1043: I’m Using GPUI For Everything

Scott and Wes ditch Electron and Tauri for GPUI, the GPU accelerated Rust UI framework from the Zed team. They get into its React-like API, styling that feels a lot like Tailwind, GPUI Kit components, the community fork drama, and the tiny desktop apps Scott has built with it. Show Notes 00:00 Welcome to Syntax! GPUI 01:24 Electron, Tauri, and the Trouble With Web Views 04:05 Enter GPUI: Why We Don’t Take Money to Talk About Tools 05:32 Brought to you by Sentry.io 05:59 What GPUI Is: GPU Accelerated, Cross Platform, Native All the Way Down 06:50 A React Like API in Rust: Divs, Flex, and Chained Methods 09:35 Components, Props, and State in GPUI Scott’s GPUI Demo Gist 11:21 How Much of CSS Does GPUI Actually Reimplement? 12:52 Flexbox Everywhere, But Where Is CSS Grid? 14:17 Where Does Your App Logic Live? No RPC Layer Needed 15:53 Render vs RenderOnce: Stateful and Static Components 16:57 GPUI Kit: The shadcn of GPUI 18:46 GPUI CE: The Community Fork and Ecosystem Tension 21:29 Apps Scott Built With GPUI: Gendo, File Bro, and Mr. Space 22:48 Why Tiny Utility Apps Should Stay Tiny 24:25 File Bro: File Automations a la Hazel 25:27 Mr. Space: Building a macOS Window Manager 27:05 The Downsides: No Native UI Elements and No Browser DOM Scott’s Rust Best Practices Skill 29:56 Testing GPUI Apps With Jev and the Accessibility Tree gpui-agent 30:39 GPUI Mobile: Experimental, But It Works GPUI Kit Mobile Docs gpui-mobile 31:52 Final Verdict: GPUI for Personal Desktop Software Hit us up on Socials! Syntax: X Instagram Tiktok LinkedIn Threads Wes: X Instagram Tiktok LinkedIn Threads Scott: X Instagram Tiktok LinkedIn Threads Randy: X Instagram YouTube Threads

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Scott Tolinski?

  Goodbye Electron, goodbye Tori, goodbye desktop apps using web views. I'm a web guy. I love the web, but I'm building desktop apps with Rust and I'm not looking back. In this episode, we're going to be talking about GPI, what the heck it is, how it can be the desktop application development workflow of your dreams. And a little bit about how similar it is to React. We're going to be looking at some code. We're going to be talking about some libraries. I'm going to be showing some stuff I built. If you have not heard of GPI, or this sounds interesting to you, this is the episode for you. We're going to be going deep on GPI. It is sick. My name is Scott Tolinski. I'm a developer from Denver. With me as always is Wes Boss. What's up Wes?

• 0:42Wes Bos?

  Very curious to hear what you have to say about this because I know that you were Mr. Tori for a while. And I know you dip into quite a few different things. This is from the folks at Zed, we should say that. I'll let Scott explain it. He knows a lot better. But very curious to see, is this something Scott is actually going to like? Or is this just something Scott's really excited about and then he's gonna... Because you were really excited about Tori for a while and then you're like, ah, you know what? Maybe it isn't, isn't it?

• 1:12Scott Tolinski?

  Let people be excited about stuff, Wes. I'm excited about things.

• 1:15Wes Bos?

  I'm glad you are because that's how we get good things, right? Like you dive in, you figure out what the edges are and then the world is a better place.

• 1:23Scott Tolinski?

  Yeah, let's talk about that because so Electron uses a web view. Electron uses like a Chrome instance and then it does everything in TypeScript or Node to, you know, run your, your actual server side of things.

• 1:39Wes Bos?

  Yeah, your backend of your app.

• 1:40Scott Tolinski?

  Your backend, yeah. And the sell for Tori was that, ooh, it's Rust on the backend, right? But the, the benefits come from the fact that it's using your native web view. And now Tori, you can use any web view you want now. It can use Chrome if you wanted to too. But the native web view part seemed like a major sell because you're not having to ship a browser web view with each app. One of the things that Electron gets chunky with is that the apps are shipping an instance of Chromium with the app. And then also Chromium can get really costly, man, when you're, when you're not working in it well. Now, we've seen like some apps like VS Code use Electron very well.

• 2:25Scott Tolinski?

