Jerod Santo: I feel like you are attacking me personally at this point, because I am resistant… Well, as Brigit will probably point out,

Codespaces is very much in the cloud, insofar as you’re literally running it in a browser tab, right? And that’s one thing. But what this dev container thing provides is not just - it’s cloud-based coding, but you’re still local with your VS Code, for instance. And of course, as we talk about the spec and what your team’s desires are for this spec, it’s gonna be more than just VS Code, it can be more than just Codespaces… Of course, Microsoft is really front-running this idea and this spec…

I do think it’s probably futile, because there’s so much upside to the containerization, to the repeatability, to the sort of like your dev infrastructure as code, like having it all right there… And then just teamwork. Teamwork tends to be what makes the dream work. I don’t know; what do you think, Brigit? Is it futile?

Jerod Santo: Puter is the internet OS

Puter is an advanced open source desktop environment in the browser, designed to be feature-rich, exceptionally fast, and highly extensible. It can be used to build remote desktop environments or serve as an interface for cloud storage services, remote servers, web hosting platforms, and more.

I’ve been around long enough to see a bunch of “desktop OS in a browser window” demos and toys, but this is the first time I’ve been impressed by one long enough to keep the tab open longer than 30 seconds. From the URL structure, to the cloud storage integration, to the developer portal… Puter strikes me as an actually viable internet-based operating system with potentially real-world use cases. And that’s saying a lot!

Oh, and it’s almost entirely built with vanilla JavaScript and jQuery… so you know the devs haven’t cargo-culted together something they can’t grow & maintain.

For performance reasons, Puter is built with vanilla JavaScript and jQuery. Additionally, we’d like to avoid complex abstractions and to remain in control of the entire stack, as much as possible.

Also partly inspired by some of our favorite projects that are not built with frameworks: VSCode, Photopea, and OnlyOffice.

Nader Khalil: Yeah, so it’s kind of funny… Before we leaned into the AI/ML workflows pretty heavily - you’re right, dev environments… Who is your target audience? People who code, right? And that’s kind of a very naive answer, for a very early stage of the product.

I think what we learned is you really want to be able to solve someone’s problem as quickly and acutely as possible, and then get out of the way. And I think that’s been a big change in direction for us. Even if you look at like the way that the product onboards - you need to have a CLI so you can run Brev open. So we used to say, “Oh, well, when you make an account, we’ll tell you to install the CLI right there.” But the user doesn’t know yet why they want to install the CLI. They haven’t had an expressed desire to open their dev environment yet. So the way that we change it is it’s just focused on getting your environment created. When your environment is created, then you see an Open tab. When you click the Open tab, there it tells you “Install the CLI, because you haven’t yet.” So the user says “I want a thing”, and then we can kind of show, and not really impose.

And so when we were thinking about like broadly dev environments, when we initially started building brev, it felt like we – you know, someone says, “Hey, my local environment is not working.” And so we’d say, “Great, we can make one for you in the cloud.” But now we’re not just introducing Brev as a tool to solve their environment issues, we’re also introducing the cloud. It’s a separate thing. And so in terms of like acutely solving the problem, we’re not doing that; we’re introducing the element of the cloud, which they have not yet expressed a desire for.

So what’s great about the GPU use cases is we’re meeting people where they are, which is in the cloud, right? They’re saying “I am trying to access an A-100 that does not exist on my MacBook Pro, and I want to get this running”, right? So the cloud intention is coming from them, not us. We’re not kind of like sneakily trying to introduce something else so that we can get them to use brev. It’s just meeting the user where they are. And the issue with using a GPU in the cloud is that they’re really expensive, and they’re really painful to get set up, and then, of course, all the dev environment issues.
So that’s been a really great focus for us, and we’re leaning in as hard as possible to the MLOps tooling; the dev environment issues are much more severe here, it makes a lot more sense for – there’s a lot more room for us to delight users by making a much better experience.

