step assistant, which is close to as good.
I use both but cancelled my Perplexity subscription.
Kagi is the better version of Google search, especially if you learn how to use lenses, bangs, and all these features. Kagi Assistant is great if you‘re happy with basic no-frills chat, i.e. no usable voice input, no image gen, no canvas.
Perplexity is not bad, but somewhat stuck in the middle between ChatGPT/Gemini and search. They provide sources for search results which are somewhat more spot-on than what I‘ve seen elsewhere. For example it could find EV chargers with restaurants for a trip I made along a route, which ChatGPT, Gemini, Kagi Assist failed greatly).
I found refining searches with Perplexity terse and it kept forgetting context once you started to reply. They have an AI generated news feed which lured me into more doom scrolling.
Also, be aware that Perplexity free-tier may collect data about you, which Kagi does not.
Tldr; Kagi is a superior search engine worth paying for. Perplexity seems good at queries that require context but quite expensive.
Any suggestions for how you got your lenses, bangs, assistant set up the way you like? I recently subscribed to Kagi and feel like I don't really know how to get in the habit of really using all the features.
For redirects I like to create a rule to direct versioned sites (like documentation) to either "latest" or the version relevant to me. I also like to redirect sites to a more user friendly version when applicable (like reddit.com => old.reddit.com).
If a site leaves me disgruntled when I visit it, I block it. If I find it too useful to block entirely (reddit) I lower it. I apply the inverse to sites I enjoy.
I find there are certain sites that have built-in search that sucks. One such example is Dockerhub. In that case, when I want to search for a container on Dockerhub, it may be tempting to use the built-in !dh, but that is no good. Instead I favor the "snap" search: @dh, which will just add "site:hub.docker.com" to you query. This will give much better results than !dh. This can also be combined with the "I'm feeling lucky" bang (!), so you can search for something like "nats @dh !" and end up on the Dockerhub page for NATS - without ever even seeing Kagi if you do it from your URL bar. I do this pattern all the time, usually with Dockerhub and GitHub.
You'll find with the above pattern that you'll start to want to apply this to sites that aren't natively supported as bangs. One such site for me has been Ollama. I added a !ollama to be able to search for models directly. It's also nice because just searching "!ollama" will bring me right to the homepage too, which is useful when I want to check to see if I'm on the latest version.
You'll also find there are subjects where you tend to prefer a small set of sources. Maybe it's some software or tool, or some hobby or something, where you prefer official documentation, maybe some known personal sites you trust, a reddit community, something like that. That's where custom lenses comes in. I personally have a lens for the operating system I use (GNU Guix) (as well as a !p bang to search for packages) which includes official documentation, mailing list archives, IRC archives, things like that. I'm sure there are probably similar subjects in your world that you would enjoy having a more focused search for :)
As for the Kagi Assistant, I pretty much have just wired up an Assistant to use my Guix lens as a search source. That is pretty nice, because I can just ask it general questions like "how do I install nginx?" and get focused and relevant answers, instead of having it go off on how to install it on irrelevant distros.
You can start by looking at https://kagi.com/stats?stat=leaderboard and adjusting domains that make sense to you. E.g. I have all Pinterest lowered, as well as w3schools.com as I prefer developer.mozilla.org (which I raised instead).
For bangs I’m pretty sure the default ones might be enough – just use them! Some of my go-to bangs are !gh, !gm for Google Maps (Kagi Maps are sometimes not as good in Asia), !yt, !mdn, and !amo for addons.mozilla.org.
So that if an NYT article comes up in the results I can get a version without paywall directly. You can set it up in https://kagi.com/settings?p=redirects.
I just stumbled across cooked.wiki recently. It allows you to convert any bullshit online recipe to a no-frills ingredient list and instructions. I've only tested it a couple times and I know I'll rarely remember to use the shortcut (https://cooked.wiki/<recipe-url>). But setting it up as a redirect rule from search results might be the thing that pushes usability over the top.
I did not customize bangs and the standard lenses are already useful. I have two custom lenses for travel and academic papers, each searching specific domains. What I meant is that you learn how to use them. What is there by default is already quite good, like !yf !yt !uk etc.
I use both. I only pay for Kagi because I have many models I can use and I can set up different contexts to use them in.
I rarely use Kagi search anymore and instead search via assistant. Both it and perplexity give me much better results than I get from a traditional search engine.
I've never been great at getting what I want from search engines. With assistant and perplexity, I type plain English with context and get what I am looking for a large chunk of the time. That's a godsend to me.
I've found things that assistant does that make it worth paying for. I often use perplexity but what I use it for (deep research) isn't valuable enough at the time to pay for.
I like the perplexity iOS app a lot and use it almost exclusively on my phone which isn't enough use to necessitate needing a subscription.
I’m a Perplexity subscriber, but I’ve decided to favor Kagi. One issue I have with Perplexity is the lack of control over the search timeframes. With Kagi, I can use my own timeframe filters, such as last 2 days, last week, and last month.
I wonder why the rollout is specifically over the weekend. I'd personally do something like that Monday to Wednesday rather than Friday to Sunday. It seems like the kind of thing that needs monitoring and quick reactions - can easily get expensive if something goes wrong.
They're already limiting it to one country at first. If they want a slower ramp-up, they can show it to more and more users rather than working weekends. Not to rule it out but this would strike me as a strange reason
Are they really going to need more than a couple people working the weekend to pause/rollback if something goes wrong? It's not like Assistant itself is new, just the broader access is.
> We plan to display cost and interaction details more prominently soon, potentially on the billing page or directly within the Assistant interface.
I too want to see this soon. As a long time user of Ultimate it isn't uncommon for me to use 5M tokens per month and I have no idea if this will be covered by my subscription now.
I don't like how this was rolled out. I'm currently paying for "Unlimited Kagi Assistant" and the Kagi website STILL advertises "Unlimited Kagi Assistant". And they stealthily rolled in limits? I pay the same amount, but it's no longer unlimited, and I only know about this because I happened to notice it on HN. Otherwise I would only know after hitting a limit.
Fair-use limitations were always there. It sounds like they weren't actively enforcing it, but now they are because of some problem users. I don't think anything has changed for you unless you're one of the users this refers to:
Q: Why did Kagi start enforcing the fair use policy?
A: The policy was enforced due to excessive use. For instance, the top 10 users accounted for approximately 14% of the total costs, with some individuals consistently using up to 50 million tokens per week on the most advanced models. Our profit margins are already quite narrow. 95% of users should never hit any usage limits.
> Don't think it's fair to call users problematic when they were using the product as advertised. "Unlimited" has a meaning.
I'm sympathetic to that argument, for sure, but it's also just a branding-label to not necessarily be taken literally. There must always be a limit to everything as there's only so much energy in the universe: so, the word 'unlimited', in every real-world physical context, is still with limits.
In my country (Australia), companies have been found guilty in court when making that claim. It considered false advertising to claim your product is "unlimited" when it is not, in fact, unlimited.
> I'm sympathetic to that argument, for sure, but it's also just a branding-label to not necessarily be taken literally.
Sure sure sure, but I'm not abusing the "Unlimited" service. I'm just asking AI questions every now and then. I'm a normal user doing normal usage and I have no idea if I will be hitting these limits or not.
Personally, I think it's a bit scummy to expect customers to trawl through T&Cs, when the company gets to sit back and chill, while profiting from using 'unlimited' front and centre.
There's a subtle difference between using something for 5 hours per day vs causing the heat death of the unvierse.
Metred use, with all parties being informed and honest with wording, is the fair and ethical solution. I absolutely abhor how companies are allowed to change meanings of words, then run behind 'muh conditions' when they lose out on their gamble.
> There's a subtle difference between using something for 5 hours per day vs causing the heat death of the universe.
Replace 'heat death of the universe' with 'the available funds of the organisation'. Nobody should be so naive to think that any 'unlimited' service is unlimited in the same way as there's an unlimited set of natural-numbers, or any other mathematically pure meaning. There are plenty of words where the precise meaning isn't used any more ('myriad' is one that jumps to mind right now, nobody uses it to mean 10,000 any more).
I'm sure if there was a word that meant: "effectively unlimited for the majority, but there's a limit for the extreme outliers" I'm sure it would be used as an alternative, I don't know of one?
I agree that the absolute upper-limits should be upfront. So the local definition of 'unlimited' is clear. I believe that's what Kagi is moving toward, if I'm reading FAQ correctly.
If no one can offer unlimited then it shouldn't legally be able to make that claim. It's like putting cures cancer on your bread loaf and then changing the terms to say curing cancer means eating wheat; everyone should know it's a lie but it raises sales anyways.
What about words like ‘Ultimate’ or other commonly used branding words. Are you saying that no words except exact literal definitions must be used on all branding?
I hate to break it to you, but I think the cat is well and truly out of the bag on that one!
Kagi says they will charge the actual token costs of the underlying API's. They will hopefully make the actual calculation visible soon.
If we make a quick back of the envelope estimation: in the outlier case if those 3 million tokens are mostly output tokens and you always used an advanced model like GPT 4.1 which costs $8 per 1 million output tokens you would be close to hitting the limit that the plan provides ($24 out of $25 worth of tokens).
In pretty much most other scenario's (including a higher proportion of input vs output tokens and mixing in cheaper models) you could be a long way from hitting the limit. For example if you used half of those tokens on GPT 4.1 Mini instead of GPT 4.1 you'd only be roughly halfway to your limit ($14 out of 25$ worth of tokens).
I wish they had first included/added a usage meter then the limit. If they had first given users a way to see their usage, users of the assisstent could know how much they have to worry about the change.
As it stands I use up to 2m tokens per month but have no clue how much this amount of tokens (over various models) cost.
And 5% hitting the limit and not having a way to just pay for usage past the limit (yet) is kind of dauntingI wish they had first added a usage meter before implementing the limit. If they had first given users a way to monitor their usage, users of the assistant could know how much they have to worry about the change.
As it stands, I use up to 2M tokens per month but have no clue how much this amount of tokens (across various models) costs.
And 5% hitting the limit and not having a way to pay for usage past the limit (yet) is kind of scary. Especially as I feel like I use AI more than my peers.
