Exactly my thoughts! I was expecting a Rust app written from scratch available at least in Qt and GTK flavors , and maybe with a third and a fourth option.
Don't they have agent swarms with infinite tokens?
They replaced a pretty decent native Swift Mac app (with only some chronic state management bugs), weighing ~100MB ish, with the Codex app that most people don't need, made in Electron, which takes minimum 5GB+ including filesystem cruft. Today I ran into a memory leak that froze the app and made the Renderer process take 5GB, then 10GB of RAM.
That was after I asked it to only touch 3 text files in its working directory, after which it sparked macOS permission prompts to access my Desktop, Documents, Downloads, and other apps' data. Thank god for the web platform and web apps.
"The American private sector is the most innovative and
technologically advanced in the world, and its scale,
speed, and capacity secure a critical offensive cyber
advantage for the United States."
Since we now have three "Codex"es, I think it's worth specifying you're talking about codex-cli/tui. Codex the hosted version and Codex the GUI are both fully proprietary I think (besides the codex-cli/tui parts they use, I'm guessing mainly the app-server stuff).
How do I get it to fix my Bluetooth if I do that? If you've been AI-pilled, 2026 is the year of Linux desktop because instead of dicking around with config files, I can just tell AI to fix python.
Bluetooth is the one thing I had problems in the past on GNU/Linux, Windows, macOS, Android and iOS. Linux is the only mentioned platform where you could technically give Codex root and let it fix it :)
I guess OpenBSD is the only OS where I never had any problem with Bluetooth audio.
Also note, that RAG and even vector search are more paths of the early days that didn't prove too valuable. Just let your agent search it directly and optionally create an index as a default entrypoint for common topics.
This will make it much easier to trap people into installing it without any isolation just to their primary user account or even make it run with admin privileges by default.
With the recent security related events we already see that it seems to be wide spread practice that many operators avoid the hassle of isolating their agents.
It is just an incredible cultural experience to see how quick stupidity gets accepted as a behavioural default.
Bubblewrap gets a bit confused inside a Docker container. If anyone knows how to run Codex inside a Docker container on Linux without it constantly complaining about a broken bubblewrap, please tell.
> If anyone knows how to run Codex inside a Docker container on Linux without it constantly complaining about a broken bubblewrap, please tell.
Why would you even use bubblewrap once it's running in a container? I've been running Codex in a container since it was publicly available, not even sure what you'd use bubblewrap for when you already separated codex from the rest of your filesystem?
With claude, the difference I noticed is that you can run their cloud enviroment from GUI app, as well as connect to remote machine. Sure, remote connection is nothing interesting, but I'm not aware if claude cli can work with cloud based enviroment.
Also claude design is alvaliable on claude gui app.
But all of these can be accessed with the browser as well, only the local machine access is something that browser version doesn't have. But for that you can use cli...
Really mess of the tools they are shipping. Waiting (not) for Claude/Openai OS
Latex rendering, inline images, inline browser showing what it's clicking on, being able to view a spreadsheet and then select a region to reference in the conversation, clickable links when it references a specific line number with mouse-over previews, interactive inline visualisations.
There's probably more I don't remember too. In my opinion TUI apps are just silly. You don't get any of the advantages of it being just plain text because it's all wrapped in funny Unicode characters and at the same time the GUI capabilities are hamstrung by being text on a grid.
I'm only familiar with the Anthropic equivalent, but the cognitive load is different as opposed to necessarily better or worse. I have friends who prefer mousing around the UI.
Tabs for long-running often-compacted sessions is a difference I've noticed in emergent UX.
I use the CLI, too, but I enjoy watchinng people use the desktop apps. It seems like the desktop apps may be a little bit better suited for human-in-the-loop work. If you're already switching back and forth between a mouse, having an IDE running may make more sense.
The best feeling has to be in the CLI, in my oppinion. A low-stakes project where you can let ultracode runaway from you in the CLI is the most fun.
I think these styles of apps will be popular for OpenAI and Anthropic as they help solve some of their moat issues. The deeper you integrate someone's files into their app, the harder it will be to switch AIs
ok, but you know these things are marketed as super products; it's pretty easy for me to use a local model to scale up almost any project on github. Why would this be an effective moat beyond "I use codex too!!"
For someone who has not tried these desktop apps: what is the the advantage versus cli codex with some MPCs and multiple folders each one with their context files?
