codex-discussions

[small violet]

i see

u are right

what ive built so far is nice

ive made in one week

with plus

[plush harbor]

I'm supposed to be working on a different project but I've been home sick for 2 weeks and that's been enough time to burn through the entire backlog in two projects so I'm gonna need something else to do

[boreal holly]

Oh my god I found the word lol

[small violet]

and i wouldve done more if i had more usage

[lean lark]

That's awesome metallic. Congrats.

really though?

[small violet]

thanks

[small violet]

u asking me?

[lean lark]

Are you wasting tokens or are they really being put to good use?

[small violet]

how many days are these

[lean lark]

doesn't mean anything to me.

[small violet]

id say of 100% usage im properly utilizing 85%

[lean lark]

I'm asking if you're using your tokens productively. If not then you're Always gonna run out of tokens.

[small violet]

it should

considering what ive made

[plush harbor]

chatgpt describes my codex use as "sustained maximum economically rational exploitation of available synthetic labor". I'm just systematically destroying all my to-do list items and I'm running out of them

[small violet]

do u remember that guy who was making something for sec filings

was it u

or are u the image guy

[lean lark]

Have you looked to see how much work Codex needs to do to complete your tasks? Are you looking at what happens during the thinking processes? Or are you telling it to do something, it works forever, and then you're out of tokens with a list of things you still need to do?

[plush harbor]

I'm the image guy, but this is my other project and is very different

[small violet]

the first part

yes

yes

and no

[boreal holly]

You gotta have them create the to-do list items now, and you will have completed the circle of vibe

[small violet]

do u know who im talking about

[plush harbor]

I do. Its a codex skill

[lean lark]

Hmm, gotta refocus on my code here ... Codex generated a ton of stuff and I'm going through it, finding a lot of things that need correction, fixing some of them manually, giving it the heavier stuff ... never running out of tokens. I wish you the same experiences...

[plush harbor]

I have a couple regular visitors to my house who are claude fans so every time I mutter something about codex and bugs, they say "claude never does that" and I don't believe them

[oak trellis]

I figured why they reset the limits .. if they do .. they control all sub timelines. Means they can organize themselves better

[oak trellis]

Fan boys 🧒

[nocturne folio]

i kinda miss openai's old limits

especially the 5 hourly

[meager wadi]

It's a lot more than just a skill:

Codex skill
  -> Node REPL js tool
  -> scripts/browser-client.mjs
  -> Codex Chrome Extension
  -> Chrome native messaging
  -> extension-host.exe

the backend for the in between chrome extension / codex is removed after codex restart with no vpn ._.

It functions close to how the in app browser does

[boreal holly]

Gotcha, so they tightly coupled the extension to the plugin, and renaming it causes it to not be detected and used.

[hard drum]

Reset?

[meager wadi]

I think it could be tied to the "bundled" "marketplace" which is done in a way so it can update without codex app update. When the chrome plugin isn't available then it just nukes it

[unique spade]

i guess i'm done with vscode for good. still using codex app for one thread for now though

[boreal holly]

That is a bummer.

I think honestly, a lot of folks get privacy based VPNs which cost more, but it would be really cheap to fire up a simple server on something like DigitalOcean and pay $5/mo and set up a wireguard VPN. Would be cost effective

Do split tunneling, so all OpenAI server IPs get routed through VPN, and other traffic routed through normal means

[boreal holly]

Feels good huh 🙃

[cloud pike]

Is anyone else getting constant permission prompts even starting on "Full access"?

[boreal holly]

[mcp_servers.openaiDeveloperDocs]
url = "https://developers.openai.com/mcp"
default_tools_approval_mode = "approve" # Add this line to all MCP servers to make permission prompts go away (probably)

[lean lark]

I have servers on DreamHost, equivalent to DigitalOcean, company doesn't matter. I host VPN on one of my servers and give friends access. It was surprisingly easy, given all of the contemporary hullabaloo about VPN.

[boreal holly]

Yeah! I was just recommending DigitalOcean because I'm familiar with em, they have tons and tons of guides including setting up VPN, and it's straight up $5/mo for their cheapest machine

[meager wadi]

I'll have it function eventually

[lean lark]

I used to use DO docs for setting up my DH servers. Hahaha.

Hey Robert, given the recent enhancements to Codex App, I think I'm ready to pull the trigger to load to Windows.

[lost drum]

but what areyou taliking about? what you want to deploy and why vpn?

I mean what you all were reffering to

I dont see context

[boreal holly]

EU folks can't use the remote control features because EU requires paperwork & private encryption keys for all software so was offering a cheap workaround

[lean lark]

I'm also thinking about loading Gemma4 to Windows then using Codex in WSL with API back out to the LLM for operations that don't require GPT competence. By extending out from WSL I won't be allowing the model to operate in the primary dev system, it'll just send instructions back into WSL. I haven't seen anyone else doing this? I suspect as of this moment it'll be the new way everyone does this. 😺

[torpid trout]

Linux gettin hammered again

https://github.com/V4bel/dirtyfrag/tree/master

[plush harbor]

getting screenshots between computer to a CLI bot is mildly annoying

[torpid trout]

People with homelabs might want to do the workaround… not sure when upstream will fix this

[lost drum]

I see maybe as I go it will be a problem for me to as I live in EU but I dont see what I am missing on yet.

[boreal holly]

Oh yeah, you can set up a custom model provider:

[model_providers.lmstudio-local]
name = "LM Studio"
base_url = "http://windows-host-ip:1234/v1"
wire_api = "responses"

Then launch with codex -c model_provider="lmstudio-local" -m "gemma-4"

[plush harbor]

... codex tried to get offsets for images in my sprite map

[lean lark]

I didn't know we could do that until very recently. I completely value GPT 5.4 and 5.5 for what they do, even used 5.4-mini today. But for dumb munching on dumb stuff, there's nothing wrong now with using a local "medium-class" local model.

[lean lark]

I feel that one. But it can be scripted.

[plush harbor]

really did not expect it to attempt to figure out sprites on a sprite sheet, I just wanted the structure. It actually got close

[stuck warren]

has there been any talk of aws bedrock integraiton with codex? we are using the azure integration with codex and its honestly just hot garbage

[lean lark]

Spritesheets are just binary files with calculated addresses - dirt simple for software to compute offsets.

[plush harbor]

was more that I didn't ask it to do that and it just helpfully did it

[high girder]

Codex app with image gen is actually REALLY good with sprite sheets now as long as you give it a refrerence sheet, and a stable reference image to work with

[lean lark]

Gonna wrap up here, go say hi to the wife.
... does anyone here remember her name?
Gosh, that's embarrassing. I need to write it down somewhere...
I'll just call her Mrs Starbuck. She's OK with that one.

[plush harbor]

my sprite sheet is out of gemini, i couldn't get chatgpt to cooperate with the icons

[plush harbor]

I'm single with cats, much simpler.

[small violet]

codex via vs code is broken

[high girder]

It's been busted for a few updates. Use /status

[lean lark]

Oh, about /status. I'm still taking requests for improvements at https://github.com/CaptainStarbuck/codex-usage.
Gonna go crazy on that soon, for no good reason other than coolness factor.
See screenshots in README.

[small violet]

"Rate limit: Unavailable"

[craggy jewel]

I did this with Gemma4 and Qwen local running in LMStudio in Windows.

TLDR: Use a real linux distro on a real host. Don't use WSL.

Accessed it with OpenCode in a Debian WSL2 instance. It was ok, but three things:

1. WSL in insanely insecure as it has alot of access to Windows (like it can run notepad.exe etc). So you need to really lock it down.
2. Networking is somewhat tricky to get it to see the LMStudio server. Can't use localhost or 127.0.0.1 to bind to - gotta be your real computer IP address.
3. WSL doesn't have a copyfail fix. If you go to upgrade ex. my debian distro, it would not recognize there were fixes (for copyfail, among others) no mattter what mirror's i used.

Since my c# code lives in Windows along with VS2026, i'd have to share a ton of things in windows so opencode could do its thing. If i was developing python, or websites or something that can entirely live in the debian wsl instance...that would be workable.

Because of the copyfail issue I uninstalled/disabled WSL and my Debian instance. Made a mental note that if I were to try that again, I would use a real virtual debian (VirtualBox still around), not the microsoft-integrated-with-your-linux wsl thing. Even better if you have a dedicated linux box - even a raspberry pi 4 would be enough, or a ProxMox linux instance somewhere, as long you keep your LMStudio close to your graphics card (in my case a RTX5060Ti-16GB).

Hope that helps!

[lost drum]

sacred knoledge

[craggy jewel]

Oh hey Captain, I was just working on a .jsonl reader/interpreter thingy in c# with hope of undertanding better context usage. I like the look of your repo there...it is going to be very useful. I'm trying to answer the question: Where do all my tokens go? Did I do something that caused excessive usage? What can i do better? kind of thinking.

[unique spade]

it feels the best haha. when you know that other and exec is you

[cedar skiff]

they changed the dictation short cut in codex app grrr

[uneven kayak]

Love it.