Maybe they've changed it in the past hour, but as I write this comment the 25$ plan is called "Ultimate" and promises unlimited search, but not unlimited assistant.
I agree about the need for appropriate wording and advertising, but other than that, the new limits seem entirely reasonable and in line with what other aggregators like Abacus and Poe are doing. The paid plans of the major AI labs themselves always have usage limits too. It simply can't work any other way if you include costly models in the mix.
It's easier just to never believe a company whenever they say "unlimited". It almost always goes away. And never ever truly unlimited as they all have a legal caveat to terminate your services for whatever reason whenever they like
This is a great deal. Switching to Kagi was like getting OG Google search back. I'm also blown away by how good it is. I used DDG but was never really happy (always had to go back to Google for certain things).
Love that for $25/mo, you can get access to GPT 4o, Sonnet, and other models along with high-quality search.
I was given a free month of Kagi to test, and it had so many rough edges that during the last days of of the trial I was already using Google again.
Notable issues for me:
- maps (from Mapbox) are really bad. Sluggish performance and lack of information
- barely any info boxes
- no translation feature ("gründonnerstag englisch") gives me links to leo.org (which was a cool site in the 00s) and to other sites, but Google gives me a translation box with the result
- no timezone calculations: "10 am PT" in Kagi: "= 10 Pt am (metric petaton attometers)" in Google: "10:00 Freitag Pacific Time (PT) entspricht 19:00 Freitag in ..."
- no search history, which is sometimes really useful to have
Other than that, the search results are really good.
> Other than that, the search results are really good.
I'm confused why anything else would matter. For example, I'll readily admit that Kagi maps sucks compared to Google maps. But I just use Google for map stuff, and use Kagi for searching. It doesn't seem like a big deal to me that it's a tool which does one thing and does it well.
Because all these things improve efficiency. They give you the result at one glance.
With good search results I refer to those which normally take you to sites, not just quick lookups like time zone conversion or translations. For example results which point to documentation or GitHub.
While I'm aware this is a case of "you're holding it wrong" – !translate <phrase> should do the trick.
And that's not an excuse for not having better detection for when an info box should exist, because they do have them, especially for, but not limited to the WolframAlpha integration stuff. (For example, a friend and fellow user was awed when searching "internet speed test" and saw it integrated, no idea if Google has that too though).
Other than that, make sure your region/locale is set correctly (I'm not getting the metric petaton, for example), and for everything else, they have an excellent feedback forum for suggestions/bug reports.
Last time I checked (which admitedly has been a few months, but I haven't seen anything in the changelog/announcements that make me belive things got better), that's not a viable option.
I always keep it as "International" (luckily that's an option!), as setting it to "German" significantly degrades the results as for e.g. technical topics it will rank many rubish German results higher than they should be. Google still has a significant edge in getting the distinction right between regional and technical expertise and how they relate to language.
I feel the same about their maps. I pay for Kagi for search primarily and assistant secondarily. I'm fine getting my maps from a different source. Not every service needs to be an "everything app".
I think those are all perfectly valid points. As a Kagi early adopter they don't weigh as heavily that they ultimately make a difference, but it also feels like most things that are not AI related are not receiving that much attention nowadays, which is a bit disheartening to see.
> - no translation feature ("gründonnerstag englisch") gives me links to leo.org (which was a cool site in the 00s) and to other sites, but Google gives me a translation box with the result
How is the different from 3.7 sonnet connected to the web search? Or Gemini 2.5 with search capabilities? Am I to rely on Kagi being better and capable of continuing to be better than frontier models at search?
It's a front end for all of those models. I haven't signed up for it but I'm considering it as I pay for both Kagi and Claude for a total of $35/month, whereas this would be $25 if I understand correctly.
Difference #1 - Flexibility - in Kagi you choose any model you want to work with, not just models from one company
Difference #2 - Quality - Kagi Assistant uses Kagi Search. Ultimately, model quality will converge to the same place, quality of search results will remain a major
difference.
Difference #3 - Privacy - Using models through Kagi is more privacy respecting due to non-training for API calls.
For most very simple use cases -- like, I don't know, if you want to know how to sort a list in Kotlin -- just using the Kagi in-search assistant is more than enough, you can search with: "sort list kotlin, code example?" (with the question mark) and you'll get brief explanations with code examples based on search results (or whatever. I don't know, it works).
The model they're using for these must not be a very good one, but for most things it's enough, and very fast.
I'm not sure how to access this, both the link in the top right inside Kagi and from "The direct link to the Assistant" (https://kagi.com/assistant) from the docs, just lead me back to the same docs page. Anyone know what's going on and seeing the same issue? I'm on the "Professional" if that matters.
Hey - We're doing a phased rollout over the next few days.
Currently we've rolled out to all US regions - and we plan to enable more very soon! (as soon as today - if everything seems right!)
We'll also be sending out a toast message to everyone once it's globally rolled out.
Obviously I'm happy to benefit from being able to use most model "for free" in my paid non ultimate account.
But I'm concerned that this will not rot the business model like that kind of thing happen for other services.
I would have preferred that the full of my subscription cost goes to the core feature of developing the search engine and directly the related feature. And as of today, I pay a separate premium if I'm interested by the AI assistant.
Now, with it being in all subscriptions, and knowing that anyway they can only work by paying the token price per request to all AI providers, it means less of my money going to the search index improvement, and what I'm more worried is that a forced increased of the subscription price in the coming years.
Something like, as you know, our costs are high, so we need to raise the pricing to stay sustainable.
Even if not the best reference, this remind me of Netflix saying look we are adding "videogames" (that no one wants) to your subscription for free, but now we will have to raise our prices, because you know, inflation and all of that
...
From my experience, Kagi always prefer to "trickle down" features to lower tiers . First they removed search limits from some plans without increasing the price. Now they're allowing to use the AI assistant, if you want.
The gist is, when you don't use the AI assistant, you still pay the base price, and that money goes to R&D, since your subscription money doesn't go to AI providers in the first place.
For example, I have no interest in AI assistant, and I won't use it. As a result, my support will Kagi won't change.
> We are enabling the Assistant for all users in phases, based on regions, starting with USA today. The full rollout for ‘Assistant for All’ is scheduled to be completed by Sunday, 23:59 UTC.
I paid for both for close to half a year to see which one I wanted to keep. I decided to drop Perplexity in favor of Kagi, because Perplexity felt like it was trying to position / portray itself as a supernatural-esque Source of Truth, where as Kagi does a better job at letting you use the tools how you want.
Perplexity is also much less flexible than Kagi Assistant. The most customization you can do on Perplexity is answer a few questions about yourself, and hope that the info you add is injected into relevant prompts (spoiler alert: hope isn't very powerful here). With Kagi, I created a lens about a year ago to filter search results down to sources I find useful relating to GNU Guix, which I use for my machines. When Kagi Assistant rolled out (I pay for Ultimate, so I have had this a while) I made an Assistant that only pulls search results from my GNU Guix lens. The practical comparison here between Kagi and Perplexity is that I can go to Kagi and search "!guixc How do I install nginx?" (or simply ask the question in the Assistant interface; the bang will bring me there from search) and I will get back the answer I want. I added info that I use GNU Guix on my Perplexity profile, and there is not a chance that my question would have been answered within the context of GNU Guix as I wanted.
Perplexity is cool, but I found Kagi to simply be more useful.
That would eliminate the one differentiating feature on the (presumably) highest margin plan they have.
Moving to a pay-as-you-go model across all their plans might be interesting, but could equally give the wrong impression to some audiences given that it's a pricing strategy usually reserved for budget brands in the consumer space and tends to scare people off.
I don't know. To me, requiring me to give them my email and then having all my searches associated with that email is the opposite of privacy to me.
Yes, Google, Bing, Perplexity and Co could do fingerprinting and try fuzzy matching to cluster my searches. But at least that would be fuzzy and against the law in many places. While with Kagi, every search of mine would be clearly labeled as coming from me.
That Google does fingerprinting across searches to cluster them and pinpoint them to a person? Are you sure you are not confusing this with cookies? Cookies are under my control. I can decide to not store them or just delete them.
> Because the source code of that feature is open and you can look into it I guess
You can't know what source code that is running on the server which you send your queries to.
I think it is interesting that you think everything Google does is kind of more ok just because you think it is limited to cookies while simultaneously being extremely skeptical towards Kagi, not giving them the slightest benefit of doubt.
> You can't know what source code that is running on the server which you send your queries to.
Based on my (admittedly cursory) reading of the ideas behind it, the idea of the IETF standard they implemented is that one does not have to trust the server.
i quote:
“With all these pieces of information, it’s possible to create a unique fingerprint by which websites can recognize you, even if you clear your cookies. They will even be able to make an informed guess if you visit the same site with a different browser.”
i urge you to consider how google makes money. it’s not at all surprising really.
If it is a switch in the UI, then that is just trusting the site.
If you trust companies with what they say they do, then why not enter your query in any search engine that does not require a login? Afaik none of them say that they will try to apply fuzzy fingerprinting to cluster your searches into a profile.
Don't give them "your" email. Give them a mail alias, which I am sure as a privacy conscious person you already have, and you are good. They also recommend to do so themselves.
However, the most important argument here is not the fact that they are legally bound to those privacy commitments (they are), but that their business incentives are fundamentally incompatible with tracking users.
For a very niche business with an extremely narrow and homogeneous user base, if they would get caught doing so, it would be game over.
The privacy pass feature is available if you don't trust, and you can verify since everything relevant happens client side.
> A note on our fair-use policy
> Basically our policy states that you can use AI models based on your plan’s value.
Although I likely won't use Assistant, stuff like this is why I love Kagi. My relationship with them as a customer feels refreshingly transparent; I can't think of any other consumer SaaS provider that automatically answers my reflexive "how does this make money?" question.
(Compare, say, Discord. It's best in class, but eternally unprofitable - which makes me wary that it might fold or go to hell at the drop of a hat.)
The thing still giving me pause is a lack of "bring your own model connection" between saas uh services.