And going back to that container strategy - if we can move between different clouds, we can also move between one local cloud, which is your actual computer. So I think the way that we kind of want to approach broader dev environments is you should be able to run something on your computer and then say, “I now have a need for the cloud. I want double the RAM. I want a GPU. I want something.” So you can start local, and then move it to a cloud. And that’s the way that I think we can ultimately broaden from ML dev environments. But this is a huge focus for us right now, and what I really want to do is, rather than think about so many of those other use cases, how do we get really tight integration with Banana? How do we get a really easy way to go from a collab notebook to something that you’re now fine-tuning on a much more powerful GPU? How do we find an interface with other clouds? That’s where we’re focused right now, and there’s a lot of work to do here.

Adam Stacoviak: So Jason, you mentioned Typesense Cloud for the first time in this conversation… Now, I assume that I see a pricing tab, and this is hosted, this is your ability to make money, your ability to resist venture, potentially attract venture… This started as a nights and weekends project… How did you get to – did you ever think you’d be here? You know, launch cloud, and self-fund… What’s the story there?

Michael Jackson: Yeah, you were saying earlier - we call ourselves kind of center stack, and I think this really kind of gets to it… Because we’re not super-opinionated about where you host; we love Fly, we also love AWS, we love Google Cloud, people deploy on Vercel, Netlify… It’s totally cool. We’re kind of backend-agnostic, we’re database-agnostic… But what we do care about is the network tab, which is the thing that tends to come in between the backend and the frontend.

Kelsey Hightower: I can almost guarantee. Guarantees are bad, because people come back and listen, like “Dude, you were so wrong…” [laughter] “You were overly-confident.” I am overly-confident that we will go back into a world where there will be one pod/VM. But to think we call a virtual machine is dramatically changing. I think Amazon hinted to this with things like Firecracker. We saw Firecracker come out… And for those that know their history, Firecracker comes from the Chrome team, where they decided that they want to isolate things at this level.

When you can isolate things at this level almost using some of the machine-level isolation, then you can actually isolate a web browser tab, Amazon is isolating serverless workloads… Now you’re in a really interesting spot, because guess what, one thing we already have - every cloud provider has a really nice hypervisor layer that most people don’t see. We already have things like VPCs, Direct Connect… So imagine saying “Let’s strip the kernel down by 80%.” Strip down the virtual machine; no floppy disks, no things we don’t need… And then just pack the container or the pod in there.

That means you could be using the Kubernetes API, but all you’re really doing is launching these lightweight virtual machines connecting to the existing stuff… And now we can take the eight years of advancements in the cloud, or VMware on-prem, and pair it with all the beautiful things we like about Kubernetes.

Brendan Eich: We don’t know. That’s a great question. I think among the early adopters, lead users, yes they get it. A lot of them are outraged by the malvertising stories that broke this spring. And it was really great for us, because we had the late March malware on the front page of New York Times, then we had on 7th April… I woke up and there’s a letter from The Newspaper Association of America, counsels us to cease and desist, but we haven’t anything yet to actually cease and desist; those words don’t occur in the body, but it’s full of threats and crazy legal theories, including that these Newspaper Association of America members own the copyright on those ads that we would be blocking. How could they own that, because it’s malware from Russia, or whatever? They don’t own the copyright; those ads are injected by JavaScript in your browser, running on your page, communicating with third party sites with ad exchanges. Nothing to do with New York Times. There’s no creative work, ensemble work that has the ads.

I think the lawyers - it’s generally the associate GCs that join these trade groups, like Newspaper Association of America, now called The News Media Association… Newspapers have been in decades long of decline, but they view the ads as ink on paper. It’s like we’re sneaking up to grandma’s porch and we’re facing the ads that they printed on the Sunday New York Times and we’re pasting up our own ads to trick grandma into transacting with our advertisers and us getting a piece of that action.

First of all, we didn’t do any such thing. We only talked about how it can be better if we did something like that. Second of all, there’s no ink on page ad the New York Times owns. The ads are third party, they’re placed with JavaScript. Another one of my guilty legacies with JavaScript is how it’s used for third party ads.

[51:02] There’s really a deep topic here. Will people appreciate it? I think mainly people appreciate speed in browsers, they appreciate safety, and we’re leading with those. Safety is a broad term, but I include privacy. People say, “Oh, you can’t market privacy”, but you can. Snapchat built up a good cohort doing disappearing messages. People care about things like secure communications. WhatsApp’s doing end-to-end encryption.

