Stop Making TUIs
sockpuppet.org
sockpuppet.org
As a developer outside of that, I love the scratch the itch apps stuff mentioned in here. I have a vibe coded SwiftUI chess repertoire builder app that fits in that same sort of space that I'm currently working on, where I'd probably not have chosen to explore the idea if not for coding agents.
I agree with the article that the terminal is an odd shape which you're fighting against historical specs, etc. The key observation I have though is that the mismatch of terminal apps and libraries to deal with this are all fundamentally built around character cells and cursor movement plus various CSI/OSC/ASC/DEC/xterm/... protocols which are complex and have varying support and interact in weird ways.
To get a good terminal UX, I think the answer to this is probably to throw all that compatibility mess away and redesign a modern terminal protocol that bakes in accessibility, regions, scrolling, selection, proper keyboard, etc.
I know Mitchell Hashimoto has a bit of a different perspective on this, which seems like it's to define some more protocol stuff and continue to build up on things. I think that probably gets to 95% pretty well. But a 100% good is a full replacement.
Why not just… a GUI framework or layout that’s meant to be keyboard driven and information dense?
If you're only targeting macOS/Windows/Gnome/KDE then the solution is easy: just grab a GUI control set and go ham.
Then there's remote access: you can spawn an X11 app through X forwarding and have a terrible laggy experience on Linux, you can use RemoteApps on Windows to have a good remote experience (but almost no other platform), you can use VNC to have an awful cross-platform experience, or you can use a TUI and have all the graphs and interactivity you need over a responsive, low-bandwidth connection on any combination of client+server.
So because TUIs look universally bad, they're better than cross-platform GUI?
You can't stay engaged if the menus don't bleep and bloop at you and confetti rains down on every click? That's a personal problem. A minimal UI is not automatically a bad-looking UI.
sounds like your terminal emulator is just crap, terminal and iterm on mac just work
qBittorrent looks great on KDE, and would look great on macOS too if they used native icons (I’ve made a theme for that but was too lazy to install it when I was switching laptops). No idea about Gnome and Windows, but probably alright as well.
So, Qt can get you a long way. But you should of course adapt your app to platform conventions and guidelines.
> TUIs run on any OS with minor patches to support quirks
So, which one is it?
TUI that runs on any OS |----------------| TUI without any compatibility mess
In any case we can surely do better than emulating DEC hardware from 50 years ago.
I really think they are. The reason for terminals’ ubiquity is precisely their age, and their age results in them standardizing and accumulating bizarre behaviors.
Even in the microcomputer era, standardizing behavior to where it’s literally everywhere someone might want it takes time (e.g. the browser compatibility wars). Re-standardizing terminal behavior is both chasing a way wider (in terms of the number of places folks expect terminals to work the exact same way) but shallower in feature complexity target compared to browsers, and would necessarily be replacing a widely adopted existing standard behavior, not providing something largely novel like the graphical web was. That’s a tall order. I am hopeful for and impressed by the efforts of folks like Hashimoto, but expectations here should be tempered.
TUIs look bad everywhere, so that's strictly worse
I would definitely be pro minimalist unicode-character-themed GUIs for what it's worth. Where every graphical primitive is basically a unicode character, without the restriction on font size.
Have you tried to write a GUI program lately? I suspect things may have changed. I can write one using Qt, EGUI, etc (Or many other options), and it will just work on Windows, Mac, Linux/Wayland, and Linux/X.
Basically, you write your application on top of a GUI framework that smooths over OS differences for you.
This is hard to square with the continuing obsession with Electron.
I thought Claude could do anything nowadays.
Querying into the DOM and working off of that means you're not futzing about holding onto a bunch of component references for the one thing you might need later on.
Web layout is also pretty easy to just get right, native GUI frameworks have a different kind of layout model that is not as amenable to "arbitrary" data (at least at first blush).
the reactive programming model also works quite well. While there are native GUI frameworks that lean into reactive programming, they tend to be doing a bunch of weird stuff that cause other issues.
And at the end of the day, when you package something like a Qt app it still often ends up being quite chunky.
In the end tho... the simplest thing is if you do web stack you get a web UI _and_ your "native" GUI in one go. Build things once, not twice. Hard to argue against that when that's presented.
Mobile has significant paradigm changes that makes using both terminals and traditional GUIs harder anyway though. I do expect the need for completely different cross-platform libraries for mobile UIs.
Also, Qt does support Android development.
The official NDK documentation on what is supported for NDK written code on non rooted devices, and the hurdles that Termux faces to keep running on standard Android images via PlayStore distribution, show why it isn't right.
I expanded on this point nearby: https://news.ycombinator.com/item?id=49400697
I think this is slightly wrong in two ways.
i) If you want platform native then reimplement the gui on every target platform. It’s as ”simple” as that.
If you need to ship to multiple platforms then preferably you need lots of plarform specific engineering in any case.
”I just want my hobby tool” scenario likely does not require cross platform support unless there is market for it. I mean rather than offering the tool on multiple platforms it should be offered on multiple languages first, perhaps. As soon as you are not writing the tool for yourself we are talking markets and distribution. Likely most people can chill out and just implement the gui for themselves (and not even publish it in githubb).
ii) cross platform quick-and-dirty way. Without screenreader support, multiple language and glyph support etc. Game ui:s look the same on all platforms. Websites look the same on all platforms.
There is place and time for deep ui engineering that respects the _critical_ cross platform issues like accessibility (screen readers etc) and different languages and scripts.
And then there is the situation where you want just few images to click. The latter is vastly simpler, fast and fun. But harder to convert to a professional quality gui experience. For small tools and MVPs this sounds like a fair tradeoff.
Invent a gui client environment that is universal, works the same way everywhere, over any kind of channel, and is already implimented and supported everywhere, and utterly weightless in all dimensions (ram/cpu/network), and then you might be able to ask that question without it being incredibly ignorant.
Today the closest you might be able to say is web/electron, which is gross on all counts. If anything the closest is X11, which is not remotely close enough.
That's why not just...
I'd say http://localapps should give an overview of all web apps running locally, with links to them. Those links should follow a naming scheme where the hostname ends with `.local` or smth.
My approach: Add entries for my localhost apps to `/etc/hosts`, and set up nginx as a reverse proxy.
Now when I go to `http://myapp.local` in the browser, `/etc/hosts` tells it to go to `localhost:80`. The http server sitting there knows which port to send the request to. I also made a systemd service that starts nginx automatically.
Now I have meaningful URLs for local apps, and I don't have to remember which one is listening on which port.
Malware will thank you.
1. In /etc/hosts (or dnsmasq): 127.0.0.1 ourapp.test 2. Caddy in front: ourapp.test { reverse_proxy localhost:8022 }
Caddy auto-provisions local certs, so you get https://ourapp.test, no ugly ports and no "not secure" prompt.
For the overview idea: a tiny local dashboard that scans common dev ports (3000, 5173, 8080...) and lists each running app by name is a ~100-line utility. This is exactly the kind of small tool I build, happy to make a quick version for you.
That same Pi won't even blink at a local TUI.
No just ignorant most likely. I can see how a child might think that an http/html session qualifies as providing all the same utility of a tty. Well it does not remotely by orders of magnitude.
I already said in the comment to begin with: "The closest you could say is web/electron, which is gross on all counts."
good dx, cross platform, fast, not ugly.
pick i dunno 2 or 3.
tui's get to check 2 or 3 of these just like any other gui library. somehow there still isn't a silver bullet here.
now i personally don't build tuis but guake style dropdown simple command and i have some other thing someone made open immediately running in ghostty? hell yea. with that said yes there are some stupid tuis out there that are anything but simple.
You can make slow things with it of course but it's not inherent.
The closest in spirit is Probably something like Dear ImGui[0] and the ecosystem of components and apps built with it, but it's not quite there for me. Something is missing.
At that point you're just inventing a new DOM variant and aren't fundamentally different from any other opinionated GUI framework. The value of the terminal for me is the minimalism. I absolutely despise all these new "features" that get added via protocols. I _want_ the boring grid composed entirely of extremely primitive character cells where WYSIWYG in terms of any underlying abstraction. The only innovation I welcome in this space are improvements to portability and speed.
My snarky response to the author's title would be that you can take your GUIs and all the janky flashing lights they entail and shove them up your ass.
What do you think of replacing the 'render' part with something that's more like a generic UI? You could keep the UX the same, yet your app would look like a desktop app (or at least close to it).
Trying to fit in the limitations of 80x25 text rendering is cool, but there's nothing in the underlying architecture that of modern graphics display that's predisposed to that, it all ends up as GPU commands.
I'm sure you could get most of the way to what I'm describing with just replacing the font/character set with one that renders things like borders in greater detail.
Hell, you could even add this as post-processing. You look at the terminal output with cellular automata like rules, and try to render UI primitives fitting that.
But in general, going with a fresh new protocol allows you to do a bunch of things semantically that are currently done as pure graphical stuff and that's useful.
From what I undestand, the terminal is essentially a concatenative text buffer extended with control commands allowing you to move the cursor (pen) to draw more sophisticated stuff, and update UIs.
The logical way to extend this would be to allow genericdrawing commands over the protocol, at which point the whole thing starts sounding like X11.
I'm sure the Unix people at MIT in the 80s saw this as the natural evolution from text terminals to graphics based ones.
Not saying this is a bad idea, but sounds suspiciously like a retread of history.
Or you might want to stop before that, as you suggested, at targeting fixed size cells for rendering, and keeping the console-like UX throughout. I'm sure there's a sensible intermediate step, but if you want to do things like displaying graphs or images (turning the terminal into something that's halfway between a terminal emulator and a Jupyter notebook), it'd be awkward to respect the character grid and the non-square characters themselves.
I'm curious what you think about how far this new protocol should go, and in what direction. You mention 'semantics', as describing behavior, not just what gets drawn to the screen, which neither X11 nor tty does.
Not sure how you're planning on doing that, can you incrementally extend the current model, or will it require a completely new protocol?
Let me get the caveats out of the way first: this only makes sense if you already live in Neovim. The whole ergonomic case is muscle memory you already have. If you're not a Neovim user, you're just buying another learning curve. Also, the language is Lua (not Rust) so it won't appeal to the ratatui crowd on that front, and there's no proper layout system yet.
Neovim [0] renders to its own buffer system (extmarks, floating windows, virtual text) rather than emitting ANSI escape codes. Under the hood it's still a terminal app, but from the UI author's perspective you're not fighting terminfo, CSI sequences, or varying terminal support. You get proper cursor tracking, scroll regions, and editable text areas out of the box.
On the UI framework side, morph.nvim [1] provides a React-like component model (h(), state, reconciliation) on top of Neovim. It's not a general answer the way ratatui is, but for the slice of devs already in Neovim (there are dozens of us!), it's a pragmatic way to get GUI-framework-like ergonomics without the terminal protocol fight.
For a sample of what TUIs in this paradigm can look like, I've been working on tuis.nvim [2] (Docker, K8s, SystemD services, process management, and more all as native Neovim buffer TUIs).
> Under the hood it's still a terminal app, but from the UI author's perspective you're not fighting terminfo, CSI sequences, or varying terminal support. You get proper cursor tracking, scroll regions, and editable text areas out of the box.
All that is not true. Take scrolling for example. Each terminal emulator implements this subtly differently (mouse/trackpad/speed etc.), scrolling will never feel native unless you push the control of it into the terminal emulator.
There's more than that, each and every point on this becomes something that you have a gap between native and the terminal emulator that you're building to run in the terminal.