The same as an OS with a UI. Discoverability, fewer actions per activity, easier navigation, faster to learn/lower skill floor, etc. Some highlights:
1. Much better multi-agent orchestration. The app supervises many agents, with different project views and built-in git worktrees, so agents can work on isolated copies of the same repo. It's easy to jump between tasks and see changes. I recall OpenAI wrote that CLI was not designed around this style of work.
2. The app combines multiple terminals, file previews, PR-review workflows, SSH connections, an integrated browser, and page annotations. For front-end stuff, Codex can run the app in Codex for mockups/prototypes, then iterate inside the app. IMHO I think this is a much faster, more intuitive flow.
3. For long-running and recurring agent workflows, I think they're much easier to manage in the app. You get automations, review queues, persistent conversation context, future schedules, and auto resume for longer running tasks. E.g. CI failure summaries, issue triage, and release briefs.
You can do some of the above, to some degree, with varying levels of success in the CLI, but I think it's miles ahead in the app.
You get GUI niceties: a nice presentation of the output like inline markdown rendering including tables and images in a nice font. Input supports drag & drop, and I can actually navigate the input field the way I am used to (pointing and clicking, using keyboard shortcuts to jump to beginning of words / sentences).
Not sure what the most modern CLIs support, but for me just the ergonomics of a GUI text field alone are superior.
I'm not ready to try this yet, since after some recent experiences[1] I've decided not to be at the cutting edge of new tools and models for a few weeks or months.
However, from the screenshots and description, I expect that you would have these advantages:
- History. Currently, if you type into the CLI (TUI) your intention and requests are captured in some transcripts, but not in conversational turns. If you want the history of what you typed into the cli/tui you would have to datamine it from transcripts. By contrast, on the screenshots you can see it is a simple conversation that's saved and easy to return to.
- Multiplatform. As per the announcement "Use Codex across ChatGPT, your editor, and the terminal", meaning that you could open the ChatGPT app on your phone and continue to interact with this. While this might not matter to you if you exclusively code while sitting in front of your computer, still, it helps you be able to stay on top of things if you are on mobile.
- Background tasks and autonomous work. It says you can schedule it to do things in the background (issue triage, alert monitoring). At the moment, it would only do this if you've asked either codex or another agent to build a harness for it that does so. (i.e. you could ask Claude to build a harness that calls Codex on some schedule or interval.)
These all seem pretty good and I look forward to starting to use it once it's been more battle-tested.
[1] notably, on the Anthropic side the Opus 5 regressions - I had to downgrade to Opus 4.8 as I found Opus 5 was unreliable. I currently use Opus 4.8 and Codex Sol 5.6.
Does this mean it is possible to talk to Codex from ChatGPT on a mobile phone now?
When I tried that with the Codex CLI version on a Linux VM, I did not get it to work. Possibly because OpenAI only supports connecting ChatGPT to a desktop installation of Codex?
I have a codex cli instance on a Debian VM connected to ChatGPT on iOS via SSH. That works since.. about a month and has pretty good UX. Before, I used the shellfish terminal and tmux which also worked really good, but to be honest (even though I'm a big terminal fan), ChatGPT on iOS sometimes has a smoother UX than a terminal.
Are you sure ChatGPT on IOS connects directly to the VM? Or via a desktop as the middle man sitting somewhere?
If you really only use ChatGPT on IOS and a VM and nothing else, then I would be curious how you set that up. I do not see a "connect to codex via ssh" in my ChatGPT app. I only see "Connect to a desktop".
Oh HOLY MOLY! Now I see there is not only a "Remote" section in the app but also a "Connection" section where you can add an ssh connection. I need to try that.
I'd love to see the stats on how many Linux users want this. Most Linux usage probably comes from WSL/SSH'ing into servers, where they would use the CLI instead of a GUI. Even for the people who run Linux on the desktop, most probably prefer the terminal anyways
Why not create a separate account on Linux just for the agent?
This security measure should also be enabled by default. Many packages REQUIRE the creation of an account, but agents do not. Interesting.
I tried it yesterday. The experience seems smoother than on Windows. I hope Computer Use will be added soon.
Also, when will OpenAI fix the issue where Xhigh and Ultra are both translated as “极高” in Chinese? It’s been this way for quite some time. As far as I know, the proportion of Chinese employees at these Silicon Valley AI companies is quite high.