  And in fact, if VS Code is slow, it's usually a third-party extension, not VS Code itself. So Electron can work and can be very fast. But in my process here, I found that the, the native web view side of things is not it. Especially on macOS with Safari, the versions, like the, the feature set sucks for desktop apps on WebKit Safari, like, or WK, WK WebKit? I think that's it. WK, no.

• 2:52Wes Bos?

  WK?

• 2:53Scott Tolinski?

  WebKit web view, web, whatever, web, webkit web view, whatever it is. Either way.

• 3:00Wes Bos?

  WK web view, you're right.

• 3:02Scott Tolinski?

  Yes.

• 3:02Wes Bos?

  I just looked it up.

• 3:03Scott Tolinski?

  WK web view, not feature-rich and not good for desktop applications. And very unpredictable in terms of who has what features because you're not shipping a specific version, right? With Electron, you can ship a version of your choice of Chrome that you know the features will be there for every user who downloads it.

• 3:22Wes Bos?

  And what most people do with these Electron apps is they flip on some things that are either experimental or they'll modify Chromium in some way so that there is support for what they want, right? Because there are inherent limitations of the web browser. So if you can sort of turn some of those things on, like one example, HTML and Canvas, huge, huge one. Like for, what is it? Remotion is the JavaScript video editor, right? They can render it out for you. If they, they couldn't do that if they were using the native web view because that would be Safari in that case. And you have to turn on HTML Canvas.

• 4:06Scott Tolinski?

  Yeah, yeah, totally. So enter GPI. The Zed team, when we, and actually I will say, Wes, this is a pure coincidence. I'm wearing my Zed hoodie today. I did not do this on purpose. I'm very like, this makes me look like I'm sponsored here, folks. This is not sponsored. They've paid me no money. They did give me this sweatshirt, but only because we had talked about Zed in an episode and they liked that. So we don't, we don't take sp- I've been seeing that more and more, Wes. Anytime we publish a video, people are like, you got paid to talk about StyleX, which is an open source CSS library from Meta. Who cares? Why would we get paid to talk about that? That's crazy. That's the most insane one to me.

• 4:44Wes Bos?

  Well, let me tell you, like, congrats or like, like clap to everybody who is suspicious because the amount of DMs that I get from people who are like planning a launch of some crappy-ass AI project and the amount of money. Like I just got a message from somebody, they're offering $1,500 just to quote tweet one of their launch posts. Like, and most, I'm not going to do it because I tried, I tried the product, it sucks. But like, like a lot of people will be like, I buy, I buy a new MacBook. I do two of those, you know? And like a lot of people will sort of throw that, throw that out the door. So good on you for thinking that type of stuff.

• 5:27Wes Bos?

  But I, I hope that my integrity is, is worth a little bit more to you guys, you know? Like I don't, I don't want you guys to stop watching because we're shills for something. Except for Sentry, sentry.io forward slash syntax. You should install it in your application.

• 5:42Scott Tolinski?

  Yes, and that's an important point. We do not accept money to talk about libraries, tools, whatever. Except, we are a part of Sentry and that's it, straight up.

• 5:53Wes Bos?

  Yes.

• 5:54Scott Tolinski?

  So you can be confident about that. GPI. They did give me this hoodie. It did not make me talk about Rust or GPI. Okay, so GPI is sick because basically it's a UI framework that is GPU accelerated. It is cross-platform, so it runs on Mac, Linux, Windows. It even runs on the, the web and mobile, but we'll talk about that in a little bit because it's not something maybe that I would do. So you could think of it as like an alternative to building UIs in like, you would use it in place of a web view in Electron or Tori. You wouldn't use it with Electron or Tori, but you would use it instead of using Electron or Tori to build a web view. Now, why would you do that?

• 6:37Scott Tolinski?

  Well, it is sick. It's fast. Like I said, it's GPU accelerated. It is native all the way down from Rust rendering to the GPU. Rendering to the GPU, rendering with the GPU, yes. So let's, let's get into it here a little bit about how this looks because it's a very web and dare I say React-like experience, which is going to make web developers happy. Again, you aren't shipping a web view for this. You're not using the DOM. You're using just these primitives.

• 7:09Wes Bos?

  Like, like, to be clear, if anyone's sort of trying to place where this is in their thing, like this is not something that converts it to native, you know? Like it's not like you, you use a GPI text input and then it converts it to like, whatever the Swift equivalent is. Like they literally wrote the entire, like you start with a blank square and then you, you go up from there. You got to implement like text wrapping and borders and shadows. And like you have to implement literally everything because they've, they've invented an entirely new way to do UI from scratch.