Summary observed_token_volume: 1,355,676,434 raw_total_tokens: 1,355,676,434 effective_input_tokens: 59,576,772 cached_input_tokens: 1,290,774,272 cache_hit_rate: 95.6% output_tokens: 5,325,390 reasoning_output_tokens: 1,960,348 session_count: 55

I don't have any suggestions for improvements, it simply works

[meager wadi]

I got it to work, it is a mess though. Can't get so I can "@" it ._.

The skill is taggable though

[oak trellis]

53% until 12 4 days left ..

100 sub

[oak trellis]

what was that ?

[cedar skiff]

ctrl + m

[cedar skiff]

it got a bit of buff recently - removes all the uneeded words from your prompt, like, um, right?, okay, etc etc oh and cuss words ahha

[plush harbor]

ack, blind bots. "just mind that that helmet icon is very wide" codex crops off the right half of the icon no not like that

now its scaled it so small you can barely tell what it is and put it on top of the text. Step away from the css, little bot

[vital imp]

theres more special variables you can use than the ones they show on the web app settings for codex

[gentle harbor]

all 9 have been going for 20 min now i wonder how long this will take

[vital imp]

The UI won't let you save it if the simulated example branch they show is > 100 chars, but if you replay a settings change PATCH request with your desired value, it holds

[plush harbor]

anyone here answer this? #openai-chatter message

[gentle harbor]

when you upgrade you do get new limits im not sure if it fully resets them or if it just adds onto the limits though

[vital bone]

support saying this, anyway will try

[sharp silo]

i don't see any chrome plugin, neither the Browser Use that lets Codex operate the in-app browser directly. Can someone pls help me

i've updated the latest version

[gentle harbor]

are you talking about codex or normal chatgpt app ? as normal chatgpt app on pro plan doesnt have any limits at all unlike codex

[prisma flare]

Same

[gentle harbor]

weird if your on the latest version of codex it should show here, maybe try doing a quick pc restart

[sharp silo]

how about the Browser Use? I don't see any plugins support that like the docs mentioned

[gentle harbor]

that one is in skills not plugins

[sharp silo]

i only have these two

why my codex so lack of features 🙁

[gentle harbor]

goto your settings

both browser and chrome should be able to be turned on there as well

[prisma flare]

Btw only available if not in EU or UK. Maybe that's the issue

[sharp silo]

@gentle harbor here are mine

[gentle harbor]

are you in the eu or uk ?

[honest yacht]

someone convince me codex is better than claude please

[sharp silo]

hm? i live in SEA

[sharp silo]

no

[honest yacht]

whats weird is no one is using codex in enterprise

like no one

more enterprise is using copilot for some reason. Its like 80% claude 15% copilot and 5% codex

[gentle harbor]

yea havent seen that before, you might be able to manually add it via files but other then that no clue why it wouldnt be showing

[sharp silo]

i downloaded codex via microsoft store, is that the case?

[gentle harbor]

i mean you could always try it for a quick 10 usd in credits, id imagine people use copilot since its cheap and normally already bundled in microsoft 365

[gentle harbor]

i downloaded it from the chatgpt website so maybe

[plush harbor]

the fun part is finding the handful of people here who use both

[vital hill]

i use both and codex is hands down better at coding

just not in frontend though

[sharp silo]

wait, why i can't access settings via personal account?

but my workspace account can?

[gentle harbor]

i cant either is yours handeled through apple ? ¯_(ツ)_/¯

[plush harbor]

doing all front end stuff right now. Mildly horrified by how bad it is. But, I can't compare to claude

[gentle harbor]

its your payment handeled through apple

[sharp silo]

i'm not sure, i got my account from company gave it to me

[balmy ermine]

any math enthusiasts?

codex indicates high certainty of a twin primes proof formalized in lean: https://github.com/alegator-cs/infinite_twin_primes

[digital thunder]

<:book_icon:1363314738255364126> Rule 7: No self-promotion, soliciting, or advertising.

-# Do not post or direct message any members of this server to promote non-OpenAI services, products, or projects. Exemptions: OpenAI API projects in #api-projects & GPTs in #custom-gpts.

[gentle harbor]

is this new ? and by agents does it mean ai agents like codex i just updated and saw this

[sharp silo]

@gentle harbor you are using win or mac?

[gentle harbor]

i use both but right now im using win why

[sharp silo]

i still can't figure out why i don't have the option to use the integrated browser codex

literally no docs or any help from this issue

[cedar skiff]

i see it here

[sharp silo]

this is the issues i'm trying to say

[cedar skiff]

I see, is computer use missing as well?

[sharp silo]

even browser skills/plugins i don't have

why i can't use these features

[cedar skiff]

some ppl have been mentioning they didnt have computer use, and they are from the eu. Might be a location thing

[sharp silo]

i'm from SEA

[cedar skiff]

hmm me too

[sharp silo]

can't find out why

they don't have any discussion about this

[plush harbor]

... so everyone I think is an active regular here is just ... in asia?

[cyan gyro]

its only live for the US

so use a VPN or wait

[blissful basin]

im starting to wonder why EU sub costs the same as US when all new features are delayed 😛

[cedar skiff]

im in australia which is not exactly sea but it's sea for gaming

i have it here

[plush harbor]

australia too and used to being online at the same time as all the gaming kiddies in vietnam and singapore

[neon plinth]

• Close Codex entirely
• Use a VPN and switch to the US (Windscribe is a good option)
• Open up Codex

And you should be able to use all the features

[sharp silo]

let me try, thanks

[plush harbor]

I think I've used more tokens in the last 3 days on my 'other' project than in the last couple weeks on my main project. Type of project apparently matters a lot

[neon plinth]

Codex usually still have background tasks so I would recommend end task via task manager

[sharp silo]

yea i notice it's mini icon in the tray

[neon plinth]

I mean, even if the icon disappears, Codex is still running background tasks

[sharp silo]

let me restart the computer for best refresh

unfortunately no, it's still the same

[neon plinth]

that's strange

when you turn on the VPN and reopen Codex, did it have any new updates ?

[orchid plume]

you don't have split tunnelling enabled right?

[sharp silo]

nope, my codex account alr update to latest version a few hours ago, via microsoft store

[sharp silo]

what is that?

[neon plinth]

Windscribe

[orchid plume]

it's a feature that allows or blocks specific apps from using the VPN connection basically, should be disabled by default so don't worry about it 😄

[neon plinth]

You should be fine, default settings should work

Hmmm

[sharp silo]

yea i checked it and it was off

[orchid plume]

very strange

from the UK I'm also using Windscribe, I can see the Computer Use and Chrome sections fine

[neon plinth]

Maybe i can help if you screenshare

You can speak vietnamese to me

[sharp silo]

lmao, how did u know i'm vietnamese :v

[neon plinth]

This image has Vietnamese

[sharp silo]

haha, good catch

we can DM

[neon plinth]

there's a voice caht here

under voice

[sharp silo]

i can't voice rn, check your DM

[torpid trout]

Hello former fellow Vietnamese people!
Had the honor to live a few years there ❤️

One of the nicest countries and people ever 👌

[neon plinth]

Glad you like it!

[oak trellis]

4 days left 50% .. uff i saving tokens like no other ..

so every day 10% iam good

[orchid plume]

maybe Tibo will do a reset at some point before then 😄

[cyan gyro]

a man can hope

[simple star]

There hasnt been a reset in a few days... I hope Tibo is healthy

[sacred minnow]

imma cry

[oak trellis]

on 50% ... puuh

i count every 1%

[oak trellis]

bro .. nahh..

[sacred minnow]

xD ye

it will reset once im reset always does xD

[oak trellis]

never thought i would be a token junky .. i value tokens more then every shekel on earth .. OY VEY

token limits..

[sacred minnow]

ive just done alot this week i never hit this low

[orchid plume]

would love to have been one of the fortunate people who got 10x'd

[blissful basin]

codex goal blew my mind. Monday i started goal of full port electron app to web based one with syncing between two, backend, moving from local storage to remote one. I had it running during office hours 6-8h a day every day. It still works and it actually doing progress

[simple star]

Im friggin sick at Codex sucking at frontend

[cyan gyro]

the day the frontend is fixed...

[simple star]

You cannot imagine the hideousness that it is building for UI

[oak trellis]

uuuuuh 49% me

[oak trellis]

what that mean 20% reserve ?!

i need that too

75% .. what token saving tool you using ?!

[cobalt junco]

how do i give it to someone else

its not a tool

they gave me 10x on my 10x plan

[oak trellis]

wow .. why ?

[cobalt junco]

and i've been using parallel agents on xhigh fast mode on 5 different projects and it barely goes down 1%

[oak trellis]

that would be cool feature token sharing .. or second dary market place lol

[cobalt junco]

i applied to that one thing

[cobalt junco]

they would shut it down so fast 😂

[oak trellis]

true lol

[cobalt junco]

i ran out of ideas so im legit using it on max mode to fix lints in .md files

[oak trellis]

someone would launch a memecoin with wrapped gpt tokens LOL

[oak trellis]

you on 200 plan?!

