They run Leta on diskless servers, just like the VPN:
>We run the Leta servers on STBooted RAM only servers, the same as our VPN servers. These servers run the latest Ubuntu LTS, with our own stripped down custom Mullvad VPN kernel which we tune in-house to remove anything unnecessary for the running system.
>
>The cached search results are stored in an in-memory Redis key / value store.
This is surprising given that they try to cache results for 30 days:
>Each search that has not already been cached is saved in RAM for 30 days. The idea is that the more searches performed, the larger and more substantial the cached results become, therefore aiding with privacy.
That's surprising because presumably they lose all results if they have to reboot the server.
With a VPN service, there's not much they have to store past the lifetime of the VPN session, but if they're storing search results for 30 days, I wonder how they deal with this? Maybe best effort is fine because they don't strictly need to cache the results, as it just provides marginal privacy improvements.
"That's surprising because presumably they lose all results if they have to reboot the server."
Strictly speaking they only lose all results, FOR SURE, if they have to reboot ALL the servers at the same time. If they implemented a system where the cached results are shared and replicated among all their servers, it can in theory be kept cached indefinitely.
> Each time the Leta application is restarted (due to an upgrade, or new version) server side, a new secret hash is generated, meaning that all previous search queries are no longer visible to Leta.
If I read this correctly, the cached data is per-instance, there would be no way to share cached data among instances if each one has its own secret hash and they are cycled on each start.
Cost that's external, too: Brave or Google are behind the results. Things would be terrible without the cache... but that doesn't mean every request needs to be cached. Can't - gotta source it.
Wouldn't want to hang onto things too long, current events run out of currency :)
> This is surprising ... as it just provides marginal privacy improvements.
Diskless does not mean SSH-less or network-less. The "data" can be pulled / pushed just the same, which is to say, Diskless, in this case, is no better than verifiably read-only partitions (like on ChromeOS & Android, for example).
Diskless (edit: with OS in initramfs) is indeed a golden standard against local persistence, but requires quite a bit of extra RAM - few GB for "latest Ubuntu LTS".
With regards to preventing accidental persistence, disk with only dm-verity partitions is as good, with extra advantage of only adding a little bit of extra RAM usage (/tmp, /var/run, ...)
For that matter, even something as sloppy as booting with rootfs wich can't be remounted rw (iso9660, squashfs, etc..) and is the only mounted fs, is also perfectly good against accidental persistence.
So running a diskless host OS for a hypervisor and then diskless VMs on top of that? Sounds like a nightmare before even considering live migrations on top. Also what if they need to reboot the VM itself?
The cache is per-instance. A cluster of Redis caches would also limit the whole cache to the RAM size of one machine, so that is a non-starter.
Mullvad swinging for the fences suddenly. They have a billboard in South San Francisco, too. Did they get a cash infusion? Why all of the sudden are they expanding? Honestly, I'd have changed the name by now...
> Did they get a cash infusion? Why all of the sudden are they expanding?
No cash infusion. We've been growing for years, just like many other VPN services. We're still quite a bit smaller than e.g. Nord and Express though.
As for our choice of advertising, we don't run an affiliate program, nor do we want to track our customers through online ads, so we're trying this instead. It's cheaper than you might think.
Sorry for hijacking the thread, but I'm too curious not to ask: is having censorship circumvention out of the box a non-goal for Mullvad?
Because there are VPNs with good censorship circumvention tech, and there are VPNs with good privacy guarantees, but I know none which can provide both. What Mullvad offers now is either a decade old stuff which is blocked even by subpar DPI solutions, or a set of (more modern) protocol bridges which are painful to setup and sometimes IP-banned.
Mullvad's mission is to make mass surveillance AND online censorship ineffective. So yes, we do intend to offer excellent censorship circumvention out of the box.
Having said that we have clearly prioritized privacy for a long time. For what it's worth we have several censorship improvements on the roadmap. Stay tuned.