• 7:43Scott Tolinski?

  Yes, and it feels like I said, very React-like, but I think that's an important point because if you are building native, native apps on desktop, mobile or whatever, you're often using like default components and then modifying them, right? Styling them, et cetera. Where this is like, you're using the GPI version of everything. Now, the, the way that this works is that there's a, a render, just very, like a, a render function similar like there is to React. And everything is done in sort of like methods and functions rather than, it is declarative like React, but you're not using the same style of React bracket style components. So you would have like a div function with a dot flex method on it

• 8:27Scott Tolinski?

  for a flexbox style thing and a dot flex column to do flex column. So the CSS ends up being very Tailwind-like in a way because you are really doing inline styles on individual divs or wrappers or things like that. And the shorthand is very, I dare I say, Tailwind-like. There's a lot of like helpers, like text XL and stuff like that, text color. Imagine that. I actually like that, Wes. The text color is named text color and not just color. That always bugged me about CSS.

• 9:03Wes Bos?

  Yeah.

• 9:03Scott Tolinski?

  Oh my God.

• 9:04Wes Bos?

  Well, you could say, well, actually it's because it also controls like, like border and several other things, but who cares? It should be text color. I agree.

• 9:12Scott Tolinski?

  Yeah, you're right. You're right. It does. It does change the color of a lot of things. You're right. But I think getting started in CSS, text color was always a weird thing. There's the concept of children. Like I said, it's very, very React-like in its presentation when you look at it, or at least React-like in the sense of you have components, you have props, you have state, you have those types of things. So I made a quick little application here. If you're watching on video, I have this pulled up. If you're watching on audio, we'll do our best to describe it. But basically you define a component with a struct. It has props, very much similar to how you have props in TypeScript. In React, you're defining the props, the types of them. You have a render function where again, you have things like divs.

• 9:57Scott Tolinski?

  And then you can pass props into components. The render for the counter, which is a component I have here, it is able to have a button with an on click. And then you have a new listener and then you're incrementing the counter. This should all feel very similar to React. The big difference here, and maybe not the big difference because there's a lot of big differences here. But when you update state into something like that, you have to run notify. And that notify is a method on the, the general app context. It's like remix. I was hoping you would say that.

• 10:35Wes Bos?

  Right? It just, you call the update function yourself in, in remix where, okay, I've done some stuff. I've mutated some variables. Now I want to re-render this specific piece. So you explicitly do that. Is that what, that's what notify is?

• 10:50Scott Tolinski?

  Yes, exactly, Wes.

• 10:52Wes Bos?

  Okay.

• 10:53Scott Tolinski?

  I was like, I put this in here hoping that you would say it's like remix. And I'm glad you bet. Really cool, you know? And in general, like, I know if you haven't used Rust before, some of this is gonna look very foreign or very intense. If you've written Rust before and you've written React, you're gonna be like, oh, oh, I get it. Because if you've done the two of them together, the little Rust things aren't going to be surprising. And then the React stuff is gonna stick out very, very obviously.

• 11:21Wes Bos?

  So have they reimplemented most of what we expect from CSS? You know, like, I saw flex, I saw, obviously, padding, borders, text color, focus. So have they just basically reimplemented everything that you need?

• 11:39Scott Tolinski?

  So far, I've been able to do the stuff that I wanna do. As far as like shadows, opacities, textures, backgrounds, repeating backgrounds and stuff like that. I'm sure I will hit edges as I try to do crazier stuff. But, you know, like, as far as like animations, here's an example in their docs for doing a rotating SVG. Let me find this. And in the syntax, I will say, folks, if you are like, coming from a React and HTML world, and you're just chaining a bunch of methods for things, like, this is tailwind class bloat on crack. Uh, can you say that? This is tailwind class bloat on steroids. We'll say that.

• 12:21Wes Bos?

  Yeah, cuz it's one per line, and you're chaining it, almost like, like jQuery chaining your methods.

• 12:26Scott Tolinski?

  Yes. So you get big, old, long files. It doesn't necessarily matter that much to me, because agents are often writing this stuff to me. And I find it to do a good job. There's like, with Spring, to have Spring animations, things like that. So in my experience, it's been very full-featured for the types of things I wanna do in CSS. Now, does it do everything in CSS? I can't imagine that it does. But so far, I'll show you some maps that I'll go with it.