People care after a crisis. Snowden changed things for a lot of people. I think as things evolve, we’ll have more concern about privacy. It’s often driven by crises and revelations. People just didn’t know they had a problem until they had one. So we don’t need to get too detailed on the economics, but I wanted to paint a picture because there is a lot of money exchanging hands here, a lot of middle players taking big cuts, very little for the publisher.

Brave cares about users first, and we think user attention is not fairly priced. We care about publishers, too. If you can’t keep a website a going concern, the web’s in trouble, so we’d like to see publishers get paid better. That’s where we think, if we get the right experiments done with user opt-in and publisher opt-in, we could build a better (I almost wanna call it) promotion system. The idea with advertising online now – Joe Marchese, founder of TrueX (I think Fox owns it now) said this: “You’re shotgunning people’s attention across ten thousand pages.” That means you’re wasting a lot of money, because first of all a lot of people guessed wrong, they didn’t go to that site. Then you’re retargeting them, which bugs them. You cross the line and they get an ad blocker. They’re lost to you. What if you could just get the right information at the right time, in the right place, to the person who’s likely to actually benefit from it and be happy with that marketing information? That’s the ideal model for advertising.

It solves what’s called “Wanamaker’s dilemma.” There’s this guy Jude Wanamaker who had a chain of department stores in Philly a hundred years ago, and he is alleged to have said - at least if I can get the quote right; it’s not clear if he actually said this - “My problem with advertising is half my advertising budget is wasted, I just don’t know which half.” Even then, he was shotgunning newspapers or catalog ads, and some of them missed the target.

Theoretically, with a very private system like Brave where your data is kept on device - we don’t see it on our servers, we use zero-knowledge proofs to transact things like payments for donations or ad impression counts in aggregate; theoretically, you could keep that data secure; you could keep your own Facebook, your own Google, you could do your own ad business. It would be a very personal ad business; it would be a “right information at the right time” business. It would not be replacing one-for-one all those indirect ads that we block. It might even be using a different channel, like a full-screen video channel or a set-aside personal mall; some people might prefer to get an email once a week with promotions. These would be really well targeted, they wouldn’t annoy you, they would give you a deep discount, because the marketing side wouldn’t have to spend for those 10,000 ads, half of which or more (maybe 90% or more) miss the target.

[54:03] That’s the big idea with Brave. It goes to search too, because when you search with Google and Google does that great result - they’re better than Bing, as I said; they’ll probably always be better. They have the oldest data set, they have the oldest machine learning that’s co-evolved with it. But what about your keywords that you type in? That’s your data. Again, Brave’s point of view is you own your own data. Not just your browsing history, what’s visible, how you open the tab from another, where you are scrolling, but also your keyword queries to search engines. And that’s a very hot data set that you should benefit from and we should protect on your device. So we’re looking at the whole picture. And when I say anti-Google, I don’t mean that in a hostile way, I mean somebody needs to build this. In a coming world where AI is everywhere, do you really need the cloud superpowers owning all your data? From your house, your cat, your own body monitors… I think there are scale advantages to the cloud and to clustering AI calculations there, but a lot of it is personal, a lot of it could be done in your home server, or even on your phone. So there should be tiers of AI and machine learning and tiers of data, where some of that data doesn’t even leave your device. Maybe only abstracted summaries or anonymised summaries leave your device. That’s the really big vision here, and I think people will build this. I see more signs startups are doing this. Instead of building some surveillance device based on cookies or search or everything in the cloud, they’re doing local computation and doing things that can be defensively secured in your pocket or in your house. That’s where Brave gets in.

Guillermo Rauch: …so that you don’t have to type it in from scratch in a prompt that doesn’t even render characters, because terminals normally mute characters. So we can see how even seemingly deficient defaults that the terminal has… Like, when you type in a password, the prompt goes mute and you don’t even know what you’re typing - now all of a sudden we leapfrogged and we made it so easy that you’re not even typing.

