#codex-discussions
1 messages · Page 63 of 1
probably need a reboot actually, i do keep an eye on it and it is pretty high atm
Whoe! He's right.
it seemed to handle some of my longer sessions w/o issue but there's a UX problem where the token counts per session when broken down into individual turns reset at compact events
I got Android to Windows. Thanks @hushed lava !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
The zone map thing, it's very rare, but if you don't wanna suddenly kernel panic and lose progress it's a good thing to do like once a week or every few days
Damn, the permission link wont work for me 🙁
no problem!
Compaction events aren't yet processed. As I said, that's another enhancement. In the next few days I'll be looking at that and grouping by Turns rather than Events.
i gotchu
that is a really strange screen shot
sorry i ever doubted you
Where did you get the memo about remote-control, @hushed lava ?
this is what I mean
which one haha
on twitter lol
the side ways cropped one
came here to share it incase some people didn’t know the work around
You have 13000 events in a session!!?!
it’s from codex, just wanted to show that it’s running from windows system 32 and not mac and covered my chat titles because you know … haha
yeah I use a custom harness
no wonder you're compacting! OK, I need to look through that. I really haven't even gone down that rabbit hole yet so it's simply not a supported feature. Soon will be tho. I need to pick through JSON props to eek out details. 🙄
Bro 😩 xhigh though?
Scanned 1541 rollout files and ranked 1526 of 1541 unique sessions by total tokens.
Excluded models: gpt-5.3-codex-spark
Rank Session Model Input Tokens Output Tokens Total Tokens Compacts Rollout
---- ------------------------------------ ------------------------------------- ------------- ------------- ------------- -------- ---------------------------------------------------------------------------------
1 019db0e0-a12b-7283-8632-a5cb4963b2be gpt-5.4,gpt-5.5,gpt-5.2,gpt-5.3-codex 2,647,122,698 7,547,863 2,654,670,561 301 2026/04/21/rollout-2026-04-21T09-30-04-019db0e0-a12b-7283-8632-a5cb4963b2be.jsonl
2 019c64ad-84f5-7483-b5dc-2561a03c71c9 gpt-5.2 2,282,049,687 8,421,258 2,290,470,945 112 2026/02/15/rollout-2026-02-15T20-20-19-019c64ad-84f5-7483-b5dc-2561a03c71c9.jsonl
3 019cad4a-5eea-70e3-9570-c082ed1db233 gpt-5.2 1,690,410,884 8,212,126 1,698,623,010 94 2026/03/01/rollout-2026-03-01T22-44-18-019cad4a-5eea-70e3-9570-c082ed1db233.jsonl
4 019e0eda-1c11-73f3-ad5a-2a4e9f4df669 gpt-5.3-codex 1,658,006,523 4,475,224 1,662,481,747 194 2026/05/09/rollout-2026-05-09T15-27-15-019e0eda-1c11-73f3-ad5a-2a4e9f4df669.jsonl
5 019db9da-34e0-7500-b0ca-7e552052c821 gpt-5.4 1,629,313,046 5,405,263 1,634,718,309 193 2026/04/23/rollout-2026-04-23T03-19-38-019db9da-34e0-7500-b0ca-7e552052c821.jsonl
6 019da538-c1b3-7b40-ab03-e161f967f844 gpt-5.4,gpt-5.2 1,403,637,518 5,258,203 1,408,895,721 70 2026/04/19/rollout-2026-04-19T03-10-53-019da538-c1b3-7b40-ab03-e161f967f844.jsonl
7 019e040d-08fd-7dc3-963c-4dea0b095c01 gpt-5.3-codex 1,327,892,714 3,275,963 1,331,168,677 175 2026/05/07/rollout-2026-05-07T13-07-03-019e040d-08fd-7dc3-963c-4dea0b095c01.jsonl
8 019ce07b-fe13-7631-a3ac-916eabdb329f gpt-5.4,gpt-5.2 1,136,859,210 5,040,450 1,141,899,660 61 2026/03/11/rollout-2026-03-11T22-19-08-019ce07b-fe13-7631-a3ac-916eabdb329f.jsonl
9 019dabcb-e292-7723-8696-5738dfbe093b gpt-5.4,gpt-5.2,gpt-5.5 1,104,608,773 4,142,387 1,108,751,160 96 2026/04/20/rollout-2026-04-20T09-49-19-019dabcb-e292-7723-8696-5738dfbe093b.jsonl
10 019ccdcf-c652-79d0-baba-3b96ab0f837b gpt-5.4,gpt-5.2 1,056,179,179 2,926,881 1,059,106,060 42 2026/03/08/rollout-2026-03-08T07-17-51-019ccdcf-c652-79d0-baba-3b96ab0f837b.jsonl
Totals across all matched sessions (not limited to the top results): input=76,532,524,329 output=307,877,047 total=76,841,203,776```
301 compactions!?! That's serious abuse. I'm calling the Bot Union! 
what are you guys even building haha
If you do number of compactions, it would be insane if it also compared before and after
3d renderers and bytecode interpreters and everything in between
I found a bug in claude code lol
i'm building a todo app
You have the boys creating a VM???
ooh I like this
branch "filter1" has alpha filtering features.
It's really good
A few of our friends here are graciously providing feedback on updates. ❤️
anyone know what it means if "Create environment" button is greyed out on website for codex enterprise account
trying to get the mobile connection set up
That's weird
You’re a Claude employee?
no, I am not
What kind of response even is that? Only Claude employees can find bugs in Claude Code now? Wow must be some of the best written software ever if only employees are able to find bugs 😂
What do u delevelop?
codex app desnt show how to connect my phone on it
can someone share what its supposed to look like
is it under settings -> connections
Seany - are you sure the app is running? Do you see it on the screen?
The Windows app...
Yes
Nothing has popped up
I've checked though all the settings in codex, I'm so bloody confused
Got the same issue ^, thought this was macOS only but its suppose to work on windows too right?
😂
Looking and thinking, Seany....
In the Windows app, open File>About Codex. What's the version?
The good news is, that version is correct and should work.
The bad news is, it isn't working... OK, hang on.
I'm in learning mode, good for both of us.
It might be the network my PC is on perhaps
You have your PS session open in terminal.
- Is that Administrator Terminal?
- Ctrl-C, then
codex --version.
I'm in hospital being a human Guinea pig
i think i got it, but its not updating chats on the windows codex app
yeah: npm i -g @digital thunder/codex@latest
codex remote control is cool but it crashes pretty often
hope it will get better soon!
The fragility of remote-control might be the reason they original said Windows support needs to be deferred. We might all be playing against the better judgement of the dev team. 🙁
im on mac + iphone but it still crashes
Get Windows and an Android phone - you'll be fine! 🤣
/goal improve stability of codex remote control, make no mistakes
See above SrHAo` : Check version of Windows app. Check Codex version from PS.
Is it related to the chatgpt version on the phone
guys i just pushed to main on the toilet
Now version 0.130
I'm dying it's 2am
Thanks for the help guys but I'll try again tomorrow ❤️
Sounds like Ireland there.
🙂
reboot
Oh aye good shout
yes, needs to be fully updated as of today.
or perhaps the official OpenSSH client needs to be installed and fully configured
Mine worked
So you pushed to the mains? Lol
Not sure why I like watching the agent so much, it's just fascinating watching it work
Plus I'm a little paranoid, I want to let it start working and go to sleep but I'm afraid I'll wake up to a massive mistake and I'll have to redo stuff cause I wasn't watching
anyone's codex acting slow?
No slower than usual, it's not exactly fast but it's a better value than faster options
just make it create interim git commits as part of the workflow
im saying the app itself
Yeah I do that already, I just have an irrational fear lol
I was just reading through a session file, cuz War And Peace is for kids, and I saw that the assistant didn't find the "gh" command so it didn't bother to report the problem or go out to GitHub to get repo data, it just quietly derived an understanding of what I wanted and proceeded to operate on that spec. 
Yes, watch, and be afraid.
Be Very afraid of what it does in the night without telling you. Mwahahaha....