See https://github.com/openai/codex/issues/8344 and https://github.com/openai/codex/pull/9640
Claude code and gemini have both had various versions of the same idea at times. I'm unsure where they ended up.
The New Terminal, the complete break from VT100, is I think the most powerful rebuttal to what I'm saying about TUIs.
My take is that the Morlocks should pack everything that is great about TUIs (and there are great things about them) and move them upstairs to live alongside the Eloi in GUI-land. Let 1000 Bloomberg terminal interfaces bloom.
But there's another even more ambitious take on this, which is to take everything that's pleasant and good about Eloi world and bring it down into the Morlock caves. I don't think you can accomplish that so long as you're drawing interfaces with punctuation characters, but there's nothing to say a terminal has to work that way; you can have a terminal with rich out-of-band-signaled UI. Not just, like, Kitty graphics, but something more like a real terminal that is meant from day 1 to work with all the complexity of a Bloomberg terminal.
You could have had it in 2010, but it's a king hell mess to put together, and you'd sort of assume nobody was going to use it (because it's a break from VT100 compatibility). But that doesn't matter anymore! We are the music makers, &c &c.
I think a lot of very smart people assume we're going to build up from VT100 to something incrementally but significantly better (Ghostty is already materially better than anything I'd used prior). But I hope those people eventually set their sights higher.
Yeah, in my hypothetical new protocol, the character cell is still there, but it's not the element that drives the main abstraction. The layer above that being more semantic is where the smarts is. So borders, interactive areas, mouse hit mapping to elements, etc. ends up being built into the protocol rather than being a cell level thing. Cells having proper borders being similar to how they allow for underline, being able to have both images and text, ...
I think it's important to do something about the VT100 +kitchen sink stack and to have that as something that still works with that in areas of the new terminal proto. I've got a slopcoded rust library in the works that's about fully mapping the full list of terminal protocols and related things in one coherent library (VT100, ECMA48, DEC modes, iTerm extensions, Kitty, ...). From there I think making that fairly isomorphic with the terminal emulator layer (I have a RIIR ghostty slopfactor there too that fits).
I think ghostty / superlogical looks at this from the perspective of doing pty things and then transfering that calculated state across the wire while doing fun stuff with windows, panes etc. with replay and that sort of thing. I think this is valuable too.
I suspect that there's possibly a way to do a bit of a hybrid on this - normal pty things with the extra sidecar of more semantic structured, but it'll make things more complex for app builders on that side of things.
(On another comment I saw you mentioned "couldn't help noticing how much of the tedious work of putting a TUI together". I think probably textual in python is probably the best of the low ceremony libraries for making TUIs if you haven't checked that out yet. I think someday there will be a good Rust library that has that level of put together-ness. I have some ideas on this for ratatui and have seen a bunch of things that are directionally right for this, but nothing that gets close yet).
I don't know why Codex and Claude Code haven't switched over; both apps have a terrible UI in comparison.
TUIs have downsides like accessibility, poor mouse support, but work really well if you focus on keyboard input. It is a better/faster input mode to begin with, especially for power users, but also for average people; you'd be surprised at how much more user-friendly an arrow-keys driven UI can be vs modern web interfaces.
I've been building my own TUI framework [1] and getting amazing results. Good TUIs are more of a GUI than a dumb text terminal.
If we had a cross-platform usable GUI framework, and more consistency across OSes, the story would be different.
Everyone always acts like GUIs couldn't possibly make HTTP requests, interact over TCP or WebSockets, or do other remote communication. But they're actually, more often than not, more well suited for that than browsers for specific use cases, which are the use cases where you'd desire a devoted GUI.
RDP works fine but it's based on a full desktop experience. Despite having the technology, we don't have the ability to remote in and run a single graphical program in a single window with the same ease.
We need something as seamless as Parallels for remote remote apps. But I guess we just settled on web apps.
What do you run like that day to day?
Modern cloud systems shouldn't even have shell access, immutable containers with telemetry data.
Genuine question, what is your alternative suggestion to accessing lower level features then? A hard time I have with GUIs is that their input/output can't be piped. It's much more cumbersome to try to automate a GUI than to automate something in the terminal.
Whilst I get that this is about TUIs, I do believe there is a reason for shell access due to automation. Good abstractions can be put into the GUI, but I'd argue that you should first build the abstraction, then put a GUI on top of it. AWS is a prime example, any of their APIs can be used w/out AWS Console.
devdraw can be thought of as a 2D rendering engine. You load assets like text, bitmaps and so on then issue draw commands to render them. There is no SSH. Instead you connect to a remote machine and export resources served by your local machine including your keyboard, mouse and display, each of which is a network transparent microservice. The window manager, Rio (and others), multiplexes access to these microservices. You can run a graphical program without a window manager, e.g. you can boot right into Doom or the Acme editor. It's seamless and can run over any pipe that can push 9P including ssh. If you look at the Sam and Acme text editors you will find they are true modern TUI programs.
Shame we had a cloud ready OS for 35+ years and no one noticed. Instead were living in the 60's with GPU accelerated typewriter emulators.
However I would rather prefer its successor, with Limbo for userspace, although probably not Tk as the main UI toolkit.
The terminal thing emulating typewriter are a UNIX thing, as other graphical OSes, presented a plain REPL, or graphical shell, e.g. Amiga DOS did not emulate any tty.
Otherwise I agree with the spirit of the comment.
Since when is pipelining a low level feature?
How do you think OSes like Mac OS (the original one), exposed their low level features, whatever that may be?
VSCode SSH extension is one of the most widely developer-used proof against this. Windows' RDP is the other one where many engineers and POS personnel have experince using them. Many electrical and mechanical engineers still use Windows RDP at my work to run simulations. The connection is responsive enough to survive their home-to-office VPN connection. I set up complex chemical simulation software that is accessed by clients of a HPC cluster over 1000s of kms away.
VScode has a really efficient protocol that communicates only that's required, Windows RDP can declaratively send UI component changes as long as one uses Win32 or WPF. It can also reasonably compress the data to even survive dial-up levels of bad data.
We use multi-attendant video conferencing day to day with whiteboarding extensions. If there is will, efficient and amazing remote-GUIs are easy.
What was worse was their mouse platform wasn't level. So every time they let go, it would drift downwards and they'd have to go through the "where's my mouse pointer" wiggle again.
The way the whole industry adopted it is understandable given the install base, but it's also probably a trillion dollar tax. I'd hate to see how much additional energy is used by billions of end-user devices running a rendering layer that is 100X heavier and more CPU intensive than anything native or TUI.
There should be public browsers and public search engines. The private incentives are too much and software built for public needs can be extremely cheaper than their private counterparts once you remove the need to support advertising services.
They're wannabe window managers.
I could swear there once was a (n)curses or otherwise text based GTK+ implementation / backend. It didn't do everything of course, and this was, I don't know, about 2003-2007, but you could compile a GTK application and run it in the console.
I tried finding a link, but unfortunately could not, can't even remember the name, I only remember building a piece of software that way that did GUI or TUI depending on how it was linked.
Especially becuase TUIs doesn't really have any of good properties that a CLI offers such as composebility. It's still a fixed UI with panes and menus which has just been squeezed into a character grid, and now forces the user to learn its pre-set keyboard bindings (which surely can be changed, but with every app having it's own config managment).
It's kinda the worst of both worlds to me. Either go all in and give me a proper text-based interface like Emacs (where you can globally configure how you wanna deal with lists of text in buffers), or just make a proper GUI with good keyboard support.
My big beef is the dichotomy of today's GPU accelerated bit-mapped displays which are burdened with this silly task of running a program that emulates a typewriter. Why TF do we not have a display protocol that can do graphics natively over SSH in 2026?
> TUIs have downsides like accessibility, poor mouse support, but work really well if you focus on keyboard input.
This assumes a typewriter emulator. Why are we not working on graphical systems where text is a first class primitive allowing us to easily draw text in any color, font, size, formatting and orientation as we please, onto a GPU accelerated display? Keyboard can still reign supreme. No mouse needed but if it is - first class support unlike crappy typewriter emulators.
Did you mean X server?
X is well past its sell-by date.
Plan 9 is text first as it was written by Unix programmers who wanted to build an OS and write programs (Rob wanted a better text editor.) It's the OG TUI. Window managers like Rio can run in Rio. You control things by writing textual messages from the command line, scripts, or code, e.g 'echo 100 >/dev/volume' to set audio volume to 100%. Someone wrote a new WM called Lola so I ran Lola in Rio in Lola in Rio because I can. Sam is very much a keyboard driven modern TUI ed (the standard editor). Acme is a mouse driven TUI editor. Both can be automated by programs and scripts. You can edit text while reading your email and chatting on irc from within Acme.
BUT Lots of stuff missing. We don't yet have GPU accel. It's a MSSIVE undertaking. And it has to be carefully designed and would likely be a built around a generalized devcompute kernel device.
When I moved from DOS to Linux, I chuck away TUI file managers and switched to pure CLI and never looked back. I could use GUI file manager, there you can really use mouse, without that jumpy-movements it has in TUI apps. But I wouldn't touch Midnight Commander or whatever.
I tried to run emacs in a terminal. Emacs' support of GUI toolkits is very cumbersome thing. But guess what forced me to abandon this experiment? Emacs is keyboard-driven, and terminals have issues with some keys, while GUI apps can deal with virtually anything.
> If we had a cross-platform usable GUI framework, and more consistency across OSes, the story would be different.
Try egui. I really like it. I'm not sure how much it is cross-platform though, hadn't tried it outside of Linux, but it is really nice. Or if you are an elm-purist then iced might be your choice. Iced lacks on widgets, it has just some basic ones, and if you need something to render a 2d-plot, then you are on your own.
Here's an example I just pulled out of my shell history:
c=3afba1a; reset; laptop "virtdev ssh liblinux -- git -C liblinux format-patch -1 ${c} --stdout" | tee /dev/tty | termux-clipboard-set
I didn't need to think to write this, I just wrote it right there in the terminal using the shell's line editor, just because I needed it, and it completely solved my problem. It's as easy as pressing up on keyboard now, and I can easily turn it into a script if needed. Nobody had to bend over backwards to add monstruous features to the apps to accomodate me. I knew what my computer needed to do, and I made it do it.
In my opinion we need more of this, not less. More unix and less iphones.
Another example from recent history, I just wanted a easy way to mount all connected drives, so whipped up:
lsblk -l | grep -i part | cut -d " " -f 1 | xargs -I{} sudo mount -m /dev/{} /mnt/{} # automount_all
Meanwhile, a shell alias/function just sits there, easy to see what it does, can control anything in your computer, GUI or not (one way or another), lets you build up your own "database" of tools that all compose together (again one way or another) and finally is easy to put under automated testing.
It’s extra mentioned that CLIs are mostly a good idea but TUIs aren’t.
Isn't it against both? Maybe I'm reading it wrong then, because things like these certainly seem against CLIs:
> Back in 1999, Neal Stephenson wrote an essay about command line interfaces that set the field of human-computer interaction back about 20 years. In it, he depicts the priesthood of Unix nerds wielding CLIs as powerful Morlocks, holding the entire computing industry on their shoulders.
> If you haven’t tried your hand at turning one of your 500 throwaway CLIs into a native app, you’re doing yourself a disservice.
Maybe I'm missed something else, but I certainly got the impression the author isn't very impressed by CLIs.
> CLIs have purposes for which they’re irreplaceable. Building a CLI is almost always a good idea. Building a TUI almost never is.