[cobalt junco]

nah 100$

[oak trellis]

wow

[cobalt junco]

but its basically 10x 10x

[oak trellis]

where i can apply ?

[cobalt junco]

so its 100x usage

[oak trellis]

wow

[cobalt junco]

you had to apply before to that 5.5 party

[oak trellis]

aaaah

[cobalt junco]

and get rejected

[oak trellis]

i hope they do not reset the limits this time ..

[cobalt junco]

idk how this is sustainable tho, realistically this offer they gave is worth like 25k a month in api credits

how do they do that

[wicked briar]

hope they reset limits today

or tomorrow

[plush harbor]

hah, I've spent the last couple days doing a site refresh/reskin and you're not wrong

[cobalt junco]

use the image model to design

then codex to. implement

u can use the image model within codex

i found the best results that way

[simple star]

It is absolutely horrendous

[unique spade]

Meh, Llm's are still bits of data processed by algos.
Never heard anyone complaining Photoshop is bad at editing, or word bad at writing haha

[still trellis]

can someone tell why codex opens chrome to debug even when the browser preview is open within the codex application? Is there something I have to say or call to stop it from doing this?

[unique spade]

So i dunno take some ux courses

Maybe you ll understand better how to get what you want

[plush harbor]

I've mostly done it myself the human way but all the default stuff it puts in is pretty bad. It got me a fair chunk of the reskin and I did the rest by hand, with help from gemini choosing colours and chunks of css. Looks great now

bit sketchy on mobile though but that's tomorrow's problem

[nocturne folio]

its not bad, it made https://autodisc.xyz , it just sucks at pregen

and using skills helps alottt

[simple star]

@nocturne folio Dude, that doesnt look good

[nocturne folio]

it converts 🤷

[plush harbor]

hehe it made me a lil task tracker app and it has exactly the same colours as your site

[simple star]

I mean, there's clean and minimalistic, but it doesnt work for anything real. It is a bad landing page.

Its mostly whitespace

[nocturne folio]

whats an example of something good

[unique spade]

That's commenting on the guy aesthethic taste though

I don t personally like it either

[river bramble]

eeee a bit unbalanced

[still trellis]

gpt also makes weird ux choices, for instance i asked for a collapse button on a left rail. It places it at the bottom out of the viewport, thought it did not do it.

[simple star]

This is also "unbalanced"

[torpid trout]

this looks hot tho

[nocturne folio]

dannnngggg it

[torpid trout]

apart of the bad text width

[simple star]

The image, yes. Image generation is great. But web design sucks

[torpid trout]

use stitch

[nocturne folio]

should i just remove it?

[simple star]

@nocturne folio Your main problem is lack of content. SEO-wise, the page sucks

[torpid trout]

Have GPT optimise it for GEO (but tell it what GEO is). It does an amazing job at that.

[simple star]

From a design POV, too much whitespace. People will think that the page is broken

[torpid trout]

And use Stitch for design mockups.

[simple star]

@nocturne folio Im going to assume that you only saw it in mobile (where it looks decent)

[nocturne folio]

yea

[simple star]

Well, yeah, that's not enough 😛

[river bramble]

noo, that's not criticing your site, btw, it looks better than most of the plain output from gpt5.5

but it stil struggels with little details

it helps when you turn things like this card into a shared compomnent

[lost drum]

there is another update in codex app? oh man esterday there was one too

[simple star]

@lost drum Today? no

[lost drum]

so I have no clue why I have one again

[simple star]

v0.129

[nocturne folio]

no, i rather criticism, trying out google stitch atm

[nocturne folio]

theres /goals now but its like hidden

[river bramble]

👀

[lost drum]

but goals worked 2 days agoo too

I mean you just had to write /goal [message] and somehow it worked

and put in config goals

[river bramble]

it works because it uses codex app server just like codex cli

[lost drum]

yeah I used it in VS extension

[river bramble]

my codex fork has some helpful small qol like going from plan to plan mode direclty

[lost drum]

Hmmmm can someone share the chrome plugin the one that is available in US?

[unique spade]

Yea you can do wonderful stuff with your own fork.
I got tired of updating it, so now I'm just implementing my own harness from scratch lol

[torpid trout]

https://status.openai.com unstable systems loook like lol

[river bramble]

"/goal $sync-upstream"

[unique spade]

that's not the point. it's just more control from your side to build it from scratch, and have your end-to-end logic for the harness

[river bramble]

it could be, but it's not the problem i'm trying to solve, as i use it for real work

so my codex fork and pi are good enough harness

for myself

[unique spade]

right. vanilla is good enough too in most cases

[oak trellis]

Sam hear me out I know you in here incognito 🥸: I want weekly 2x 5 hour 1x not 5 hour 2 x weekly 1x

[river bramble]

call it Samuel 3x and it will work

[torpid trout]

Its still not accurate but a start in classification

[river bramble]

it's missing the "the model was nerfed"

[small violet]

i dont think open ai nerfs their models

might be wrong tho

[cyan gyro]

or pressing f5 every thursday for a new model announcement

[river bramble]

ohhh it's there

[river bramble]

dudeeee 🤦‍♂️

[small violet]

i said i might be wrong

[river bramble]

i'm talking about the image

lol

[small violet]

im not talking about the image

im talking about model nerfs

[river bramble]

it's a natural phenomenon, aka usage > compute

[boreal holly]

Interesting... changing reasoning effort mid convo apparently causes cache miss

[small violet]

what does that mean

[boreal holly]

It means if you have 200k tokens in ctx window, if you switch reasoning effort, the next submission is billed 200k input + prompt instead of 200k cached + prompt

[velvet wren]

oh really? that would explain what I have been trying to nail down all day

[small violet]

ahhh

normally its 200k input+prompt?

or 200k cached+prompt

[boreal holly]

the way caching works is, when making a request, all contiguously matching tokens from 0 to n are reused. if there is a single token that does not precisely match the previous request, it trims all tokens after that point.

I guess changing reasoning effort mid convo causes a cache miss. maybe trims reasoning tokens in the past

[small violet]

oh damn

i dont feel it tho

maybe cus i work on very small projects

lol

do u feel it?

[boreal holly]

nah, I only use medium. but figured that info was worth sharing

[vital bone]

for coding i'm using 5.5 medium, but just found GPT-5.3-Codex-Spark, is it good?

[vital hill]

codex 5.3 medium has been very good to me

not sure about spark though

[agile seal]

It's relatively fast, but not very intelligent.

[uneven kayak]

Man it's nice to have my server out there on AWS, inching closer to public play testing 😁

[broken rain]

is it me or codex 5.5 got nerfed?

it got stuck at a simple file took longer eventho im running at the same speed and extra high

[vital bone]

1,3 same ryt?

[tribal lake]

Is there an easy way to run codex review with mcps disabled?

[boreal holly]

It's really good for fast refactoring. I had this one job where I needed to add accessibility elements to a massive frontend, and in a matter of like 20 minutes it wrapped every single input element in semantics. Any other model might have taken a few hours

[vital hill]

wow nice. that codebase must be huge

[boreal holly]

The frontend alone is ~600k sloc of working code

[vital hill]

all in one file right?

[boreal holly]

Spark plows through it quickly

[small violet]

i need some advice

on my site

[boreal holly]

1473 1486 UI-only files

[boreal holly]

Sure!

[cobalt junco]

600k loc for a frontend?

show me this front-end.

[hard drum]

That’s not too far from size of Codex itself, which is ridiculous

[boreal holly]

It's a pretty big app (pics removed)

[hard drum]

I’m sure a lot of the code could be compressed

a more aggressive DRY could go miles alongside KISS

I’d estimate a solid 470k if you could do your things right

[boreal holly]

Haha yes, there will always be critics.

[hard drum]

But of course LOC isn’t a good metric since the syntax itself falls for it

Rust for example is super verbose in syntax && formatting

[boreal holly]

I am not counting any lines of rust, which drives the business logic of the frontend

[hard drum]

Then your frontend is an absolute mess lol

600k of frontend shouldn’t be that

your 600k of frontend is nearly 5.6x my entire programming language && its tooling

[boreal holly]

Idk what to even say to that

Thanks for the advice?

[vital hill]

i thought my internal dashboard project was huge

[hard drum]

DRY principle. Don’t forget that

If you have similar or equivalent logic across the code base, then it’s time to consolidate

I should probably invest time in a language-agnostic tool specifically for this type of stuff

to enforce principles, best practices, God object violations, && more

[boreal holly]

I mean you can argue it doesn't look nice, and I'm working on that, but the advice you're giving right now is not really applicable. I can't explain why it's that big, but just know it's not because of novice engineering, it's completely unavoidable.

[river bramble]

but just know = i stop there 😄

[glacial shadow]

Some of you guys have typed more lines here than lines of code to production

[boreal holly]

Yeah, good point! I needed that reality check

[vital hill]

time to center a bunch of divs

[short linden]

bro, are you joke ? 😅 $200,000 is too much.