I'm not sure what it is, maybe it's the Codex app itself, but I'm on Windows 11 and whenever I have Codex working on something that requires interacting with Android Studio, my computer lags like hell. It keeps lagging and the lag increases, sort of like a memory leak, until it gets so bad that it's running at half a frame per second and I have to restart the machine.
It builds until the lag is too intense to do any testing, takes about 5 or 6 hours to get to that point but it happens every time like clockwork
What kind of "interaction"? Are you watching processes to see if it's firing up a new instance of studio for every new request?
It just builds gradle and pulls logcat from an AVD, nothing fancy
I explicitly instructed it not to start a new emulator and not to open Android Studio, if it's not already running, it stops and asks me to start it
thinking...
Love mobile app
Anyone have a problem reconnecting to codex desktop from the mobile app after it disconnects due to computer going to sleep?
Well I think it had something to do with a stale HyperV setup I had from yesteryear, after I disabled testing mode in Windows, I think the lag might have stopped. But it's hard to tell because the AVD runs like crap even when it's working normally lol
But the overall PC lag hasn't come back since then so knock on wood
Android logcat can be huge. I'm wondering if it's making a lot of queries?
Those emulators are really slow anyway.
hmm did they removed 5.3 codex ?!
Ah that's a good point, there are a LOT of extra entries about screen resolution and stuff, I can probably get it to quiet down a bit
What I'm thinking is that these session--.jsonl files that I'm working with can get really big. And a single text file that needs to be parsed constantly is gonna start to lag.
Yeah makes sense
5.3-codex is still available in VSCode extension under Other Models.
had a bug it said custom .. now it came back hmm
hmmm, I'd restart VSCode if that's what you're working with - or just Reload Window.
also its ultra slow now
guys i think macbook + iphone might be the ultimate tech
It's a very good combination, best ecosystem available if you need that sort of compatibility. It's a no brainer for corporations.
But I'm a bit salty that MacOS has proper computer use while I'm relegated to the CLI 😭 lol
It's so fascinating to watch the agent hit a runtime error then actively fix it, test the fix and keep working on the thing it was working on before. I can't stop thinking about when I was nearly this fascinated by simply being able to generate snippets to copy.
Codex Windows was just using my Chrome browser to surf around ... that's almost "Computer Use" 🙂
Yeah it's useful for sure, and that gives me hope that they'll open it up for other apps as well
Why don't we call Windows apps "programs" anymore? It just occurred to me.
But that's not really related to Codex lol
I just remembered something I did in 2024 that's causing me breaking issues now while on the toilet. Valid method of working if you ask me
Well the lag is back, spoke too soon
Actually might take this laptop and find a good solid brick wall to throw it up against lol
And I checked the logcat, it's not unreasonably long, but that shouldn't matter anyway because the agent only pulls what it needs
WYSI
However, just to be on the safe side, I changed the logcat buffer size from 1024 KB to 48 KB
■ Error running remote compact task: stream disconnected before completion: error sending request for url (https://chatgpt.com/backend-api/codex/responses/compact)
That's been happening to a lot of people over the last couple days
that right there is why I only use gpt-5.3-codex
it's the only model I can never get to do that
it started last week for me
every once in a while when you get a compaction failure like that you can try switching the model to 5.3-codex, manually running /compact and then switching back to whatever you were using before it if succeeds
I also burned through 11% of my pro 20x WEEKLY quota in 5 mins having Codex read some files on 5.5 xhigh?
Heh I lose track of time watching the session scroll by 😂
seems like an exaggeration somewhere in this comment
actually, not really
guess who here has the most issues with weekly deficits?
&& i don't ever use 5.5 xhigh
even on low/medium, issue persists
i’m not joking, seriously my context window was at 137k when the task was completed but my weekly quota dropped by 11%
just reading a few documents wont burn an entire 5 hour window
Are you sure you arent on plus?
i just upgraded to pro 20x last week and burned through the whole thing in 24 hours
doing what?
yes i’m sure
doing what?
If any of you with token issues want to get insanely nerdy, I can lead you to some data that might help you to figure out the issues. Any takers?
building what else would i be doing
normal usage wont burn like this. something else is going on
sure
i agree, my claude 20x account lasts way longer
well thats your first clue codex 20x lasts at least twice as long
and its on double usage atm
i wonder if you can even spend them in 1 day without hitting 5 hours limits
me
1 prompt and after 5 mins of running I was at this
doesnt make any sense, seems like that account is on free teir
with a high turn limit hermes will lay waste to a codex acct on xhigh 😂
or a low turn limit
so i've heard which is why i think something is going on. I emailed OpenAI, but no response from a human yet.
if you spent 12% in 5 mins that sounds like free account usage i agree
i know, but I'm paying $200/mo right now and not getting it
double check your local codex is actually connected to the account youre paying for
using a little normal codex cli right now and it seems more gimped today than ever
i just built a full fivem server with codex..
it's the right account
How big is all your documentation
nothing else to do except contacting support i guess
the context window hit 137k after that one prompt... that's the only task I gave it. I've been sitting on a dead codex for the past 6 days
What was task
context not an issue. during documentation it fills up fast. issue is 12% used on one taslk on pro 20x
i'm building a composable runtime UI, it was prep reading files before I hit it with a goal.
@hard drum
OK, nerd, here ya go. 🤓
Copy any session file under \home\codex.codex\sessions\2026\05\13 or wherever your .codex folder is to your tmp folder - simply the name to session.jsonl.
Run this utility to convert it to .json so it can be read, otherwise it's kinda tough. YMMV
https://github.com/CaptainStarbuck/codex-usage/blob/main/src/utils/jsonl2json.js
(You already have it if you have codex-usage)
A corresponding file is created in the same folder, session.json.
Open that format, and get yer nerd on.
See if you can see a pattern in commands under "type": "custom_tool_call", or excessive "type": "custom_tool_call_output".
Look for huge encrypted content under "type": "reasoning".
Please DM me with any patterns you find that seem like the assistant is going in circles or processing huuuuuuge amounts of data.
Have fun! 🤓
it would have to be 10 encyclopaedias to burn 11% at double usage
probably 20 or more who knows, its no 137k context though
havent tried yet, ill wait till the next compaction error for that
Wait so it’s working right now?
already down 83%
Unless you tell it to “poll something periodically that’s been working for me”
i definitely get more bang for my buck with claude at the moment
But what is your workflow like
spec driven is how we work
Sure, and did you deploy it on an existing codebase?
Or is it building one from scratch
Has it been reading a lot of code?
5 mins worth lol === 11% weekly allocation
How many files/loc did it take in?
try removing the installation_id file and logging in again, it shouldn't be burning anywhere near that fast
Hi, can i ask for chatgpt's android version rightnow
idk 15, this is not normal. i've been using coding agents for over a year now. i've never seen quotas burn this fast.
It's certainly a bug or some configuration issue, you can't use a single thread and burn 11% on 20x in 5 mins while only using 137k the maths is a few orders of magnitude off.
I'm running Beta, but I have 1.2026.125(19)
ill try this next once that stupid compaction issue happens again
( You know you're getting old when you want to disable your own VSCode extension and you can't remember it's name 🙄 )
same goal
now the permissions things is buggy, it's on full access right now and keeps stopping to get my approval to change files
I'm wrapping up for the evening here in CA/USA.
@trim rapids I've been pursuing compaction. It's fascinating. Please DM me to let me know exactly what you're looking for. I'm sure I can get it now.
Night, all!
what's everyone else building here?
When Android controlling Windows Codex host?
So the chat has been going on for 7 days?
Between the resumes ig
nah just started the goal like 2 hours ago or whatever
^
remember when people use to respond to customer support?
tried still showing the same numbers as before
Tks , but this android app version 1.2026.125 didnt have codex remote 🙁
How i can access to beta
had a couple of goes on the new app, it's great. Feels a little sluggish, but i guess that's expected.
You don't need beta. It should be available.
I see this rn, i must to delete my app and dowload this again
I sweer i didnt see that in my last chat 🥹
Is codex on mobile only for Mac users?
idk im a mac user
guys i just pushed to main from the bar
No, android available
Using a iPhone with regular laptop.