> CLIs have purposes for which they’re irreplaceable. Building a CLI is almost always a good idea. Building a TUI almost never is.
You're also arguing for CLIs which are absolutely great. But TUIs provide none of the composability of CLIs nor the malleability of GUIs
So while I'm not fundamentally opposed to graphics, I'm absolutely opposed to the "average user" culture surrounding it. I don't care about the "normal person", I don't want "user experiences" that hide the computer, I want all of the computer's primitives exposed so that I have full control.
Except for, you know, every serious professional application ever.
I was visiting a radio test lab in last February for our new product. They use a professional software which was written in WPF. The assistant there was flying with the keyboard shortcuts.
I recommend every self-describing Unix or terminal fan to watch videos of photo or video editors / VFX people on YouTube. They are hella fast.
+ seamless cli integration + network portable + graphics (!!) with modern emulators + always themable + scriptable + works with any navigation modality + you can copy & paste entire segments of your window easily + easy to consistently theme + did i mention entirely network/os transparent? + blazing fast
tongue-in-cheeck mostly, but i assure tfa, the tui serves useful functions. i spend >80% of my time in terminal, and having a pane to do some task a gui would otherwise be needed for is a sheer blessing. being able to essentially move my session to a laptop is wonderful. no mucking with screen sizes or weird X11 wonkery (if you still use x11). what would blow my socks clean off is a tui/gui+gui/tui backend pair so we can have our tui and you can gui it tooie.
If you run an emacs server you can connect to it with an emacs in TTY and one in GUI mode.
Stop building user interfaces in general. A command line interface is best. If someone doesn’t want to learn how to talk to the computer, they can ask a chatbot to drive the CLI for them.
Edit: the only TUI anyone should be willing to learn is their text editor, vim or emacs, because that’s fine for 99% of tasks.
A command-line interface is a "language" like any other. Each tool has its distinctive keywords, many have complex non-standard grammars (e.g., openssl, ffmpeg). And for more complex tools, the learning curve for this language is hilariously steep. I've been using ffmpeg on and off for two decades and I still feel quite helpless with it.
I think your argument makes sense, but it's just an argument for a standardized language, whether that's GUI, TUI, or CLI. Unfortunately, every other program or class of programs tries to innovate in some weird way. There's no reason why switching from Photoshop to GIMP should be so confusing, but it is.
They're an absolute torture to work with over a dodgy connection, which often may not even be your fault. Looking at you AWS.
My life is spent inside terminal and Emacs. Every time I reach for the mouse, I consider it a major fail. Like: big time failure.
Now at times I'll use the mouse: for example when I'm designing a 3D part to then 3D print to fix something around the house. In that case a mouse makes sense. Or for gaming: a mouse makes sense.
But for most of what I do, even just moving my hand away from the home row position to reach for the mouse is a massive fail.
Meanwhile a GUI's developer has to decide to grace me with the ability to even open more than one window. "A tabbed interface will be sufficient!" -- yay, I'll never be able to view two screens of info at the same time.
It's not true for AppKit, SwiftUI, GTK, Qt, WinUI, or Electron, and I'm not sure what's left.
vscode, electron app, no problem opening multiple windows. Geany, gtk app, no problem opening multiple windows.
Can you give an example where this doesn't work?
But the point is that the developer has to build the app in a way that lets you do this.
Once again, there is no default with the tools I listed. You decide on how you want windowing to work and then build it.
At this point, show me examples of needing a mutex to "block" a multi-window default.
Just like you can start vim several times you can start gvim several times and have a few windows. This is the default behavior, and if the developer wants to change this for some reason, they need to put in some effort.
Maybe macOS is special? I'm not familiar with this platform.
Even in SwiftUI which has the simplest way to support multiple windows with zero work, the developer has to use WindowGroup, and if they didn't then you don't get to open more than one window.
TUI apps are multi-instance by default since the user can invoke the program in different pty sessions. The developer would have to actively fight that.
GUI SDKs generally require the developer to pick a solution. And it's easier to default to the simpler solution where you have one instance, one data model, one window.
Whether things might change in the future isn't relevant because we're talking about how things are today, not speculating about the future. And the time to change one's opinion is when reality changes, not ahead of time based on how you think things hopefully will be.
I don't know which GUIs you have programmed but for Windows and GNU/Linux this isn't the simplest option. It requires quite the effort to actually enforce single instance.
The merging behaviour is also in principle possible for TUIs, btw, it's just that nobody does it. But with something like tmux is would be easy to set up that all subsequent calls to your binary actually connect to the already open session launched by the first instance.
It looks like macOS Finder tries to make it difficult to run two copies of a .app simultaneously, so I imagine many programs don't check for this case, but this is Hacker News, so we can use our hacker skills to try and do what they don't want us to: find the exe inside the bundle and run it directly from Finder, or run it from the terminal using its path. (This works for me with Ptacek's mdv app. Two copies of it end up in the dock.)
https://www.electronjs.org/docs/latest/api/app#apprequestsin...
and handle this event:
https://www.electronjs.org/docs/latest/api/app#event-second-...
I believe macos is the only OS that enforces single instance by default. If you start xterm five times on linux, you get five independent xterm processes.
You're wrong about GTK, Qt, WinUI and Electron though. They default to "do nothing" which eliminates the ability to even know that another instance exists. You have to manually program that.
Imagine if there were no terminal/pty system. Then what, I'd have to use Claude Desktop (doesn't support multiple windows) and Finder?
So no, it's not the only argument for TUIs.
They are usually tied to specific user accounts, it's true. But that's something some people made up, and it doesn't have to be that way.
But I don't think I've run into a TUI that I couldn't have multiple instances of.
For a TUI to do the same, it wouldn’t just need to find and focus the window, but would need to trace it through the myriad ways that a TUI can be displayed. I may run a TUI inside a docker container, as part of a tmux session, over ssh. If a TUI somehow managed to focus the existing instance through those layers of wrapping, I would be very surprised, disgusted, and impressed.
So I think it’s that the end goal of “redirect user to existing instance” is infeasible, so nobody bothers to enforce a single instance, since the rest of the steps aren’t possible.
Emacs Server and tmux would both like a word. Make the current tty show the app output.
There is an opencode plugin for this opencode-pty.
Terminals + browsers is all I want for most apps.
Native GTK/QT is really good, but I only want it for "system" things like a clipboard manager or file manager settings app and things like that.
Pretty much all GUI frameworks I have worked with behave the same, and you have to actively create singleton behavior.
Then you have Mac OS which makes it hard two create multiple processes of a GUI app. Other OSs don't necessarily behave the same.
https://en.wikipedia.org/wiki/Tab_(interface)
Same with docker containers
Same with concurrency
Same with Notepad
Same with exec
Same with bsd jails
Same with copies of data in a for loop
Same with recursive functions
Same with numerous copies of image viewers showing the same jpg and txt files
...it's all arbitrary containers and arbitrary recipes to encode and decode binary
I’ve always appreciated the positive side of TUIs and appreciated less the positive side of GUIs. I didn’t grow up with a computer at all and in one of my first adult jobs I had to use a TUI at Papa John’s for punching in orders, it was 20x faster than anything I used after it at other restaurants that were GUI driven (keyboard speed wins for me, fit my brain perfectly). I think I learned it in 5 minutes. I was just a normy trying to pay bills.
I dunno why we gotta be insulting to people’s preferences, the article is genuinely bad about this.
I agree that I found the article to be poorly written. Mainly because it takes a long time to make its point. I didn't really need to see a million examples of the author's vibe coded slop before getting to the topic which the title suggested the article would be about.
Agreed. One would hope the article would go into detail of how to make a great GUI (including tips on how to make sure it’s easy to navigate with a keyboard).
A GUI can be keyboard-driven and as fast as a TUI. The sad truth is that the status quo for GUIs is absolutely terrible, and most were rushed out and not properly tested to check if they're really usable without a mouse.
And TUIs are only keyboard driven. So the choice for speed if you don't need graphics is always - TUI (or CLI).
Plenty of TUIs support mouse to various extents too. Claude Code is an obvious example (try the scroll wheel for example).
I agree with you that TUIs are keyboard driven first though - and so you can be certain it will work well.
Not so. GUIs absolutely should be keyboard-drivable, if they're written to the human interface specs of whichever platform they're hosted on. That doesn't mean they don't support being mouse-driven or even that mouse isn't the default, but in my experience if the developer has taken any sort of care with the interface then keyboard interaction will be at least as good, if not better, with the keyboard.
> And TUIs are only keyboard driven.
Not true either. There's a line in my .vimrc saying "set mouse-=a" specifically because I want to turn off the mouse in that particular TUI app. If a TUI app doesn't support the mouse they've just not added that support, it's not that it's not possible.
Absolutely not.
The problem is discoverability. You need to go out of your way to learn shortcuts.
FWIW, most of Windows XP's utilities and included software was perfectly usable with a keyboard alone.
This is the key point. No platform UI paradigm focuses on keyboard navigability and speed, all focus on mouse and visual effects for discoverability in a pursuit to dumb down the user experience enough to make it usable for everyone, breaking power users in the process (and increasingly also accessibility stacks).
Every time I use Claude Code, I'm wondering why the hell I'm regressing to the level of a machine. I paste an image that I want fixed, and I don't see the preview of it. I can't cut or edit text somewhere in the middle of a long sentence like in a normal user interface. I can't visualize any helpful graphics, not even Latex. I'm forced to interface like a machine when I'm a human being.
Why can't developers understanding that GUIs and TUIs both have their usage, it just depends on the task?
I'm fine to use vim over ssh into a server, just don't force me to edit fine typography with it.
As far as I can tell, practically everyone arguing for TUIs thinks Claude Code is a horrible example.
Though there’s been some recent progress in alternative models, the fact is native frameworks have some catching up to do. Some in particular (SwiftUI) are downright awful
We are users.
> Why can't developers understanding that GUIs and TUIs both have their usage, it just depends on the task?
Isn't that the case for most TUI apps? They tend to serve quite specific purposes. Nobody writes a WYSIWYG word processor in a TUI. Because you can't and even if you could somehow with sixels it would work horribly bad.
In a dual role, I would say that the most impactful aspect (being a developer) matters more.
> Nobody writes a WYSIWYG word processor in a TUI.
But at the same time people write all sorts of other apps that could hugely benefit from a GUI. I seems like if the output itself is not graphical, TUI-enthusiasts won't acknowledge the advantages of a GUI layer, while only bringing in arguments why they can implemented poorly.
GUIs have a much higher surface to be done wrong, so standards are higher. But there's still a decision that needs to be weighed carefully, and many developers seems to simply jump on the TUI bandwagon just because it's easier.
But a TUI needs dedicated understanding of terminal toolkits. That's not easy. A GUI is much more easy mode with electron, a bit of javascript and a GUI toolkit to design the UI, easy peasy. To me the TUI is the hard option.
I really enjoy TUIs though. I wish there was something like vscode in TUI.
Now I use Codex UI which is miles better than any other tool I used.
Once you go down this path, it is hard to go back. Very few tasks, apart from things like Photoshop and Revit, really require me to be at a GUI these days. And with MCP a lot of that is going away.
On the Linux/BSD side, most toolkits are of not great quality, you'll inherit all sorts of subtle bugs and quirky behaviours. Most of the software that's of _great_ quality doesn't use the toolkit, but interfaces with the compositor directly. The effort for developing GUIs under such circumstances is dramatically higher than a TUI, and a TUI is usually "good enough".
I really would prefer GUIs for a lot of things. But when a TUI takes a few weeks to make, the equivalent GUI would take a few months to make.