I think their backend data probably shows that very few Chinese users are using Codex through the official way, and the number of those who have switched to the Chinese interface is even smaller, so there’s not much incentive to fix this issue.
You have "send feedback" button in the app. Something tells me that it's much better place to report issues with the app rather than comment section on unrelated website.
Access to your computer is the entire point of the app. Codex computer use can script and control any application on your computer.
You can give it any task on your computer and it can automate it.
Such as creating a photo on Photoshop, creating 3D models, editing videos, or browsing any website without scripting.
Honestly, you're missing out. The productivity gains I've had with these tools is unbelievable. I understand the risk though, so if you've got government sensitive data on your computer, or no backups, don't use the "full access" option. Babysit what it does instead.
Depends how much you are using it - for a casual user the $20 tier gives you just about enough use for hobby projects etc.
Or if you are doing full time development then you just need to pony up for the 20x tier if you want to use it.
Running Codex locally removes so much friction and is so fast compared to either running it on the web (where it has to build environments online etc which consumes time) or compared to copy-pasting chatgpt that it's just a no-brainer.
I don't know about Chatgpt but I used Kimi (they have a Chinese-only plugin that does desktop control - works fine if you enable it even if you don't know Chinese). Observing how it works, it's mostly using the accessibility features of the user interfaces - it fetches a sort of DOM for the desktop applications. Only rarely it has to take a screenshot.
Literally everything from generic coding agent to shopping.
I run it in a MacOS vm, fwiw, for certain tasks. I usually hand it a repo and say “install all build dependencies” and it’ll click through installation GUIs where necessary. I wonder if the TOS of an application is enforceable if a bot agreed and a human never saw it.
It can UI test web apps if that’s part of your job, but it can also just use the browser for tasks. I’ve had it research flights by checking different dates automatically and compiling the results for me. I’ve heard others use it to handle building a shopping cart for instacart/grocery delivery.
Probably the hero usecase for most people would be app/web development. It can control either a browser or a device simulator and can iteratively build and test the application, by clicking real buttons.
I upgraded my workflow a few months ago from "copy-paste things in and out of ChatGPT" to "use an agent that edits my project files and runs tests on its own" and the ergonomics are just so much better and enables automating bigger tasks. I still monitor everything it does and do manual adjustments so I feel ownership of the code.
But isn't that use-case solved a harness like OpenCode or Codex? OpenClaw and Hermes are a bit different beast although they can also be used for development in a more holistic manner (e.g. automated GUI testing).
I'm just wondering about other uses where such a system would be useful aside from asking it via Signal or WA for the weather.
- Electron based app: Electron is a framework sold on the basis of enabling rapid cross-platform development at the cost of performance.
- Frontier AI company: AI is sold on the basis of enabling rapid development
- App was released in February & took 6 entire months to port to Linux
Don't they have agent swarms with infinite tokens?
That was after I asked it to only touch 3 text files in its working directory, after which it sparked macOS permission prompts to access my Desktop, Documents, Downloads, and other apps' data. Thank god for the web platform and web apps.
Give it a full desktop in a VM if you want to, just not direct access to your system.
Also in the news: https://www.whitehouse.gov/presidential-actions/2026/08/expa...
Since we now have three "Codex"es, I think it's worth specifying you're talking about codex-cli/tui. Codex the hosted version and Codex the GUI are both fully proprietary I think (besides the codex-cli/tui parts they use, I'm guessing mainly the app-server stuff).
Or we are just assuming that was a PR stunt, which it almost certainly was.
Either way, this shit isn’t getting near my system.
I guess OpenBSD is the only OS where I never had any problem with Bluetooth audio.
Also note, that RAG and even vector search are more paths of the early days that didn't prove too valuable. Just let your agent search it directly and optionally create an index as a default entrypoint for common topics.
With the recent security related events we already see that it seems to be wide spread practice that many operators avoid the hassle of isolating their agents.
It is just an incredible cultural experience to see how quick stupidity gets accepted as a behavioural default.
A VM would be even better, but people could presumably run this in one if we manage to convince them of higher sandboxing in general ?
Why would you even use bubblewrap once it's running in a container? I've been running Codex in a container since it was publicly available, not even sure what you'd use bubblewrap for when you already separated codex from the rest of your filesystem?