• 12:52Wes Bos?

  Does it have Flexbox and Grid or or Grid? Or what's the layout thing? Because that's the one thing about React Native, is that they don't have CSS Grid implemented, right? You gotta do everything in Flexbox.

• 13:02Scott Tolinski?

  Yeah. Where's Grid? Grid, grid, grid, grid, grid. I'm trying to, the docs here, what I need is a big, old Reference docs, is what I need.

• 13:10Wes Bos?

  Yeah, the doc, there doesn't seem to be docs. There's just a bunch of examples.

• 13:15Scott Tolinski?

  I have mostly used Flex to do everything, so, well, that's a good question about Grid versus Flex. Yeah, I'm seeing just Flex everywhere.

• 13:24Wes Bos?

  Yeah, me too. I'm not seeing any Grid. So that's a bit of a limitation.

• 13:29Scott Tolinski?

  I guess, but, I mean, it's still like, I don't know, before CSS Grid, you were doing everything in...

• 13:35Wes Bos?

  Yeah, we were able to do a lot of these things. There's probably like, something like a subgrid or whatever, you'd run into, be like, ah.

• 13:42Scott Tolinski?

  Yeah, I'm sure there...

• 13:43Wes Bos?

  But, and again, you could probably also measure things, and you're in the Rust code, right? So you could probably do a lot of the stuff programmatically. Or, like, here's another question, is can you implement your own things? Like, if you wanted to have fancy border?

• 13:59Scott Tolinski?

  See, Wes, I knew you would ask me a bunch of questions that would stump me. I've, the apps that I've built, I'll show you, the apps that I've built have been very vanilla-looking. So, I mean, I've styled them to a degree. But you're always one to push it in the place of fancy border.

• 14:17Wes Bos?

  Okay, well, here's another question, is, like, obviously handles layout, handles styles, handles events. Like, we saw, like, when you click on things, you can do things, it handles state. At what point, like, if you want to do things on, like, is there a backend? Or is all of your, your logic just mixed together? So, say, for example, you wanted to, like, process a video when somebody clicks on something. Normally, in Electron land, you would use something which would call a function that lives on a backend, right? Like, some sort of RPC you'd have to implement. Yeah, it would dispatch that, and then your, your backend would then fire off that function. What does that look like in GPUI land?

• 15:01Scott Tolinski?

  Well, it looks like you don't need, you don't need GPUI to send a message to something else, because it is just in the Rust code, right?

• 15:13Wes Bos?

  You're already there.

• 15:13Scott Tolinski?

  If you wanna call a function that does something, you're already right there. Yeah. So, in that, in that regard, you're eliminating the need for like, a RPC layer there, right? And, and again, GPUI is just the UI stuff. So how you use GPUI can be up to you. Yeah.

• 15:32Wes Bos?

  Oh, yeah. Your, your actual implementation is, like, this is not like a full stack thing, right? You typically would...

• 15:39Scott Tolinski?

  It's a UI layer, yes.

• 15:41Wes Bos?

  Yeah, you would put this on to your existing Rust application, or you would then choose how you want to architect it. Whereas something like Electron is a little bit more opinionated.

• 15:51Scott Tolinski?

  Yes, exactly. One thing I thought was interesting about this, Wes, there's two different types of render in, in this. There's a render and a render once. So like, a render once could be used for like, badges and more like, static stuff that doesn't need updating. Well, was like, wasn't there like, a React version of that before that didn't have state? Was that pure component? Or am I like, going nuts?

• 16:11Wes Bos?

  Yeah, like memoize it. You can obviously still re-render a memoized component, but, or, um, yeah, what was it? There was something in Babel as well, where you could just like, pre-compute it at, at like, build time. So that's another one, where you just don't have any logic. It's just, at build time, you figure out what the HTML would be to be rendered.

• 16:31Scott Tolinski?

  It's been so long, I forgot.

• 16:34Wes Bos?

  Macros, Babel macros, that's what they were called.

• 16:36Scott Tolinski?

  Macros, yes. And then there's also just this typical render, where the component has state, state changes, and then context notify, request an update rather than, like, that function's more like, kind of like remix or React, like you'd expect there. So, folks out there might be looking at this and being like, great, now I gotta reimplement every single React thing? We're fine, I'm, I'm, you know, whatever? Well, fear not, there's something called GPUI Kit, which is been a, uh, man, the very best thing for me in this. So GPUI Kit is a component library. And this thing has so many components in here, from accordions, alerts, dialogues, bubbles, badges, carousels, dialogues, I mean, just like, anything that you could possibly need.