Good luck exhausting Claude. As long as you keep feeding in quarters, it will keep hacking Qt.
Flutter is an absolute nightmare to install and work with. Even now the few bleeding-edge distributions where it's available are lagging behind many releases because of the pain that it is to maintain it. I don't want that level of pain as a pre-condition to building my software.
I don't want to forward a whole desktop environment just because I want to use a small application I have written for myself.
TUIs are much more portable, easier on eyes, can be information dense, and resistant to lassiez-faire UI library changes on platforms. They don't need GPUs to render, megabytes of RAM to run, great amounts of bandwidth to access from afar.
I'll write a library for the functionality, and will slap a TUI on top. You want a GUI, write it yourself. Everything is GPLv3 licensed. You can do whatever you want as long as you respect the license.
They do these days, because you still need a terminal emulator and those are then basically GUIs in themselves. Ain't the 80s anymore :^)
Most, if not all these systems have Matrox G200 or similar low end 2D integrated graphics which are just there to see the text during boot, nothing else. I sometimes SSH from these systems to my own systems to use TUI or CLI tools.
If I want to go a step further. We have a couple of real VT320s.
Working at a datacenter has it perks, it seems. :^)
That's awesome, I have one too and I fell in love with it. I fixed up the old gif320 program to run properly on modern machines a while ago and made a C# port too:
https://github.com/ldyeax/gif320
https://github.com/ldyeax/Gif320Sharp/
How do you deal with modern programs not respecting TERM/LANG capabilities? I found even using man pages on my machine would sometimes result in the terminal getting caught up on some invalid byte sequence emitted from some unicode that snuck its way through or such.
I ended up writing a program I called VT320 Translator, or vt320t for the name of the binary, which in broad strokes runs your target program in a background terminal and sends curated updates to the VT320 while passing through keyboard input. (You can kind of approximate this with just screen and some other things but it doesn't end up being very good IME) This lets me do some neat things like translate unicode characters into DRCS (Dynamically Redefined Character Set) sequences that you can actually see on the VT320, use optimized heuristics for screen updates tuned to the exact hardware, etc.
At the core is a vtparse-based VT320 emulator that strives to be as accurate as possible with an accompanying diff generator from one emulator state to another. If you're interested I could get into more of what I've done here and even get the code onto my github - it's been private for a while because it's being used in an upcoming game demo but I could easily separate out the non-game-related parts that make up vt320t.
If it's not a problem, can you share the vt320t source code on GitHub, if not you can share it privately and it can stay between us. I'm not in gamedev, even remotely, so the game parts has no utility for me. It's completely up to you. I'd love to read some low level code talking with vintage hardware. You can find my contact information at my homepage, which is on my profile.
On the other hand, our VT320s doesn't do anything fancy these days. We generally use it to talk with our networking devices like switches and routers over RS232 since they can communicate really well with each other. While we have lots of hardware to tend and projects to run, maybe I can hook-up one of the VT320s to something for fun.
Thanks for your lovely comment and spending time for all of these. It's greatly appreciated here.
Oh, also your homepage is great. I liked it a lot.
Here you go, just finished getting it all together. I do plan to add the game stuff here relatively soon - I'm just saving it for a certain demo to be released, but I did include a screenshot, as well as some pictures of vt320t running with real hardware.
> Oh, also your homepage is great. I liked it a lot.
Oh, thank you! I'm probably even going to put a WebGL/whatever Unity exports to these days for web up there when I can so the terminal emulator can be used on there.
Cheers!
I love a GUI in many cases! But if I'm building something, I personally want it across Linux and MacOS at least (personal and work, respectively). TUI answers that problem quickly, easily, effectively, and efficiently.
I'm actually amazed that we have never gotten to a really good UI framework that is cross platform. I really just want something like SwiftUI that's truly cross platform, personally. There are a million arguments to say that we do have cross platform GUIs, but all of them lack in some major way.
Just curious re: the obvious example of "The Web" - do you consider it "not good" or "not really a cross-platform UI framework" as such?
It's also generally resource intensive, compared to a solid native UI.
I would say that, if you don't need more native control, the web is the best cross platform interface we have, by far. But not having those system API really starts to limit it for certain tasks IMO.
When I just want a graphical utility, I build a single HTML file that includes all of the assets required, and all of the JavaScript so I can run it offline (but generally host them on my site so I can get to them from anywhere). These tools are generally text in and text out that I want a GUI representation for and don't need in the terminal.
One example - wxWidgets exist, and wxDart exist, but getting to something half as usable as Flutter from there is almost impossible. I defy you to even build a layout engine that works on all three platforms. After that, you still have to find answers for theming, spacing, responsive behavior, component composition, inspection/debugging, hot reload, accessibility conventions and on and on.
No, I don't think I will :)
It's great for admin tools, alongside CLIs.
It's portable, it's fast, it's discoverable, it has zero bullshit effects, it's as secure as my shell, it's efficient.
And what is this point about agents making it easy to generate GUIs ? They are equally good (if not better) at generating TUIs using any of the great frameworks for this.
Not really?
> it's fast
It's not - most terminals are ridiculously slow, and each TUI framework requires its own hacks to do partial "repaints" using ANSI escapes, because clearing and redrawing the entire screen (yes, just a screen of monospace text!) is unworkable.
> it's discoverable
With no ergonomic way to build a menu bar, display hints/tooltips, or even support modified keyboard shortcuts well across platforms? I don’t think it is.
> it's as secure as my shell
Extraordinarily insecure, in other words.
> it's efficient
As above.
It is, everywhere I need it. Never had to change a single line across the various OS/Terms I used. Can't say the same for web apps or even so-called cross-platform GUIs. Been there, done that.
> It's not - most terminals are ridiculously slow, and each TUI framework requires its own hacks to do partial "repaints" using ANSI escapes, because clearing and redrawing the entire screen (yes, just a screen of monospace text!) is unworkable.
I don't care about these details. I care about the app feeling snappy and responsive. It sure does. Again, I wish it was the case for GUIs.
> With no ergonomic way to build a menu bar, display hints/tooltips, or even support modified keyboard shortcuts well across platforms? I don’t think it is.
Discoverable as in "you can start the app and figure it out along the way without prior documentation reading". And we do have hints, menus and what not. Trying to nitpick ?
> Extraordinarily insecure, in other words.
Are you implying the app running from my ssh session is more exposed than a web app or a API endpoint for a GUI app ?
Like all things, TUI is not a panacea and has its place.
I will just stop here. No hates for macs, but your typical cli/tui software devoper is concerned about making their software run on almost anything with an actual os. As soon as I see references to specific operating systems or platform specific libraries my bullshit meter goes off.
I was expecting an article about methology and tooling to develop guis, got an apple fan telling stories about ai use.
>But remember: I’m not really building applications for other people to use. I’m building them for me.
What's the point then? What's the point of writing an entire argument proselytizing towards application developers and ending with saying you're only talking about applications you're writing for yourself?
Let say someone, as windows or linux user, clicks on this article claiming he/she should stop "doing TUIs", what take away message should he get from it? "GUIs just looks nice but sucks you have no swiftui"? Personally, I was expecting a developer's angle on GUI programming.
Apple deprecated carbon (which was a thing when gtk1 was around). I don't think you have an option for this on their ARM hardware.
QT1->N code has the same problem, the older libraries are not shipped on most modern linux.
I do absolutely understand if you're going to do static compiles, that can work around that problem, but that arguement nullifies everything, since you can run/write/execute anything in a turing complete system, if you complain you're just not dedicated enough.
y'all are slow learners if that's an issue for you, but React doing major breaking changes at least once a year is totally fine.
The same reason webui and js is so popular!
No they don't. Many components are SwiftUI only.
Once we have robust automated computer use, that becomes the verification loop, and this problem will almost surely get solved.
Why? Because just look at the examples on that page.
Or for Go coders there's BubbleTea: https://github.com/charmbracelet/bubbletea
Why? Because TUI!
Please, don't. One of the main benefits of TUI is being able to freely copy text, I hate opening Textual application and being dropped into this canvas like state.
I guess it’s a matter of taste ¯\＿(ツ)＿/¯
I love the visual style. It's especially interesting to set the colour scheme to the orange/yellow glow of reaaaly old school terminals.
It’s ridiculous how far we have fallen behind the Terminal GUIs from the early 90ies. Look at the Turbo Pascal IDE and business apps built with it. Windows, window management, scrolling in windows, rich dialogs. Everything there. While many modern TUIs are just lists with numbers you can select. Everything is pretty rudimentary
I was missing a real terminal GUI library for some software I built. The rudimentary libraries did not work for me.
That‘s why I took Free Vision (the Turbo Vision derivative for Free Pascal) and ported it with the help of Claude Code over to modern Delphi, then modernised the object system and added lots of features and from there we ported it to go.
There are still lots of bugs and oversights (some caused by me not really knowing how original TV worked) that I fix as I go along using it for real projects.
https://github.com/oldwired/fv-delphi-modern https://github.com/oldwired/fv-go
This pretense that the entire value of source code is the "spec" / "requirement" is alarming.
The value of the code is how well-formed the spec is + how well tested it is in the real world. So prompt / mockup cannot replace code.
Sure, an LLM can fix it, but that assumes your original spec sufficiently described the user requirements and wasn’t just a transcript of the original session trying to figure out how how to interface with all those dependencies in the first place.
The one advantage that a terminal has, is that commands can be concise and scriptable. But that only really applies to CLI.
A TUI is just a poor man emulation of a GUI, without any advantages.
I salute you, heros.
Please, make more TUIs and web apps, so they can live in my terminal or my browser.
Once you get the key bindings for screen or tmux thoroughly memorized it's very easy.
I have a windows box that never shuts down and i remote into it and use it GUI mode to do everything you said but with a proper GUI interface. I also use the terminal but your feature (constantly up + resume + multi device) does not need TUIs to even work.
My reaction to this is basically about the same as the fictional Ivan Chesnokov
(beware, crude language, profanity, etc)
But with a worse UX?
As a devils advocate in response to all the comments claiming TUIs are fine, maybe consider TUIs mostly do not come with many of the advantages of using a terminal CLI app - you can't pipe data into them, out of them, spawn them and collect the output, etc like you can with the usual stuff you run in a terminal.
IOW, they are a reduced functionality of a GUI - resizing works poorly, scrolling works poorly, UI affordances are poor (sure, you can grey out a button, but drag-n-drop hardly ever works between your various terminal tabs, and the smallest UI element is a full character block, unlike a GUI where you can have indicators that are smaller than a character, mini-windows, etc), lack of widgets, etc.
The lightweight argument is not so good either - look at Claude Code, after all. If you want lightweight GUIs, you can do that too.
The only argument that stands is the ssh one, but if working via a remote connection, I'm not going to want a full functionality application anyway - why would I prefer a wordstar/wordperfect interface to a complex document over a GUI interface? For simple documents, sure, TUI all the way, but when I have multiple panes (outline, minimap, status bars, toolbars, etc), it performs poorly in a TUI.
TUIs are making a resurgence, sure, but only in the context of developers and development.
While a mouse can theoretically use all pixels of a screen, in practical terms a keyboard has a much larger input space than a mouse, and if I can keep both my hands on the keyboard, I can work a lot faster.
At least that's my experience as a (starting) greybeard.
And you present the division of this simple unification of modalities as a benefit, but don't explain why it ought be many.
So far it's all just preference.
I'm not going to deny that the terminal has a ramp, is a lifelong learning curve. But I personally find it much more earnest, much closer to what is really happening, with less intermediation, and the way that terminal conposes with other things is usually unbeatable. TUI for humans, jsonl event sourcing for machine-to-machine.