[01:03:57.26] I think by continuing this mindset, you can imagine that a lot of other modifications can be made to this workflow, or enhancements can be made to this workflow, some of which could even involve the internet. I don’t know if anyone has explored it yet, but for example I mistype a lot of commands, and what happens immediately after I mistype a command is I type it correctly. And once again, if we analyze the output of the terminal, it’s fairly easy to understand that a command was wrong because you get something like “Command not found” and then you see that the user types in a command that has a Levenshtein distance that’s not too far away from what he typed before, and it succeeded. So you can built your own little dataset of spell check, so that you can render some inline suggestions, or perhaps a keyboard shortcut for typing in the right command when you got it wrong. That could involve the cloud, for example. It could involve saving a dataset in the cloud, or logging could involve saving your logs in the cloud. So all of a sudden it’s now possible to go in and do XMLHttpRequest, or do open real-time connection to get some data, or change the behavior of the terminal as different actions or different things happen. You could imagine a collaborative terminal experience, right?

So something we’re adding to further expand on this notion of HyperTerm being a platform is we’re adding a little dropdown, non-invasive, maintaining our simplicity, a little dropdown that you see when you hover the tab, just like you see the X sign, where if you click it you would see options. And those options will be populated by plugins. So you could imagine, “Turn this session into a collaborative session”, and you get a link that you send to your co-workers and they can see what you’re typing on the terminal. You could imagine a lot of different additions that are not very invasive, yet they can enhance a lot of your workflows.

Gerhard Lazu: I took my sweet, sweet time… And I even did like a Cloudflare Pages experiment. I was using Wrangler, by the way; that’s up and running, you can go and check it out, nightly.changelog.place. Nothing to purge. I also did like a Cloudflare R2 experiment. There were like some issues with index.html… And if you go to this pull request, you’ll see – obviously, there’s a Dagger pipeline, which is captured as code, as Go code. There is a GitHub Actions workflow that runs all of this… There is an NGINX conf. So we are using NGINX to serve this. There’s something called Super-chronic, which is a modern way of running a cron tab.

Sunil Pai: And oh, it killed me, but it was great, because by day two, day three, I had all the [unintelligible 01:14:33.29] I had the pitch very clear, I knew how to do it. And I knew what I was raising for, I knew where the money would go… I was so clear about that. I have a tab open in my editor with a blog post for how to raise a pre-seed round, by the way, like “Here are the three things you need to do.” I’m going to post it I think within the next few days.

So don’t get me wrong, I was still super-surprised and happy and grateful at how well it went. I was like “Oh my God, I have money in the bank in 2023.” My leader, Sequoia, is such a great firm. I have such a great partner in Sequoia, Bogomil… He’s in the US, actually. I had to register – I feel very blessed and grateful for the cap table that I have. The CEO of Cloudflare is on it, the CEO of Vercel and Netlify is on it, Sequoia is on it… And some great, great angels, and smaller funds. And I really, really need that for the future of this company. Like, if it has to do well – it’s a journey. I have to pick these people. So that happened… And now I get to spend it on building a company. I’m happy to answer more questions, but that’s sort of how it happened.

Ryan Dahl: That’s right. Of course, you can take Deno - the executable that you download, that’s free - and you can wrap that up in a Docker and you can send that out to AWS Lambda and kind of have a very similar experience… But that is not what we are doing internally. We are not running Deno in some Docker containers. We actually are kind of using the V8 isolate… We were talking about how this is a secure sandbox before. So instead of having multiple Linux VMs running for all of the different tenants on this web server, we actually have all of these little V8 isolates running. And these, as we mentioned, are secure sandboxes, so we don’t have to worry too much about hackers being able to see other people’s requests and whatnot… And the nice thing about these is they’re super lightweight.

This is essentially like opening a tab in a web browser. So think of a web server with a bunch of tabs in it, and each of the tabs is for a different tenant on the system. We can respond to requests very fast. We have cold start times in less than ten milliseconds. I’m not sure the exact number. And we do all sorts of nice optimizations.

Obviously, it only runs JavaScript. This is not a Linux VM, so you cannot run your Python process in this. But because we’ve built explicitly on top of V8, we can make all sorts of very nice optimizations, like Cloudflare has done, to speed the response time.