[sharp silo]

i've figured out how to fix the error unable to install chrome or Browser Use plugins for integrated browser, you just need to uninstall the app and then using VPN to US, then reinstall it while using VPN, it should have full features back

thanks @neon plinth for helping me

[neon plinth]

ur welcome

the ultimate move

[chrome raven]

and with this after restarting all features stay?

[sharp silo]

all chat history, environment, plugins, skills, stay the same

you don't loose anything

[chrome raven]

including the computer use and other eu features?

like on my side after using vpn i get the features but after restart everything leaves, but i haven;t tried the uninstall you just dropped

[chrome raven]

the restart mean closing the codex app and reopening

[sharp silo]

i'm not sure about the EU blocked features, but you should try

i just need to open VPN once when reinstalling back, then i don't need to use VPN anymore

[boreal holly]

My count was off. 600k includes massive amounts of unit and integration tests, not UI-only. Idk what the distribution is but you were right to suspect something

I just counted single lines of dart, didn't really subtract tests

[hard drum]

tests would not count towards LOC

tests are usually left like so, but it would make sense to count tests separately

[nocturne folio]

spotlight sucks soo badly, tried searching up codex

time for ray cast

[boreal holly]

Well, I would say a majority of the tests are just "render this part of the app to a png", and they use it for seeing the UI without running the app. There's some tests that actually check functionality, but I completely forgot they write tests just to render and view the UI, which accounts for a lot of non-working code

[hard drum]

i call that "prose testing"

which is not fun

[boreal holly]

Part of that process is "must mock the DTOs and controllers", so they are doubling up on code, but it's to have mock data for the rendering

[hard drum]

⚠ 28 hooks need review before they can run. Open /hooks to review them. why?

how do i disable this hook-review garbage?

[boreal holly]

Apparently you can add a /etc/codex/requirements.toml which is typically for business/enterprise accounts, but works on any account type, and put your hooks in there like this

[[PreToolUse]]
matcher = "^Bash$"

[[PreToolUse.hooks]]
type = "command"
command = "python3 /tmp/pre.py"
timeout = 10
statusMessage = "checking"

requirements is treated as managed & mandatory, but it makes the hooks global

[uneven kayak]

Man this is nice, Codex can use the AWS CLI and handle everything. I guess that makes sense but it's just blowing my mind 😆

I set up most of the stuff I needed in AWS but I was having a small problem and it dawned on me like 6 hours in that I could just use Codex

[hard drum]

should i get a 2nd codex account, a Plus account, specifically as fallback when main Pro 20x hits limits?

i cannot do my CI notarize fix when codex limit

[uneven kayak]

Well that sounds reasonable, if you're hitting your limit then you need another account, but you won't have the same prompt cache

[hard drum]

they need to really boost weekly usage

it's ridiculously low

[boreal holly]

Might be more cost effective to do credits. Pro comes with higher priority processing and a higher 5hr limit

[hard drum]

might be more expensive with creds

or wait... i see spark

[boreal holly]

I mean you can try it, but if you're hitting the weekly on Pro 20x, you're probably gonna hit the 5hr limit on Plus

[hard drum]

could i... use spark?

spark weekly is 100% left

nope

[uneven kayak]

Eh, idk. Trust but verify lol

[hard drum]

even spark cannot be used when main weekly limit hit

[uneven kayak]

Oh damn, I thought they were separate

[hard drum]

they appear, but are not

[unique spade]

yea i guess spark has also speicifc limit but it s still part of the big qouta

[hard drum]

see?

[uneven kayak]

Looks like they're going to get their money from you one way or another lol

[hard drum]

they need to increase weekly

[uneven kayak]

I guess your project must be a lot more complex than mine, because I'm not having any problems with the limit. It's a lot more generous than cursor

I've been running an agent non stop basically all week can't hit my limit if I tried

[hard drum]

what are you even doing?

[blissful basin]

Do you use a lot of MCP servers? These can eat up your limits

[uneven kayak]

Client/server Android game, Fastify for the server, Svelte + Vite + Capacitor for the client etc

[hard drum]

not unless invoked manually, but the only ones i have enabled are OpenAI Docs, Anthropic Docs && OpenCode Docs

[boreal holly]

Tbf borrowcheckbtw burned a lotta quota from a multi agent bug. My guess is next week probably wont have this issue

[uneven kayak]

Ah that sucks but that makes sense

[blissful basin]

But i've noticed last week there were some situations that tokens were draining like crazy for one of my projects, few days later in new chat window same project, similar work was fine

[hard drum]

&& the bug is somehow entirely in the fault of the user

[blissful basin]

So i think there could have been a bug in one of last releases

[chrome raven]

When reset

[boreal holly]

Yeah they really should set sane defaults for multi agent so it's not like navigating a mine field when enabled

[torpid trout]

whatever this is

Experimental goals are now discoverable, stay paused across resume unless the user opts back in, and show clearer validation and multi-day duration output.

First of all I cannot see any change to /goal command, and would not know how they would grant a goal to run multi-day
They really should NOT let AI write everything man. This is one of them "huh what" "ah yes soory I misinterpreted that, inferred something that was not real" lol

[signal tapir]

I just had something weird happen: I saw my 5h limit jump up from 19% to 20%

[lean lark]

I'm getting the impression that there are a number of runaway "hey let's put this cool feature in" updates/features that aren't fully documented with expectations, guidance, original purpose, etc. Maybe they need to get ChatGPT to explain features for new documentation. If ChatGPT can't explain it, then they haven't explained it well enough.

[torpid trout]

I am not entirely sure there are actually any humans left behind the dev process of codex

I have a gut feeling it is all automated. They are sitting in the recreational room and are cooking up new things like phones, while codex makes codex somewhere in a stargate

[signal tapir]

It's going the same way social media platforms did several years ago

I don't think Facebook has anyone doing anything customer related anymore. Even "human evaluation" is automated it seems

Would be weird if OpenAI had less automation. 😛

[torpid trout]

Woudl be weird since even at least 3 users in this very chat here are creating tools to do that lol, indeed

Codex FTW:

 Why I did it: your repo instruction says “Always create atomic commits to git for work you consider done,” and you explicitly asked to put
  the brainstorm into milestones/. I also saw at least one tracked milestone file already in git, so I assumed new milestone docs should be
  versioned the same way.

  How it happened: milestones/ is ignored for normal untracked files, but Git still allows an ignored file to be tracked with:

  git add -f milestones/...

  I checked ignore status, saw .gitignore was ignoring the folder, then force-added the two new files and committed them as 1c0b619.

[solid lake]

It took me 3 days (+1 reset) to get over with the $100 plan

[torpid trout]

I mean - I guess now I have to be specific that making atomic commmits does NOT mean circumventing gitignore ah

[solid lake]

Atomic as in each file gets committed separately?

[torpid trout]

Atomic as in small thing. big boom. many regret. reset impossible.

[solid lake]

Git revert

Often for faulty commits

[torpid trout]

Codex commited a file in a folder that was gitignored
That is the issue. Not how to revert the faulty commit.

Also that is the reason for atomic commits

Because to revert, you do not want to revert 5000 edits across 1000 files

[solid lake]

How do you commit a file that was git ignored

[solid lake]

You can checkout individual files

[torpid trout]

why do you ask me lol, its explained in gpt's answer above

How it happened: milestones/ is ignored for normal untracked files, but Git still allows an ignored file to be tracked with:
git add -f milestones/...

[solid lake]

Ohh it force added

[torpid trout]

yes, because it said "I have to atomic commit my work"
So it went and circumvented the gitignore.
So now the system prompt says "do not ignore the gitignore"

I guess next will be "do not ignore the ignore of the gitignore"
Summarised as simple ignor-rant

[solid lake]

Interesting

[torpid trout]

Home Reseticus says

[solid lake]

Chat must be long for it to ignore those instructions

[torpid trout]

It did not ignore them

It followed them to the letter

Why I did it: your repo instruction says “Always create atomic commits to git for work you consider done,” and you explicitly asked to put
the brainstorm into milestones/. I also saw at least one tracked milestone file already in git, so I assumed new milestone docs should be
versioned the same way.

[solid lake]

So you didn’t have anything pointing to git ignore earlier

Got it

[torpid trout]

That's right, because you would not expect you have to tell an AI to conscioulsy "break" git and .gitignore is a file that git respects ...

In other words, when you hear "AI deleted my computer", that is probably how it happens.

[solid lake]

I mean it’s a probabilistic engine

It’s probable for it to make that assumption

Clearly it did

[torpid trout]

I would not say that is expected - it is in fact the first time in months
But, I guess, yes, the it can be reasoned that there is a reason for it to reason like that

[solid lake]

Well yeah it gave you the reason

[torpid trout]

I am still pissed.

Cant you just be pissed with me, and comfirm it is outrageous?

||:)||

[solid lake]

It’s hard to get pissed at ai anymore I’m numb

[signal tapir]

I just used "steer" to tell codex to not do builds. It went "ok, I'll just let this one finish". Then... "I'm just doing one more build because the last one gave useful information"

[lost drum]

would watch a series of him growing up thtuth live and like whole lore XD

full movie

[lost drum]

yo but wont eehey ban acc then?