Why does Codex on mobile need the desktop app? Literally Replit is an example of mobile vibe-coding without a desktop app connection.
Because Mogman said so
there is a distinction here, the mobile app isnt a cloud instance it's a direct connection to your personal desktop instance. That is its purpose. There is also a codex cloud that can do what you suggest it connects to github and you can control it from chatgpt.com.
The cloud run instance cost something like 4x
think of the new mobile feature as an ssh connection with a nice ui
has anyone been seeing Codex glitching today? Like baby talk kinda glitch?
The standalone authoring tool is now at [tools/vector_authoring/index.html](</mnt? no, need windows abs path?>)
Wait file link instructions absolute path like /abs/path, but Windows absolute J:\? In markdown link target can be J:\?
Developer says absolute filesystem path. Use [tools/vector_authoring/index.html]
(J:/bevy/duhrng/tools/vector_authoring/index.html). Better. Final. Ensure no weird. Need mention open. Also mention tests. Need final no citations. Let's draft.
Need maybe "open from file explorer" not "copy".
Final under 70 lines. Use not too many bullets. Use file links with absolute path J:/bevy/... clickable? likely. Ok. Wait "Never tell user to save/copy this file" no. "You can open" okay. Need mention build web warning.
Final.```
"Ensure no weird"
I thnk you failed there dude.
saved like 20 tokens in that output
good
i was supposed to take an hour nap but then woke up 5 hours later smh
is it available for windows?
Have you been using the 'caveman' skill that was popular recently
Has anyone on Android managed to connect Codex to a Windows PC? For me, it just loads endlessly and nothing appears on my PC. Thanks in advance for your help 🙏🏻
Guys I see Claude have Trimble Sketchup MCP. Can we do that with CODEX?
wen fix openai wen
why codex consumed so quickly.
Check if you are on fast mode
no i'm plus. and normal use .
Is subagents active?
i even don't konw how start subagent.
What client are you using?
mac os codex software
Then I'm afraid someone else is going to have to take over helping you. 😛
Last time I was using up my quota too fast I had a huge AGENTS.md
oh thank you i just delete out my .codex but i have used up it now
Hi Guys, a question: Can i use Codex App with free Account and if yes can i also use Mobile Codex with the same account? i tried already. On Windows it said it is connected but on my Pixel it stucked on configure Codex screen. now im wondering if it is beacause my free plan or anything else?
i wanted to figure out how Mobile Codex works with Codex on Windows togheter
you can use codex with a free plan but for the windows part they said "Support for connecting your phone to the Codex app on Windows is coming soon."
oh i didnt saw that thank you i will wait then
I wish OpenAI would fix the frontend design issue
use one of the skill packages
small amount of configuration to fix it
I am using it, it still sucks
at least compared to something like claude
which ones are you using?
Uhmm, lost most local chats after connecting to mobile? Anyone?
Ugh I don't want to go to work and deal with life, I just wanna vibe code forever 😂
The issue is that you are expecting Codex to do the frontend design for you. Codex will follow extremely precisely what you tell it to do. Claude will do whatever it wants. That's the difference between a tool for a software engineer and a magic for a vibe coder 😄
And Codex has much higher code quality when compared to Claude
GPT handles UI if you're extremely detailed and explicit, I haven't had too many issues that couldn't be easily solved with a quick followup
i doubt they are using the same skills i am thinking of
Yup
taste skill and impeccable solve it
Also yes, skills allow Codex to behave like Claude, but Claude can't behave like Codex even with skills
you have to spend a little more time on it, but its fine
Care to share it?
Just out of curiosity
you can google those ones they are battle tested packages
I've felt that very much yes
Only issue I've had with Codex is that when I make obvious errors in the prompt, he goes with them as instructions, even though they are obviously mistakes :/ at times it's following a bit too much instructions lol
Has anyone been experiencing this issue as well? Otherwise I might need to go through help center
Heyy, anyone here having problems with Codex failing to run commands while working on a Flutter/Dart project?
When will official linux codex app available
which commands?
i have a couple of flutter projects i work on regularly and it is consistent
Did anyone got Codex mobile to work on android?
Somehow the Computer Use disappears in the latest version of Codex app, anyone experience the same thing?
I can, any problem you encounter?
When I scan the QR code in the Codex App, I get this
I suspect that URL should make the app open, and not the browser
I think you don't need to scan the qr code to enable this feature
may try going to the ChatGPT app on android
and see if there is Codex
There is not
Is it the latest version
Yes
wait
where is the :
isn't the complete link should be com.openai.chat://codex/open?
there is not : in your link
Is it even out for Android aka Windows?
Aka not an Apple User?
Nah, not working
Android+macOS works for me
Well Windows is Not out yet ig
i think so
Macos
what you used to scan the qr code to obtain this link?
I think this should not be a http link
maybe your qr scanner intepreted it as a http link
while it is actually not
i think you should use another qr code scanner
randomly get one from Play Store
opening this link in the scanner should directly go to ChatGPT app
but not browsers
@white fulcrum I used a normal scanner. Doesnt really matter, no? It's just to convert an image to text
I was using the Chrome scanner
Let me try a different one
Holy Hell. That was it
SHAME ON YOU, GOOGLE!
Sanitizing QR codes, are you f kidding me
@white fulcrum Thanks
who made this ?
Yeah, now it works, awesome
im on the chatgpt desktop app and they are telling me i can download the chatgpt desktop app
i also have the codex and mobile app
just bloat thats annoying
Do you still have Computer Use this plugin in the latest version of Codex?
Ok. Thank you!
There is no /goal in the Codex App?
mmm
but there won't be completion as you are typing it
I guess it's expected, since I need to edit the config.toml for the CLI, after all
I will try it...
Yeah, you are right, it still works
Will the session started with /goal appears in the sidebar after you close the app?
Before in the CLI, I encountered the issue that it was not appearing in the list of resumable sessions
I did it directly on the MacOs app, not the mobile app
I had to resume it by typing its uuid
In the mobile app, I see no visual cue telling me it is a goal, except the original prompt
Does the session started with /goal appear here?
I dont even know where to find that view
yeah that has 3 links to things I can't use, wasting screen real estate
yes
great
I tried to use /goal some days ago in the CLI
the session just disappeared
which was quite frustrating
I have it working for 30h already, my goal in CLI
Even though I would like for it to pause a goal, at times, to ask questions it may come up with
Hello. A task that took 2–3 minutes just a few days ago now takes over 20 minutes. Is this happening to you too? Codex Desktop 26.506.31421
@thorn sail May very well depend on the server load. Codex is not equally fast every day
also depends on account tier
I have the PRO plan
You can get goal working in codex app by saying "use create_goal tool..."
Hey guys just wondering if i am the only one who got that “codex” icon on their phone but not on the laptop?
Am using the same account
- updated
I just find it weird that the feature got on my phone but not the laptop
I only have chat gpt icon on phone
ok hold on do you mean the one in the left side bar?
let me check on my pc
On my phone
i have it on desktop by im 99% sure that is the cloud instance version
I charges like a bull
Is that on codex app or web
that is on web, codex app is codex o.0
Oh okay i might have misunderstood you
i think you can scan the qr code to have codex on phone
Bc i have the icon on web too
I was talking about that new mobile feature thing on the codex app
i´ve seen recently someone was posting here some repos for using multiple accounts for codex, anyone know does it preserve history when i switch beteen accounts?
yeah so to use it you use that icon in your phone then you do the auth setup and it's like ssh connection to your local codex app on your pc, but with a nicer ui than tmux or what ever
and easier to setup
So the codex app wont have that icon
Oh, understand now, i have that setup. Used it a bit today. Sent a couple of prompts
it's a little flakey, disconnecting and having to reconnect etc
Yeah you are right
I just found out that this feature is only for macbooks for now
Does anyone know how I can connect to codex on a headless server via a codex desktop app on Mac with an Android version of ChatGPT app?
Ah that explains why I haven't been able to get it to work lol
Yeah
OpenAI is really falling behind with Windows updates.
it is a bit sad that tools like this are so much delayed on windows when both type of users pay same sub
And the most stupid thing about it is, there are vastly more Windows users than Mac users.