> drawing on a couple of decades of established UX conventions for how they are displayed
I tend to not just click random apps open. I tend to have done some research and found what options there are on my own. I tend to know what I'm looking for, by the time I'm going to run something. In terms of getting there, finding stuff: the best convention I know of is aptitude, and dselect before it. Both very fine very old tui systems, and clear cuts above any app or web store, imo, for its power, directness, clarity, intent, and script ability.
But then a load of others will complain about electron apps.
Further you then have an os within an os. You need to remember to look in your bookmarks, not your start menu. And you're basically throwing away the window manager too.
I want to run web apps in my browser. I understand why Signal isn't a web app, and I'm willing to put up with that for security, but nobody else gets that excuse. Web apps belong in a browser tab.
Native UIs simply don't have the same flexibility.
Edit to add: I didn't bring my laptop; I'm working from Termux. Surprisingly effective! The code is all in a folder syncing with my phone via SyncThing, so I can try out the sites I'm working on directly on the phone.
Terminals are great for CLI. CLI integrated into GUI/TUI is absolutely amazing, but 99% TUI apps do not work like that, they're just awkward crutches foreign to the terminal, so you get no upsides and all downsides. Examples of good CLI integration I regularly use: Kate, Zed, Dolphin, Total Commander, pi. Only one of them is TUI, and honestly it's much worse off with it, since it has to display a ton of formatted text and sometimes pictures. It's really inconvenient and just looks like cargo cult rather than a necessity.
So you can't even immediately switch to your app and don't see it in an OS-integrated list?
The main thing I like is that TUIs are guaranteed to be navigable by keyboard. So many non-TUI apps don't include shortcuts at all for critical actions. It makes the first week of using the app more efficient and the entire rest of your life less so. Add on to that, I can probably run a TUI on my server. Or my cloud instance. And in my sandbox. Over SSH.
So: do make TUIs and bind their keys to Vi-like shortcuts as much as possible and I will use your app every time over a GUI app.
Pro tip for Mac users - install Karabiner and bind right-Alt + jklm to cursor movements. Immediate VIM navigation through the whole OS.
I don't disagree with you, but in ye old days you had guis that were navigable by keyboard.
And these were discoverable. That one program you use every day, you can spend the time to learn the short cuts. You can still get things done by clicking around though.
I suspect the underlying issue is one of limitations. Terminals are limited in what they can do. Guis and the web, you can do so much more, so people do.
I think another issue is touch screens. Guis and the web have moved to optimise for those. It decreases information density, forces a move away from the old standards to new much weaker ones. Tuis can't really do touch, so they don't have to make those tradeoffs.
There are real advantages of TUIs: less CPU; sometimes I want to be using a personal TUI on a personal dev server quickly from a work laptop at lunchtime and an ssh + tmux + TUI is perfect; most modern TUI libraries are significantly less painful to work with than most modern web app frontend libraries; I can change the font size of my terminal easily if I need to, more easily than native (but not web app). But by far the biggest, is they just generally tend to be very fast and responsive compared to everything else.
While I get your point, a lot of vibe-coded agent TUIs (Claude Code, GitHub Copilot CLI, …) aren't exactly fast, either.
The only one I use is Copilot (inside of a VM). ALl other applications I use are GUI/CLI/webapps.
- yazi as a file explorer
- lazygit for git
- tmux/zellij as multiplexers
- fzf left and right for a bunch of utils (terminal history, switching between projects in zellij/tmux when working in multi-project workspaces, etc)
- neovim as a text editor
- glow for markdown reading
- currently on pi for my home harness, CC at work
Claude etc do make sense as TUIs too, but I don't like they way they were implemented. They mess with basic terminal features like copy/paste or scroll when there's no reason to. And they're somehow heavier than actual GUIs.
People are unable to do good GUIs. The web shows this; completely disregarding any UI guidelines by web people established since 80s is a proof that most people have no idea how to make GUIs.
I don't know how to make a good GUI. That's why I need good templates and strong guidelines, discipline and spend long time on design. Most people don't do that.
TUI is much, much easier. It's also possible to do bad TUIs, but it's easier to do a good TUI than to do an average GUI.
I'm currently trying to create a GUI version (immediate mode library) of a CLI tool of mine and it's mostly a shit show and three times the work of a data-in-data-out CLI tool. Good learning experience, though. I should note that I don't care whether the GUI version has a "native feel" or not (if you have the time to do so, be my guest), only whether it's functional and reasonably fast, since otherwise that's even more work for an open-source tool. Though, when I undoubtedly make the more interactive TUI version later on (ratatui just seems too nice not to try! :-)) it will probably bring a similar complexity development wise.
Then again I'm not into agentic coding at all so maybe the joke's on me. On the other hand, at the moment I need to learn about the data I'm processing, the GUI library/framework I use, and being able to help the people I work with troubleshoot the equipment that collects said data. I'd miss out on most of that if I coded by prompt. Perhaps I'll try later, perhaps not.
Keep making TUIs, or GUIs for that matter.
Sometimes I can use `du` to check what uses so much space, but sometimes ncdu is better for me.
And the best part - I can use both when I ssh to another machine, using the same command line, without thought. While GUIs need added thought, to switch from CLI to GUI on the remote one would need to use a different interface, or have X11 forwarding on (which is also a good option) and x11 libs on the other side.
Or use some other connectivity. ssh is simple, works on most servers I have access to.
Exactly. I'm currently on vacation and has checked in on long running Claude Code sessions over ssh every few days. I could have used X11 forwarding or VNC if I needed a GUI app, but the connection I'm on is currently slow and unstable, and using text mode has been far more palatable, especially since I can use tmux (or in the case of Claude Code, just using "claude agents" and let it background the sessions itself).
> I barely think about these things as “apps” (I have no intention to distribute them). They’re artifacts of me making my computer do stuff for me, the way I want it to. As a Unix nerd, I’ve always been able to do this, in the language of the command line. Now, it’s just as easy to do that kind of work with graphical interfaces.
And so the author (second sentence of the post)
> I built my first serious Mac application a few months ago, and since then I’ve built more native UI thingies than in my entire career prior to that.
I still build my personal stuff either for the command line (not even TUIs) or in a HTML page but I can see the point of creating a native app. Unfortunately there are some drawbacks: I would have to create the app for GNOME Debian and for Android, while a web app runs on both, and I would have to go all in on vibe coding because my last serious GUI programming has been in the 90s with Motif. The latter point is more or less the point of the author (the LLM does it for you) but it means that programming personal stuff will cost money paid to the LLMs companies. More and more money at each price hike and the multi thousand dollars box to run local models only makes the expense happen immediately instead of diluting over a long period.
You can use them remotely over SSH connections, and they can survive disconects (within tmux). No VNC, or X11 forwarding needed.
They take less memory. compare vscode vs nvim.
Super fast to launch.
They can integrate with other TUIs and shell apps, shell-scripts, pipes
and lastly they do not contain Ads. lol
First, a TUI is usable with only the keyboard. Moving the mouse is a waste of time, and if you are using an application multiple hours a day, it can lead to being more productive.
Second, a TUI has no useless animations, elements, etc. It's not pretty, you have on the screen only the stuff you need to do your job, and you are not distracted by useless elements.
Third, a TUI works beautifully in a client/server world, where you have one central computer and dumb clients that connects to it, and the connection can be a simple serial connection as well as a network connection. It can be rendered even with an ESP32, making it suitable for embedded HMIs and all kind of computing scenarios.
Lastly, the user can choose the font, the colors, the size of the font that are best for him, usually a monospace font that it's surely better to read information multiple hours a day.
To me it's the opposite, GUI interfaces were born to make it intuitive to use a computer, but are not as efficient as TUI in terms of productivity.
I've used the same systems you describe, except in gui form.
And the thing is, with this kind of application, everyone has a different 20% that they spend 80% of their time using. Guis can be as good as the TUI in the 80%, and very handy in the 20% where you aren't so well versed.
The issue, is that guis seem to have dropped the ball with regards to efficiency.
Is it? Many TUIs these days don't allow me to configure my usual key bindings for moving my cursor within input fields. Meanwhile, GTK offers (used to offer) a way to customize key bindings across all GTK applications. (Unfortunately, they removed that feature in recent versions. Now I understand why people always get so upset about the GNOME devs…)
While I don't use most of them, but I can see why TUIs they are popular around devs. But in this post it reads like this:
"I don't like slop TUIs because they are popular, but please look at my vibe coded slop GUIs and I like vibe coding them and you should too"
One of the worst, if not part of my top ten worst blog posts I have ever seen on this orange site and absolutely do not listen to that horrible take that the author just wrote.
Intentional or not, the rhetorical style of the title, plus the subtitle opening on the word "weird" before it's defined, has an ostracizing effect on me. I guess that it could have an ostracizing effect on others, though I can only speak for myself.
GUI apps are only platform independent if you(r framework) make them so.
It's a quite common illusion that TUI apps are "automatically" platform independent. Even knowing which terminal emulator intercepts which shortcuts and how they handle width change, control character, mouse events, images, etc and design around them is not a trivial task.
On the other hand, making platform independent GUI apps haven't been easier. Just use Electron. Yes, performance blahblah, but thousands of Electron apps work on millions of machines just fine.
They don't support standard GUI shortcuts. They don't support drag-and-drop. I can't double-click a file and have it open in a TUI. They don't integrate with spotlight metadata. They don't ship document icons. They have terrible support for accessibility APIs. And so on...
Ctrl c does 2 completely different things. Terminology is different. Etc.
Problem is. Terminal applications standardised in the 70s. Guis in the 90s. Although that's being undone by web apps, and the influence of touch screens.
Then why isn't it done more often? I suspect part of the reason is that most gui frameworks are not really designed for making dense economical interfaces.
> you probably don’t need a user interface on prod. You need a command line interface on prod that
I don't need one, but I often want one. Being able to run vim, htop, etc. on servers is quite useful. Also, it isn't just prod servers. It's also nice to run tuis on VMs and containers that don't have access to graphics.
> I’m not really building applications for other people to use. I’m building them for me. TUI affordances are an awfully big hit to take to get Linux users of programs I don’t even want to publish.
Ok. That's fine. But then why are you trying to convince other people not to make TUIs? If I make an app you find useful, wouldn't you prefer me to make a TUI you can run on your Mac to a linux-only GUI?
Because it tends to require a lot of thought and effort along totally separate engineering pathways than most of the actual functional coding. Good UI design is not easy, and the skills to do it don't directly overlap with good coding skills. It's hard.
> I suspect part of the reason is that most gui frameworks are not really designed for making dense economical interfaces.
Don't blame the tooling for a broad failing of many developers. It's because constructing a genuinely dense and economical interface for most apps requires considerable time and effort, and frankly, most developers are crap at understanding how users other than themselves use applications. Those that aren't crap at it tend to have worked in large teams with dedicated UX people to help drag them away from their preconceptions.
Put differently, a GUI application seems to subject you far more to the whims of Apple or Windows than a TUI and perhaps even a different run context. Maybe the permissions fiddling for this is insignificant. Is there a clear user contract from Apple or Microsoft about this?
Naturally, this is a non-issue in Linux.
Me: you exist so I can launch iTerm2, Chrome, and VS Code
MacOS: oh, my god
I can't ever imagine using walled garden graphics api in 2026. Particularly for work tooling
They're fast, efficient, work across SSH, in a tmux, and survive the almost daily browser upgrades.