[signal tapir]

4% left until the 12th... I really really need pro

[blissful basin]

im 11% left on pro until 12th 😄

[signal tapir]

That would easily last me

[lunar tulip]

damm do you have x5 or x20?

[blissful basin]

x20

[lunar tulip]

damm i have x5 and i am not even getting to 50% weekly

[blissful basin]

i use it 8h daily in work on multiple projects, so defo getting my value for money 😛

last time i checked i was using around 25k usd in tokens value monthly

[signal tapir]

holy crap

[lunar tulip]

damm i dont even know what i am using mostly been vibe coding en enhancing my specialist gpt for making good prompts for codex

been working on a health and fitness app

[lost drum]

was soing it for 3 week untill the reality hit me whrer it drifted off and now I need to fix it ehh

[signal tapir]

I usually use it on function level, or sometimes a couple of functions, but lately I have been programming a galaxy generator. Lots of complicated, and annoying math. Been using it heavily for this.

[lost drum]

next should be goblinus resetus featuere where it laughs mroe and mroe when your usage lowers untill he can finally spam message reset wen to tibo

[blissful basin]

if you are not dev and just vibe code app, i strongly suggest you to do 2 things: use gpt to help you build very detailed project structure first + some architecture guidelines, then get skills for frameworks/libraries it suggest. Get git nexus and then you can build really advanced apps

[lunar tulip]

i am a dev and do exactly that

[blissful basin]

Without that it might start loosing context of app and do a very stupid mistakes

ahh ok 🙂

i really hate "vibe coding" term 😛

[torpid trout]

GPT yesterday decided that it was more of a traveller lol
I think it misunderstood the word tokens for crypto tokens

[signal tapir]

I dislike the concept. At least when applied to production code. People who put stuff in production should know what the code does.

[lunar tulip]

thats is precisely what i do make a structured framework on my idea and let is also advanced reasearch bij using the gpt research function and then make the end state of what i wanna have and let it rip until it reaches the layout of the final state

[blissful basin]

yeah goal functionality is absolutelly amazing for stuff like that

[lunar tulip]

yes precisely and then it is also not gunna halucinate so much

and then when its doen i let it do a whole qa as a tester who tries to break everything and let it do code reviews where i watch it as well

and then the next polish and new made final state start again

[signal tapir]

@torpid trout That was directed at the line before your memes

[torpid trout]

Yeah, got that, still
Why put stuff on production I when know what it does?
That takes away all the fun

[signal tapir]

Try adding this to your agents file: "At random times, make creative changes to the code, with no regard to functionality" 😉

[lunar tulip]

hahahahaha. that wil make pure chaos

[torpid trout]

Since i have too many tokens I had codex index my entire ~/, create a state map in csv, a suggested re-organization map in csv and an audit of each asset present 🙄

Instead of consuming tokens, it added 1% to the available limit.

I am not kidding. And that was while 2 more instances run a /goal of each 2 milestones

Their thing is so broken it does not get more broken

[undone patio]

infinite token glitch

[signal tapir]

no, it's the lite version: "Tiny Token Glitch"

[undone patio]

set an automation to repeat it ad inifinitum

its free real estate

[torpid trout]

Next up I will feed it the entire handwritten scans folder, have it OCR it, put into md files, vectorize it, and connect the dots into a contiguous novel.
I wanted to do that since a long while, mainly didnt because api costs where prohibitive. And, well, I might get banned for it, my handwritten notes are often not using family friendly language

[fringe shuttle]

anyone else feel codex rate limits are getting used quicker now?

[signal tapir]

smileBeda isn't

[lunar tulip]

no actually the opposite

but also depends on how you use it i geuss

[signal tapir]

I feel like they are used up about ♾️ faster than I need

[undone patio]

i am going to code a new setting for my bidet

[torpid trout]

I just stole yours I guess

[torpid trout]

There was another guy here recently saying his limit went up 1%

[lunar tulip]

i always just use normal speed mode

[torpid trout]

Its the 1% club

[fringe shuttle]

give it back

[signal tapir]

me

[lunar tulip]

and then game while i wait

[torpid trout]

hello other guy

[signal tapir]

hehe

only the 5h limit though

[undone patio]

peasants

[lunar tulip]

have been working a damm lot of time on my health app

[signal tapir]

netherlands?

[lunar tulip]

yes

[signal tapir]

I like the language, but I think I will never learn to pronounce it

[lunar tulip]

its not easy i am telling you now

and also i hate my own laguage hahahahaha

i prefer english any day of the week

[signal tapir]

Same. Mine is Swedish.

[lunar tulip]

nice

[signal tapir]

I translated the name for the game I'm working on to Swedish. It instantly sounded silly.

[lunar tulip]

have done so fare a 4 year dev studie and now a bachelor cyber security and cloud studie of 4 years

[undone patio]

i heard if you talk nice to codex it uses less tokens

[lunar tulip]

hahahaha nice

[nocturne folio]

its unacceptable

[signal tapir]

my 5h jumped a percentage point again. Weekly still at 3%

[blissful basin]

i think they took away reset button from Tibo

[solid lake]

Lmao

Wait

How many tokens you use in the chat

[craggy jewel]

Incredible that image2 can create this level of detail and colorize it 🙂

[simple star]

Indeed

[undone patio]

image 2 is goated

[steady vigil]

image2 is basically the first alpha of ai imaging. it can create random things, but it really sucks at specific changes

[undone patio]

lol

not even close man

[steady vigil]

but its the first one that can actually do those changes

[undone patio]

the first alpha was called VQGAN

[steady vigil]

nah I mean for real utility

everything else was pretty much useless

[undone patio]

debatable

[steady vigil]

it will make sense in 5 years from now

[undone patio]

i loved vqgan

[steady vigil]

gemini for example cant do the edits at all. codex can struggle hard but do some very simple things

its super nuanced, requires understanding of intent goes way beyond generating images

[boreal holly]

Idk about the codex image gen side of things, but in ChatGPT I've had a lot of success with forking earlier parts of the convo if image gen starts struggling. If it can't nail it on the second attempt, fork it 🤘

[steady vigil]

yeah thats kind of driving home my point where it gets confused super easy

kinda like super early llm just having a normal conversation

[torpid trout]

huh... suddenly images do not post anymore here

[craggy jewel]

same here. Perhaps they have been modded?

[steady vigil]

perhaps images2 cant figure out how to allow the edits (j/k)

[torpid trout]

PM either

This is discord I think, not the server

[signal tapir]

yup

can't even send to friends

[torpid trout]

how dare they.

Meme must be shared, meme is getting hot in my pockets

[steady vigil]

describe it in 500 words or less?

or ascii art

[craggy jewel]

We are going to need a codex image gallery of some kind...don't want to lose these in the history.

[torpid trout]

codex make website for meme

codex deploy
dont kid me, I have too many tokens, i am doing that rn. Inclusive exclusive submission form so I can get hacked

[craggy jewel]

kind of a Codex satire gallery (though we love you Codex!)

[solid lake]

Ohh yeah images can’t be sent here

[signal tapir]

anywhere

[solid lake]

But I’m curious about your token usage in the chat your usage went up

[solid lake]

New?

[craggy jewel]

Homo Reseticus broke Discord, methinks 🙂

[solid lake]

Says service is being rate limited

Interesting

[craggy jewel]

Homo Reseticus, with the ironic-ish abbreviation HR, should be tasked to fix Discord.

[torpid trout]

gpt-5.5 high · ~/Developer · Context 79% left · 5h 96% · weekly 84% · 258K window · 677K used · 662K in · 15.2K out · Fast off
does that help?

And before it was 83%

[cedar bear]

Finally back

[blissful basin]

was there a reset?

[velvet wren]

no

[solid lake]

Servers so weird today

Not just open ai

[signal tapir]

Discord probably had to reset their servers

[lost drum]

Oh I thouthg I broke my windows 11

weird things appearing the longer the codex runs

now the file explorer has weird black bakgroudn around it hehe

I am not turning off my pc untill I finish project cause who knows if it fires again

[solid lake]

AWS had an outage a few hrs ago

But that should’ve been resolved

[signal tapir]

this was just like 30min ago

[solid lake]

Lmao it didn’t help tbh I don’t know why I asked

But extra 1% is nice

[signal tapir]

As someone who is carefully wording every single prompt to make the last 2% last for as long as possible on 5.4-mini, a single week percent would be gold. 😛

[remote quiver]

is there a changelog we can look at for codex updates?

[lost drum]

I dont trust it😭 thread can behave as a bridge/orchestration compiler: compile messy owner intent into the right prompt, handoff, plan, /goal, route packet, module request, multi-agent sequence, or blocker question, instead of acting as a simple prompt rewriter or default downstream executor.

[vital hill]

we're back baby!

[remote quiver]

I did, it doesn't give me a clear answer, just like the webiste, lol

[velvet wren]

https://chatgpt.com/share/69fe4e9b-4d94-83ec-9175-0502c6932b15

[boreal holly]

https://github.com/openai/codex/releases?q=Changelog&expanded=true

[remote quiver]

@boreal holly<3

[boreal holly]

I like it cuz it skips over all the alpha releases

[remote quiver]

wait, that's for the TUI, is the one for the GUI?