The number isn't even close
Laziness is not a good excuse
even codex in IDE is basically unusable on windows without WSL
They shot themselves in the foot when they decided to use the apps for Windows ecosystem. That locks them into an update review cycle.
(got much better though in last few months, however it still struggles with powershell)
Yeah it's basically useless without WSL. Makes me wonder if I should switch to Linux
i was using it on linux past half a year and was ok, however app is defo not optimized because it was literally cooking my laptop
but quality wise, much better than windows, so if you are fine with linux then switch
I've used Linux for work a lot but didn't run it at home for very long, always switch back to Windows for my games lol
now changed to mac and im in general happy but still change was for wrong reasons, if i could use it on linux/windows without any issues i wouldnt even consider going for mac
There is no planet upon which I buy a Mac until every component is user replaceable
i have my gaming pc with windows as well, linux got better for gaming but it is still too much hustle to get everything working
i just stopped coding on my main pc and using laptop for everything coding related
but again, this is a bandaid not a solution
you can always get a dual boot for windows/linux
I was so excited to see the mobile app update, then I just discovered it's only for MacOS and that seriously pisses me off. Why? Two reasons. One, they didn't specify that at all, not anywhere I can find. Two, Mac needs to come second because Windows has more users therefore it has more customers and more revenue.
Probably, it's the same app
I see the option to connect to Codex desktop in the mobile app
On Android
So now I'm still looking for a remote access solution that doesn't require remote desktop
nobody in this space will put Windows before macOS while the vast majority of devs use a Mac
The vast majority of devs use Windows or Linux.
i´m actually curious about stats for that, because that might be the case in US, but from experience in EU i think there is much less mac users
It's strange because all the devs i know use macos, it must be a cultural differences where larger population countries outside of mine use windows. Almost no one i know uses windows for dev.
Yeah it's probably a cultural thing
I don't personally know a dev that isn't on Mac
No dev I've ever met uses MacOS, most of them swear by Linux
Well, in person that is lol
in europe apple mobile phones are popular, but for work i´d say maybe 30% of people i encountered use mac
but this really depends on company as well
Now i was only person in company using linux before change
I guess I'm in the minority. Only about 18% of devs use WSL lol. There's a good reason for that too...
but in previous one was mostly windows -> macos -> linux
this is the take i get from chatgpt
Windows wins globally because a huge share of developers are outside the wealthy Apple-heavy mobile/startup bubble. That includes lower-income countries, universities, enterprise workplaces, government, outsourcing firms, and corporate IT.
and split was probably around 70-30-<1%
That pans out with recent studies
The Stack Overflow 2025 survey shows the global split clearly: for professional use, Windows 49.5%, macOS 32.9%, Ubuntu 27.7%, with 31,569 responses to that question. Stack Overflow’s respondent base is global, with the USA, Germany and India as the top three responding countries.
Except for the 1% on Linux, that's a bit low
However windows became awful for software development in recent years
half windows and half macos and linux
It's never been fantastic for software development unless you're working with .NET
it was only me and other person in team who used linux in previous corpo 😄
Actually I have to be wrong, the stats don't lie. Maybe it's not so terrible for development.
it also strongly depends on field, probably much more linux users in cybersec for example
My intuition says the ai crowd are on mac at openai
which also goes under IT
that was only reason i changed to mac because wanted access to latest tools
Yeah I believe that is true, seems a lot of AI companies use MacBooks
Except Google of course lol
didnt elon threw temper tantrum last year that he is banning usage of mac in his companies ? 😄
It's crazy to think how far Google has come. Once a search engine, then an advertising platform, now an AI company.
Sounds like Musk lol
i dont remember exactly why, but i think that was something about integrating openai, but i might remember wrong
I agree, it's a shame. I've ended up switching to an unofficial Linux port of the app, going to use that from now on pretty much, via Codex on the mobile
I mean I understand that focusing on a single OS for initial rollout of new features makes perfect sense, I just wish that focus were on Windows 😭 lol
true, but the problem is that it's not just been one feature, it's been multiple features. Computer Use, Chronicle, and without some manual changes to config.toml, also Remote
Yeah that's a good point. Quite a few updates have been pushed for MacOS since the last Windows update.
yeah 🙁
Well it's still extremely early on, they only released Codex desktop for Windows in March
Anyone knows why they are not giving some love to linux user base? I guess CLI is good enough for Linux users...
prob consider Linux users to be terminal mostly 😂
I'm using an unofficial port on Ubuntu desktop at the moment, working well
Yeah when I think Linux, I think CLI
What, the codex app is closed source last i checked, unless i am outdated.
can be patched to work, let me get a link to the repo I use
assuming automod doesn't remove the link - https://github.com/ilysenko/codex-desktop-linux
Interesting.
add a few lines under [features] in config.toml and you can also enable remote control
Github links are allowed
And wow that's cool
yeah wasn't sure 😄
Leave it to the linux gremlins to be on top of things...
haha yeah, well if support isn't officially added, then someone will surely find a way to add it themselves
even has computer use support
That's impressive, am i right to say that they never released the code for the app?
yeah it's closed source
doesn't mean it's impossible to modify though, it's just minified and such
Don't take this the wrong way but I hope it becomes redundant lol. They'll release they're own Linux desktop app eventually.
I agree, I would prefer official support
It's known this things can get redundant. But it's the transition that hurts users where they don't add support to it...
another interesting one is Codex++, which again "patches" Codex desktop app so that users can more easily extend Codex with their own features
They should have open sourced the app... I am clueless as to why they chose to make it closed source, if someone have a better view please educate me.
I agree, but I assume it's down to competition
Well I'm home from work, back to work on my game. I'm in the process of running Codex through 18 extensive plans to fully realize the UX, and it's slow work but it's looking absolutely fantastic so far.
impressive
once my ubuntu devbox has finished being fully setup (Codex is doing the work for me 😂) then I'll be working on a few projects, guessing it'll be after lunch now
installing all the essential things like perl, python, php, docker, etc
Dude this game is shaping up to be so much more than I ever expected. I have four tiers of three separate upgrade systems, multiple end game systems, an entire mail system, client/server telemetry and back end player data authentication. That's only some of it 😆
damn 😮
But that's all done pretty much. The game is functionally complete, now the vast majority of the remaining work is UI/UX.
Already spent over $100 generating images with GPT Image 2, it adds up quick when I'm testing and iterating
But I think it's coming together nicely now, so there won't be a lot of testing anymore, just need to generate a few hundred assets 😂
pls share some video or something of what you created really curious
I'll post a new video today for sure, the one I already posted in #codex-show-and-tell is extremely outdated at this point
I'm still sleeping here but... If you want to use codex on Android with Windows, open a terminal and run codex remote-control. The the device will connect to Windows.
Have fun!
In Windows you need the latest codex-cli:
npm i -g @OpenAI/codex@latest
anyone getting "Control other devices from this Mac" feature working? I just get this error
Need to go back to sleep. See y'all in a few hours. 💤💤💤
That didn't work but it may be because I didn't update the CLI first. However, I found a solution for Codex desktop that works!!! YAY!! I'm connected!
All I had to do was add this. Probably only need the middle two.
remote_connections = true
remote_control = true
workspace_dependencies = false```
to ``[Features]`` in config.toml
then I restarted Codex and it instantly connected to the mobile app
so that makes me wonder why this isn't just on the settings page like in MacOS
Omg yes. This is beautiful, it's so nice to have mobile notifications.
What is the best prompt that codex checks my project for security (skills?)
could try this:
https://chatgpt.com/codex/cloud/security/
oh hell no 😂 😂
I dont have the security button
hmm not sure about that o.0
Lol I'm just tired of staring at screens all the time
It does get tiresome i have a gym in my office i get into that when i need a break
Very nice
this behind me in the office, don't mind the towel as a curtain ahha
I deleted my phone from my mac instance of codex because they wouldn't connect so was going to re-pair them, but then discovered you can't remove your mac from yoiur phone codex app. Anyone know how to do a new pairing in this state?
pet should be the main ui
I noticed something with /goal today, saw goal limited and ended the goal right away
What have they done to our beloved goal
I wouldnt be surprised if they did
I mean right now you are just going to dump that /goal usage somewhere even if you dont really have any good reason for it 🤣
just to use the usage you get/still have left near the end 😄
might promote wasteful use
If anyone has my problem, just log out of your phone app and back in.