And they are easy to run as separate users without VNC or sandbox hell.
Like do you want to think about X/Wayland isolation, or do you just want to get shit done?
I was alive back then already, and was a relief to finally be able to afford GUI powered computers.
- keyboard can be used for pretty much everything
- less visually busy (no/less animations, background images/color variations. often close to plain text on plain background)
- more information dense (fewer things hidden behind hamburger menus, modals, drawers, etc)
- more/more easily customizable and documented
- more often can provide data directly into the terminal that I'm working in, so don't need to copy/paste through windows etc
these aren't necessarily impossible for GUI or universally true for TUI, but maybe I would guess just more an artifact of there being types of people who prefer working in the terminal who are then more likely to make TUIs, and I share their preferences more.
- They encourage developing simpler, information-dense, and keyboard-driven interfaces, which I tend to prefer in general.
- I find them more fun to write, and less aggravating to maintain and debug, than interfaces built with GUI toolkits.
- Most software I write for personal use is going to be run in both a Linux and Mac OS environment (and a lot of the time on a headless server to boot), and a TUI often has less fuss than other options (at least, IME).
- I think monospaced text in a terminal just looks cool.
Anything that lends itself to keyboard input, piping, text manipulation and a bunch of other things are so much faster if you're equally proficient in both input modes.
YMMV, of course, I'm just sharing my own experience and perspective.
It is, of course, completely subjective, but I know there are a fair number of people out there who agree with that statement.
If there was a powerful, information dense, configurable docking/tiling, multi-application-composing GUI framework (all things a good terminal can do) then yes, build everything in that. I’d love to see it.
Until then, TUIs are vastly better than another Electron app or Mac app using UI frameworks that are trying their hardest to look and act like basic phone apps.
So even professional tools all follow industry design and product trends which are too simplistic, don’t give you the info you need, and are optimized for abstract brand design or chasing metrics for the company, and not optimized for “how do I let my user accomplish their task as fast as possible”
If you want to automate from the perspective of what's on screen, tuis have no metadata. A button in a gui does have metadata associated with it, so you could automate 'click button x' where as in a TUI you'd be limited to clicking 'position x,y'
For GUIs you have to handle the particular metadata of a given framework, if it exposes metadata at all for the controls. Otherwise you're dealing with raw pixels which is an instant nightmare, though not so much now there are visual LLMs, but you're still paying for the tokens and/or extra processing.
I enjoy the usability and self perceived speed of not having to leave the console or reach the mouse to move a pointer when a cople of key presees is enough.
It's a matter of choice, I know, but I also think there is a genuine case for text over graphics. I see it everyday on businesses that keep old cobol tpvs at the cash, the speed at which the clercks check inventories or perform other crm activities is unmatched to mouse or touch interfaces
- Some are very good tools that have no clear GUI superiors. Vim, Emacs, Mc for example. GUI wrappers over these don't make them better.
- Efficient, meaning fast and limited in scope so you can learn the whole thing.
- The muscle memory you develop is transferable to any new platform or OS you use.
- Often free and OSS so they can't be taken away, making the muscle memory a long-term investment.
- If you work in the terminal for any other reason there's less context switch.
The low bandwidth / high latency lossy link problem is still a major issue. Cell phone service is sufficiently shitty in the area I live in that I have to use mosh to get a usable remote connection many times a year. Mosh is also quite useful while travelling! Got a crappy connection on a train or airplane? No problem. I have so much muscle memory with the mutt email client that nothing else has ever come close. Where's the equivalent of mutt's limit command in Thunderbird?
Try getting a usable VNC session when your cell phone hot spot is hitting 2-30 second ping times (yes -- thirty seconds) with only a handful of packets getting through.
Version 2 of the Amiga's operating system did a great thing in version 2 of AmigaOS which explicitly spelled out what conventions to apply (see Amiga User Interface Style Guide at https://archive.org/details/amiga-user-interface-style-guide). Modern GUIs that have weird meaningless buttons are infuriating. Even moreso when they lack obvious keyboard shortcuts.
I love how I have to configure font size and family only once in the terminal emulator and not for each program/GUI toolkit again.
Of course, there is no way to replace graphical-heavy programs (GIMP, QGIS etc.) with TUIs.
One thing I think which is overlooked is the raw feeling of speed a good TUI can give you. The lowest grade consumer laptop could probably download my little dictionary program and still get true search-on-keystroke performance, because there is virtually nothing slowing down that path, no animations, no nothing. Now most of that comes down to the data structures involved and making sure nothing ever has to go over a network, but I still appreciate the sheer simplicity of the approach.
It's also nice to know that I can probably recompile this same exact code 50 years from now and still get it to run exactly as intended, since GUI frameworks have waxed and wanted in popularity, but terminals truly don't seem to be going anywhere.
Recently, I posted a GUI mp3 playing program written purely in Tcl/Tk using shell scripts for some things. On my 2011 desktop, filtering for a specific mp3 across all the mp3s stored on my computer (about 80GB of mp3s) takes single-digit milliseconds. I can only imagine what a modern desktop, with an SSD, will do!
GUI apps don't have to be laggy. My Qt and Lazarus apps aren't laggy at all, and are far more snappy and responsive than the clear majority of TUIs I use (like CC).
- A lot of TUIs that run on your own machine and have a mature GUI alternative are indeed not necessary. I once tried to replace Spotify with ncspot, which was exactly falling into the "TUI is more advanced" trap.
- The real target of this article is people who treat TUI as a kind of identity, not people who have to work in a terminal because of their job and therefore have plenty of good reasons to use TUIs.
But the title and the general tone are still too clickbaity, which is kind of annoying. In reality, apart from a small portion of TUIs that were created just to follow the trend and clearly have better GUI alternatives, most TUIs are really just side tools that grow out of one premise that we have to work in the terminal.
And even those flashy TUI tools still make life easier for people who work in a terminal. If you prefer GUIs, fine, you don't have to use them, but that doesn't mean they have no users or no value.
That’s exactly what I thought too when I read the article. When the author mentioned the essay In the Beginning Was the Command Line I instantly knew it. I thought to myself the audience of the article must be the people who had read the essay and allowed it to influence their taste, as well as the people who might not have read the essay but conversed with people who did.
You like GUIs and want to build/use ones that works in MacOS, that's perfect go ahead and build as much GUIs as you want. However, this wouldn't work for me as I know how bad developing native cross-platform GUI applications is.
The skills are useful, thanks for linking them! They'll be useful for my future projects like this. FWIW I've had good luck with using Claude Design first to do mocks.
At the same time, I think there's still the "is this worth doing a hyper-customized version of this software, if I'm also then responsible for fixing the bugs?". But I think increasingly as we improve our agentic workflows and make them more self-contained loops driving the result to completion, this becomes kind of a non-issue.
In general, I do think LLMs are bringing about a new age of hyper-customization, which I welcome.
The whole point of a TUI is that it can operate under no-GUI constraints.
I can SSH to any server and use a TUI because SSH can handle that. Running an entire GUI and setting up VNC/RDP would be horrendous by comparison.
Alternatively, a TUI can be a part of an interactive CLI command where it maintains context to your CLI session. For example, I have a tool that helps wrangle AWS accounts and regions where the CLI momentarily drops into a TUI to help pick between regions and accounts if not already specified in environment variables. If I had to type all that out every time it would be very annoying. Hitting some arrow keys or searching with a few letters beats the CLI experience, and a GUI experience is not possible.
It seems like an absolute slam dunk for AI. Create an integrated AI screen reader app that observes the screen and talks to you. It wouldn't even need to be a huge model. It could be a small local model that can just read and interpret UI elements and feed the results as tool use calls into an accessibility layer. It'd probably be a custom trained accessibility model, could probably be based on an existing small computer use model fine tuned for the purpose.
This seems like the "accessibility solution to end all accessibility solutions" that totally takes the burden out of the code and puts it in one place that can be uniformly improved.
This is the way. It just wasn't possible 5 years ago.
Now it’s “native apps for 5 platforms done almost for free” era for anyone who cares.
See Meta’s new AI desktop app (no, I won’t install anything from Zuck on my computer, but I appreciate what they did).
Can you elaborate?
cmd+U scroll to beginning
cmd+I scroll to end
cmd+u page up
cmd+i page down
cmd+j one line up
cmd+k one line down
cmd+- font size decrease
cmd+= font size increase
cmd+0 font size reset to default
cmd+3 default font
cmd+4 alternative monospace font
cmd+[ grow left window border
cmd+] grow right window border
cmd+{ shrink right window border
cmd+} shrink left window border
cmd+9 switch background and foreground color
- cmd+space to open spotlight search
- type the name of the app, enter, it opens a new window with clipboard contents, i inspect it with keyboard hotkeys only
- cmd+q to exit or cmd+tab to switch back to the prev app
You can also just use the system console with tmux or GNU screen if you absolutely don't want any graphics at all.
The TUI programs usually do not consume a lot of resources. They are also very fast and optimized for keyboard-only navigation.
[1] https://en.wikipedia.org/wiki/Text-based_web_browser [2] https://github.com/browsh-org/browsh
People write code using the platform X because they like it. It doesn't make sense to try to stop this
My stances:
- Dev tools need a CLI at minimum, TUI for complexity
- User facing needs a GUI, CLI for power users
If you're building a TUI for a user facing thing, then yeah, you're doing it wrong, but if your target audience is devs then yes, PLEASE do a TUI, and make it nice.
I understand the sentiment that TUIs is not as accessible, sucks to use. I still do not like the fact that we should "stop making TUIs" altogether.
Let me give you an example from my carpenter friend in Cananda. He has a saw (TUI) and electric one (GUI).
He uses the saw in diff ways than his electric one. He sometimes extend the saw by attaching it to end of a long stick to cut tree branches he cannot reach. He can do it with the eletric saw but requires much more effort.
But if he needs to cut down the tree, he uses the electric one, and sometimes finish off (trim) with the regular saw (so the tree falls in certain direction).
The gist is, TUIs can be used to do things quick and dirty, and easy/cheap to make/buy. GUI even with AIs, still not as cheap as TUIs to build.
TUIs has a different usage in our field, where it's used more for automation/piping. If you have GUI, it's hard to pull it off unless you build the functionality (which coulda been a command in TUI), and expose it via GUI. the Authros is focused more on consumer side than builder side. With AIs we will all be builders using AIs.
Lastly, shooting down people building TUIs is saying, no more innovations, by building upon other's TUI inspirations.
Let people enjoy things.
- I'm team GUI! The author outline elegantly why. Sometimes I put command-line-like things in the GUI; I generally view it as a superset of CLI, with vastly richer capabilities. This is kind of interesting in the bioinformatics/structural bio space, which I feel is inherently visual, and benefits from 2d layouts. Most of the tools are CLI and based in I/O of stdout and text-based files. I am taking a different direction!
> Back in 1999, Neal Stephenson wrote an essay about command line interfaces that set the field of human-computer interaction back about 20 years. In it, he depicts the priesthood of Unix nerds wielding CLIs as powerful Morlocks, holding the entire computing industry on their shoulders. The Eloi use GUIs like Microsoft Word. Because this is high-test fan-service, “In The Beginning Was The Command Line” has become one of our field’s sacred texts, despite very little of it holding up 25 years later.