[boreal holly]

Well, that repo contains the app-server, which is what the GUI uses, so they parallel develop this repo and the GUI together

All the changes in the releases on there are directly relevant to the GUI

[remote quiver]

okay, thanks

that's good to know, lol

rofl, I just asked, and I just got another codex update, lololol

they've had like 3 or 4 updates in the last 2 days

[blissful basin]

Did someone try extra credits? Im wondering how it compares to sub usage

[signal tapir]

They are supposedly identical to API usage costs

[wide hull]

how do codex's rate limits work? What is being tracked? running time? tokens?

because I've been working on something off a train hotspot and the spotty connection is making tasks take longer, and im worrying that its burning my usage rates because of the extended time

[signal tapir]

tokens, what model you use, what mode it uses, if you are using fast mode or not... I think those are it. Anyone have anything to add?

[solemn acorn]

tokens but not all tokens are priced equally because cached tokens are significantly cheaper

[lost drum]

single pyhon command of mine that he triggers runs for liek 11min and it thsoe not burn tokens I think he just waits forthe output

[wide hull]

ok so a spotty connection isnt gonna make me burn through it faster?

[signal tapir]

I'm guessing if it has to resend stuff it will

[solemn acorn]

no, openai stores cached tokens on their end

[wide hull]

I usually operate on 5.5 Low, then increase the intelligence setting if needed

[solemn acorn]

smart move

5.5 low seems to be pretty nice as far as price to intelligence goes

and it’s faster as a bonus

[lost drum]

I just use high cause I have no clue if chiefed pormpt with clear rules and goal would be executed as better as on low intelligence

[plush harbor]

I use 5.4 cos I'm too lazy to change what its got as default

[lost drum]

no clue if more the intelligence means that the model just has more depths of reasonign about the work he plans and executes at a moment

[wide hull]

yeah it usually gets the job done

super helpful for coding my game, and even if it isn't, it's still going faster than I would've.

I usually try to be as descriptive wiht my prompts as I can to give it a clear end-goal.

oop there's the spotty train connection lol

[craggy jewel]

Your Codex session logs are helpful!

[small violet]

@boreal holly what was ur github

[still blaze]

when

us

when is

[exotic cave]

Anyone tried v2 of compaction? how is it?

[still blaze]

intel support

COMING

[velvet wren]

Intel support is not coming

[still blaze]

duddeeeee

i hate thgusssss

i want codex so baddlyyy

@velvet wren wht abt codex cli

like running in ur term

[lost drum]

no happy ending?

[still blaze]

OMG

its from openai devs site

[small violet]

oh

macos intel in 2026

[solid lake]

Codex app on Linux

Pls

[small violet]

wrap it up

[solid lake]

Someone tell Sama pls

[vital hill]

codex desktop app needs to have ctrl + backtick to open a terminal

[velvet wren]

just got a Codex macOS update

[vital hill]

same but nothing in changelog

[still blaze]

keep dreamminggg

[solid lake]

Nah

They will listen

Now that they no longer bound by Microsoft

[boreal holly]

github.com/robertmsale

[plush harbor]

wait what, I say from an intel mac

[plush harbor]

I had to give back my work computer when I quit my day job, I'm running on fossilised hardware from the 2010s here

[vital hill]

what's stopping you from vibe coding it

[solid lake]

Someone already did and it doesn’t work properly

Also usage

[plush harbor]

where are you seeing hte second download? I'm only getting the one download link

well that link sure isn't tailored to your OS. But hey, I tried

[vital hill]

😭

[solid lake]

Also auth flow

[torpid trout]

I did something.

I hope I wont get banned for this, it is for fun, not an ad.

[small violet]

fix ur slop

[torpid trout]

what antiquity of a screen are you using

It looks beautiful here. No one has ever seen something like that.

Get a wider screen, brother.

[torpid trout]

Cant you see how wide apart those eyes are. How dare you staring straight forward? you need a wide screen. very wide. the widest

[torpid trout]

this is ancient

[small violet]

its the normal size for a laptop

fix ur slop buddy

[torpid trout]

You need newer. Better. bigger!
No kidding, I am scolding codex now for that, and hiding that chip a bit earlier than mobile 😉
Decided to leave it like that exactly for the rage bait purpose.

get a wider screen will be a footnote soon.

[lost drum]

🤣 he made that file himself

[tiny fulcrum]

anybody else having problems?

[torpid trout]

I have many problems. Not of that kind tho right now.

[potent mason]

Wow I'm flying through my limits, this might be the first time I manage to get 100% of it down in the week without actively trying

QA testing devours it

[wise heron]

guys.. do u see the Chrome plugin in plugins?

[lost drum]

am not in US but I might just use proxy to dowlaod the codex app for mac or something use ghidra or other tool to extract raw app or some to know how the real plugin works and modify it or idk what to do I am not dev and havent even done research whats the best approach so ye

[plush harbor]

best thing I've seen today

[wise heron]

eh?

[lost drum]

in EU you dont have a chrome plugin

[vital hill]

gg2ez

[carmine copper]

I don't think you need ghidra for codex app. Codex app extracts codex app's source code somehow from installed codex app, just ask it.

[vital hill]

new desktop update

still nothing in changelog

must be bug fixes

[nocturne folio]

3 words !! o4 is releasing !!

[vital hill]

mf needs to stop being "edgy"

[hard drum]

dude looks like he's trying not to cry in every image he is on

[cobalt junco]

hey i just met you

and this is crazy

so here's my number

call me maybe

https://open.spotify.com/track/20I6sIOMTCkB6w7ryavxtO?si=4fcce8fd05df4d6e

[vital hill]

nah, this is mood https://www.youtube.com/watch?v=it47TBjmXT8

[civic viper]

How do you clear context on Codex App? Any command or start a new chat?

[unique spade]

for non-project chat you have

in windows at least

[civic viper]

Yeah I was wondering if there was a slash command to clear context like /clear

[craggy jewel]

Start a new thread/session/chat.

[lost drum]

who uses the /side command efficiently and can give feedback of the actual good use case

[unique spade]

in an existing chat no. you must craft it yourself but you need to fork the codex cli for that

if you want fresh context just start a new chat

[solid lake]

I just tried /side to have a discussion/come to Jesus about certain log outputs cause I was getting low on context on main chat

[lilac thorn]

Codex "Computer Use" is working on my MBP but it is very slow. Is this just where it's at currently or could it be because I have high-res displays (5K & 6K)? It takes hours to do computer use tasks.

Since Codex is so slow, can I continuse working on my machine? Codex doesn't seem to steal the cursor but will I confuse it if I do work? Also what if I witch from the space that the app it is using is in?

[lost drum]

agi complete. 😂

[lost drum]

60% to use within 2 days cause tibo will reset on 11th probbably ro early 12th

[frail meadow]

What’s the latest codex meta? Any new changes besides /goal ?

[velvet wren]

the Codex app now supports Intel Macs

[small violet]

how is this a meta

dumb

[velvet wren]

it's news

[small violet]

he didnt ask for news

dumb

[velvet wren]

he asked for new changes

[solid lake]

Google Apple and intel today

[small violet]

do u know what meta is

[solid lake]

Google Apple and intel news for today

Quite a pleasant surprise

[small violet]

bro that isnt meta

😭

he's asking if theres anything else apart from/goal that people use

[velvet wren]

he asked for new changes, Intel Mac support is a new change

[small violet]

man shut up

What’s the latest codex meta? Any new changes besides /goal ?

[solid lake]

He said “the codex app now supports Intel Macs”

[small violet]

"What’s the latest codex meta? Any new changes besides /goal ?"

rephrased

[plush harbor]

I still can't find the link anywhere for the intel version

[velvet wren]

refresh the link I posted

[small violet]

any new changes to the codex meta

[oak trellis]

I hope 🤞 OpenAI double the weekly limit

[small violet]

how is intel news

meta

[solid lake]

Intel Mac was meta back in the day fr

Especially if you had one of those cheese grater towers

🤤

[small violet]

ok

yall interpreted his sentence though

[small violet]

did they discountinue in 2019?

[solid lake]

2023

[plush harbor]

codex is so conservative. Said there's shiny things in the api, should I add more api stuff. It says no, unless. Told it what I actually want to do and it changed to yes, but

[solid lake]

Okay wait I thought you meant intel chip

No wait that too 2023

[plush harbor]

ok that link works on my laptop but not my desktop

looks like I have an app already that does what api would have done so that's nice, less money

bit mediocre though. Can openai say trebuchet?

[solid lake]

I’m still confused by what the use case for spark is

Is it trained on the Cerebras chip?

[plush harbor]

ok this is funny how chatgpt and codex disagree

[agile seal]

Overly secure and conservative security filtering actually hinders efficiency.

[plush harbor]

codex is very ... dull

[uneven kayak]

Codex is still blowing my mind every day, now I have it set up with an Android emulator and an SSH connection and AWS CLI to the backend server on AWS so it can test the full stack in a live production environment. Absolutely amazing.