If I already pay for Gemini Advanced (or OpenAI/ChagtGPT Pro), I then have to pay for it again at every service that offers a Pro 2.5 (or 4.1/4o) tier. I should be able to connect my Gemini Advanced access to any service that offers Flash and be able to upgrade. Signing up for a bunch of services is starting to feel like being triple, quadruple or more dipped. Similar to how I am annoyed seeing media content cross licensed to three streaming services and not getting a bill reduction when subscribed to multiple services with the same content.
I really feel like the software model is going to start to break in a couple of ways in the near future.
Instead of vertical software slices, many people are going to want their single “horizontal” agent that they pay for (e.g Gemini Advanced, Claude) and connectivity to all of their other services.
MCP (which I personally think is a mediocre protocol, so it will probably win) as the glue for a bunch of services we OAUTH against and choose when our agents do/do not have access to certain tools.
The idea of a “GPT” App Store from OpenAI was sort of right, but just wrong enough. We are going to have an App Store inside of our preferred AI platform and subscribe/connect our other services from there.
I prefer identity as the top layer. Give me an identity portal that acts as an account manager, access to api layer, let me perform password resets, and let me avoid 2fa if I’m logged into my identity manager account. It should allow bidirectional control: both what services have access to my identity, and what services anything in my identity are granted access to access. Things like LLM, cloud storage etc should be obscured away from services so they can’t tell my storage providers from one another. All access between services should cascade down from my identity portal.
Who is going to pay for identity as a product and not just expect it as a free feature of another product suite?
Why not both? Use Apple or Microsoft or a third party identity provider.
I mean, that sounds basically like Oauth provided by the LLM provider, right? The provider just needs to build out the features.
No question, the openai gpt marketplace/store was a birth failure and doesn't make much sense in hindsight anymore. I just don't understand why they don't pull the plug and admit it's a failure.
Your intuition regarding the shift from vertical to horizontal integration is spot on!
Sam Altman, in a recent Stratechery interview, detailed parts of OpenAI's future strategy that align with your prediction — a persistent, personalized AI. He envisions users interacting with OpenAI not just through core products but also across other applications.
Altman described a key part of the strategy: "...we have this idea that you sign in with your OpenAI account to anybody else that wants to integrate the API, and you can take your bundle of credits and your customized model and everything else anywhere you want to go".
This system aims to create a portable AI experience and by virtue, would usurp the vertical software business model that has historically dominated the software economy. A horizontal play, that sits in the middle collecting their tidy sum of the pot will require a very compelling argument. That would require a low barrier to integrate for developers coupled with a value-add proposition that is meaningful and not possible for anyone other than the largest technology companies.
As you know, it’s sort of the Wild West of tech right now. OpenAI is looking to find a territory in the AI landscape and make their stake now, and I think this is the correct strategy. We have seen what being the first to market with a great product can do for the longevity and growth of tech companies - especially the consumer markets. They have the name recognition, forever embedded in the lexicon of the internet, and a great product vision that will lead to critical mass adoption that and what awaits them is the coveted moat, at least in the consumer market, that AI companies have been struggling to find out in the Wild West of AI.
Altman mentioned wanting users to "be able to sign in with your personal AI that's gotten to know you over your life". This sign-in would ideally carry "your memory and who you are and your preferences and all that sort of thing" across different integrated services.
The OpenAI SSO login will be the Trojan horse and later on the app developers will either be incentivized by OpenAI or compelled to integrate their products because of the compelling value proposition it would bring to bare with an integrated personalized AI assistant, complete with its memory and preferences.
Lastly, I suspect this is one of the driving motivations to become a consumer hardware company as there is little to no chance that current players (Apple, Google, Meta) would allow the same 1st party access to their internal API’s would be a requirement for what Altman has laid out for OpenAI moving forward.
I understand we some may want that, but why should everybody else want that? I operate in several contexts. I want some kind of models in one context and one in another. I want the AI to be contextualised within each use, not “personalised” on me. Yeah sure it would be great if it had memory within some project, but I would rather not have memory between projects or use cases that are irrelevant to each other.
The industry solution to this is to create an open standard for AI memory.
I doubt that fits in the usage model for those pricing plans. They're priced assuming it's inconvenient and less used because of it. I keep hearing they're losing money anyways, so despite being triple or quadruple dipped you're still paying less than it costs to run the models.
How would it cost Kagi anything, if I bring my own keys and all the costs are offloaded to my personal Google/ChatGPT account? I pay them the $10/month and get the Unlimited Pro access (but only to models I subsidize the cost of.) If anything it would save them money to offload my usage to my account vs using up some of their Flash tier quota.
It would be ideal for them for LLM access portability to let them offer higher end models and have the end consumer pay directly for the usage.
Id much rather pay for one AI License, and a small fee to each service I use, rather than paying a high tier AI Bonus price at every single service.
I think they might be talking about the cost to the model providers.
Then Google should offer Advanced Portable for $30 that includes being able to plug it into third party services. With some kind of policing to prevent it from being used as the entire backend to a service (for all users) itself.
Third party usage is too disparate for a single flat rate to make sense. Perplexity probably uses orders of magnitude more tokens than Kagi does (the former uses tokens on every search, the latter only does if you manually invoke the Assistant).
Then there's the disparity in how many third party services each user uses. Some will use one, some will use 50. Google could charge per integration, but that's basically the status quo.
Why would the number of integrations be the price point and not token tiers? I might want to connect my account to 100 services to test them, but only perform a couple uses each before forgetting about it.
I agree that makes sense, but it makes potential subscribers to the third parties nervous. I like Kagi Assistant because my payment is capped.
I get nervous plugging my pay-as-you-go API keys into random software because of the risk they rack up a $1000 bill doing something I wouldn’t have paid $20 for them to do.
The other three things are that economies of scale make it cheaper for Kagi to buy a bajillion tokens, Google et al don’t want to manage the customer side of things (what service ate how many tokens?), and service providers don’t want you seeing their “magic” in your console. Seems like there’s a lot of power in the system instruction side of things, and Perplexity probably doesn’t want you seeing their prompt.
So let me lock it down. When I add the key, let me tell my provider this service is allowed 3% of my capacity before it’s locked out.
> I keep hearing they're losing money anyways
Hearing from whom?
I've paid for a monthly subscription with Discord for years
They also have ads in the app and they have other monetization features...
Revenue != profit. Discord, like almost every 2010s SaaS startup, has revenue (from your Nitro) - but not nearly enough to cover expenses.
> Revenue != profit
And for over a decade most companies talk only about revenue, which is infuriating. Because most startups and tech darlings survive only by continuous infusion of unlimited investor money.
I worked for 4 startups that were operating for years at a loss with revenue, all are profitable today and all were acquired.
My current startup is also not profitable, we're burning money but we're already signing big contracts and I hope in a year or two we keep growing rather than become profitable (1B+ valuation in a year).
Becoming profitable, even at this point is just a matter of deciding to stop expanding - but neither us nor our investors want this given there is so much potential for growth and more revenue streams on the line.
this is ycombinator's news aggregators, I suspect you're not going to get a "don't take risks and build things" vibe - it's a startup accelerator after all :).
> all are profitable today and all were acquired.
They are either profitable or acquired :)
> Becoming profitable, even at this point is just a matter of deciding to stop expanding
Yeah, growth at all costs is one of the defining factors.
> it's a startup accelerator after al
The only business models for Y Combinator startups are:
- run indefinitely long on unlimited investor money
- get sold to the highest bidder at some nebulous market valuation
Becoming profitable never enters the picture :)
> They are either profitable or acquired :)
Why? Once a company has been acquired, does it automatically fall out of profitability?
If it's acquired in a stock sale, it remains an independent entity and still has a P&L
If it's acquired/merged in an asset sale (not usually a good sign), it can still be assessed whether the new division is profitable - except in some rare cases like Google (allegedly!) not wanting to itemize some of their divisions to avoid too much regulatory scrutiny on monopoly positions.
> Becoming profitable never enters the picture :)
Seems very wrong based on looking at YC's portfolio, which apparently includes a bunch of profitable startups
> Once a company has been acquired, does it automatically fall out of profitability?
It becomes a part of the company that bought it?
> Seems very wrong based on looking at YC's portfolio, which apparently includes a bunch of profitable startups
It contains very few profitable startups. Those are the exceptions.
> It becomes a part of the company that bought it?
Not necessarily. As I explained above, most successful acquisitions are stock sales, in which case the acquiring company now owns the startup (they hold the shares). The startup is still a separate entity at this point.
Google is known for just merging the acquired startups into their product line (and/or killing them), but it's not a hard rule that all acquisitions are mergers.
For example, AFAIK Livestream is still a subsidiary of Vimeo (ie wholly owned, but separate): https://en.wikipedia.org/wiki/Vimeo_Livestream
So Livestream can be profitable or not, separately from whether its acquirer is.
> most startups and tech darlings survive only by continuous infusion of unlimited investor money.
Investors making long bets is a good thing, I’d argue.
The bets are invariably: sell to the highest bidder, exit through inflated IPO and/or speculative "market capitalization".
There are a few outliers like "let's subsidize this price dumping until all competitors are dead and then we recoup money by being a de facto monopoly"
They are profitable
And besides my point was that it's pretty clear what their monetization is and that it's not some mystery
> They are profitable
Citation needed. With this type of company and business model, it's highly unusual to be profitable, so the burden of proof is on those who say it is.
They are apparently not profitable (yet):
> Discord has raised a total of about $1 billion in funding. It has more than $700 million in cash on its balance sheet and the goal to become profitable this year, according to a person familiar with the matter.
https://www.theverge.com/2024/1/11/24034705/discord-layoffs-...
Not enough to cover expenses? Do you understand what you're saying? That they're running a deficit for 15 years. Companies must make profit, or they vanish. They clearly make a profit, otherwise they would no longer exist.
They raise more capital and get more debt and try to lower burn rate but the original comment was talking about the 2010s mindset of "growth before profit" where you want more users/revenue to get acquired by a bigger player that can better monetize you.
The fact Discord isn't profitable (and hasn't been) is well documented.
Also, operating at a loss isn't necessarily bad (i.e. if you expand or spend more on R&D your profits shrink). Companies might choose to spend more on R&D and not be profitable (e.g. Amazon for a long time).