• 17:20Scott Tolinski?

  It feels very like ShadCN of GPUI. Something's not loading for me, but I've been using this thing to great success. On every single GPUI project I've been doing, I've been using GPUI Kit. And then using that as a basis to style my stuff with. Because it, it just gives you these primitives that are actually nice and extendable. They're not too designed, but they're just designed enough to look decent, you know?

• 17:47Wes Bos?

  Input, output, pagination, pop over, combo box, command, input, all the stuff you'd expect. All the pieces that you'd expect to be able to use in an application.

• 17:59Scott Tolinski?

  Interestingly enough, Wes, it looks, the, the components look very similar to my graffiti UI stuff, which means that I like, have a lot of thoughts on how to style it.

• 18:09Wes Bos?

  Stick an agent on that this afternoon and get everything ported over.

• 18:13Scott Tolinski?

  Yeah, that's a good point, yeah. But it does, it gives you the primitives, all the input types and stuff you need. And one thing that I really loved was the, the data table and like, virtualization out of the box and everything like that into this. You know, like, sorting and tables just work and are nice and easy. Like, when I think of GPUI before GPUI Kit, I was thinking, okay, now I gotta build every component. And to me, this is the major unlock with this thing, because there's just so much in here. And it's all really good. So if I need something, I just got it right here. Hey, real quick here, I wanted to mention, because we don't talk about it in this episode, about GPUIC-E or the community addition.

• 18:56Scott Tolinski?

  Now, the community around GPUI is a bit fragmented at the moment, because, uh, community members feel like the Zed team, the team that created GPUI, is only adding things to GPUI that serve them. So there is a fork of GPUI that brings in all of the changes from GPUI. And according to a user on Twitter here, Cameron Campbell, "CE has some cool features not present in GPUI, like declarative style transitions, corner smoothing, AKA squirls, blur, background blur, gradient support on shadows and borders. And honestly, all of this stuff sounds awesome."

• 19:40Scott Tolinski?

  However, I do wanna highlight that in the conversation about GPUI Kit, which is a library that we talk about extensively in this episode, the creator of GPUI Kit, or the maintainer of GPUI Kit, says, uh, in this, in terms of not supporting GPUIC-E strictly as even like a flag, because GPUI Kit has a version pin for GPUI. So they say, "I was never optimistic about them from the beginning. These projects create problems without fixing them. I don't think it's worth wasting time on them. Only projects validated by real-world applications will generate continuous real-world demand

• 20:24Scott Tolinski?

  and drive the framework's improvement. GPUI has Zed, GPUI Component has Longbridge. These frameworks weren't created out of thin air. They were built upon real-world project needs, which is why they have achieved their current success. Just like my initial choice of GPUI, I was drawn to the way that Zed team works and GPUI's design philosophy. Currently, GPUI is indeed facing slightly slower process due to Zed's knees, but it's actually been continuously improving." Basically, they are saying that this is why they will not support the GPUIC-E, i.e. it's just a community fork, and they don't want to add things into their application.

• 21:05Scott Tolinski?

  Now, whether or not these points are valid or not, some people disagree on all of that. It's still a conversation that's being held, and one we don't really get into in this episode, because I have really personally only stuck so far with GPUI as well as the GPUI Kit side of things. Just wanted to clear that up for anybody who wanted to understand a little bit more about the situation. Here, let me show you some apps that I've built with this. I've built a few apps with this. One of this is called Gendo, and it shows every project that's on... I wish I could make these bigger, I don't have the font size, and there's some ugly stuff about these. Shows every project that's on my Mac or my MacBook Pro.

• 21:46Scott Tolinski?

  And then for things like this one, it's a, the, this is Battle Mode, which is the, the... App we did for our CSS tournament. It shows the version of, okay, I actually have three copies of this on the computers I have. I have two copies on my MacBook Pro and then one on my Mac Mini. So basically, it's here for me to be able to, and then I can see like the the status of all of them. And I can transfer them, I can star them, I can open it and stuff like that. This is an app I'm just working on for myself because now that I have multiple computers running multiple dev process and stuff like that, it's getting tough to keep track of everything. The versions, which one's up to date, what's where, what's everything like there.

• 22:27Scott Tolinski?