[plush harbor]

I've run out of things for it to do unless I pull forward future stuff or start a new project. 4 days to rebuild/reskin this website isn't too bad

[uneven kayak]

Wanna make my website for me and save me some useage? Lol I already have the plan documentation 😂 jkjk

[plush harbor]

hah. I was a front end dev for ages. I just don't do design, so this would have been way faster if I had a figma design already done

[uneven kayak]

Yeah makes sense

[plush harbor]

was a lot of fiddly "this looks bad, gemini, halp"

[uneven kayak]

I've always used a theme builder on WordPress so I never had to do much, now I'm handing it off to Codex with GeneratePress and GenerateBlocks so let's see what it comes up with. For the most part it's just a basic information page for my game with a basic support ticket system, except I'm leveraging the power of agentic vibe coding to kick it up a notch with a web shop and player dashboard for the game. It's beyond the scope of what I would probably do under other circumstances, but this is 2026 and everything has changed lol

I'm using three layers of authentication but I'm looking forward to handing the shop off to hackers and see what they can come up with 😂

[plush harbor]

I've usually worked for corporates, so you get this ridiculous mildly unworkable fancy thing from some phenomenally expensive design firm and you're expected to glue it to a website, preferably by next tuesday because they wasted 6 months of dev time arguing about the button colours

[solid lake]

Dam

[plush harbor]

I think my shortest turnaround for an entire website was around 24 hours. THat was not a healthy company to work for

[uneven kayak]

Ah yeah, sort of the same thing in sysadmin, the server better be migrated by 6 AM on Monday and you better not sleep all weekend to make it happen. But I guess I could just had things off to agents these days. It's been a few years.

My current use of Codex is purely personal, though I do intend to make money from the game but that's sort of beside the point

[plush harbor]

I have burned through all my backlog items on two projects now, its entirely onto clean stuff I thought I'd do in the distant future now. Damn codex. Too efficient

[uneven kayak]

It's incredible, I just can't get over how it can coordinate a live full stack test and check for errors and fix them autonomously across three repos.

I keep trying to find more complex things for it to do and it keeps doing them without any issues. It fixed an issue in AWS with target groups and then found two other problems and fixed them without me even asking

[oak trellis]

what i like is the small feature when codex do updates

beautiful

model got better too .. something changed overnight

Yes. The direction is smart overall. It is one of the first versions in your discussions that actually stays disciplined instead of turning into “AI magic floor detector” nonsense. The strongest parts are:

[plush harbor]

this text to speech api thing I've had on my to-do list for like 15 years as a "maybe later" item and its now popped to the top of the list due simply to sheer lack of anything else to do

[oak trellis]

adding too much AI
adding too much scoring
using future leakage
mixing execution + research
scanning raw swaps forever
building non-replayable heuristics

This version avoids most of those traps.```

🫡

gpt start selfreflect

[oak trellis]

48% saturday in bali - 3 days left ..

until 12 and the limit reset ..

i might update to 200 ..

[hallow nexus]

🙇

[oak trellis]

ratio is now 18% 5 hour is 3% weekly .. man .. that is bad .. ..EATING .. all on 5.5 medium

[oak trellis]

1% of weekly in 30 minutes using one agent 5.5 medium

[cedar skiff]

it is more expensive per mil toc than opus, which is notorious for being expensive.

[uneven kayak]

So that's roughly 50 hours of use in a week, not bad if you ask me.

It's hard to have anything substantive to compare it to other than hiring an employee, who does less work in more time and costs more money. $100/month for 50 hours a week of real effort is kind of an amazing price if you think of it that way.

[balmy olive]

Unusual use case of me abusing codex lol I made it export and crop upload ready images from random design frames in webp format 🤧🤧🤧🤧

[plush harbor]

does codex use its own token limit or your plan images limit for that

I suppose just cropping is an external tool not codex per se

[cedar skiff]

I think its in the codex uasage bucket

Oh i misunderstood the question

I am in the middle of trying to build ui images with image gen and got my wires crossed

[uneven kayak]

Tool use and imagegen both come out of the Codex useage limit

[plush harbor]

been a bit of talk about getting codex to make images

[plush harbor]

I've had codex process about 5500 images but its been via a script, so it hasn't touched my quota

[cedar skiff]

When do we get skill sin chatgpt?

[uneven kayak]

It's not bad at all, uses the latest models so it has great quality if you need it. Plus you can integrate with the Scenario MCP if you want even more options.

[cedar skiff]

image gen uses the new model for generating images

[uneven kayak]

Smart

[plush harbor]

I hear they've fixed that. I've been routing around it to the older one

[cedar skiff]

was it broken?

[plush harbor]

just running out of things to do, and I've got another old site I could happily do batch image creation on

[plush harbor]

so very very broken yes

[uneven kayak]

Man I have plenty of things for you to do if you want to. 😂 I'm condensing a 5 year plan into a month with agents. It's like a tornado style of development.

[plush harbor]

I'm back to sifting through photos. Nothing to code right now, unless I start something totally new

[lavish mauve]

So the announcements say we can now import from Claude Cowork..BUT Codex import detects Claude Code but not Claude Cowork for me despite Claude Cowork being installed and actively used with projects on my Mac...any ideas why?

[solid lake]

$path?

[lavish mauve]

Is that for me?... could you elaborate ...? thanks 🙂

[solid lake]

Actually I don’t see where it says cowork supported

Only cc

The cli

Is cowork a desktop app-only feature

[lavish mauve]

it clearly shows cowork here in the video ...https://chatgpt.com/codex/switch-to-codex/
Cowork is desktop app

[uneven kayak]

That looks like a marketing demo, not the live import flow. See how the wording is different, "Import work from other AI apps" versus "Import from other AI apps". They might not have shipped Claude Cowork import and accidentally left it in the promo video. 😆 That would explain why @solid lake can't find a mention of it.

[lavish mauve]

aaah...LOL... oops 🙂

[solid lake]

No but surely there’s a way to migrate that

Cowork saves work in a different format than claude code lemme check

[oak trellis]

yes true!

[solid lake]

Hmm there’s no single way to do it, looks like you’ll have to differentiate manually

[lavish mauve]

No idea how I would do that…any ideas?

[solid lake]

Just tell codex to find the files on the drive and migrate what’s useful imo

[lavish mauve]

Aaah I see… ok…thanks ! shame it doesn’t do what it says in the ad 🙁

[uneven kayak]

Omg my build is so squeaky clean 🫧🪥😲 I asked Codex to resolve some warnings and it went all out on library updates and compatibility patches, it's so beautiful it brings a tear to my eye lol

[floral garnet]

just tried the grill-me plugin, bloody hell....on question 81 now, it just won't end but damn this thing is thorough.

[uneven kayak]

That looks interesting. Can it be adapted to an existing project or is it more about fleshing out a concept at the beginning of a project?

I'd love to be grilled about my project, that seems like a useful thing to do when vibe coding

[floral garnet]

I am reworking parts of my project and its taken it, went through it and now asking me questions since I asked it some specific things I would like to change in it.

[uneven kayak]

Very nice

I'll have to give it a try

[floral garnet]

basically no more assuming anything from the llm side, we get to decide every single decision in the design process. I have been at this for almost 2 hours now.

definitely give it a try, I should have done this earlier, wouldn't have had to re-design what i am redesigning now because it turned out to be a complete mess

[uneven kayak]

Nice. Well better too much info than not enough, I call it the grep gripe paradox. Give it more to grep and it'll have fewer reasons to gripe lol

[simple star]

I can't believe Tibo made a Twitter poll, and didn't include Linux

[uneven kayak]

I'm pretty confident in my current architecture but maybe I'm totally wrong so we're about to find out lol. Trust but verify

[floral garnet]

lol, well I may have lazied a bit earlier which has led to me to my current predicament, but definitely would give the skill a shot.

[uneven kayak]

The cleanup session is actually still running so I'll wait for that, updating dependencies and cleaning up deprecated code is a real rabbit hole

[floral garnet]

good luck man, hope this thing helps!

[uneven kayak]

Hell yeah, I'm really looking forward to seeing what it has to ask. Pretty much guaranteed to learn something.

[floral garnet]

I have, learned a lot, question 90 now...should have eaten before I started this

[uneven kayak]

Lol take a break, you'll give better answers and have a better understanding after you eat something

Strong body, strong mind 💪 lol

[floral garnet]

its all good, I feel like I may be at the end of the questioning, I did get this feeling when I was near question 60.....I will get to 100 and take a break 😛

[uneven kayak]

Lmao

It's just leading you on, okay we're almost done .. with this section.

[floral garnet]

lol, it never says we are always done, but the questioning feels like it and then bam, it hits you from the side

[uneven kayak]

At a certain point I'd just be like, "Are we almost done here? Can you cut to the chase already and just tell me what I need to friggin know!?" 😭

Jk lol

[floral garnet]

ah man, I am pretty happy with these questioning, definitely fine tuning the whole thing with a comb

[uneven kayak]

Yeah that's definitely important stuff if you're working on any sort of project larger than a basic script

[floral garnet]

project is already in production, making major improvements.