I already see shadowsocks which is nice. I'm still forced to use V2ray and xray-core in some rejoins though so I route traffic from my device -> xray -> my server -> wireguard mullvad. Works for now I suppose. Also been experimenting with routing small amounts of traffic through the syncthing relay network since they have relays running locally which may be in less restrictive provinces
Interesting. Try reaching out to Mullvad's support as well if you haven't done so already. If I'm not mistaken they conduct censorship circumvention experiments from time to time together with customers. I'm sure they'd also be interested to hear about any long-term resilient low-bandwidth channels you've found, such as the syncthing relay network. Those are very useful for bootstrapping and configuration updates.
Hey. Silly thought. I used to have the idea that Mullvad is the only VPN I trust because the founders seemed ideologically motivated (I guess from some interview I read, don't remember for sure). But advertising seems to undermine that view. Maybe I was just naive.
Hi! I used to think that the product should speak for itself, only grow by word of mouth, and that it was wrong to do any advertising. Part of me still thinks that.
On the other hand we ran a very political advertising campaign one-two years ago when we protested a new EU law proposal. We plastered Stockholm's airport in billboards targeting EU politicians and journalists. We published a book and sent copies to several hundred politicians. It was quite a success. Incidentally our office was raided by the Swedish police a month later - the first time in 14 years.
I really appreciate your feedback. Are you able to pinpoint more exactly why you feel that our advertising undermines trust in our brand? Is it simply the fact that we're advertising at all?
Our marketing team works hard to ensure that our advertising doesn't make security guarantees we can't keep, or sell the product through fear-mongering. I feel that we've found a set of advertising messages that work, but clearly it still causes some unease and skepticism.
Perhaps it's simply a worry that we'll change because Mullvad is growing up and is no longer an obscure underdog?
I really hope they don't change the name, I like the name "Mullvad" (Mole in Swedish) and "Leta" (Search in Swedish) and everything doesn't need to be Anglo centric in the appeal :)
Although the society is almost zero privacy, it have historically had some funny IT figures for privacy and digital issues so people searching up for the background of the name might stumble upon it.
I'm guessing they won't change the name. It's a similar branding strategy as ikea, with "funny" nordic (specifically swedish, but other brands have done it with norweigan and danish too) names that for some people makes it sound quaint and quality.
In (American, at least) English, there's a very common pattern of vowel reduction on unstressed syllables, resulting in "schwa-ification" [0][1] where all such vowels become indistinguishable from each other.
In this case, we say "duh lorr uhss" instead of "do lor ez". The second one doesn't sound like clitoris at all, but the first one.. okay it doesn't sound similar to me either, but it's closer at least.
I have to say that, the vast majority of the time, the way I've heard and said the word "clitoris" doesn't rhyme at all with "Dolores," so I wouldn't have been able to guess it either.
They were one of the earliest to adopt bitcoin and monero payments--if they didn't convert all those payments immediately to cash, they're probably sitting pretty right now.
They also have a partnership with Tailscale that can't be undersold.
I'm not sure how much it adds to their bottom line for each sale, but my corp was using the Mullvad VPN addition to tailscale to do global testing by our developers.
IE; "is something blocked, do we detect GEOIP properly" etc;
> there's definitely been a lottery win or a series A
We have neither won the lottery nor taken on outside investment. We've been growing for years, and we've reached a point where we can afford campaigns like this. It is an interesting experiment by our marketing team. Still, I think people on HN overestimate the cost of campaigns like this.
My concern is that when they can advertise to the extent they do, to what extent can they really be trusted? Anything that popular is going to be a target by law enforcement, and we really have no way of verifying any of their claims.
Yeah, this advertising to the masses push makes me queasy. It has the reverse effect on me than was intended. Weird brand self-harm for a privacy/data hygiene oriented company.
Same, but on the train at the DC airport. I liked that they align their actions with their mission. Physical ads like this are perfect way to advertise a privacy tool, as their ads respect user privacy.
I had to switch to iVPN last year (similar ethos), because Mullvad became pretty much unusable due to blacklisting and laggy DNS servers.
I'm assuming it has something to do with the push in recent years to expand their userbase, but they don't seem to be able to keep a clean enough pool of IPs like the big popular ones to cope. I know all VPNs struggle with this but it was getting ridiculous, where every single server in a country would receive infinite re-captcha.
iVPN is a great choice in terms of security, they also use STboot, but I think you're just flying under the radar with their IPs because they struggle with the same problems as Mullvad.