We should not take Sci-Fi and creative plot devices as predictions for the future. They can be inspirations and motivations for your own creativity, if tempered by realistic expectations and discipline. Neal Stephenson is especially prescient (Certain concepts from Fall; or Dodge in Hell and Anathem are hitting home strongly right now!). He spins vibrant, speculative yarns, mixing concepts, and estimating trajectories. The story referenced he is one that's in particular easy to see what parts did and didn't pan out. (And what Neal's more recent reflections on it say)
I think there is a bit of cultural identity (Which we are genetically inclined to insert in all sorts of places!) going on. There is a certain cyberpunk fun in SSHing a terminal, pipeling stdout/in together, and Viming. I suspect there is a high overlap between people for whom Linux is part of their identity, TUI use is as well. I.e, it's not about a practical weighing of merits; standard identity/tribal-based choices.
FOr use of TUI as a general file browser/command-executor... I've made my own. It's a GUI, and has built-in terminal for executing commands, with a visual file browser. And most importantly: Shortcuts to execute commonly-used workflows, history, bookmarks etc. All of which I feel like are glaring things missing from standard terminals. (PS, Bash etc). And having a terminal as one part of a multi-window program feels like a good use of my screen space!
Its also a bit funny how the author show cases a bunch of apps that only exist on a very specific setup, IOS Mac devices while TUI can exist everywhere a terminal can reach.
The right tool for the right job, noone would srsly use a TUI photoeditor but for many things the simplicity and constraints that a terminal introduces condenses design.
I got this article about UI density open since weeks and meant to read it https://mattstromawn.com/writing/ui-density/ But from what I could interfere so far..more UI frameworks, more white space, more wasted space.
Also..TUIs usually allow me a wide variety of colorschemes out of the box which is nice
1. CPU/memory performance on cheap or throwaway systems has a niche benefit. Mostly poor people. I'm still usually on an ancient Thinkpad with a 2nd gen i7. It runs native apps really fast to this day. Almost all GUI cuz I agree with the OP but my apps are TUI by default to keep them lean and fast.
2. Security. While I don't aim for it these days, it was much easier to make textual apps securely than GUI apps. Secure OS's from the 90's already secured console apps. TX, Nitpicker, and EROS made progress on GUI's but there's high complexity still. I'll note the OP's idea of GUI front ends is basically what we did to isolate the GUI part in a dedicated partition with messages it sends checked by the secure component.
Other than those observations, I'm with OP where I'm tired of TUI's if GUI's are that easy now. I considered trying it with some lightweight, cross-platform frameworks. Anybody tried some with cheap AI's?
EDIT
> CLIs have purposes for which they’re irreplaceable. Building a CLI is almost always a good idea. Building a TUI almost never is.
Oh, right. I can see that, but I go the other way: When I have a CLI, don't give me GUIs; give me files. Actually, just give me files and daemons. I'll do the rest, thank you!
I like chocolate I like vanilla
We run an internal agentic development harness we experimentally built for working on large-ish projects. first version started as a TUI, but it quickly became unusable for the large scale projects it was meant to work on (essentially PRD to PR, with planning loops, task decomposition loops and dependency aware execution / review loops). The TUI was too state-bound for another agent to drive reliably, and too cramped for a human to review at scale. We ended up replacing the TUI with a headless cli, and every interface we've built for the harness has become either a text artifact (mostly for agents to work) or a web view (for humans to drive and review).
With X based apps, you can run GUI apps through SSH. That was all the rage for light remote access to resource hungry desktop app a few decades ago...
But that being said, in my opinion, a few limited TUI apps are useful, but most of them nowadays are young "cool kids" that wants to be cool by using "terminal apps" without really having really embraced the spirit of the terminal and commands.
For them, using GUI apps would look like a basic computer user and not an expert nerd.
The magic of the terminal, is not that you run things in almost black and white with low graphics, but that you were supposed to use "simple" commands that you can scripts and pipe one through each others at will.
Since I wrote my own terminal multiplexer that puts terminals on infinite zoomable canvas [0] and I just can't get enough of putting weird stuff like asciiquarium, cmatrix or tty-clock all around (tried peaclock but it's slightly too fiddly for my taste). There are fun projects just to run in TUIs like astroterm or weathr, but I'd say that the space is still open to terminal toys.
The only problem I have with TUIs is that they're boring after a while but something that can be solved with often theme switching.
[0]: https://race-term.com (commercial)
Well, that's a new one, ha ha (guessing "tripping balls" on GLP-1).
(And here I am too, about to be glipping balls… like everybody else, I suppose.)
For something to become a GUI app (or a web-based UI) the GUI needs to offer something more valuable than a TUI can offer in terms of interaction.
I have one TUI app that wrote a while back to keep track of what I work on. It started off as a "wizard"-style UI. You choose project, then subproject, day, hours, activity and then hours and comment in a sequence of steps.
Arrow keys -> enter -> arrow keys -> enter .... -> comment -> enter
It takes about 2-3 seconds for me to do. And this is a time-sensitive task: if it isn't fast the user will not do it as often. There is a web interface to do the same thing, but the backend is slow. My TUI utility speeds that up by caching (between sessions) the things that the web UI has to look up (slowly) every time.
Then I wrote a more involved UI for it that "looks more like an application". Result? Now we were up to 15-20 seconds of interacting. Oh year it could do more and it certainly looked better, but it was less efficient so I noticed that I would postpone using it to record what I was doing. Because it now had friction. Lots of it. So a couple of weeks later I just did a hard reset and went back to the wizard style, minimalist version.
I would make a web UI, but that would burn 2-3 seconds just to navigate the browser there, juggle tabs (which I have too many of already), find the right browser to open the tab in etc. Once there do I keep it for "faster access" or close the tab? Keeping it is almost always the wrong thing.
There's a reason I have perhaps 15 different tools in my garage that can attach to a given fastener size. And there's a reason for that. Not all 19mm bolts are treated the same way. Some are low torque and live in tiny spaces. Some require precise fastening torque. Some need a lot of force to come loose - some will only come loose with force and I'm okay with destroying them. This is why I have so many tools that all do essentially the same job: apply rotational force to a piece of metal that has a standardized interface.
Similar tasks are not always best solved with the same tool because the context and the details do matter. Same goes for UIs. There are some things TUIs are very good at. Not least portability and remote usability.
UIs are also useful for information that is best represented in a UI. keep using those too.
You can bend each to be more similar to the other, and that's good too! Sometimes there is a good use case for that.
I'll just copy a comment I made about this in another thread:
TUIs are just GUIs that use a grid of characters instead of pixels. They are strictly worse than true GUIs, by definition. The only thing I liked TUIs for is running over SSH, but can any of these newer ones really run over SSH with any kind of decent performance anyway?
I think people are confused and think they like TUIs because they like them being keyboard driven etc. But this could all be done with a GUI.
The other factor is probably just fashion. Similar to how some kids are now listening to music on cassette tapes, which are objectively worse than other media in almost all respects. The less cynical take is it's like vinyl: it does come with compromises but gives us back some of the things we lost over the years.
The actual interesting text-based interface is the CLI. I've seen a few examples of TUIs that really should be a CLI and would be much more useful as such.
On the other hand, they are easier to build in a cross platform manner.
Linux is probably the most difficult platform to build GUI for, because of all the fragmentation, X11, Wayland, all the different flavors of GUI toolkits, driver issues, window managers issues, font issues etc.
TUI are a bit like web apps in that regard, using the terminal instead of a browser to abstract the platform.
a lot of boomer and millennial dev users love TUI bc that's what they grew up with. part of A\'s early success is hit that sweet spot and triggered a culture shift late last year.
as those age group phasing out, web ui will be dominant
There's more on this here: https://sockpuppet.org/blog/2026/05/12/emacsification/
maybe the answer is... you just don't need features like scrolling? this guy is trying to fit a square peg into a circle hole
Terminal, coding, and even prompt writing skills are language skills.
If you cannot use a terminal properly you are basically an appliance user. Like a toaster operator.
TUIs are not for everyone by design.
I’m with Tom on a lot of what he says. This is a nice reminder to take another shot at this.
It's inherently cross-platform (msys under windows). It allows to copy any part of UI. It doesn't require dependencies other than curses.
Back in the late 1980s, there was no question—the Mac changed everything, and everyone agreed that graphical was the superior interface. But we in the Unix world were slow to convert fully, because of a couple of problems:
* designing GUIs is hard, especially for programmers;
* writing GUI code is hard (and often tedious).
So good GUI apps really came from teams of elite designers and developers. The best ones all worked for Apple. There was an acceptable second tier that all came from major companies. If you wanted to write a custom one-off tool for yourself, however, you'd be in for some pain and the result would be jank.
But LLMs solve both problems. Point Claude at a description and some mockups of an app, and it will give you that app, or at least a slick-looking prototype that is suitable for your use if not the wider market's. So we no longer have an excuse. ALL our apps should be graphical, because creating graphical apps is now easy.
No, they didn't. The Mac's GUI benefitted in certain specific applications, like DTP, early on, but the Mac was an also-ran for most serious productive work for a very long time, and even when GUIs became standard in the PC ecosystem, it wasn't until the late '90s that productivity apps decisively moved from the DOS-based TUI world to Windows GUI applications.
> So good GUI apps really came from teams of elite designers and developers. The best ones all worked for Apple.
The folks who actually managed to build interface paradigms that successfully reconciled complex business functionality with sufficiently easy-to-master GUIs worked mainly for IBM and Microsoft. The CUA paradigm was the winner here.
People seem to forget that Apple had next to zero penetration into the business computing market, and nearly went bankrupt, back in the '90s.
As in, using graphics, vectors, but with a fully text-based UI?
Also the original Metro design as in WP 7/8/8.1 or Windows 8/8.1 relied heavily on the pure text, and the only place populated with a lot of icons was that iconic tiled start screen, although this is probably not what you mean.
Like Brooke Shields, I'd rather nothing comes between me and my LLMs.
Sure it looks pretty but it’s impossible to drive from the keyboard without mutant hands and savant level of memory for arcade combos. That means the entire OS relies on being able to drive the UI from a positioning device and thus you have to have the coordination to piss on an ant off a moving train. Which is terrible when the input devices are a touchpad, which varies so utterly frustratingly depending on whether it’s in the laptop or over Bluetooth. That and a mouse which is designed by a complete psychopath. That leaves you with third party options which require apps to not suck and they still suck. Scrolling on Logitech options anyone? Even LinearMouse can’t fix that shit.
Compare windows which works absolutely fine from any keyboard or any mouse and is discoverable and consistent. Also no mutant hands required.
TUIs are closer to windows than macOS. And that’s a good thing.
I did the last ten years on a mac and decided I was just hurting myself. Give me windows or a TUI. I notice Linux desktop environments tend to copy windows with respect to keyboard and discoverability too.
There's a shorthand that derives from the Phoenician string instrument : nabla . Latex has it as \nabla
Who said we're building them because we have to? I like TUIs, and often prefer them over GUIs, depending on the case.
You want it graphical: Make a webapp.
You don't: Make a TUI.
What exactly is the use case for native apps? Even lots of legacy code can build to WASM now.
(yes, I'm exaggerating a bit, but not much)
I'd take TUI apps over electron apps everytime as long as the UX is good.
Nowadays, you are more likely going to struggle with plugging your laptop into a 320Volt outlet than have a terminfo problem.
TUIs are a creative constraint.
TUIs are an invitation for programmers.
TUIs are doing "less" to render.
TUIs don't need to be "responsive".
Hello World is TUI.
Making GUI apps has been way too hard on all platforms since forever. Frankly I think it's gotten worse for 20 years. People started fleeing to web wrappers like Electron to escape the horrors of native UI development.
So if that was the reason why you made a TUI, take heed. LLM's can chew through UI frameworks for you.