There's probably more I don't remember too. In my opinion TUI apps are just silly. You don't get any of the advantages of it being just plain text because it's all wrapped in funny Unicode characters and at the same time the GUI capabilities are hamstrung by being text on a grid.
Tabs for long-running often-compacted sessions is a difference I've noticed in emergent UX.
The best feeling has to be in the CLI, in my oppinion. A low-stakes project where you can let ultracode runaway from you in the CLI is the most fun.
1. Much better multi-agent orchestration. The app supervises many agents, with different project views and built-in git worktrees, so agents can work on isolated copies of the same repo. It's easy to jump between tasks and see changes. I recall OpenAI wrote that CLI was not designed around this style of work.
2. The app combines multiple terminals, file previews, PR-review workflows, SSH connections, an integrated browser, and page annotations. For front-end stuff, Codex can run the app in Codex for mockups/prototypes, then iterate inside the app. IMHO I think this is a much faster, more intuitive flow.
3. For long-running and recurring agent workflows, I think they're much easier to manage in the app. You get automations, review queues, persistent conversation context, future schedules, and auto resume for longer running tasks. E.g. CI failure summaries, issue triage, and release briefs.
You can do some of the above, to some degree, with varying levels of success in the CLI, but I think it's miles ahead in the app.
However, from the screenshots and description, I expect that you would have these advantages:
- History. Currently, if you type into the CLI (TUI) your intention and requests are captured in some transcripts, but not in conversational turns. If you want the history of what you typed into the cli/tui you would have to datamine it from transcripts. By contrast, on the screenshots you can see it is a simple conversation that's saved and easy to return to.
- Multiplatform. As per the announcement "Use Codex across ChatGPT, your editor, and the terminal", meaning that you could open the ChatGPT app on your phone and continue to interact with this. While this might not matter to you if you exclusively code while sitting in front of your computer, still, it helps you be able to stay on top of things if you are on mobile.
- Background tasks and autonomous work. It says you can schedule it to do things in the background (issue triage, alert monitoring). At the moment, it would only do this if you've asked either codex or another agent to build a harness for it that does so. (i.e. you could ask Claude to build a harness that calls Codex on some schedule or interval.)
These all seem pretty good and I look forward to starting to use it once it's been more battle-tested.
[1] notably, on the Anthropic side the Opus 5 regressions - I had to downgrade to Opus 4.8 as I found Opus 5 was unreliable. I currently use Opus 4.8 and Codex Sol 5.6.
When I tried that with the Codex CLI version on a Linux VM, I did not get it to work. Possibly because OpenAI only supports connecting ChatGPT to a desktop installation of Codex?
If you really only use ChatGPT on IOS and a VM and nothing else, then I would be curious how you set that up. I do not see a "connect to codex via ssh" in my ChatGPT app. I only see "Connect to a desktop".
Oh HOLY MOLY! Now I see there is not only a "Remote" section in the app but also a "Connection" section where you can add an ssh connection. I need to try that.
Thanks!
You can give it any task on your computer and it can automate it. Such as creating a photo on Photoshop, creating 3D models, editing videos, or browsing any website without scripting.
Or if you are doing full time development then you just need to pony up for the 20x tier if you want to use it.
Running Codex locally removes so much friction and is so fast compared to either running it on the web (where it has to build environments online etc which consumes time) or compared to copy-pasting chatgpt that it's just a no-brainer.
I run it in a MacOS vm, fwiw, for certain tasks. I usually hand it a repo and say “install all build dependencies” and it’ll click through installation GUIs where necessary. I wonder if the TOS of an application is enforceable if a bot agreed and a human never saw it.
It can UI test web apps if that’s part of your job, but it can also just use the browser for tasks. I’ve had it research flights by checking different dates automatically and compiling the results for me. I’ve heard others use it to handle building a shopping cart for instacart/grocery delivery.
Probably the hero usecase for most people would be app/web development. It can control either a browser or a device simulator and can iteratively build and test the application, by clicking real buttons.
People need help with shopping now?
For serious stuff, is part of the shopping flow.
I'm just wondering about other uses where such a system would be useful aside from asking it via Signal or WA for the weather.
A senior engineer in an Open Source project told me "I haven't looked at code in months"
In which case what TF is it you even do ?