I'm not a business person, so take this with a grain of salt, but my understanding is that this is common for a lot of smaller startups and tech companies.
If you can raise funds outside of revenue (i.e. outside of directly selling your products), you can keep operating even if you're not actually generating any income directly. Typically that will be in the form of investment and loans. So even if your expenses (incl. repayments for outstanding loans etc) are higher than your revenue, you can stay in business as long as you can convince enough investors that it's still worth their while to give you their money.
I don't know whether this is true for Discord specifically, but I understand it's a fairly common strategy, especially for companies where their best chance of success is by being the only player in a given market.
It took Uber 15 years to make their first profit
https://www.theverge.com/2024/2/8/24065999/uber-earnings-pro...
This is particularly an ironic comment given that this is the forum run by a company that invests in companies that are not yet profitable and are often not profitable for years if ever
> Companies must make profit, or they vanish.
Oh, buddy. That's how it's supposed to work, but that is not how it works at all.
What do you mean?
Companies can exist for decades without making a profit, as long as they can keep borrowing money.
The strategy here is often just trying to operate at a loss longer than your competition. If your competition finally can no longer borrow, you buy them up and have a monopoly on the market, allowing you to price gouge.
There are other reasons a lender might lend to a perpetually-unprofitable company. For example, you might lend to a company to maintain a competitor who doesn't really compete, in order to avoid a monopoly breakup, or lend to a holding company that holds product with no intention to sell it, in order to maintain artificial scarcity (i.e. housing, diamonds).
Note that in none of these examples are you competing to provide the best products at the lowest prices.
This is a big misconception (though not one I’m used to seeing in the HN crowd).
Companies can operate at a deficit for many years without vanishing, usually because they have venture capital funding or investor backing.
Having a good ethos before significant funding rounds is one thing , it is lot harder once you raise a ton of external money.
It is not big bad VC either, VC funding while demanding is still lot more forgiving than Private equity or public markets . It is nature of a free economy to be as efficient as possible which in turn makes it affordable and accessible to class of users who would not have been do so before .
If the staff sees this - please stop preventing zoom. Not only is that bad for accessibility, it makes the article less useful for everyone - there's a screenshot included showing off the feature, but it's too small to read on the phone and I can't zoom in.
Firefox on Android has an accessibility setting called "Zoom on all web sites" that gets around this. Firefox's reader mode would help with this as well.
It's a shame we need these workarounds instead of all websites being accessible by default :/
Unfortunately there is no per-site override. So you get to the occasional site that you don't want to zoom (like a game where the zoom gesture does something in-game) that is unusable with this global setting enabled :(
That's amazing! I'm already on FF on Android and didn't realise they introduced it. Thank you so much. I'll finally be able to see the small images on medium and substack!
The original android browser had something much better, zoom until the size was right, double tap to reflow to fit
The new one has an extension that reflows on zoom. https://github.com/emvaized/text-reflow-on-zoom-mobile, it is quite useful.
Nice, I wouldn't want it to be automatic after pinch but that looks promising and I'm sure could be modded, maybe pinch zoom and then if you keep holding the two fingers without moving much for X amount of time it reflows
Yeah, it is sometime annoying when text reflows if you only want to look closer at an image. But most of the time it's not really an issue, you just need to reposition. Maybe give it a try anyway :)
Opera Android does support reflowing.
Hey - One of our engineers has fixed this. Please try again!
I can look into it. Can you clarify what you mean? Article zooms in normally for me (Orion browser/iOS).
OP might have meant that pinch and zoom is locked in and that’s why they can’t zoom into that screenshot?
Edit: I can zoom in perfectly in my browser - Safari on iPhone - here.
Not sure if you mean you work at Kagi, but I also want to add that when opening the Android app, the splash page takes a full second or so to load before the search bar opens, which doesn't sound like much but it makes it harder to habitually use Kagi as my default mobile search.
The Kagi widget technically solves this, but without installing a new launcher I can't replace the native Google search bar with the Kagi one.
It's blocked on Firefox on Android. Another comment quoted the meta tag which disables it.
Got it. For some reason Bear Blog defaults to this setting, not easy to change from dashboard it seems.
Awesome, thanks for checking it!
What browser prevents this actually? None of the browsers (even mobile) I just quickly tested just... worked? No extensions.
I think Safari just outright ignores the maximum-scale property.
I can zoom in
Android 14 Firefox 136.0.1 (Build #2016078447), hg-e7956a4db6c5+ GV: 136.0.1-20250310180126 AS: 136.0
ublock origin enable zoom in all websites
Edit: I know this is not what you are asking for, but try opening the image in a new tab. Can you zoom in there?
https://kagifeedback.org/assets/files/2025-04-17/1744906741-...
Yeah, I know the workarounds. This is more of a "complain publicly to maybe fix this specific case, but mostly raise awareness for people who will build the next thing".
This has been bugging me as well.
You can open it in a new tab and zoom there.
I can zoom just fine on mobile Safari.
So what?
His complaint is easily verifiable, and valid:
<meta name="viewport" content="width=device-width, initial-scale=1.0, maximum-scale=1.0,user-scalable=0">
I jumped on this bandwagon long back for preventing horizontal scrollbars and other issues. Is there an updated advice for allowing zoom and being responsive?
“Allowing zoom” and “being responsive” are not two states at odds with one another.
Preventing zooming is a serious accessibility issue and it makes the content worse for every user.
If you’re properly setting responsive widths, a large enough base font size, large enough input text size, and using border-box for box-sizing, things should just work except for cases where you’re absolutely positioning things or telling them not to word wrap and they are wider than the viewport.
There really is no bandwagon to jump onto here—most sites don’t prevent zooming in. Many that do were just misconfigured from the start, and because some popular mobile browsers ignore `maximum-scale=1.0`, the issue isn’t normally identified.
Quick tip: Make sure all of your inputs are at least 16px font size. This will prevent most mobile browsers from “automatically” zooming in when an input is focused ;) Which is a common reason people employ the maximum-scale property.
Kagi employee here. I hate it when sites block zooming. Didn't realize our blog was doing it :S
Our site's been fixed, and I opened a PR upstream.
Kagi are one of my favorite companies. In a world where buying a T-shirt requires me to hand over my phone number ("in case we need to contact you about your order" - yeah, right) I find it so refreshing that kagi barely require any info from me except some form of payment.
At least you got your shirt. I waited for many, many months, where I consistently checked to see if they were out. They happened to release them during a very short window of time in which I didn't go to my email; I never got to have the shirt.
https://store.kagi.com/
I had to send them my address…for my free t-shirt ;)
Great t-shirt quality btw
Yes
The shipping companies seem to demand it from the sender.
But they never use it so just fill in a random number.
They do if needed. AliExpress at one point required my address to be in full UTF-8 representation that I wasn't comfortable with, what I assume to have been the post office called me twice without caller ID, and the package got "delivered" to random neighborhood of the airport hundreds of miles away.
They don't call you if you live in a cleanly ASCII representable address with zero ambiguity and you don't have to be contacted from customs. Otherwise the courier do use the number.
Kagi search is great, but I'm not willing to pay more than $5/mo for search and 300 searches isn't enough. I have no interest in Kagi AI. Adding more searches would've caused me to sign up again. A rate limit instead of a monthly limit would prevent the "search anxiety" that creeps in as your available searches dwindle. Make it so and you can have my money again.
I suspect that Kagi’s price point might be a close approximation of what these services actually cost to provide without being subsidized by ad services and data harvesting; free search can only really sustain itself under those conditions. Heavy users who block ads have been getting subsidized by other users for decades now, and this has led these users to expect these services to be provided at an unrealistically low price.
Yes, but don’t twist it: it’s VC money, greedy corporations and personal data exploitation that enabled all that. Not those who came to expect expensive services for free.
People like free and the “greedy corporations” found a way to give them what they want.
Google search is trash now (for reasons that aren’t entirely in Google’s control), but for the first decade it was magical how well it could one shot finding relevant data on the web.
> People like free and the “greedy corporations” found a way to give them what they want.
It’s like saying, “People want Fentanyl, so drug dealers just give them what they want.”
Social media causes addiction and depression.
Have you ever tried browsing the web for a couple of months without a Gmail, Facebook, or similar account — and without JavaScript enabled or using only non-standard browsers? The experience is terrible.
Kagi is not funded by big VCs, please read here: https://blog.kagi.com/safe-round
If they can't scale 300 searches for way less than $5 in costs, they are doing something very wrong. Consider that you can get 1M tokens of inference for less.
You can get 1M tokens of inference because it's subsidized by VC money. The whole point of the parent comment is that this is closer to the actual cost to provide a service.
He could be right, I have no idea what search costs.
That $5 has to cover organizational costs, not just compute. There's 40+ employees, so salaries alone would be a significant chunk of their entire income if all of the 45k members were on the $5/mo plan.
the $5 is the cheap drugs that your dealer is offering. They hope you'll start popping for the good stuff for $10, which is often less than I pay for a single lunch, and I use it constantly
I also use the $5/month plan, and the few times I have exceeded the 300-search limit, Kagi simply renews my monthly plan early.
For me it's just happened a few times towards the end of the month and I was happy with how they handled the situation.
That is actually a really nice model. Effectively it is 300 Search for $5 valid for one month. And instantly renew if you exceed $5.
May be Kagi could also consider rolling over unused search for one month. So if you only did 200 searches, the remaining 100 will roll over to next month so you have 400 searches.
The unused searches likely subsidize the people who use the full 300.
At some point this is becoming a really roundabout way to implement 1 search = 1.67 cents
An AWS-style pay-as-you-go plan would indeed be an interesting approach here.
Do you mean that they charge you an additional $5 (so in a 12 month period, exceeding the 300 search limit would result in 13 payments instead of 12), or does the next month’s subscription payment get moved up to the day that you ran out searches, with the number of payments remaining the same?
They update your monthly subscription's bill date to reflect the reup on 300 searches for the date of the 301st search.
So they charge you an additional $5?
For every 300 searches they charge $5.