But I solidly believe TUIs are often better than GUIs even if they were equally easy to make. They liberate your app from having to follow the fracture and fashions of GUI's - Liquid glass, Windows 8, SwiftUI, WinUI3, QT, GTK or whatever. A TUI allows muscle memory that you learned on Solaris in the 90s to work fine in windows 11 in 2026.
Easier to test too. Try testing your beautiful claude design React app end to end
What the what now?
You are free to build anything you like, and whatever you build won't affect me at all.
The author starts with a bunch of vibecoded slop apps that show exactly what I hate about 'modern' GUI apps, a lack of options and wasted screen-space. They look hip but are harder to use to me.
And now libghostty has caught up with a full implementation of the protocol in a second terminal emulator. I guess that mitchellh has plans for use in Superlogical... https://hachyderm.io/@mitchellh/117135178412268410
It’s much easier for the clanker to loop on testing / improving the TUI than a real GUI. It also means it can work on the app without an API/REST/HTTP/Java/Script ball of mud in the middle.
>[...]
>But remember: I’m not really building applications for other people to use. I’m building them for me.
Giant caveat to the whole argument and it's buried 80% of the way down. Granted, he alludes to that a little earlier, but if other people using your application are a completely irrelevant outcome, you're an unusual case.
I mean, sure, I make tiny one-offs, but those barely even have something you could call a CLI. I push as much complexity into command-line options and config files as possible.
Does anyone else find that this reflects a really obnoxious attitude, regardless of whether it would work?
Anyway, it seems like the premise is that people would only ever build a TUI because making a GUI was hard, and now it isn't because LLM slop will be good enough. But I'm firmly convinced that fails on both counts.
The primary value statement of TUIs is that they work over SSH with remote systems.
But now, we're hearing rumbles about the limitations of the "modern" terminals with regards to working well with a "rich" TUI. How a raw tty doesn't really cut it since it, at a minimum, can't detect key presses, only actual character/byte patterns. How there needs to be a new protocol, a "better" terminal client.
Of course when you go that route, then the current terminal emulators will need to be updated, or you'll be required to install a different terminal emulator that supports the new protocol. But that kind of defeats the theoretical, historical ubiquity of a TUI. It would not surprise me that many modern TUIs only work with an ANSI terminal.
Back in the day, we had systems like Visual Basic, Power Builder, SQLWindows. GUI/language systems particularly well suited for Client/Server DB development.
For a time, Firefox actually presented itself as a modern incarnation of these tools. Using XUL and JS for UI development, built in utilities to talk back to servers over HTTP. It was a platform for "Rich Internet Applications(tm)".
It didn't really take off, it wasn't documented very well, kind of buggy. It also required Firefox as a client. The potential was there, but not quite realized. Ajax hit the web browser in full force and that pretty much was the last nail. Now, you could use "any" browser for more interactive applications.
Of course, we advance to today, where many sites work in any browser as long as its Chrome. We have advanced applications that are no more than a Canvas element, with everything else being reinvented from scratch. "Have bitblt, will travel."
The browser is the closest thing we have to a universal remote GUI, but we all know the issues with it for many contexts. As an application platform, it may be ubiquitous, but that doesn't mean it's at all lightweight.
We used to have lightweight, remote GUI applications. Rootless X Window applications. Folk don't necessarily want the entire desktop, they just want an app, with some fields and buttons. You could slap them together with TCL/TK. Remember dtksh? We're not talking about trying to write Adobe Premiere or AutoCAD in these things. We just want some simple utilities, fields, icons, maybe a chart. Scrolling tables and a menu.
But we can't do that today, not readily, not easily.
So, we're kicked back to 1978 and the rise of the Smart Terminal, instead of 1984 and the rise of the X Terminal.
If folks are going to write out a custom client for a TUI, then may as well go the extra 10 feet and make it a remote GUI client. Reinvent all of the wheels as they go round, and round again.
TUIs, at least to me and some friends in my circles, do something that native applications don’t: demand focus from the user.
Sure, modern terminals and TTY environments have visual tabs and virtual terminals that let you quickly toggle between multiple apps, and you can of course manually code your screen to split apps across a single “window” if you really want to get into the guts of things.
But for several folks in my close circles (myself included), the fact that each TUI occupies an entire screen by default, forces us to focus on what’s going on and what tasks we’re trying to accomplish in a way native UIs don’t. On the desktop I can flit between Discord and Telegram and multiple games and several Firefox windows full of tabs and the video encode I’m working on and hey did that bank transaction go through let’s check my email except UGH music streaming went off the rails so let me go back to my client and-
You get the idea. For folks who lack the neurotransmitters to maintain consistent focus or wave off distractions, a terminal environment acts as an accessibility device in and of itself. btop on 1, irssi on 2, Hermes (Qwen 35B) on 3, with 4 as my primary workspace and 5 as a fallback. Simple. Clean. Efficient. Lightweight.
I generally concur with the author that we should be “summoning” (I love that word for reasons that deserve their own essay/comment) more native UI apps that are as information-dense as, well, Bloomberg Terminal. There’s no reason we can’t build something as powerful and focused for the everyman for, say, inventory collections, or K8s constructs, or network architecture, all in a native UI. There’s nothing wrong with the GUI, only that at some point we let ourselves get suckered for the narrative that GUIs have to be easy, simple, and opaque, as opposed to transparent and (logically) complex. We can and should build GUIs that flow logically like the commands they’re executing behind the scenes; hell, we should be able to open a console log of every action we took in the GUI and the associated command(s) it executed, so we can better learn to transpose from GUI to CLI ourselves and build more capable or complex automations.
GUI isn’t bad. TUI isn’t bad. What’s bad is presuming one or the other is superior to everything (or for everyone) else, or worse, assuming one cannot function in a given way comparable or equivalent to the other.
Lists a bunch of things .
Fails to make any clear points .
Fails to give real reasons for the few claims it makes .
I can see how LLMs help.
I wouldn't want to maintain software for other people entirely vibe coded. I think I might think again about native app development for my own needs, if it's kind of disposable.
People talk shit about LLM writing. Well, here's a prime example of why I like LLM-speak. Does this article have personality? Yes. Attitude? Yes. Is it structured and accessible? No.
I'll take clarity over personality any day of the week.
Of course, I didn't mean to imply that there's a dichotomy between personality and clarity. There's a dichotomy between ordinary LLM writing and personality.
Here's an organically grown summary:
• TUIs suck because they are primitive, often buggy, hard to write and aren't accessible. Their existence isn't due to any real technical advantage but more because UNIX historically had a very bad UI toolkit (Motif) which established a 'culture' of TUIs.
• It's now easy (on macOS) to write native apps that use SwiftUI and give a much better GUI. You can just vibe code them.
• Some specific skills, features, templates etc are linked which look useful if you agree with this approach.
• You don't have to give up remote access because there's no reason the GUI has to run on the same machine as the thing it controls. A native GUI can just SSH in to a remote machine and run non-TUI CLI tools to control it. Lots of apps have worked this way and it functions fine.
• On the other hand, TUIs are portable (ish). The author concedes this may sometimes be useful.
I prefer my articles to be coherent without having to wade through pages worth of irrelevant material.
- All those GUI windows look the same. How the fuck do I tell one app from the other?
- The abysmal corner radii are a clear indication of terminal macOS. Pun fully intended.
i.e. ignorance is bliss in the world of typography.
The total abandonment of the form-follows-function principle on the part of purely visual designers, who worry about what software looks like without considering what it's for, has led to 10-15 years of cumulative usability degradation in almost all categories of software. TUI and CLI software remains one of the last bastions of people actually designing interfaces for functionality and usability.
Just because there have been some crappy industry trends doesn't mean the answer is to turn away and permanently regress to the confines of a terminal emulator.
And sure, for programs for one go ham on the ui.
No, for the love of god never stop making TUI's.
Who tf cares about what the author who appears to be new to building native desktop apps (and vibe codes them) says about terminal apps?
I am not writing my code 3 times for each native platform even for agents and then dealing with signing keys just for it to break on a OS update. Terminal apps for developers survive all of this.
The entire article is just a massive rant with woefully weak points. I guess agents really does amplifies both Gell-Mann Amnesia and Dunning-Kruger in others.
So one reason to write this post is just to send a bulletin to developers like me, for whom it wouldn't have even occurred to build native UI before. Native UI is now a thoroughly solved problem. All "off the rack" user interface is solved now.
But that's not the biggest thing happening here.
What's really going to destabilize us is what this says about computer usage and computer programming. When I was a little kid, in the mid-1980s, I imagined all sorts of new different things I could do with a computer, if we ever got one besides the ZX81 clone that plugged into our TV.
I had to grow up to become a computer programmer to learn that one doesn't simply tell a computer to do new things, that there's an elaborate ritual to build anything useful, and it takes years to get comfortable with those rituals. And like most other programmers, and really craftspeople of all stripes, I came to appreciate the rituals and the specialized knowledge. They're part of my identity, so I tend not to question them.
But there's always been this dividing line between computer users and computer programmers. It's rarely disrupted. It happened once with spreadsheets (our profession has a sort of Kubler-Ross thing going on with the fact that Excel formulae are the world's most important programming language), and maybe just a little bit with HTML in the early web. Other than that, we've all been pretty siloed.
That's obviously about to change, in a more significant way than it ever has before. Computers are going to work the way I assumed they did when I was 7. The line between programmer and power user is going to dissolve.
I'm not interested in what that does to our profession or the question of whether or not there will always be a need for serious software craft or engineering or whatnot. Totally valid question, but not where I'm coming from.
I'm interested in what systems look like in this new world we're heading into. What is an operating system in a world where most applications are summoned by the person who's going to use them? What even is an application at that point? What are the interface idioms when we're truly no longer constrained by text inputs to compilers and program building tools? Does everything look more like Smalltalk? Like a Lisp Machine? Or like something weirder? It's gotta be something. It can't possibly be the case that the shape of computing we all count on today is going to survive the next 15 years; it'd be like driving steam-powered automobiles.
That's the fun question.
> Our field has a weird relationship with terminal and command line interfaces. The time has come to re-evaluate it.
Why? Why is infatuation weird? Why should we reevaluate it? The whole premise sounds like a "give me a hot take on XYZ".
(I built itter.sh so I'm biased)
/s
AI slop spam.
I don't want to use those AI sloppers. They often are not based on any intelligent design.
A good counter example is, in my opinion, htop. Many people use it, despite its ncurses-nature. It simply does what it wants to do well: give an overview over resource usage.
Yes, AI slop can spam-generate a clone, but why would I want to use that over the real thing? It makes no sense.
> I’m not packaging this application up. If you want it, just screenshot this section of the post and give it to Claude. It’ll build something useful.
I don't want this spam slop anyway, but notice that the author became super-lazy. Rather than providing source code, he tells you to copy/clone it via Claude spam. So this human is now too lazy to do anything useful. Skynet won. He has been AI slop spam absorbed, without even noticing. Just like in the movie Invasion of the body snatcher (great movie, both the original and the first remake, though I liked the remake more, because Sutherland was in his prime back then).
The author continues to show more AI slop spam created useless things. None of which is interesting. I am surprised people now blog about boring AI created software and call it engineering. So, nah, don't stop making TUIs. Instead, design them better.
Having said that, I think ncurses is the wrong tool in general. We need to be able to design both TUIs and GUIs in one go without barriers. Every time I have to use ncurses, I do indeed curse.
and it "worked" - the audience he found for this is bigger than the previous one. So this continues until the AI psychosis become so high it start to turn people off.