I hope it takes context into account and has some semblance of understanding what is an important, substantial goal and what is just filler to use up the weekly usage.
So it might let a well structured goal go ahead, but it might say hold on a minute if you try doing something dumb
I didnt even notice at first
@cedar skiff Better like this
Aw hell yeah, codex all the way
Really guys?
I’m disappointed.
Imagine the nightmares
I disappointed myself
Bruh…
I was wonderin.... 
Beda got his mojo back
I just noticed that the Windows ChatGPT app (just updated) has the Codex link that opens the browser to the Web Client. That's fine but I woulda thought that if the App is installed that it might prefer that, or maybe offer a choice between App and Web.
Sounds like an enhancement?
Today gonna work through how to get Codex App / (+Phone! 🤩 ) to use both Codex environments, the one in Windows and the one in WSL. Next trick ... there are multiple WSL environments/distros here ... 🙂
I feel like the correct thing to do would be making ChatGPT Atlas the one and only desktop app, move the features from the standalone into Atlas (e.g. connecting to IDE/terminal, the Alt+Space mini chat thing).
Atlas has profiles just like Google Chrome, so I have both my plus and pro account profiles logged in. It just seems like why have 2 separate apps when they can lean into Atlas
Atlas is amazing imho. They absolutely nailed it with that one
I am having issues signing into the codex app on android.
I made sure the chatgpt app is updated on my phone.
When I press codex on the app and click "I'm signed in on desktop", it just says waiting for desktop... and nothing happens on the desktop app for it to connect.
does anyone have a fix for this?
I'm on windows 11
I have the impression that Atlas has been a failure. Zero buzz, no channel here, no announcements. What's the point?
Viktor, scroll up or search for my mention of the word "update" or "remote".
thank you!
I’m using codex in Mac. Neither did I get a QR code in Mac nor did I get any notification about Codex in iOS app. I see Codex though but when I click on it’s blank. I’m using Teams subscription. Is this rolled out to everyone?
are you in the EU
Nope APAC
ah only for mac right now. that's a shame
that feature might be localized to the US for now
No Victor - I said exactly how to make it work.
Strangely when I use my personal account the codex update appears (remote control)
But when I use my teams subscription it doesn’t show at the moment
As per the announcement it’s rolled out for all plans
That's a shame... I think it's their most useful consumer software outside of Codex. But I guess it is a whole entire web browser where some folks just want ChatGPT with minimal system resource usage, so I guess the split is inevitable
FAQ / Pin : Connecting to Codex Windows from Android :
Update the Windows app and the Android app.
In Windows you need the latest codex-cli:
npm i -g @openai/codex@latest
Then open a terminal and run codex remote-control.
Now on the phone, connect to your desktop.
The the device will connect to Windows.
i'm glad i'm on a mac, everything just works ™
I'd have to look at it at some point. I have no idea what they have done that makes it special. I don't feel compelled by their Marketing or public buzz. I'm highly vested in FireFox and Chrome is always the first integration point for extensions, so I use them. I'm not inclined to add yet a third browser.
So those are the reasons for this consumer not adopting, right or wrong.
It's not that the Windows interface doesn't "work". They didn't announce Windows compatibility yet. A way was announced by "someone?" to get it to work un-officially. So until the company makes it official, where we can/should expect it to "just work", we need to share notes. There's no fault. This is just how the world, um, works.
It's Google Chrome, but ChatGPT can pilot your web browser. More importantly if you have multiple accounts you can add them as profiles. I think it's really handy 🙂 maybe not for everyone. I had been a long-time Safari user - not Firefox or Chrome or any other browser - but Atlas struck a chord for me
Im still indignant, that Google sanitized my Codex QR code
doesnt seem to work for me
This may or may not help:
In Codex App, open Settings > Configuration > Open config.toml
At the top, add:
remote_connections = true
remote_control = true
Save, exit.
Stop the codex remote-control process.
Exit and restart the Windows app.
Restart codex remote-control.
Check your phone.
Codex remote on Linux was a blessing
I should include a qr to the website in future memes. its assistant told me today that it is sad and idling because no one challenges it 
1 day remaining with my pro sub oh man
hey guys, is the new codex on phone thing is also for windows and and android?
OMG,, dude, just look up a few lines...
cuz my codex in pc is up to date but it wont show anything in the app to connect
scroll up
remote_connections wasn’t a feature for me. I was using the cli to list them. Maybe it’s hidden from there but I am on latest
How about a reset for all?
uh oh
got it but is it possible to connect without a terminal open? maybe just opening the codex app on my pc? cuz it only works with the terminal
do not trigger that please. its tickling
Already did it
Should give at least 12 hours before the reset
its making salad, to be honest
Not sure what that model is today but thats not 5.5
This should move the odds in prediction markets
At the moment this is just a work-around. Codex Android > Windows App hasn't officially been announced as being supported. It's just a terminal window - you can minimize it.
ok ty for helping good sir
Any time - thanks for looking up!!!!! 🙂
I love ai ❤️ 💙 💖
time to turn fast mode on, reset incoming https://x.com/thsottiaux/status/2055316274394300829
@bright swift Activating /fast !!!
Reset wen!?!?
/fast
/goal improve forever
🤤
5.5 xhigh --yolo /fast /goal subagents
the great reset in 12 mins
Hey Robert, when peeps talk about comms performance, I believe it was you who mentioned using websockets with the app server?
for me at least
Yes, websockets enables communication with OpenAI's backend to be measurably faster.
The other thing that hurts performance is long threads. I had one with like 14k turns, it would quite literally take up to a minute between each turn. Seems like the app server does something with the whole entire rollout log file each turn that affects performance. My solution was a "warm handoff" skill where an agent can spawn their successor and archive themselves.
Good approach on the handoff. I'm thinking a lot of the delay is simply in parsing and transporting the huge JSONL session file on every turn.
I just read in https://developers.openai.com/codex/app-server that "WebSocket transport is experimental and unsupported."
Ohhhh that websocket setting! Yeah it's experimental. Been using it for a month now with absolutely zero issues
Oh, right, I see the difference - brainfarted....
I think the primary benefit of app-server --listen websockets mode is that multiple clients can connect to and interact with the app-server. The problem with stdio is that it's really not easy to share the app-server with other processes
Something else, the session files are bloated with two schema and it looks like many transactions have two objects, one for each form. So your 14000 turns gets 28000 transaction objects. 🙄
Yeah, and in theory all they need to send is the objects since the last compaction + new input. Would make sense to me to have an in-memory cursor to that position and not read + send the whole entire file
I thouught that's the way it was going to work. Compaction puts all user prompts into an array, then follows with an encrypted blog of (I believe) relevant responses, and that becomes the new context window. But it's not truncated there to move forward, the session file still contains all of the data. I'm hoping they only send the compacted data-forward as we're saying.
did they make app-server sync better with the Desktop/TUI for the Codex Mobile app ?
At some point I'll try to look at the wire to see how much they put on there but I got real stuffz ta do too...
No idea if they even touched it.
It's in the codex-cli package, and I believe that Was updated in 0.130 with a fix for remote-connection. Dunno.
My guess is technical debt and backwards compatibility. They don't wanna change the existing shape. The ideal scenario would be inserting an event with the compiled AGENTS.md + all user messages up to that token limit + compaction summary, giving them a stable cursor position to track in the jsonl. I bet you they recompute the whole entire file so they can find the user messages within the token limit, and that's why it performance degrades
Ask Tibo. 🙂
Oh yeah all the infrastructure they've been putting in app-server is directly relevant to the new features they're releasing
I had to fork Codex TUI to add my own TS SDK 😕
so i can programmatically access the TUI app
Plugging directly into the TUI app? What's in there that's useful?