I've been paying for the $10/m for quite awhile now and it's one of the better QoL subscriptions I have.
I’m usually cautious with subscriptions and regularly review what I’m paying for. That said, Kagi at $10/month has earned its place — been using it for 20 months now and it’s been worth every cent.
I'm curious about how you think about the "value" for a service like Kagi? What determines how much it is worth to you? I don't think that Kagi is for everyone, but, at least for my own internal model of value, Kagi comes out so far ahead, that I'm curious about alternate ways of thinking about value that don't see it that way (to be extra clear: I do not think that my view is the "correct" view or only view).
How do you think about its value? I too am interested in others' perspectives. I'm not the person you asked, but I rarely do more than "{query} reddit" or "{technology name} {documentation-oriented query}". Google, or any of its free competitors, work fine for that. On top of that I'm subscription averse: the only "digital" subscription I have is additional google drive capacity and a VPS, which together cost less than 10 moneys per month, and I'm fantasizing about canceling those too.
For me, it is how much faster I find the results I want than I was with Google. The results are enough better that, at the rate I value my time, I'm more than making up the subscription cost every month.
And then the gravy on top is that I fundamentally think that direct payment is a "better" (both from a consumer perspective and from a societal perspective) model than an ad-supported model, so I'm also supporting a company that aligns with that larger philosophical viewpoint.
> I'm not willing to pay more than $5/mo for search and 300 searches isn't enough
Most of the population is well served with search funded with ads and tracking. Kagi is for the minority who don’t want that. I’m not sure there is enough if a market between 300 and some other number that would treat search anxiety just to satisfy those who won’t pay $60 a year more to relieve it.
$10/mo is very cheap for such a high-quality service.
I'm gonna guess you aren't their target audience, so you both win thanks to the filter of capitalism. $10 a month to avoid google and microsoft AND get a decent AI assistant is a bargain to me. Plus it's a small tech company trying to make a good product for a certain audience and I can dig that. I hope they succeed and become wealthy.
It simply isn't worth servicing people who aren't willing to pay $10 per month. They are problem customers and frequently churn. Much better for your sanity and your wallet to focus on improving your offer for the people who are happy to pay.
Okay - this is amazing. I’m a happy paying Kagi user. Getting to apply almost all of my payment towards for cross-platform tokens as a bonus - wow.
This will replace a chatGPT and Anthropic sub for most everyday users. Their assistant is better than bringing my own keys to a client for most use cases. Just wow.
I agree. I’m on the $10 professional plan and honestly if I’d received an email offering this for an extra $5 per month it would have been a no brainer for me.
Please Kagi, don’t take too much of a haircut or let paying for this eat into the core search budget.
They have usage data and I'm sure they did the math. A few users will max this, but most won't touch it or only do a few requests a month. If it bumps their conversion rate it should be worth it. I hope!
How do I get this to work? When I try to access the Assistant I just get the help page: https://help.kagi.com/kagi/ai/assistant.html
Are you outside the US?
Q: I can not access Assistant!
A: We are doing staged rollout beginning with USA, full rollout scheduled by Sunday, 23:59 UTC. This will include other regions and even the trial plan.
Still not working
I missed that, thanks. I will try it on Sunday
Should be out globally today actually
Still not working. I am logged. I have free account
You have to login first.
For some reason instead of redirecting you to login kagi.com/assistant redirects you to the wiki rather than a login page when you're not logged in.
It's right there in the article:
> An important note: We are enabling the Assistant for all users in phases, based on regions, starting with USA today. The full rollout for ‘Assistant for All’ is scheduled to be completed by Sunday, 23:59 UTC.
I missed that, thanks. I will try it on Sunday
Well in your defence, the HN title does say 'available to all users'
Good point - how would you word the title?
Kagi Assistants to become avalailable to all users this week.
I was also confused, the title is very misleading and US-centric.
Outside the US as well. You can get it to work now through a VPN. Not all models are available at the moment.
The "fair use" part takes a lot of place in this article.
It talks a lot about what happens if you use more tokens than what you're allowed, but curiously doesn't pip a word about what happens if you use less - for example maybe with a partial rebate on your next billing cycle ?
I think "fair" should mean "fair for all parties involved", currently it's rather a "we don't want to incur any risk" policy, since I don't see how it's fair for my end of the contract. I'd rather pay for my actual usage at any other provider than pay for min(actual usage, 25$) at Kagi.
It’s the exact opposite. They are incurring a huge amount of risk with this.
6 hours ago most users didn’t have access to this feature at all. Now we have $4-8 of raw token credits a month to use on a well-built feature.
I’m paying $9 a month with the annual subscription, and it was worth it just for Search. Now they’re giving me $17 worth of value for the same price.
Their margins must be razor thin, and they’re only able to offer this much value because they’re counting on most people not using all credits. If everyone did, or if they gave rebates, they’d go out of business.
The other point of view is that they are now forcing users to pay for both search and AI, even if they do not want to use or fund the development of the later. You used to be paying 9$ for search, now you are paying 9$ - x for search and an unknown amount for AI.
Probably a lot of subscribers that will never use it, like myself.
For myself at least I agreed to the original price.
If more features are added and I don't use the but other people do, as long as the features I pay for stays, I'm happy.
As an existing happy subscriber to Kagi, this statement is illogical.
I currently pay for x. Soon I’ll get x + y for the same money.
That’s better.
I am also an existing happy subscriber to Kagi. I currently pay for unlimited X. Now I pay for limited X, where I can't even see if I'm approaching limits or not. Anyway, my main point is that I'm getting LESS for the same money.
No! Are you wilfully misrepresenting this?
You currently pay for unlimited X.
Now you get unlimited X + limited Y.
More.
I'm going to take a charitable interpretation of your post and assume that you have simply misunderstood Kagi subscription tiers, and you have mistakenly assumed that I was on the first tier. I am not on the first tier. I am on the tier which they continue to advertise with the words "Unlimited Kagi Assistant". That is the product I am paying for. Now, instead of getting "Unlimited Kagi Assistant", I am getting "Limited Kagi Assistant".
Search isn't getting limited is it?
And the GPT features always had some limits.
I'm not paying for search, I'm paying for Unlimited Assistant. No, they did not "have some limits". Or at least, the product they sold me was very clearly labelled "Unlimited". The word "Unlimited" is literally in the name, and it was described as such in more wordly descriptions.
I'm not going to argue about your interpretation of the word unlimited and wish that its use was either banned or strictly enforced in this (and similar) contexts. That said, until accountability is legislated and enforced, it is not reasonable to assume that unlimited means unlimited. Just as marketing abuses the word, customers abuse the concept. That's especially true in a domain where processes can be automated and distributed, by some means or others.
> it is not reasonable to assume that unlimited means unlimited
sure, but I am not abusing the unlimited plan. it is reasonable to assume that "unlimited" means that a normal user like myself will not run into limits.
If you’re not abusing it and are just using it like a normal user, you won’t hit the limit.
As far as you’re concerned, nothing will change, so I’m not sure why you think this is bad for you.
You seem very confident I will not hit the limit. So presumably you know where the limits are? Where are they?
It’s all written in the blog post
No it is not.
The article claims that 95% of users won't be affected when enforcing the policy. Assuming the claim is true, not affecting 95% of users is pretty much the definition of not affecting normal users.
That policy is there because some users were incurring >$500 in AI costs in ultimate.
Out of tens of thousands of users, the top 100 were using 48% of total AI costs.
Kagi is bootstrapped, it can't just hemorrhage money like this the way the VC-subsidized AI startups do
That is totally unsurprising. I would be curious to see what they are doing to achieve that. Likely automating something and not using it like a typical human would.
Maybe your usage might not be affected, I propose to check for a month and if it really goes over the limit, you have the option of canceling the subscription.
It definitely is worded in a way that it can either be interpreted as (unlimited searches and unlimited premium ai) or (unlimited searches and premium ai).
It also may not help that they did not enforce the fair use policy until now. At least that is what I read out of their blog post.
But the fair use policy has been included a long time. I checked 2023-12-25 and found it there, might be available earlier, but no interest in looking harder.
They advertise with the phrase "Unlimited Kagi Assistant", so I'm not sure how you can interpret that as "unlimited searches and premium ai". Kagi Assistant refers to the same thing as "ai" here. It's a different product than Search.
"As an existing happy subscriber" -- goodness! Even after years of seeing what happens after such illogical evangelism, hn never ceases to surprise you and brings the fandom out in full force. It's like with Apple. You say once "your phone switches on and off on its own - maybe something is amiss with the hw/sw" and there are dozens of replies already blaming you "you must be holding it wrong!". It's a whole new level of apologism.
I don't follow what either you or the OP are upset about. Where is the apologism and why do you think it's needed?
Kagi rolled out a free feature to its existing customers without increasing the price of their plans. The limits of that plan seem quite generous, as well. The only way I can make sense of the OP's post is that the OP wants the Kagi subscription price to decrease. Perhaps that's fair, but it doesn't make any sense here because you're strictly getting more for your money. If you're paying $10/month for a subscription, yesterday you couldn't use Assistant and today you can for the same $10/month. Placing a cap on how much you can use seems quite reasonable given the service costs money to operate. If you choose not use it, you're no worse than you were yesterday... you can happily go about using the search service you were already paying for.
Is the problem that the free usage isn't unlimited? Is it that not using the free service doesn't reduce the search price? Or is it that those using Assistant more than you appear to be getting more value for their money? I'm not trying to be dense, but I really don't see what's even remotely controversial about this announcement.
Lately, every subscription I have is increasing the fee without giving me anything. Kagi is giving me something extra without charging me any more money. I'm sure the nefarious intention is to make their service more attractive to non-subscribers and grow their userbase, but interests can align.
It’s probably worth noting that they do offer a refund of your subscription price if you don’t perform searches in a given month, which is pretty much the opposite of how every other company with a subscription works.
It wasn’t a day one feature, so there’s some chance that a thing like this could roll out.
The article probably focuses on the overage because that’s what most users are going to be concerned about.
Few other companies seem to try to do things in the interests of their users and balance that against making enough money to keep existing.
This feels a bit like manufactured outrage.