Yes, it only works better because the obscurity to IP ratio is good. It could easily be as bad as mullvad if they became more popular. But as I understand it the really popular VPNs address this with huge pools of servers and IP cycling?
One other issue I had with Mullvad that put the nail in the coffin for me was randomly laggy DNS resolvers, they would get fixed just by the time I start investigating it, but it kept happening... I say this as a mostly happy user for probably 7 years, but then found myself having to turn it off more than on to be able to access most sites.
> where every single server in a country would receive infinite re-captcha.
What does that even mean? Have you also disabled cookies?
Typically it's a Cloudflare captcha if you're doing that, not a re-captcha. And afaik pretty much everyone gets this treatment with zero history. Welcome to the modern web.
They’re referring to the situation when a service has blacklisted you, but will pretend they haven’t and give you captcha after captcha to keep you busy.
Yup, I found a shortcut to determining this is to use the audio option, which will instantly admit you are blocked due to "suspicious network activity" rather than make you solve stuff - i guess because of accessibility?
Cloudflare recently started holding stackoverflow hostage as well. "Weird" OS + "weird" browser + cookie autodelete = www is hell, even on clearnet. I hate cloudflare so much it's unreal, including everyone who works for them, for enabling this nonsense.
I think it's a new/old marketing strategy. Make it interesting enough that people see and notice it but don't understand what it is, with the hopes that you go out to figure out what it is. A brave strategy, but since it's still around, I guess it works sometimes.
It incentivises people to ask each other about it. But you need high pervasiveness of the ad for two people to both have seen it and ask each other about it.
I still didn’t know what it was. Went to the headline link and had no idea, typed in “what is this”. Still no idea. I had to read the hn comments to discover it is a search proxy..
> However, Leta is useless as a service if you use the perfect non-logging VPN, a privacy focussed DNS service, a web browser that resists fingerprinting, and correlation attacks from global actors. Leta is also useless if your browser blocks all cookies, tracking pixels and other tracking technologies.
In other words everyone can benefit from it. I don't know any browser (not talking about obscure browsers like lynx) who can completely resist fingerprinting.
Mullvad makes a fingerprinting resistant browser. It uses tricks like displaying the content inside a smaller window to mimic popular laptops and phones.
Pretty much the only way to use google search as an HTML webpage instead of a JS web application these days. It's great. It reminds me of the scroogle.com proxy days.
I use it for all but my retro machines, which is a shame. I know Mullvad is a 'privacy' company but I really wish they'd acknowledge that HTTP+HTTPS is more robust to governments' censorship than centralized CA TLS only. HTTP+HTTPS would allow my non-bleeding edge TLS retro machines to search again.
So how do they make money? Are they hoping to convert users to their VPN service? Or are they just trying to stay under the free tier Google API limits?
Leta is the supported search engine of Mullvad Browser which is a privacy-centric version of Firefox that integrates with Mullvad VPN. Think Mullvad Browser:Regular Internet as Tor Browser:Onion websites. So this is part of an ecosystem for their VPN subscribers.
(I'm a Mullvad customer, not Mullvad directly, but that's how I use their browser and Leta.)
In the past, Leta was a service that was only accessible to paid Mullvad users. I'm unsure when they started allowing general access, but that's initially how it made commercial sense.
Many workplaces use a corporate firewall, and on the admin panel, they can enable-disable categories of websites, like "Porn", "Adult themes", "Gambling", "Social", "Video streaming", "AI", etc. One of the categories could be "VPN", *.mullvad.net can fall into it, and it could be that they disabled that category. At many workplaces, it's against the rules to circumvent the company's monitoring, and so, many of such technologies are banned.
I don't think that .ai is automatically filtered in this case, it's more of a case by case basis. But it's killer nevertheless. "Adult themes" for example is a large umbrella at OpenDNS, and for example I wanted to check the lyrics of a song I was listening to, and it was hosted on darklyrics.com. Nope, couldn't visit, because it's Adult Themes.