  So like, one of the things I'm going to be doing is organizing things and making it a little bit more of like, here's a system, even managing catty domains and stuff like that. So I built this little app. It's it's nice. It's it's just a little personal software here. Again, it's going to connect to any computer I have in the network. I also built file. Oh, yes, continue.

• 22:48Wes Bos?

  I got a question, though. So like, people are probably asking like, like, why would we want to do that? We we answered the fact that like, like, obviously Chrome and whatever, that's they're they're like hogs and they're they're fairly large applications as well, right?

• 22:59Scott Tolinski?

  Yes.

• 23:00Wes Bos?

  But like, like, this is perfect for tiny little utilities where you don't want to have to ask somebody to download an absolutely massive thing. Like, I just updated DaisyDisk, which is a little app I use on my computer to scan, find out. Like, I always had a space and I need to find out where is the space being taken up, right? And DaisyDisk just updated itself, eight megs, eight megs. When when's the last time you've downloaded an app that's been eight megs? That's tiny. I'm very happy about that. And like, versus like, somebody, there's probably some Electron version of DaisyDisk that's 267 megs. And every time you have to update the thing, it's just this massive thing sit on your computer.

• 23:44Wes Bos?

  I have probably like, not to mention RAM and resources of just these things running in the background.

• 23:50Scott Tolinski?

  Yeah, totally. And and I 100% agree. It's a little desktop app and I want it to be thin. I want it to be light. I want it to be fast. I want it to be all those things. I don't want to have, for every little idea I have, I don't want to have a separate Chromium instance running. And this takes it from something where I would have to have that to something that I don't very, very quickly because we're able to make it look good and fast. I'm actually redoing my video recorder right now because I want that to be as lean as possible, right? So moving that into to this would be really nice. Another one that I have is File Bro. And File Bro is an app that does like file automations à la Hazel.

• 24:32Scott Tolinski?

  So basically, hey, if this screenshot is like this and the name starts with this, then move it into here. I'm going to use Jev. Well, I am already using Jev, but I'm going to be using Jev even more in this bad boy. And it's going to be able to set up automation. So again, something you want to have running, but something you want to be very nice and fast, like file duplication checkers. This is all just temp check test data. But this is like file duplication checking and like merging of files and stuff like that. So to me, this is like a perfect use case for this type of thing, right? A nice, small, personal utility app that I want to be, I don't want to run. Like, okay, so File Bro, I will say originally was an Electron, right?

• 25:16Scott Tolinski?

  And now it's like, I can have all this duplication checking. I can do deep analysis of files. I can do all of that just directly in Rust in the application. And I found it to be very nice. So this is File Bro. I've also been working on a window utility library because I just make too many things, right? This is Mr. Mr. Space. I was tired of, you know, like Aerospace came out and Aerospace is great. It's a window manager that locks everything in. But I found it to be like the automatic window locking in. I find that to be relatively obnoxious in daily use. What I really wanted was something more that I could say, all right, here's a shortcut, pin to the left, pin to the right. And obviously, like, in here, you can see command failed.

• 25:59Scott Tolinski?

  Obviously, Raycast does that very well. But I also wanted the ability to send things to different spaces, send things like in Omarji, you you have your window, you can send it to space number two. You can spend, send it to space number three. Or I want to be able to automatically have every window I have automatically fit into the the desktop at once, right? Like those types of functionality. I'm not going to have that in Raycast the way I want it to be or develop these layouts. So I built this little app and it uses macOS native spaces. So like, I can send something to another spacer in here. It's really pretty slick. And agents can use this too, by the way. I could, I could prompt an agent saying to send a window someplace and it's going to know how to do that.

• 26:43Scott Tolinski?

  So it it's a pretty cool little app, but just like little fun things that I've been building with GPUI. And the cool thing about this app is that I only use GPUI, Wes, to do the settings menu. So I have the setting, this is just the settings menu. So this doesn't have to be open for the app to run. So you can use GPUI to build like full on things or just little nice looking settings menus, right? So.

• 27:06Wes Bos?

  Wow.

• 27:06Scott Tolinski?

  Yeah.

• 27:07Wes Bos?

  It's cool. As what about like downsides? Obviously, you just showed one, which is that you couldn't, you couldn't increase the size of it. Because like, that's something that comes for free with the browser, right? And you obviously need to implement some sort of scaling. I'm sure maybe that exists, but you have to then hook it up, right? Well, any other downsides there?

• 27:30Scott Tolinski?