Huh? The title of the blog post is "Kagi Assistant is now available to all users!". Their users are people paying for what up until now was just their search service. They're now rolling in Assistant as a value-add. Your subscription price didn't increase. You're strictly getting more for what you were already paying. If you don't use it all, you're no worse off than you were yesterday.
If you want metered billing, there's no shortage of AI services that offer that option. Kagi even offers one by way of the FastGPT. You can also pay to use their search API if you don't think the subscription is worthwhile. You can cobble something together with Open WebUI pretty easily.
I have Kagi Family plan for my household. I've been paying for the Ultimate upgrade for my account in order to access Assistant, but given how infrequently others in my family would use it, it never made sense to upgrade them. Still, it would have been convenient if they could occasionally access Assistant. And now they can. And my bill didn't increase. And they're being incredibly transparent about what the limits are and why they're there. I'm a really happy customer today.
That is a fair point. Considering the alternatives and realities Kagi is way too cheap for the life improvement it provides.
Is it fair enough to ask your favorite restaurant to lower the bill because you didn't eat the two last franch fries ?
If it's a subscription and someone else can eat the fries, of course?
Yeah I concur.
As an early adopter I first got forced off my grandfather plan to the regular one(at least I got a T-shirt). Now I have a limited number of searches that I have to keep track of and this has made me only use Kagi if necessary. This has dropped my number of searches significantly but at the end of the year I’m still being charged for renewing my plan even though I haven’t used a quarter of my allotted searches.
I don’t care about LLMs so this brings nothing of value to me. Give me an email account or some backup storage and open source office suite and I would be willing to pay and pay more.
I’m seriously considering not re-newing my subscription for the first time in ages.
The $10 plan has unlimited searches again.
And this isn't anything that has changed recently. It's been unlimited since 2023.
Then don't renew. Nobody is forcing you to pay for the service, and from what you wrote, it sounds like the service is not what you need/want (anymore).
How do you use kagi and not know the basic plan has unlimited searches and has had it for ages?
It doesn’t, I use the starter plan, that’s the lowest paid tie.
The basic plan has 300 searches.
I subscribed to Kagi a month ago.
It's great.
I'd love for BlueSky takes a subscription model, too, so we don't have to think about advertising or sustainability or all that jazz.
Also means the company's interests much more closely aligned with users interests, rather than advertisers interests.
That's why I moved to a paid mail service myself. I used to have a free one, but it was always lurking in the back of my mind that they'd be selling me to advertisers. (Plus, the service sucked)
Is there a way to disable this? We have a strict no-ai policy. Even having it available will be an issue and I may not be able to use Kagi at work.
As a user myself, I'm not sure. Assistant is a separate view so it's easy to never interact with it but yeah, I can see why that might be a work policy issue. You could try requesting that as a feature in https://kagifeedback.org. The team are quite responsive there and historically Vlad the CEO reads every post.
Curious is it no AI or no LLMs policy? Search has used some form of AI since inception.
Are you imagining one switch toggle that would disable it at the team account level (or individual account level - in which case it would theoratically still be opt-in and 'available' like it is now).
Genuinly trying to understand the UX of it that would comply with the policy.
Settings > Search > Manage search AI features
Has the option to turn off Auto AI feature.
It might be an good idea to post on their support forum.
Can you describe the why of the policy and if you are ok sharing the industry?
I am also curious if you have other restrictions on information sharing, API usage, and what reference documentation to use.
I unfortunately don't make the policies, or always agree with them. But, I do have to follow them.
Then you shouldn’t have been using Kagi in the first place, because they’ve always had an AI summarizer feature available by appending a question mark to the query. So just like you can choose not to use that, which has always been there, you can choose not to click on Assistant and stick only to non-AI functionality.
FastGPT & Kagi Translate have been available for a while, so if you have a Kagi subscription you've already had AI accesss.
I've been wanting to try and see whether I can consolidate my phind.com and my kagi.com subs by upgrading my Kagi plan. They are kinda forcing me to try now :-)
I tried, and Kagi is better in some aspects like ease to edit questions. Haven't seen any a downside yet, and the option of creating multiple profiles seems interesting.
Anyone used both Kagi assistant and perplexity and can share how was the experience?
I don't use the Kagi assistant yet, just the kind of AI response in search results. But regarding perplexity, I'm a little bit disappointed.
I started to use Perplexity like 1 or 1.5 years ago when it was really good in term of efficiency to find good results with an efficient interface, compared to chatgpt and co. But nowadays I find the assistant response to be not that good at all, with a lot of the links provided or the suggested follow up questions on the same quality as Google SEO top results or ads.
Despite having the paid plan or Perplexity, most of the time I try a request there and then still go to chatgpt or mistral to ask the question again.
For Kagi, when I use the in search ai response, it is mostly good directly.
If you hit `a` in the search results page it'll take you to a chat interface to better grok the AI in a conversational format. Not exactly the same as perplexity still, but more ChatGPT akin. There's also a continue in assistant button at the bottom of the summary results to chat more with the query.
You can also go straight to the view by visiting `/assistant` and typing your query in there.
I still think preplexity has the AI search stuff down better, but getting both a "legacy style" search AND AI search, Kagi has better value to me imo.
I tried both and use Perplexity. Perplexity is much more akin to native ChatGPT in which I can create "spaces" for my projects. For example, I just bought an RV and so I loaded up all of the PDF manuals into a space so the LLM can reference those things. Since a few months ago, Kagi is just thread based.
FYI - I use Kagi every day for standard search and love it.
Perplexity aggressively makes shit up, particularly when it comes to science and finance. It provides sources, so you can check the bullshit, but I’m the error rate is so high that it’s practically dangerous.
Just typed this up elsewhere in the thread: https://news.ycombinator.com/item?id=43726582
Examples were useful to visualise the difference, thanks.
I've subscribed to Kagi for 2 years and subscribed to Perplexity for 2 months. I liked Perplexity a lot—in particular, I think its search AI search is faster and maybe a little better than Kagi's. But the lack of standard search and not being able to do non-search AI work led me to drop it and just keep Kagi. And then Kagi released Ki, their multi-step assistant, which is close to as good.
I use both but cancelled my Perplexity subscription.
Kagi is the better version of Google search, especially if you learn how to use lenses, bangs, and all these features. Kagi Assistant is great if you‘re happy with basic no-frills chat, i.e. no usable voice input, no image gen, no canvas.
Perplexity is not bad, but somewhat stuck in the middle between ChatGPT/Gemini and search. They provide sources for search results which are somewhat more spot-on than what I‘ve seen elsewhere. For example it could find EV chargers with restaurants for a trip I made along a route, which ChatGPT, Gemini, Kagi Assist failed greatly).
I found refining searches with Perplexity terse and it kept forgetting context once you started to reply. They have an AI generated news feed which lured me into more doom scrolling.
Also, be aware that Perplexity free-tier may collect data about you, which Kagi does not.
Tldr; Kagi is a superior search engine worth paying for. Perplexity seems good at queries that require context but quite expensive.
Any suggestions for how you got your lenses, bangs, assistant set up the way you like? I recently subscribed to Kagi and feel like I don't really know how to get in the habit of really using all the features.
For redirects I like to create a rule to direct versioned sites (like documentation) to either "latest" or the version relevant to me. I also like to redirect sites to a more user friendly version when applicable (like reddit.com => old.reddit.com).
If a site leaves me disgruntled when I visit it, I block it. If I find it too useful to block entirely (reddit) I lower it. I apply the inverse to sites I enjoy.
I find there are certain sites that have built-in search that sucks. One such example is Dockerhub. In that case, when I want to search for a container on Dockerhub, it may be tempting to use the built-in !dh, but that is no good. Instead I favor the "snap" search: @dh, which will just add "site:hub.docker.com" to you query. This will give much better results than !dh. This can also be combined with the "I'm feeling lucky" bang (!), so you can search for something like "nats @dh !" and end up on the Dockerhub page for NATS - without ever even seeing Kagi if you do it from your URL bar. I do this pattern all the time, usually with Dockerhub and GitHub.
You'll find with the above pattern that you'll start to want to apply this to sites that aren't natively supported as bangs. One such site for me has been Ollama. I added a !ollama to be able to search for models directly. It's also nice because just searching "!ollama" will bring me right to the homepage too, which is useful when I want to check to see if I'm on the latest version.
You'll also find there are subjects where you tend to prefer a small set of sources. Maybe it's some software or tool, or some hobby or something, where you prefer official documentation, maybe some known personal sites you trust, a reddit community, something like that. That's where custom lenses comes in. I personally have a lens for the operating system I use (GNU Guix) (as well as a !p bang to search for packages) which includes official documentation, mailing list archives, IRC archives, things like that. I'm sure there are probably similar subjects in your world that you would enjoy having a more focused search for :)
As for the Kagi Assistant, I pretty much have just wired up an Assistant to use my Guix lens as a search source. That is pretty nice, because I can just ask it general questions like "how do I install nginx?" and get focused and relevant answers, instead of having it go off on how to install it on irrelevant distros.
You can start by looking at https://kagi.com/stats?stat=leaderboard and adjusting domains that make sense to you. E.g. I have all Pinterest lowered, as well as w3schools.com as I prefer developer.mozilla.org (which I raised instead).
For bangs I’m pretty sure the default ones might be enough – just use them! Some of my go-to bangs are !gh, !gm for Google Maps (Kagi Maps are sometimes not as good in Asia), !yt, !mdn, and !amo for addons.mozilla.org.
I also have this redirect rule:
So that if an NYT article comes up in the results I can get a version without paywall directly. You can set it up in https://kagi.com/settings?p=redirects.I just stumbled across cooked.wiki recently. It allows you to convert any bullshit online recipe to a no-frills ingredient list and instructions. I've only tested it a couple times and I know I'll rarely remember to use the shortcut (https://cooked.wiki/<recipe-url>). But setting it up as a redirect rule from search results might be the thing that pushes usability over the top.
I did not customize bangs and the standard lenses are already useful. I have two custom lenses for travel and academic papers, each searching specific domains. What I meant is that you learn how to use them. What is there by default is already quite good, like !yf !yt !uk etc.