These alternative search engines really feel like they're fighting the last war. Web content is so reader-hostile that you need a tool to extract the answer/information you're looking for and not just give you a link to the page.
I don’t actually. I have read far too many AI summaries where the llm combines data about two different people with the same name creating a biography of someone that doesn’t exist.
And once the use of chatbots in this role becomes widespread- don’t think for a second that companies won’t sso the thing until it’s about as useful as current search.
I had an issue where Slack AI combined multiple people I work with into a summary that was negative in tone. And of course there was no way to provide feedback on this harmful behavior.
Where does it say how it handles user information - what it collects, how long it's retained, what it's used for?
I would expect Mullvad to say they collect none, but is that said anywhere? Is there any privacy policy?
Edit: All it says is that they protect us from Google and Brave:
> When a search isn't in the cache, our server (leta.mullvad.net) queries the search engines on your behalf. Only the search query is sent; no personal data is shared.
and
> Returned search results contain only direct links to the final destination. All tracking elements and third-party content are removed to protect your privacy.
It's not a publicity stunt when they're using the technology they helped develop to run their search servers completely securely and without any stateful data.
I wonder how well the caching works. The FAQ says 30 days, so you might be getting a pretty stale result. That combined with Google's "fun fact: 15% of all Google searches have never been searched before", makes me wonder how identifying these queries can be.
This isn't really privacy or security focused unless 'trust' is a component of security architecture.
Make no mistake, Mullvad Leta knows what you searched for and who you are.
/Theater/ has no place in privacy.
The right way to do it, short of FHE, is to encrypt the query client side, pass this to the proxy which does not pass the source IP, which passes this to the search engine for decryption. Search results are encrypted and pass thru in the reverse:
Client (encrypts) -> Proxy (passes thru no IP) -> Search engine (receives, decrypts, performs, and encrypts results) -> Proxy passes encrypted blob of results back to user -> Client privately reviews private search results.
Edit: private.sh tried this in the past but unfortunately was shuttered with the end of gigablast.
Mullvad has built trust over many years. There is always someone who knows what you are searching for. The search engine will not accept an opaque blob of encrypted data as a search term, after all.
Agreed that the conclusion is that not all parties want to increase privacy. Thus there is at least one party that does not want to increase privacy. But we already know that google does not want to increase privacy. Thus this does not show that mullvad does not want to increase privacy.
If the encryption library is loaded over the web, then it provides no added security. You are still trusting them. Web client side encryption is theater.
This is a bit of an aside, but I see this take a lot and I think it's subtly wrong.
Web client side encryption eliminates fully passive snooping on the server side, but of course does nothing for actively subverting the served encryption code. This makes things a bit more dangerous for the snooping party as it's possible that the backdoored encryption code will be noticed by someone, and it's at least possibly a legal defense - the government might have the power to compel you to hand over data on your server but not to backdoor your code.
This isn't a huge technical difference, but it is a difference, and especially with the legal angle I think it's an important one.
Awesome. Maybe it's just my imagination, but it seems like there is much less crap and the relevancy seems to be much higher even then I choose Google as the underlying engine.
I don't care much about that anymore because their VPN service has really gone bad. They are great in terms of privacy, but in every other aspect, they suck. Their VPN randomly disconnects again and again, once even without the killswitch being activated. They are getting blocked from websites much more often than other VPNs, making the service barely usable while costing a lot more. Plus, there are many other minor issues. I really hope they improve because I want to keep using them
Disabling Wireguard obfuscation and quantum resistant tunnels fixed the disconnects for me, which is fine for my use-case, but they shouldn't be enabled by default if they're causing issues
Brave has a subscription tier that offers storage rights. But it's ~9x the cost of their normal Pro subscription. I have a hard time imagining that the cost works out in their favor (discounting the possibility of a special arrangement) with how long the query stream tail is in web search.
I don't understand why Google or Brave are cooperating with this, they don't earn anything. And if they're not, what prevents Google blocking Mullvad IPs?
Leta used to require a VPN subscription. They probably figured the cost weren't that high and the possible increase in future customer by offering it for free would at least cover or even outweight that cost.
Interesting solution to let the user pick which search engine to use. Sadly Bing is shutting down their API, it would have been great to be able to use that as well.