  The the biggest downside one is the the non-native UI element. So like, if you're going to be like, oh, I want this to look really Mac native using Liquid Glass, but then on, you know, Windows, Windows native or whatever, like, this isn't going to be the thing for you because you're going to need to re-implement Liquid Glass and GPUI. And and that could be a major downside to some people and it could be an upside to other people because maybe you want to have full control over how it looks. This is actually kind of Wes Boss-coded, this this theme I got going on right here.

• 27:59Wes Bos?

  Kind of looks Cobalt II.

• 28:01Scott Tolinski?

  Yeah. I want to be able to have control over how it looks, me personally. And that, I mean, you can, it's not like you can with native apps. But like, again, I I don't want necessarily the cookie cutter Mac app when I'm doing it. But either way, you're not using native UI elements. So that's a thing, right? There's no browser DOM. So even though there is a project for getting GPU apps running on the web, it's going to be doing so via Wasm. I don't believe it's going to use the real DOM at all. I don't think that's possible really unless there's like a translation layer. So as far as I've seen, GPUI just renders it with Wasm instead. So there's no browser DOM. That said, there is a full on accessibility tree in these apps, Wes, and I actually use that accessibility tree in a Jev thing I built, I'll show. And then it's three, it's still young, right?

• 28:46Scott Tolinski?

  You got GPUI Kit as a component library, right? You have this as a component library and there's a couple others too. This is the best one that I've found, but it's still very young. So the fact that this exists in this extent is is pretty sick. But because it's so young, you know, you'll be less resources or whatever. I have found that AI does a pretty good job of of writing this. I will say that. I use a Rust best practices skill to make my AI do a good job of this. I'll link this up in the thing. And overall, that's been fine. But the the biggest downsides are really, again, it's it's not a right, once deploy everywhere platform in that sense, right?

• 29:31Scott Tolinski?

  You're not going to get a React native style native app out of this as well as a web app as well as a desktop app. You're pretty much locked into whatever you're doing. And primarily, I'm using this for native desktop app stuff. On the same frame as the, it's not using the DOM, it's not using that. If you want AI to test your apps, you're going to be looking at a computer use rather than like a browser use type of situation. However, I did create a Jev-driven native GPUI testing project here. It requires a small patch to GPUI and just a little bit to your harness here. But it uses Jev to computer use via the accessibility tree in GPUI

• 30:15Scott Tolinski?

  to be able to go through and click and do all that stuff. So it doesn't have to take screenshots to test your application. So I'll I'll link this up here again. This is very early and it's working for me. But again, early, small project. So if you're looking to build like fast desktop apps, maybe cross-platform, you don't care about either deploying to the web or you're willing to put up with some headaches there, I think GPUI is a good choice. I've mentioned mobile a few times. There is GPUI mobile. It does exist. It seems to work. It is experimental. You're rendering everything with like Metal via WGPU on iOS. So it's not using, it's not like a conversion layer like React Native.

• 31:00Scott Tolinski?

  But people have it working here. There's docs and examples of it working. You're giving, you're you're not getting to use all that native iOS good stuff, but it works.

• 31:10Wes Bos?

  Yeah, you'd have to like, I I would imagine there's like a lot of bridges you would have to build in order to like, same with React Native, you know? Anytime you want to use any of the native UIs, whether it's it's Maps or like NFC or any even just like the file picker. The one thing that drives me nuts in some of these like apps that are not built natively is that like, they just like, here's a file picker. It's just a a listing of every folder you've ever had. And like, there's, I don't have any of my shortcuts, none of my favorites are in there. And it's just, it's impossible to find what I'm looking for. So there is a bit of a trade off there, but you can, I'm certainly sure you can, you can write bridges to the native Swift there.

• 31:51Scott Tolinski?

  Yes. So would I use this to to do a native mobile app? No. Would I use this to do the web app? No. But I'm using this a lot to do personal software, desktop apps, stuff that I I don't want to have just a 400th Electron app on my computer for. I think a lot of people can recognize that as being a cool use case for this. So again, agents are good at Rust. At least as far as I've found, agents are good at this. If if you're vibe coding desktop personal software, man, GPUI is the way to go if you're talking to me.

• 32:22Wes Bos?

  Beautiful. All right. Thanks for schooling us. Let us know down below if you want to use GPUI or if you plan to use it, what your thoughts are. We'd love to hear it.

• 32:32Scott Tolinski?

  Peace.