I use both. I only pay for Kagi because I have many models I can use and I can set up different contexts to use them in.
I rarely use Kagi search anymore and instead search via assistant. Both it and perplexity give me much better results than I get from a traditional search engine.
I've never been great at getting what I want from search engines. With assistant and perplexity, I type plain English with context and get what I am looking for a large chunk of the time. That's a godsend to me.
I've found things that assistant does that make it worth paying for. I often use perplexity but what I use it for (deep research) isn't valuable enough at the time to pay for.
I like the perplexity iOS app a lot and use it almost exclusively on my phone which isn't enough use to necessitate needing a subscription.
I’m a Perplexity subscriber, but I’ve decided to favor Kagi. One issue I have with Perplexity is the lack of control over the search timeframes. With Kagi, I can use my own timeframe filters, such as last 2 days, last week, and last month.
I wonder why the rollout is specifically over the weekend. I'd personally do something like that Monday to Wednesday rather than Friday to Sunday. It seems like the kind of thing that needs monitoring and quick reactions - can easily get expensive if something goes wrong.
Possible that they see lower usage on weekends.
They're already limiting it to one country at first. If they want a slower ramp-up, they can show it to more and more users rather than working weekends. Not to rule it out but this would strike me as a strange reason
Are they really going to need more than a couple people working the weekend to pause/rollback if something goes wrong? It's not like Assistant itself is new, just the broader access is.
On the other hand a huge number of countries have the whole Easter holiday off. Plenty of time to read these articles and sign up to stuff.
Lower WE usage. "Let's see if it crash"
Hm, maybe not lower per se (though I can believe that), but less critical (less likely to be work related)?
It'd be nice if you could see how much each request actually cost in relation to your plan, and to have some type of easily accessible meter.
A lot of AI providers operate in black box territory when it comes to limits, which is quite annoying.
> We plan to display cost and interaction details more prominently soon, potentially on the billing page or directly within the Assistant interface.
I too want to see this soon. As a long time user of Ultimate it isn't uncommon for me to use 5M tokens per month and I have no idea if this will be covered by my subscription now.
How do you know how many tokens you've used? Or is it a guesstimate?
The following should say your daily token usage:
https://kagi.com/settings?p=consumption
And what is the cheapest model to use? Llama 70B?
https://help.kagi.com/kagi/ai/llm-benchmark.html
gemini flash, 4o-mini, mistral small for super cheap
qwen-qwq for strong models
Thanks
I don't like how this was rolled out. I'm currently paying for "Unlimited Kagi Assistant" and the Kagi website STILL advertises "Unlimited Kagi Assistant". And they stealthily rolled in limits? I pay the same amount, but it's no longer unlimited, and I only know about this because I happened to notice it on HN. Otherwise I would only know after hitting a limit.
Fair-use limitations were always there. It sounds like they weren't actively enforcing it, but now they are because of some problem users. I don't think anything has changed for you unless you're one of the users this refers to:
Q: Why did Kagi start enforcing the fair use policy?
A: The policy was enforced due to excessive use. For instance, the top 10 users accounted for approximately 14% of the total costs, with some individuals consistently using up to 50 million tokens per week on the most advanced models. Our profit margins are already quite narrow. 95% of users should never hit any usage limits.
I'm really surprised that more people didn't jump on the unlimited usage of Claude3.7, tbh.
Don't think it's fair to call users problematic when they were using the product as advertised. "Unlimited" has a meaning.
> Don't think it's fair to call users problematic when they were using the product as advertised. "Unlimited" has a meaning.
I'm sympathetic to that argument, for sure, but it's also just a branding-label to not necessarily be taken literally. There must always be a limit to everything as there's only so much energy in the universe: so, the word 'unlimited', in every real-world physical context, is still with limits.
Read the T&Cs should always be the the advice.
In my country (Australia), companies have been found guilty in court when making that claim. It considered false advertising to claim your product is "unlimited" when it is not, in fact, unlimited.
> I'm sympathetic to that argument, for sure, but it's also just a branding-label to not necessarily be taken literally.
Sure sure sure, but I'm not abusing the "Unlimited" service. I'm just asking AI questions every now and then. I'm a normal user doing normal usage and I have no idea if I will be hitting these limits or not.
Personally, I think it's a bit scummy to expect customers to trawl through T&Cs, when the company gets to sit back and chill, while profiting from using 'unlimited' front and centre.
There's a subtle difference between using something for 5 hours per day vs causing the heat death of the unvierse.
Metred use, with all parties being informed and honest with wording, is the fair and ethical solution. I absolutely abhor how companies are allowed to change meanings of words, then run behind 'muh conditions' when they lose out on their gamble.
> There's a subtle difference between using something for 5 hours per day vs causing the heat death of the universe.
Replace 'heat death of the universe' with 'the available funds of the organisation'. Nobody should be so naive to think that any 'unlimited' service is unlimited in the same way as there's an unlimited set of natural-numbers, or any other mathematically pure meaning. There are plenty of words where the precise meaning isn't used any more ('myriad' is one that jumps to mind right now, nobody uses it to mean 10,000 any more).
I'm sure if there was a word that meant: "effectively unlimited for the majority, but there's a limit for the extreme outliers" I'm sure it would be used as an alternative, I don't know of one?
I agree that the absolute upper-limits should be upfront. So the local definition of 'unlimited' is clear. I believe that's what Kagi is moving toward, if I'm reading FAQ correctly.
If no one can offer unlimited then it shouldn't legally be able to make that claim. It's like putting cures cancer on your bread loaf and then changing the terms to say curing cancer means eating wheat; everyone should know it's a lie but it raises sales anyways.
What about words like ‘Ultimate’ or other commonly used branding words. Are you saying that no words except exact literal definitions must be used on all branding?
I hate to break it to you, but I think the cat is well and truly out of the bag on that one!
Ultimate is a fair term as with most branding terms. Unlimited is a lie, ultimate separate offerings between basic and standard.
You can't say you get 10 apples for a dollar and only give 9. You can say best, ultimate Apples because they are not quantifiable.
I have no idea if I will be limited or not. I can see I have about 3 million tokens used per month. But I don't know where the limits are.
Kagi says they will charge the actual token costs of the underlying API's. They will hopefully make the actual calculation visible soon.
If we make a quick back of the envelope estimation: in the outlier case if those 3 million tokens are mostly output tokens and you always used an advanced model like GPT 4.1 which costs $8 per 1 million output tokens you would be close to hitting the limit that the plan provides ($24 out of $25 worth of tokens).
In pretty much most other scenario's (including a higher proportion of input vs output tokens and mixing in cheaper models) you could be a long way from hitting the limit. For example if you used half of those tokens on GPT 4.1 Mini instead of GPT 4.1 you'd only be roughly halfway to your limit ($14 out of 25$ worth of tokens).
I wish they had first included/added a usage meter then the limit. If they had first given users a way to see their usage, users of the assisstent could know how much they have to worry about the change. As it stands I use up to 2m tokens per month but have no clue how much this amount of tokens (over various models) cost. And 5% hitting the limit and not having a way to just pay for usage past the limit (yet) is kind of dauntingI wish they had first added a usage meter before implementing the limit. If they had first given users a way to monitor their usage, users of the assistant could know how much they have to worry about the change.
As it stands, I use up to 2M tokens per month but have no clue how much this amount of tokens (across various models) costs.
And 5% hitting the limit and not having a way to pay for usage past the limit (yet) is kind of scary. Especially as I feel like I use AI more than my peers.
Maybe they've changed it in the past hour, but as I write this comment the 25$ plan is called "Ultimate" and promises unlimited search, but not unlimited assistant.
I agree about the need for appropriate wording and advertising, but other than that, the new limits seem entirely reasonable and in line with what other aggregators like Abacus and Poe are doing. The paid plans of the major AI labs themselves always have usage limits too. It simply can't work any other way if you include costly models in the mix.
It's easier just to never believe a company whenever they say "unlimited". It almost always goes away. And never ever truly unlimited as they all have a legal caveat to terminate your services for whatever reason whenever they like
Afaiu, with Unlimited you have more tokens available and more models to choose from.
This is a great deal. Switching to Kagi was like getting OG Google search back. I'm also blown away by how good it is. I used DDG but was never really happy (always had to go back to Google for certain things).
Love that for $25/mo, you can get access to GPT 4o, Sonnet, and other models along with high-quality search.
Why the f don't blogs have links to the actual homepage
I love Kagi and it's easily worth the money.
I even use an Ultimate plan to try out the different models but I use it so rarely that it's probably better to downgrade.
I wonder, what's the upside for the Ultimate plan now? Just better models?
I was given a free month of Kagi to test, and it had so many rough edges that during the last days of of the trial I was already using Google again.
Notable issues for me:
- maps (from Mapbox) are really bad. Sluggish performance and lack of information
- barely any info boxes
- no translation feature ("gründonnerstag englisch") gives me links to leo.org (which was a cool site in the 00s) and to other sites, but Google gives me a translation box with the result
- no timezone calculations: "10 am PT" in Kagi: "= 10 Pt am (metric petaton attometers)" in Google: "10:00 Freitag Pacific Time (PT) entspricht 19:00 Freitag in ..."
- no search history, which is sometimes really useful to have
Other than that, the search results are really good.
> Other than that, the search results are really good.
I'm confused why anything else would matter. For example, I'll readily admit that Kagi maps sucks compared to Google maps. But I just use Google for map stuff, and use Kagi for searching. It doesn't seem like a big deal to me that it's a tool which does one thing and does it well.
Because all these things improve efficiency. They give you the result at one glance.
With good search results I refer to those which normally take you to sites, not just quick lookups like time zone conversion or translations. For example results which point to documentation or GitHub.
While I'm aware this is a case of "you're holding it wrong" – !translate <phrase> should do the trick. And that's not an excuse for not having better detection for when an info box should exist, because they do have them, especially for, but not limited to the WolframAlpha integration stuff. (For example, a friend and fellow user was awed when searching "internet speed test" and saw it integrated, no idea if Google has that too though).
Other than that, make sure your region/locale is set correctly (I'm not getting the metric petaton, for example), and for everything else, they have an excellent feedback forum for suggestions/bug reports.
