Could you elaborate? I mean, I get the usual criticism of web/Electron-based desktop applications (slow, includes a whole Chrome engine, non-native UI, …) but Claude Code isn't one of them?
Search for twitter drama where Anthropic engineer says basically the following [1] and gets shamed by what AAA game developers are able to render at the same time.
tldr; the ui you see in the terminal is react-based in claude code. As for opencode just see their repo on github.
Looks like there're two main approaches to AI-first development: (i) favor slow responses to produce an upfront high-quality result, (ii) favor quick responses to enable faster response-test-query iteration. And, based on comments read here, seems Codex isn't too fit for the later. Optimally a developer should be able switch between the two approaches depending on problem at hand.
I usually just edit ~/.config/mimeapps.list directly.
I can see how that tool can be useful, but only if it's included in official repos. Editing mimeapps.list is simpler than the hassle of downloading and building this tool.
At first I thought it was going to be some kind of solution to force all linux apps to adhere to the XDG Base Directory Specification, until I realized this related to a different specification altogether (XDG MIME Applications specification).
Strange. Middle-clicking the link opens a HackerNews frontpage, but copypasting into a new tab shows the article. Presumably the server shoos away referer links? To reduce load maybe somehow? Or maybe something's weird in my own configuration idk.
No one on earth has so far managed to get xdg default apps work on Linux. I've been failing since 19 years personally. If you've really succeded then congratulations!
> No one on earth has so far managed to get xdg default apps work on Linux.
I've only been using Linux for a few weeks but what am I missing here?
I set a bunch of mime types in `~/.config/mimeapps.list` which are assigned to desktop apps and they all open perfectly with `xdg-open` or when I launch them through a file manager.
For me currently, when trying to open a `text/markdown` file, there's a disassociation between what my file manager (Caja) runs (own bin/emacs script; was under the impression it was auto-creating a .desktop file), what mimeapps.list have (emacs.desktop), and what `xdg-open` runs (Firefox for some reason).
> also of note, we (mostly) don't allow ~/Desktop as the download dir for security reasons
This isn't an XDG issue. It's a chromium engineers being silly pricks that think they know better than the power users who obviously went out of their way to create such a configuration. Also I bet it would work if you set your XDG_DESKTOP_DIR to ~/Download/
It's when I want to use a non-systemd, no-DE environment that xdg-stuff becomes very annoying, but that's usually because applications assume a certain setup rather than any fault of xdg. eg. Wayland is very stupid about requiring a certain xdg setup to run at all.
These days I pretty much write this kind of software in ratatui with LLM when I need it. But the idea is nice. I like that. LLMs are the return of the value of the ideas guy!
But this seems human-written? Then it is interesting. Thank you for sharing.
Nice, but problem with all these AI coded TUI's is we will have hundreds of them, best to stick to the built in linux commands, add aliases/abbreviations (fish) if required, do you need a TUI for everything? Sometimes the answer to "Should I write this?" Is no
I do agree with some of your sentiment. But by that logic, nothing would ever be made.
The same goes with aliases. Why not just use the actual commands. You give it your best shot, and sometimes something good comes out. And sometimes it's crap. That's life.
And I made it for fun and to learn something. And it wasn't AI coded. It's like 200 lines. I wanted to learn termbox2.h a bit more than I already had.
Yea I was pleasantly surprised by how simple the code is when I read it. Honestly a great example of what termbox2 is capable of. Very nice!
And now I know about termbox2, which looks very cool. Looking through the example projects[1] in the README I also found ictree[2], which does exactly what I was looking for yesterday (turning the output of `find` into an ncdu-like/interactive tree interface). I didn't manage to find something for that through googling around or asking LLMs, but thanks to you posting this here I did, so thanks!
I absolutely fail to see the problem and I think the whole "best to stick to built in linux commands" is an utterly dinosaur-esque take that can, and will, and should, go extinct in the age of AI-assisted coding.
It is interesting that somehow every conversation now pivots to LLM's. It's almost like people are paranoid or something. I have no issue with AI. But it should be used carefully when learning/working so you don't miss the little details that usually make a big difference later. But to each their own.