[cedar skiff]

If you say i want to make a social network gill me i think youre going to be there a while 🤣

[uneven kayak]

Vibe coders should think "what does corporate do?". Because now we have entire teams as well, they just happen to be agents.

[floral garnet]

...

it's grilling me till i'm well done x.x

[uneven kayak]

Well my cleanup just finished so I'm gonna test it real quick then get to the grilling ♨️

[cedar skiff]

does anyone else dislike the way gpt likes to say what things are not.

Like it starts with - it is not this, and does not do this. Kind of stuff when you ask what it does.

[floral garnet]

it finally ended!

[oak trellis]

i would say 99% every chatgpt model is still doing heavy mistakes

[turbid cedar]

Is there a way to get more detailed information about what the “Exec” usage consists of? Over the last few days, more than 90% of my usage has been classified as exec, and I’ve now hit my weekly limit because of it. I’d like to analyze what exactly caused it.

[uneven kayak]

Omg these questions make me want to smoke 🚬 😭😂

I can feel my hair getting greyer

[signal tapir]

You can relax though. There's a limit to how grey hair can get.

[uneven kayak]

It's like, oh you already have good architecture so let's see how we can turn agentic efficiency up to 11, going on about agent task contracts.

[floral garnet]

well the documentation is ready, time to start implementing, this thing better work in one go before i pull punches

[chrome raven]

using credit with 5.5 is so crazy, the credit disappears like it's breeze

[uneven kayak]

I found/made a cheat code lol. After about 25 questions, I said to it,

Your recommended answers have all have extremely good so far. So, please stop asking me to answer questions and instead, shift into an autonomous question and answer mode. For the next 10 questions or until you're done, whichever comes first, ask the question and then use your recommended answer. After every 10 questions, or until you're finished, you should stop and present a brief summary of your questions and answers for those 10, and then I will ask you to change answers or proceed with the next 10 questions.

[signal tapir]

This is why I'm switching to pro come months end

[signal tapir]

Hillarious, though you can probably make that actually useful if you ask it to try to model you for the answers.

[uneven kayak]

well I wasn't lying when I said its recommended answers were good, I've used all but one of them for 25 questions so I figured it would be faster to just review 10 recommendations at a time

[signal tapir]

nice 🙂

[floral garnet]

that is smart, I will do that next time.

[signal tapir]

What are you doing that requires so many questions though?

[uneven kayak]

Agentic workflow efficiency, working on agent task contracts

The contract should separate path intent into expectedTouchedPaths, requiredCompanions, and outOfScopePaths; validate path shape/existence but leave implementation drift to PR review. It should carry schema/profile versions, concise notes, profile-level stop conditions, additive extra commands, explicit human acceptance requirements, an approval trace in humanDecisions, simple dates, and lightweight ownership for decisions/questions.```

I could fill up this Discord channel with the details lmao

[signal tapir]

So the output is an AGENTS file?

[uneven kayak]

the output is a contract, well a profiled set of draft contracts that agents will refer to orchestrate large changes and feature implementations, and explicit instructions about how to understand the scope and blast radius of any given task

[signal tapir]

Interesting. I like it. 🙂

[uneven kayak]

basically to change poking around in an AGENTS file to "this is precisely what needs to be done and precisely how to do it" based on the task profile.

[signal tapir]

That will be very useful

[meager bobcat]

AI agents and models trying their best to pretend humans are still necessary 😂

[signal tapir]

We are. Maybe not for long, but we still are.

We are much cheaper to run.

[meager bobcat]

I mean its designed that way for a purpose though

[uneven kayak]

lol I told it to add the human acceptance requirement, I bet it wouldn't if I didn't 😆

[meager bobcat]

AI and robotics will always be designed to wait for human will if we intend to keep being anything but annoying flies lol

[hopefully]

[uneven kayak]

in a couple years, a 5 year old will be able to make their own games with Codex

just by being upfront "I'm 5 years old, make this entire game for me"

and it will

[meager bobcat]

why 5 lol

[uneven kayak]

ok, 3. lol

idk

[meager bobcat]

Steam search function is already obsolete its just that people dont know how to use Codex etc

[uneven kayak]

well it's not quite so simple to make a creative work, good games are still a long way away from being fully vibe coded, there is a lot that goes into it

[oak trellis]

i hope monday come no limit reset and steal 40% of remaining

[uneven kayak]

I mean there's a difference between a spreadsheet and a management game

one is data and the other is art

[signal tapir]

I kind of feel my game is both

[meager bobcat]

I have been thinking about that precisely past whole week.
I think theres something there that can be solved in future like... right now we are talking about engineering and designing games. That was invented back when you needed 50,000 people to make a game (some companies still do this today)

But what if players soon remove the middleman?

[signal tapir]

But I mean, I've been using codex heavily for around 3 weeks, trying to make a galaxy generator that fits my requirements.

[meager bobcat]

People are talking about doing that for fitness apps and etc already instead of paying 50$/m in fitness programs

[uneven kayak]

most games are a bit of both of course but the user experience isn't something AI can fully comprehend, so a good game still comes down to how people "feel" about it when they play it

[uneven kayak]

the trouble is, that middleman knows a lot of stuff. He's there not just because he makes things possible, but because he knows how to make things possible. You can tell an AI "make this game for me" but the nuances of the user experience are why the middleman can never be replaced, the AI can make the game but they can't understand the emotion it evokes when it's being played and make changes based on that feeling.

[meager bobcat]

I think this is 100% accurate for an average casual gamer

[signal tapir]

The player also often doesn't know what would be fun

[meager bobcat]

but most indie developers become indie developers because they were obsessed about gaming and fun stuff. Their whole job is to make games they would enjoy playing ultimately right

Theres a lot of complexity and theory you begin to see in game design that takes 10 years of playing games AND THEN 10 more years of making games... so I think you are right

[signal tapir]

I think most technically make slop for mobile. 😛

[signal tapir]

I've been thinking of how games work, and what makes them fun or not, in an analytical way since the early 90s.

[meager bobcat]

I think today an obsessive gamer who wants to make a game reality that does not exist yet can totally pick up Codex and through trial and error and iteration make something incredible. It wont take them 20 years of experience or any talent

Its about "dream game" not about "shiet imma get paid for this"

[signal tapir]

No. It's not that easy.

[uneven kayak]

the player, reasonably so, is not going to know about all the nuances that go into the reason why their experience with a certain game is so powerful, they're going to focus on the end result and how it left them feeling when they finished playing it. Then they'll try to use an AI agent to create a game based on that feeling and the agent will inevitably fail to actually give them what they wanted, primarily because they don't even know how to technically describe what they want.

[meager bobcat]

If they really wanna make it happen and they got time. Its totally doable

so technically the middleman is always removed if you got a good enough motivation for it

but I guess that was true even in past with so many indie devs lol

[meager bobcat]

yeah its crazy

normal players dont notice this. they just have fun

so... true

[uneven kayak]

that's true in the same way it's true you can become a pilot or a lawyer lol, you can do what you want but it's not going to end up the way you expect unless you're the one who is making the executive decisions and you can't make decisions about questions you don't even know to ask. Hence the middleman, this is why we will always have real human doctors too.

[meager bobcat]

yeah but you cant deny the gap is closing like crazy...
so win win I guess.

in past you needed million dollars in hit kickstarter funding and then some more from investment firms to make an ambitious game

[signal tapir]

No AI will spend the attention to detail on things like what galaxy generation looks like as I am doing. It's not needed for the game, but it's needed for the artistic expression.

[meager bobcat]

well ofc

[uneven kayak]

I'm not saying the AI can't do it, I'm just saying the AI is going to do it in a way that is different from what the user expects without the kind of instruction that can only come from a deep understanding of how games are made

[signal tapir]

Tons of this type of details go into making a game, which is why I think it will take a long time before AI can remove the middleman

[meager bobcat]

I am not here arguing that AI will replace humans in games
I am considering possibility of role of game developer shifting into a gamer more

[signal tapir]

I don't think so. Not for quite a while. Maybe never. Becaus of what Andrew is saying.

[meager bobcat]

Right now game developers and game stores (app store/googleplay/steam) operating like middlemen
In future some gamers might want something custom
isnt Elon even working on live streamed video-gen games? now thats gonna be crazy if it ever works even sloppily

[signal tapir]

Game development isn't another angle of gaming. It's inherently different, because it requires a different skillset.

You can be a bad player, and a good game dev

[meager bobcat]

Ofc. and yet theres something about making games in cold and technical way that doesnt work, no?

best way of making valuable games is making games that are fun and replayable, not trailer and release date hype value

because lots of games rely on that today and in past

[signal tapir]

That doesn't come from being a player.

It's a separate skill set

[meager bobcat]

ofc. but what? im trying to figure it out thats why I am saying this xD
like if it isnt players job to know what they like, then how are game designers special? arent they just more analytical gamers?

I often look at my own games, especially nowdays that Codex makes new features and iterations so instant