> make sure your region/locale is set correctly
Last time I checked (which admitedly has been a few months, but I haven't seen anything in the changelog/announcements that make me belive things got better), that's not a viable option.
I always keep it as "International" (luckily that's an option!), as setting it to "German" significantly degrades the results as for e.g. technical topics it will rank many rubish German results higher than they should be. Google still has a significant edge in getting the distinction right between regional and technical expertise and how they relate to language.
Kagi not saving search history is a big selling point for me. I don't want yet another tech company keeping tabs on me.
And I wouldn't care if they dropped maps, I pay for kagi for search and the assistant.
I feel the same about their maps. I pay for Kagi for search primarily and assistant secondarily. I'm fine getting my maps from a different source. Not every service needs to be an "everything app".
I think those are all perfectly valid points. As a Kagi early adopter they don't weigh as heavily that they ultimately make a difference, but it also feels like most things that are not AI related are not receiving that much attention nowadays, which is a bit disheartening to see.
> - no translation feature ("gründonnerstag englisch") gives me links to leo.org (which was a cool site in the 00s) and to other sites, but Google gives me a translation box with the result
How about a custom Bang for dict.cc?
Bang shortcut could be "dcc" with URL: https://www.dict.cc/?s=%s.
Then you can translate using "!dcc gründonnerstag".
When did you try it?
Because they have been working on all those issues. They even have their own translation now.
How is the different from 3.7 sonnet connected to the web search? Or Gemini 2.5 with search capabilities? Am I to rely on Kagi being better and capable of continuing to be better than frontier models at search?
It's a front end for all of those models. I haven't signed up for it but I'm considering it as I pay for both Kagi and Claude for a total of $35/month, whereas this would be $25 if I understand correctly.
https://raw.githubusercontent.com/aboxwithrocksinit/test-buc...
Difference #1 - Flexibility - in Kagi you choose any model you want to work with, not just models from one company
Difference #2 - Quality - Kagi Assistant uses Kagi Search. Ultimately, model quality will converge to the same place, quality of search results will remain a major difference.
Difference #3 - Privacy - Using models through Kagi is more privacy respecting due to non-training for API calls.
For most very simple use cases -- like, I don't know, if you want to know how to sort a list in Kotlin -- just using the Kagi in-search assistant is more than enough, you can search with: "sort list kotlin, code example?" (with the question mark) and you'll get brief explanations with code examples based on search results (or whatever. I don't know, it works).
The model they're using for these must not be a very good one, but for most things it's enough, and very fast.
I'm not sure how to access this, both the link in the top right inside Kagi and from "The direct link to the Assistant" (https://kagi.com/assistant) from the docs, just lead me back to the same docs page. Anyone know what's going on and seeing the same issue? I'm on the "Professional" if that matters.
Hey - We're doing a phased rollout over the next few days. Currently we've rolled out to all US regions - and we plan to enable more very soon! (as soon as today - if everything seems right!)
We'll also be sending out a toast message to everyone once it's globally rolled out.
they're doing a rolling out, so try again later.
Obviously I'm happy to benefit from being able to use most model "for free" in my paid non ultimate account.
But I'm concerned that this will not rot the business model like that kind of thing happen for other services.
I would have preferred that the full of my subscription cost goes to the core feature of developing the search engine and directly the related feature. And as of today, I pay a separate premium if I'm interested by the AI assistant.
Now, with it being in all subscriptions, and knowing that anyway they can only work by paying the token price per request to all AI providers, it means less of my money going to the search index improvement, and what I'm more worried is that a forced increased of the subscription price in the coming years.
Something like, as you know, our costs are high, so we need to raise the pricing to stay sustainable.
Even if not the best reference, this remind me of Netflix saying look we are adding "videogames" (that no one wants) to your subscription for free, but now we will have to raise our prices, because you know, inflation and all of that ...
From my experience, Kagi always prefer to "trickle down" features to lower tiers . First they removed search limits from some plans without increasing the price. Now they're allowing to use the AI assistant, if you want.
The gist is, when you don't use the AI assistant, you still pay the base price, and that money goes to R&D, since your subscription money doesn't go to AI providers in the first place.
For example, I have no interest in AI assistant, and I won't use it. As a result, my support will Kagi won't change.
Duplicate: https://news.ycombinator.com/item?id=43723964
Sounds good. I use duckduckgo's AI assistant a lot these days, which also guarantees that the data is not used.
how do you actually get it? none of the links work and I am a paying user...
just takes me to documentation.
> We are enabling the Assistant for all users in phases, based on regions, starting with USA today. The full rollout for ‘Assistant for All’ is scheduled to be completed by Sunday, 23:59 UTC.
Thanks! I totally missed that, I wish they would announce when it is out rather than when it isn't :)
Good idea 1 year later. Perplexity is on top of the game
I paid for both for close to half a year to see which one I wanted to keep. I decided to drop Perplexity in favor of Kagi, because Perplexity felt like it was trying to position / portray itself as a supernatural-esque Source of Truth, where as Kagi does a better job at letting you use the tools how you want.
Perplexity is also much less flexible than Kagi Assistant. The most customization you can do on Perplexity is answer a few questions about yourself, and hope that the info you add is injected into relevant prompts (spoiler alert: hope isn't very powerful here). With Kagi, I created a lens about a year ago to filter search results down to sources I find useful relating to GNU Guix, which I use for my machines. When Kagi Assistant rolled out (I pay for Ultimate, so I have had this a while) I made an Assistant that only pulls search results from my GNU Guix lens. The practical comparison here between Kagi and Perplexity is that I can go to Kagi and search "!guixc How do I install nginx?" (or simply ask the question in the Assistant interface; the bang will bring me there from search) and I will get back the answer I want. I added info that I use GNU Guix on my Perplexity profile, and there is not a chance that my question would have been answered within the context of GNU Guix as I wanted.
Perplexity is cool, but I found Kagi to simply be more useful.
The difference is that Perplexity is subsidized by $0.5B in VC investment and are losing money.
Kagi can only grow at the pace of incoming revenue.
It would be nice if all models where available on every plan too.
They give more for free, and your only thought is "sure would be nice if they gave even more for free"?
Honestly, everything should be free, and they could just make money from ads (which I’d block with my ad blocker).
/s
That would eliminate the one differentiating feature on the (presumably) highest margin plan they have.
Moving to a pay-as-you-go model across all their plans might be interesting, but could equally give the wrong impression to some audiences given that it's a pricing strategy usually reserved for budget brands in the consumer space and tends to scare people off.
If you try openrouter you'll see why they have to charge $25/month for the best models. Pay per use and you'll intuitively feel the price.
> would be nice if all models where available on every plan too
Would be nice if I had a lay-flat intercontinental jet.
Second that. Given that they have their fair use policy it should be in their interest as well I would believe since they have a baked in margin.
"Privacy by default"
I don't know. To me, requiring me to give them my email and then having all my searches associated with that email is the opposite of privacy to me.
Yes, Google, Bing, Perplexity and Co could do fingerprinting and try fuzzy matching to cluster my searches. But at least that would be fuzzy and against the law in many places. While with Kagi, every search of mine would be clearly labeled as coming from me.
There is a feature where you can search anonymously, using IETF's Privacy Pass standard: https://help.kagi.com/kagi/privacy/privacy-pass.html
That looks super complicated. And hard to tell if the cryptography works as intended.
Why would I go through all that hassle if I can just type my query directly in other search engines?
Because the other search engines offer no privacy and relentlessly track everything you do across the internet?
How do you know they do that? And how do you know Kagi does not?
Whatever the engines secretly do, why would I use one where on top of that I have to actively tag every one of my searches with my email?
> How do you know they do that?
Because it has been widely discussed for years.
> And how do you know Kagi does not?
Because the source code of that feature is open and you can look into it I guess
> Because it has been widely discussed for years.
That Google does fingerprinting across searches to cluster them and pinpoint them to a person? Are you sure you are not confusing this with cookies? Cookies are under my control. I can decide to not store them or just delete them.
> Because the source code of that feature is open and you can look into it I guess
You can't know what source code that is running on the server which you send your queries to.
I think it is interesting that you think everything Google does is kind of more ok just because you think it is limited to cookies while simultaneously being extremely skeptical towards Kagi, not giving them the slightest benefit of doubt.
> You can't know what source code that is running on the server which you send your queries to.
Based on my (admittedly cursory) reading of the ideas behind it, the idea of the IETF standard they implemented is that one does not have to trust the server.
yes, even if you clear your cookies:
https://www.malwarebytes.com/blog/news/2025/02/google-now-al...
i quote: “With all these pieces of information, it’s possible to create a unique fingerprint by which websites can recognize you, even if you clear your cookies. They will even be able to make an informed guess if you visit the same site with a different browser.”
i urge you to consider how google makes money. it’s not at all surprising really.
Aha has this person never seen the JS google/CF run for bot detection?
The browser APIs spit out so much info about the host system on demand and that doesnt even consider the other tricks they use.
Duckduckgo and yandex are search engines.
I mean, it's pretty easy to use. It's a toggle switch in the UI.
If it is a switch in the UI, then that is just trusting the site.
If you trust companies with what they say they do, then why not enter your query in any search engine that does not require a login? Afaik none of them say that they will try to apply fuzzy fingerprinting to cluster your searches into a profile.
Honest question: Is private search achievable with your worldview?
Don't give them "your" email. Give them a mail alias, which I am sure as a privacy conscious person you already have, and you are good. They also recommend to do so themselves.
However, the most important argument here is not the fact that they are legally bound to those privacy commitments (they are), but that their business incentives are fundamentally incompatible with tracking users. For a very niche business with an extremely narrow and homogeneous user base, if they would get caught doing so, it would be game over. The privacy pass feature is available if you don't trust, and you can verify since everything relevant happens client side.
Maybe their privacy pass is useful then?
https://help.kagi.com/kagi/privacy/privacy-pass.html
Unfortunately the assistant officially does not work with Privacy Pass, for some inexplicable reason.
How is requiring an email "the opposite of privacy" when making a one-time disposable email takes like 5 seconds?