It is just getting tiring that people assume more and more that things were written with AI for everything. It's like, OMG, can you stop it for a second. And who cares, really. Do your due diligence, check the code and decide for yourself. But maybe, this is just projection. Or a nice way of insulting/dismissing people, which I find quite funny.
And like you said, the age of AI-assisted coding is already here. There is beauty in piping core utils together and being really productive with them. No doubt about it. But there are also new ways of computing emerging, and we should learn about that too.
Thank god it’s not another React, Solid, Typescript, what have you, web abomination inside the terminal (claude code, opencode, I’m looking at you).
Bravo!
> claude code, opencode, I’m looking at you
Could you elaborate? I mean, I get the usual criticism of web/Electron-based desktop applications (slow, includes a whole Chrome engine, non-native UI, …) but Claude Code isn't one of them?
Search for twitter drama where Anthropic engineer says basically the following [1] and gets shamed by what AAA game developers are able to render at the same time.
tldr; the ui you see in the terminal is react-based in claude code. As for opencode just see their repo on github.
[1] https://news.ycombinator.com/item?id=46706040
That's where codex-cli shines (rust)
the startup time is crazy, you can start writing as soon as you hit the command
(I don't use codex, just noticed that it's crazy fast)
Sadly the fast startup time is overshadowed by the slow response time of the codex agent
I don't get it. Even in ChatGPT I always use Pro model with maxed-out thinking budget (selectors available only on web and windows).
Let them cook as much as they can.
Looks like there're two main approaches to AI-first development: (i) favor slow responses to produce an upfront high-quality result, (ii) favor quick responses to enable faster response-test-query iteration. And, based on comments read here, seems Codex isn't too fit for the later. Optimally a developer should be able switch between the two approaches depending on problem at hand.
I usually just edit ~/.config/mimeapps.list directly.
I can see how that tool can be useful, but only if it's included in official repos. Editing mimeapps.list is simpler than the hassle of downloading and building this tool.
Just by reading the title, I’m sold! This should be very useful specially if you are not using a desktop environment that manages the default apps.
I always alias open to xdg-open, it’s so useful to open a file directly from the terminal.
That was the exact reason for it. I made my own window manager for fun and was missing a simple way of changing default apps.
Feature suggestion: The ability to add/remove more specific mime entries such as video/mp4
I haven't thought about this. That makes sense. I will add this.
At first I thought it was going to be some kind of solution to force all linux apps to adhere to the XDG Base Directory Specification, until I realized this related to a different specification altogether (XDG MIME Applications specification).
Making home read-only can get there, and I can't find the project that forced using XDG paths. I think it used LD_PRELOAD or somesuch trick.
edit: https://soc.me/standards/defending-home
Strange. Middle-clicking the link opens a HackerNews frontpage, but copypasting into a new tab shows the article. Presumably the server shoos away referer links? To reduce load maybe somehow? Or maybe something's weird in my own configuration idk.
They explicitly check if the referrer is hackernews and do that.
I had the same issue as you. Had to copy and paste the link.
There are also replacements for xdg-open with more sane (in my opinion) configuration, like jaro: https://github.com/isamert/jaro
No one on earth has so far managed to get xdg default apps work on Linux. I've been failing since 19 years personally. If you've really succeded then congratulations!
> No one on earth has so far managed to get xdg default apps work on Linux.
I've only been using Linux for a few weeks but what am I missing here?
I set a bunch of mime types in `~/.config/mimeapps.list` which are assigned to desktop apps and they all open perfectly with `xdg-open` or when I launch them through a file manager.
It is documented in the XDG specification https://specifications.freedesktop.org/mime-apps/latest/file....
For me currently, when trying to open a `text/markdown` file, there's a disassociation between what my file manager (Caja) runs (own bin/emacs script; was under the impression it was auto-creating a .desktop file), what mimeapps.list have (emacs.desktop), and what `xdg-open` runs (Firefox for some reason).
Older desktops don’t follow the specs and xdg-open does different things based on the desktop, so they indeed can get out of sync.