Oh man, there's an easier way 🙂
codex app-server --listen ws://127.0.0.1:4200
Then
codex --remote ws://127.0.0.1:4200
Now you have a TUI connected to an app server, and you can write a programmatic websocket client that targets the same ws://127.0.0.1:4200, you will see the stuff in the TUI immediately
Also I think if you do codex app-server generate-ts > codex-sdk.ts it'll give you all of the websocket event shapes
I have implementer / verifier loops ya
plan mode, fast mode, switch models/reasoning, etc.
and sync's with Codex Desktop App?!
Andrew, I believe it would be easier to start the listener process, then attach and send() those directives to the app server ... rather than forking the TUI.
All i need is real life scripts like dishers
Yeah Robert, how would we get the desktop app to target that specific server?
Then um I'm set
Now that is a different story. Idk how the desktop app does it. I think they bundle their own codex binary, and I don't know if it's listening on websockets or not. If you can get the desktop app-server to listen on websockets, or figure out if it does already and what port it's using, you can target that with your TUI and custom client
i was originally using pi
but it doesnt support fast and plan mode out the box
ok fine i'll take just TUI for now. giving your advise to Codex
thank you 🫡
My first app targeted making a app server then a update with the codex server ended up making a conflict before but mostly having the app server you just need reading it into threads
Now remote qq imma wait for Android
I think this is key to what I'm trying to do now, which is to get the Windows app to toggle between the Windows environment and a WSL environment. I think once it opens a WSL environment it would connect to a port there, probably a localhost loopback, so wsl "mydistro1" might open 4291 as the app-server port, to which the Codex app would connect.
I don't see any evidence of that, no connection options anywhere for this. This is the reason I was lookig at the docs ... cuz, you know, otherwise why would anyone read docs, rite?
If I were doing the codex-usage utility as a production app, I probably wouldn't source from session files. I'd probably open a connection to the app-server as we're discussing, and send in commands to get the desired data. This stuff is fascinating, but I really need to back out of Nerdsville.
I fired up the desktop app, and it looks like it does not listen on a port. On the other hand, my custom setup listens on 4200.
If OpenAI didn't add a websocket mode or anything, the only way I can think of is extract the .asar, have it execute the app-server like mkfifo /tmp/app-server-stdio && codex app-server ... | tee /tmp/app-server-stdio, repackage in a new electron app. That way the desktop version gets its stdio stuff, while there's a fifo file you can sorta read from as a poor man's websocket 🤡
when will windows phone linking for codex come
As an aside Robert, why are all of those sessions files open, except maybe cuz you're running the reports? Otherwise that's just baggage.
We have another winner...
Here in discord, search for this from: captainstarbuck faq
thanks
My thing "resumes" all threads immediately with config & base instruction overrides. On macOS/Linux there's no lock contention for opening a file. Many processes can open the same file and even write to it so no issue there 🙂
kk - Aite, I need to code something. I've blown through my quota of curiosity tokens for the day. Catch ya on the other side.
Is the phone-to-PC feature available for Windows yet?
Here in discord, search for this from: captainstarbuck faq
I personally haven't had time today. Makes me sad cuz I've been wanting this for a long time. But the internet is exploding with commentary about it, all good, if that's any guide.
But like, what actually is a good situation to use it?
Like if you're not on your PC, you're likely to be unable to test the product anyway
So...
You do stuff on your PC, maybe in VSCode, and you have repo's cloned and open and folders on your PC with branches checked-out.
Codex in bed
And you're using Codex to do things in those folders....
Now you use your phone and talk to Codex in exactly the same way.
Codex while just doing anything tbh
You're not using the editor, it's making the changes in your PC
Yeah, and I've told my wife that I sometimes can't sleep at night cuz I think about code, but the PC is in the other room. But now I can tell Codex to do some stuff in a project. And now she hates me even more. 
Yeah and it’s a nice way to keep track of long running processes
Right, because all of the time we used to spend doing things, we can now spend watching a bot do them. Thats Progress! 💪
My wife called me a nerd last night, I was so excited due to Codex mobile lol
nerd!
I have been using codex past one month. But the problem is it hit limit faster how can I improve it so codex works longer
@brittle reef What model do you use? What subscription do you have?
hey guys I am doing my first ai purchase and planning to get the entery tier of codex but I can not see the pricing plans for codex
Gpt5.3 at high intelligence
It's built into your ChatGPT subscription, assuming you're using the UI or VS code sort of plug-ins
If you're looking to have API access, at that point you're setting up an API key and paying based on token usage
@brittle reef Use 5.5 at medium, without /fast
I have business plan
I'm currently using antigravity + gemini cli, which i got for free in a promo offer!
holy
Codex now hits limit so fast compared to gpt 5.5 on 20 dollar plan
88% weekly limit used in one session
lol i am planning to get codex entry level but now i am thinking of cursor which one is worth it? my whole day will be fixing bugs planning features and reafctoring code across 2 repo bcakend and frontend
Try them both on the lowest plan and compare?
And i'll say it again, if you don't have the moneis than you should choose another hobby.
Cursor will be analogous to ag, codex will be different so it’s more efficient to make .codex with ag before you start chat spot picks up skill etc
Dunno about cursor limits rn but gpt go might be low limits rn
Consider plus
Or Kimi
Cheap
And good till a certain complexity
Is Codex down or is it me: unexpected status 404 Not Found: Unknown error, url: https://chatgpt.com/backend-api/codex/responses, cf-ray: 9fc453798c5c20a2-IAD, request id: 3673ad46-f6bd-4507-959f-74f705e9514a, auth error: 404
yeah im getting that
Well here we are then.. misery loves company. How's your day going?
sensational
Nice! Mine is almost over and its Friday! I can't wait to BBQ!
i won $300 in poker last night
sweet... IRL or online?
So are "smart glasses" illegal in vegas?
cant see how they wouldn't be
is it only me or the codex app doesn't allow you to change permissions? i'm on windows
i click on permissions and nothing happens
Not work for Windows?
why do updates only keep releasing on mac? Chrome and moble only mac we hate mac
give us windows
😭😭
Brutal
Objective: implement milestones/milestone-18-runtime-report-contracts-context-hygiene.md end to end and stop only once you implemented and then reviewed your implementation at least three times.
But necessary, gpt slacking.
do yall use the codex app or the cli
I pay 200 USD/month just like everybody else I expect same features
cli but soon my own orchestrator, however still buggy af
what features do you add with your orchestrator
too many, then I stripped it down again
such as
I have a "requirements" system for this that works flawlessly
But basically it allows you to create workspaces > projects > repos, then either manually or via git intake "work items" and it lets you do a bunch of stuff with that such as "start" or "stop" or "groom" or "steer" and so on.
Something like that
I use cli
My thoughts are due to Windows is such a bad development experience, I'd say it's a pain to even develop the features for windows, while MacOS has a much better development cycle
Not my problem though is it? If Anthropic can do it OpenAi can do it too.
Yeah, soon when my orch is ready, this won't be an issue anymore either.
It has the full foolish loop of verifying each and everything
bro cracked the mainframe
For sure, but also could be due to analytics? Could be 80% MacOS users, not sure, but understand the frustration
80% mac user cuz windows experience is awful wsl automation has never worked correctly and features arrive with weeks/months of delay.
Only hipsters use MAC, real devs use Linux, only reason I'm on windows is beacuse app wasn't on linux when I started using it
From my experience of developing on a windows machine over 7 years ago, it's terrible, so I could see it taking extra time + development for windows features
If MacOS allowed for more gaming, I'd sell my gaming computer in a heartbeat, due to windows has sucked for years
Mac is the golden grail for agentic llm for two reasons:
- it is out of the box very safe
- if you run local LLM it is out of the box extremely performant with the whole metal suite that comes with moderns mac specially for traininig very very useful
Hipsters may use mac, but not only hipsters use them, people who have the cash to buy a machine to get s**** done do too
And, not everyone has a "real beard".
That's a classic noob argument anyway: a good sailor does not need a good boat to win a race, beware.