The "search engine in question" is using the google api. If the search engine in question serves it means that google actually serves it. Thus I am a bit suspicious of the motivation behind this kind of question; why do you pose it for mullvad and not for any other search engine?
I don't know anything about the subject. But in my mind it kind of made sense that serving illegal stuff is somehow worse if you can't identify the users (eg if police has a warrant) compared to if you can.
This would've been a great product 10 years ago. I've unapologetically not had to use a search engine in almost a year (or at least can count on 1 hand having to use it) since GPT models have come out.
What do you use to look for products, and businesses? I also use chatbots much more, but these are two categories where I found search engines to be much better. But I haven't really looked for an alternative either.
More than once people at work have asked me for help after not solving their problems with ChatGPT, and the solution was to google and hit some stackoverflow answer.
The situation hasn't changed for most of us. None of the people I've talked to over the past couple of years have stopped using Google, none are using LLMs for anything other than translation (or helping proofread their English) or simply for wasting time.
FWIW, since we're exchanging anecdotes, LLMs have been completely useless for me. I try them every 3-6 months and always return to Google disappointed.
What they're saying is that if you had already closed off all possible methods of being tracked, you'd gain nothing from this service, since whatever other search engine you choose to use instead would by definition not be able to track you.
I assume if you block all tracking technologies it does not offer anything more than what you already have? Because then your queries cannot be tracked?
Doesn't !g just redirect you to Google? From comments, this is proxying and potentially caching from Google. Having an intermediary is potentially of value.
Right, one that by my understanding “pools” searches, in a way. As their blog put it in 2023:
“Mullvad Leta uses the Google Search API as a proxy, caching each search. These cached results are shared amongst all users, reducing costs and improving privacy. This service is user-supported and doesn't rely on ads or data selling.”
That short period of time is likely to pool a group of users. Even if not, using Mullvad as a personal proxy for Google is a better privacy-conscious decision than using Google directly.
I presume that's because calling it privacy-focused is considered editorializing. I'd at least hope it can have "(search engine)" or similar because I had no idea what it was before clicking.
>Browsers, OSes and Knowledge Engines (search, llms) should 100% know about my tastes, my intelligence level, my interest level and absolutely render output based on my needs and input/output preference.
What about the people that own the "[b]rowsers, OSes and....." etc, etc.
I was dumb enough to buy more than 30 days worth of mullvad once. They changed their terms of service to remove port forwarding. Because I'd paid more than 30 days ago, they wouldn't refund me anything.
Screw mullvad. I'd have to be a damned fool to to ever trust them again.
I don't hold it against them, but I got burned by that change too; but it was entirely reasonable, allowing inbound provides abuse opportunities which degrade their primary service reputation.
Totally agree. Everything should be Americanized as much as possible so that it conforms to American sensibilities and is easier to use and understand for Americans.
Who cares about languages and culture of few dozen people who does not live in AMERICA.
Am I (native Swedish speaker, so perhaps ignorant of secondary connotations here) missing something that should be obvious? Is mullvad inappropriate to some readers or is it just an odd name?
No, I don't think you are missing anything. As an English speaking American, it just strikes me as a strange name that I wouldn't immediately associate with a search engine. Note for example that the parent spelled it wrong despite that being the focus of his question. But there is no second level of meaning or innuendo that I'm seeing.
>We run the Leta servers on STBooted RAM only servers, the same as our VPN servers. These servers run the latest Ubuntu LTS, with our own stripped down custom Mullvad VPN kernel which we tune in-house to remove anything unnecessary for the running system. > >The cached search results are stored in an in-memory Redis key / value store.
This is surprising given that they try to cache results for 30 days:
>Each search that has not already been cached is saved in RAM for 30 days. The idea is that the more searches performed, the larger and more substantial the cached results become, therefore aiding with privacy.
That's surprising because presumably they lose all results if they have to reboot the server.
With a VPN service, there's not much they have to store past the lifetime of the VPN session, but if they're storing search results for 30 days, I wonder how they deal with this? Maybe best effort is fine because they don't strictly need to cache the results, as it just provides marginal privacy improvements.