I’d have to look into your specific case but `gio mime` and `gio open` do the right things.
> what am I missing here?
There are gotchas, for instance Chrom,{e,ium} insists on XDG_DESKTOP_DIR != XDG_DOWNLOAD_DIR.
See this bug report from a confused user: https://issues.chromium.org/issues/41076564
> also of note, we (mostly) don't allow ~/Desktop as the download dir for security reasons
This isn't an XDG issue. It's a chromium engineers being silly pricks that think they know better than the power users who obviously went out of their way to create such a configuration. Also I bet it would work if you set your XDG_DESKTOP_DIR to ~/Download/
This is XDG base directories, not XDG default applications. They're two different freedesktop specifications.
Great. I must be living on the moon then. I guess gnome work great there since it manages this part
Gnome has done an amazing job at this, I agree. You don't even notice this issue.
It's when I want to use a non-systemd, no-DE environment that xdg-stuff becomes very annoying, but that's usually because applications assume a certain setup rather than any fault of xdg. eg. Wayland is very stupid about requiring a certain xdg setup to run at all.
I probably haven't. :) They are a nightmare indeed. But it does help a little.
These days I pretty much write this kind of software in ratatui with LLM when I need it. But the idea is nice. I like that. LLMs are the return of the value of the ideas guy!
But this seems human-written? Then it is interesting. Thank you for sharing.
Yeah, I did write it myself, yes. It's actually a very simple program.
I heard so many great things about ratatui. I am, however, not well versed in Rust. Only did a couple of toy little things in it.
Looks neat!
Nice, but problem with all these AI coded TUI's is we will have hundreds of them, best to stick to the built in linux commands, add aliases/abbreviations (fish) if required, do you need a TUI for everything? Sometimes the answer to "Should I write this?" Is no
I do agree with some of your sentiment. But by that logic, nothing would ever be made.
The same goes with aliases. Why not just use the actual commands. You give it your best shot, and sometimes something good comes out. And sometimes it's crap. That's life.
And I made it for fun and to learn something. And it wasn't AI coded. It's like 200 lines. I wanted to learn termbox2.h a bit more than I already had.
Yea I was pleasantly surprised by how simple the code is when I read it. Honestly a great example of what termbox2 is capable of. Very nice!
And now I know about termbox2, which looks very cool. Looking through the example projects[1] in the README I also found ictree[2], which does exactly what I was looking for yesterday (turning the output of `find` into an ncdu-like/interactive tree interface). I didn't manage to find something for that through googling around or asking LLMs, but thanks to you posting this here I did, so thanks!
[1] https://github.com/termbox/termbox2#examples
[2] https://github.com/NikitaIvanovV/ictree
This whole tool is much more a display of what termbox2 can do in couple of lines tbh :) You are so right about that.
It's an amazing library. And all that juice in one stb-style header file. You just gotta love it.
And if you are interested in such small libraries, I have a Github list with a bunch of them that I found.
https://github.com/stars/mitjafelicijan/lists/stb-style-mini...
Let people make and use what they want, you don’t have to use it.
The nice thing about AI coded TUIs is they're so easy to make you can make the ones that suit you yourself and ignore the rest.
reading the code, what make you think it was vibe coded?
I absolutely fail to see the problem and I think the whole "best to stick to built in linux commands" is an utterly dinosaur-esque take that can, and will, and should, go extinct in the age of AI-assisted coding.
It is interesting that somehow every conversation now pivots to LLM's. It's almost like people are paranoid or something. I have no issue with AI. But it should be used carefully when learning/working so you don't miss the little details that usually make a big difference later. But to each their own.
It is just getting tiring that people assume more and more that things were written with AI for everything. It's like, OMG, can you stop it for a second. And who cares, really. Do your due diligence, check the code and decide for yourself. But maybe, this is just projection. Or a nice way of insulting/dismissing people, which I find quite funny.
And like you said, the age of AI-assisted coding is already here. There is beauty in piping core utils together and being really productive with them. No doubt about it. But there are also new ways of computing emerging, and we should learn about that too.