I might argue that if you want to use an OS you won't use linux, because if you do, you will be fixing the OS rather than using it
Developing ON windows is not the same as developping FOR windows and it's less awful than 7y ago thanks to WSL, I just emulate Linux & code on it
I'd say to develop for windows, they are most likely developing on it for testing purposes
MAC out of the box very safe is old story, not anymore.
wrong. this is not about "hacker"
In my experience, I've never had a single virus on MacOS, multiple on Windows
ok fanboy 😂
Its about llm destroying your entire system and such
call fanboy whomever you will lol, wrong person to try to provoke some anger, all you will get out is a meme
sandbox 🤷
You know there is an alternative for everything, and then there is an ideal fit for everything, and as well a plethora of colorful "other options", right?
Viruses exist for MAC only reason they didn't before is low usage, same for Linux
Should i use this or opus
I been fighting with this one agent about putting real navigation into the app for like 3 days straight, and it would double down and draw highly elaborate and fake maps, and fake it even harder each pass. Once I got the requirements thing dialed in, that review agent is so thorough I finally have real GPS navigation 😭 building an orchestration layer is the way!
If you like your OS so much stay on it and stop complaining lol
Gotta love when it fakes it lol
I dont want to need to buy expensive crap Apple oS just to work with an app because openAI cba to properly dev their tools to work for other systems unlike their competitors
Mac is not expensive crap
Mac is good
Mac just works
What do you pass it a MD file of requirements to a standalone thread that reviews based on that?
Or more elaborate?
I had fiddled around with JSON payloads instead of plain text to leave it less room to wiggle but hadn't really huge success with it and it felt more like polluting the context with lots of {} lol
H I P S T E R
linux is probably top
Like I said understand the frustrations, but Discord isn't going to get it fixed, submit a support ticket
And at this point, as a former Claude Code power user, I would not call Claude much competition anymore sadly
Gaslighting, bad decisions, bad models, and overall bad treatment of customers has caused a massive downfall
then leave it lol?
What do you want people to tell you, that you shall have what you want?
Well, have what you want!
That's an OS that is not supported, so that's what you wanted 🙂
No offense
When you pay golden it's better but it's like MAC: way too expensive for what it is
Try out xAI, at the low price of $300 a month, might work on windows
IDC about what you think or say tbh I just want the app to work on windows. You can keep experssing fanboyness for Apple ecosystem made for boomers all night long for all I care.
you guys got nothing better to do
We got it, you cant πay and want to πplay.
Thats a freeloader, since you like so much calling people names 😉
With AI, no lol
I can pay I just have too many working brain cells to become an apple user
The Requirements are JSON like this:
{
"id": "technician-native-mapbox-navigation-lab-proof",
"title": "Technician Native Mapbox Navigation Lab Proof",
"active": true,
"enforceOnTurns": true,
"requirements": [
{
"key": "nativeLabFirst",
"statement": "First prove native Mapbox navigation in a lightweight iOS-only design_lab_native/native navigation lab before wiring into the production technician page. Do not require the Ezra API stack, auth, Rinf/Rust runtime, or full app shell for the first proof.",
"severity": "blocker",
"verificationMethod": "manualEvidence"
},
{
"key": "realNavigationSdkRequired",
"statement": "The navigation surface must use the real native Mapbox Navigation SDK or an SDK-backed native platform view/plugin. It must not be a Flutter-drawn fake map, CustomPaint route, static image, scroll-view imitation, or plausible mock map.",
"severity": "blocker",
"verificationMethod": "diffReview"
},
{
"key": "mockOnlyEzraData",
"statement": "Mock only Ezra-side job/customer/workflow/controllers in the lab. The map, route line, camera, puck/location indicator, maneuver guidance, and trip progress must come from the real Mapbox navigation runtime.",
"severity": "blocker",
"verificationMethod": "diffReview"
}
]
}
I have a form that I fill out, but the orchestrator can set a payload like this on one of the agents and it forces them into structured output mode, so to finish their turn they have to respond with claims for satisfying each and every requirement, which gets sent to a very angry and adversarial review agent.
BTW, off topic of windows, etc.
Do any of you use Warp Terminal (Logged out version)?
windows terminal is what i use, its like the only good thing about windows
Suspected as much...
I will try it again with the verification workflows I guess
Did not like it at all for "normal convos" between agents
Too deterministic, kind of chokes the llm on what its good at
BTW, I am just going to leave this here basically like a token of gratitude
Fedora is my os of choice btw, I just hate Apple and everything it stands for 🙂
Do you need a meme specifically with fedora to digest the subliminal messages?
I hate Apple as the company, but you can't beat their ecosystem connectivity
About your token wastefulness concern, the schema uses "anyOf", so mid-turn commentary is a simple string property, but the end of turn is the full structured payload. And rather than being an array of requirements, it's a strongly typed keyed object, so they cannot just omit one of the requirements 🙂
Phone/iPad/Computer all synced, windows could never
Hear me out:
mac experience is roughly: You pull your mac screen open, press for Touch ID and its unlocked.
On windows: you boot up, wait 30 seconds on SSD, enter password, try to open an image editing software and get redirected to a 199$/month paywall roflmao
You can't have any arguments against the Linux, 90% of the net runs on it bruh
Then BSOD
Still beats entering a closed ecosystem made for boomers
Mac 🤝 Linux
I think Linus would have the same feelings towards MAC as he has towards Nvidia
You know Windows is a closed ecosystem made for boomers right
Linus has had a good relationship with nvidia as of lately
I can pretty much do whatever I want on a Linux or Windows. Same can't be said for Mac
You can use Windows on mac
Can't make spaghetti
but you cannot run what you would like to, right, hence the discussion here ongoing now since a long while?
given enough RAM and an Arduino board I think it could
Good point lol
Linus Torvalds used to use a macbook air to develop the Linux Kernel 🤡 now his main machine is a ThreadRipper, and the macbook is for arm64 built testing
Could probably cook some bacon on my GPU lol, it runs hot
well cuz OpenAI is cutting corners & not doing a correct job, hence the discussion.
right.
you know who cuts corners? Opus models lol
The correct job is support the OS that can do everything 🤣
Wish I could use claude app with codex
You can
Yeah
OpenAI created a plugin for Claude Code, so you could use Claude Code, with the Codex plugin and use SSH
OpenAI put mac first. This tells you something about the friction of developing for the competition
Tells me OpenAI devs prolly Apple fanboys too 😉
Hey, I've always hated Apple, but their products are superior most of the time, especially for development
Their products are overpriced, but worth the money for me
They’re money fanboys if they could they’d let Sam make money off windows users too
But Microsoft is a garbage partner
Only use case I would agree mac is superior is video editting, for dev I don't see how it's better than a linux
Video editing on mac is super overrated and limited
Actually 🤓
Well, I do more than development, so Linux is still very limited
Compared to windows its not and linux ==> GL depending on your hardware
This right here -- exactly what Antrhopic screwed up on, ignored users for months, then said it wasn't their issue, then said it was
Oh man, you need Warp terminal lol
nice, sublimetext! 😄
Windows desparately needs tmux
you sir made my day 😂
Warp Terminal is 1 program, with unlimited tabs (directories/projects), and each window can have as many terminals you want
And it's customized: I open a tab, it opens 3 terminal windows:
2 side by side
1 at the bottom that auto opens codex
Ghostty is macOS only
he doesn't have Ghostty!!
not that I particularly like windows terminal but at least it tabulates
I use it to read JSON lol
I used it for years, but due to lack of some feature maintenance on language packs I went to vscode... that just works
But I still miss it, really nice ide
When I need to I just use terminal in vscode or on my fedora, seldom use powershell except for ssh connections to some vps where I run contained experiments
Warp has a file viewer / editor and file tree
I just want to be able to launch commands from my mobile Codex app while I'm forced to be with my wife and kids is that too much to ask? 😭😭
The time wasted socializing 😪
Force them to be at your computer
Problem solved
Yes, I also like how the commands in powershell semantically make sense. Like if I wanna print the contents of a file in bash I'd write
cat $HOME/some-file.txt
What the heck is cat? What does an animal have to do with printing a file? Powershell is much nicer
Get-Content -Path "$env:USERPROFILE\some-file.txt" `
| Out-String `
| Write-Output `
| ForEach-Object { $_.TrimEnd() } `
| Tee-Object -Variable fileContents `
| Select-Object -ExpandProperty Length `
| Out-Null; $fileContents
So much easier to read and understand
because it (cat) does more than just print
But you know that, you are a bearded man 😄
Mac is more about
You will do it zis way
you will not ask questions
you will consume ze slop
its not going to make sense but we love it


Just that cat is not something apple wants to shove down your throat
it actually shows how you can switch from a mac to a linux driven machine effortlessly, because everythign is basically the same, just changes position on screen, if even.
Might there be a reason so many linix based distros go full on the "looks like mac"?
just remote to your pc then 😄
I still have 10X Codex until next month -- offering 10% of my weekly usage for $150, can create any project you require
.... Just kidding, but 10X Codex is a cheat code, unlimited usage lol
right I forgot about Unix
it's too late rn
cat
unix
gotta make sense
right
Mac
distro
what?
gn guys
Pretty much. If you're only running a single agent at a time, you basically have to lose sleep to use it all lol
Codex might be a bit slow but it's a good value
openai launched codex mobile on iOS and android yesterday
I supposed I haven’t got it yet
Push notifications are really nice to have
||that is breaking news, put it in a spoiler please||
Linux will be the last OS in the world, to get Codex app
Maybe Windows 3.11 will get it, before Linux
OpenAI and the ceo annouced it actually if you are not kidding
But it’s a funny joke
Lol
||I am kidding.||
@codex is he really kidding?
I am challenging codex with a new meme based off https://tenor.com/view/shit-jeff-goldblum-jurassic-park-gif-15006020
But this is a hard nut, it keeps complaining it is breaking its internal understanding of the world somehow
Like the "they told me I cannot do that"
Gotta tweak my wording I guess
lol
OH, praise be
Lol that is a pile of propaganda
Not as long as I do not add the url right?
You can have codex on linux though
The only thing really not propaganda is "Pro Stand $999" lol idk what they were thinking 😂
@hybrid hare Not what I said
I still remember that moment... a crowd full of fanboys, and yet, even them couldnt believe it
I just shared a thing to lol community that I made a script for little advantage
And the monitor expelled me…
I didn’t even share how did I make it
I make it with codex Actually
Im building this, next week. Im going to order an ESP32
1,5M Antena, I owned my block 😂
Though it’s not moral
And I was banned indefinitely from that community
Too strict…
And it’s not some cheating thing I suppose
I just learn some tiny information for myself with the script
It doesn’t influence the game balance actually I just know which side I’m on ahead of the game on
😭
Banned for cheating you say? Lol jk
Wow that's beautiful
Totally doable too, great idea
Well
Any kind
I was not discussing about the script itself then
They only ban any ways to spread the script but not the activities of cheating
There are lots of softwares outside which enable many to do more than my script can do
If it works well... I might even considering selling 🙂
Codex… bring me the smoking gun
Evidence on what
Anyone tried out OpenDesign? Pretty good with codex/kimi. Im done with claude design now lol
I like VSCode again
I had been using vs code just for codex on it until the codex for windows released
I use stitch/claude desin
Well I might have spoke too soon, the game still need more work first.
so nice when it follows the guidance
"Avoid making unrelated changes ... unless that's exactly what I want."
"Do as I want, not as I say."
"and don't make any mistakes"
Ah mistakes. How naive I was to think Codex was doing everything perfectly. I'm debugging some svelte reactivity problems sigh
Open Design is open, chose whoever provide you want
Try it
I asked for this level of complexity though, I knew what I was getting into when I started making an optimistic UI design
I just want a playable game though 😭 😂
I am not refactoring into C# for a Unity migration, ain't happening lol
I've dug my Svelte shaped hole and now I'm gonna lay in it
I did start using PixiJS though, it's everything I need and nothing I don't.
So now the underlying framework and state management is handled by Svelte but PixiJS adds a WebGL layer for UX
At one time I was somewhat of a C# expert. But I never really got into Unity even though I really wanted to write cross-platform games with it. Maybe some day. More recently though I have a ChatGPT Project open where we discuss the nuances of HTML games on the Canvas. I dunno, seems like a decent platform. Can't use Java in browser anymore...
To ensure that's on-topic - I do plan to shift from ChatGPT to Codex now that I can get it to actually do coding and save it into a growing project.
I'd say my familiarity with C# is pretty basic, no pun intended
That's VBASIC.Net to you...
I know the structure and syntax about as well as JS, I wrote a few mods for games made with Unity so I had to get kind of familiar with it
But to code at the level of my project without an agent would be unrealistic
So I'm definitely hitting above my weight class lol
Hey, so it's more of a partner or a hired contractor than a tool. That's cool.
Games are a work of passion, labor of love, and fun as hell. Use what ya gotta use.
Well it's like a junior dev who knows a lot of technical stuff but lacks common sense, so in some ways, it's a terrible employee and I have to be a terrible boss lol
I learn a lot from the things it attempts and the changes along the way, it's not like it's just "go do this for me" and I never think about it again
With Opus 4.7, I have a bit more freedom. It catches stuff and thinks ahead. But with GPT 5.5, I have to be extremely specific and detailed
So that extreme specificity by its very nature necessitates an understanding of what it is that I'm actually doing
Funny you should say that now because at the moment I'm dedicated to rewriting AGENTS.md to be ultra-specific about details that it frequently does that irritate me. That said: (1) too many instructions are just bad. (2) As soon as I get this right I'm going to migrate to Skills.
Yeah it's best not to clutter the AGENTS.md, I have separate docs for all sorts of stuff and directives to review relevant documentation before touching related files
Plus a bunch of custom skills for the project. When I encounter a situation that warrants it, I have the agent write a skill right then
Oh speaking of AGENTS and Skills. The sessions JSONL files have all of the Developer instructions that make Codex what it is. Following that preface it loads our system-level AGENTS and then our project-level AGENTS. It also loads Skill headers. Those files provide great insight for anyone who wants to know why Codex does stuff. It's all in there.
Yeah that is a great read, I reviewed that a couple weeks ago. It's helpful to understand what the agent actually expects
Just rember that AGENTS and Skills (and related docs in the folders) are Directives. Other files referred to by them are guidelines that the assistant doesn't need to follow. The AGENTS directives need to emphasize that other specific docs are authoritative. We really do need to baby these tools, and the more we have to tell them the dumber they get. I look forward to this tech evolving in this area.
That's good imo, best not to clutter up the agent with too many directives or it'll start skipping things
Which is why it's important to trim the AGENTS.md
I only have the most important stuff in there, followed by references to informational documents and stuff that only applies under specific circumstances
Plus the "When in doubt" section of course
And as I've done in the AGENTS files but not in Skills yet, you've essentially already broken it up into Skills, so a migration to Skills is a great fit.
you can also add your own input to the system prompt using developer_instructions in config.toml. This is pretty powerful. Instructions in there get weighted in a much heavier way.
I didn't know that - Thanks dude!
I don't care what Andrew says about you - you're OK!
Should I try it
For anyone else running Windows and WSL, with VSCode and the Codex Extension : When you create a new vscode-workspace for a project that's all in WSL, do you save the workspace file in Windows or WSL, and why?
skills are really powerful, the fact that codex actually respects skills and is able to follow their instructions consistently is what makes it win over claude. When i moved from 4.6 to 5.3 i had to have heavy language at the top of CLAUDE.md to make it follow skills on the first turn, then to get it to follow on any other turn i had to use orchestration and explicitly tell it when to read and use them because it's ability to proactively use them is terrible. On the other hand codex just doesnt forget - assuming you have a short concise and well written description.
Skills make such a huge difference in the ability of agents that one thing alone makes the difference between claude and codex huge. If you never learned to use skills you would never know the difference though.
Your experience with Claude is similar to mine with the GPT 4 series - or as Sam Altman said : "GPT 4 is the worst model you'll ever have to use."
I gave up on a project that required intense agent instructions with v4. Planning to resurrect that now that 5.5 is (IMO) so reliable.
We're all benefiting from this industry's willingness to embrace competitor standards like MCP and Skills, and hopefully DESIGN.md.
gpt 4 was terrible, claude was way in front back then