#codex-discussions
1 messages Ā· Page 69 of 1
If it was, then why you need so many controls? For the fun of it?
such an easy fix
you can make it so much better..
lmfao yall dont know duce about design
Alright, you can do it :)
hokay buddy. i guess i dont know
i know i can lmfao
you dont lol \
that is horrible
mad cluttred
ugly background
cheap looking buttons
brother brother lmfaoooooo
lmfaooooooooooooooooooooooooooo
looks like the knock off fab filter
are you offended?
omg guys I just got a week's reset again
@orchid echo leave it. Not worth it.
true
sorry but when did they disable js_repl in code code base?!
I had codex dig in the source and it says that feature flag is gone (not usable anymore)
Yeah they did, i was talking about it when they did. but forgot when it was. they now have some code_something.
code_something?
https://github.com/openai/codex/releases/tag/rust-v0.132.0 zero words on deprecation of js_repl
They removed it in 0.128.0 and added code_mode.
congrats, another botched release announcement, gosh.
ok looking... but I recall that code_mode now
It was never released.
Even if announced, I have a ticket open on that actually
I don't know if they still work on code_mode, but it's a feature you need to add to your toml. But it's not the same as js_repl btw.
magic @lean lark
#19410 Remove js_repl feature @fjord-oai
yeah, go figure
I was sad to see js_repl deprecated.
0.128.0
no replacement
needa a bit more tweaking visually though and it should be all set
For my node heavy development it was a nice addition.
yea gotta chill bro it might bann you for 7 days or perma ban
we dont wanna loose you broski
Worth over it
Kind of boiling here rn lol
Seriously, get a grip on those release announcements and specially... review what actually is being released
I would be ready to bet that was some smoke idea of codex itself to remove it lol
i need that random reset badly lol
You think codex is conspiring against us? o.O
i already got timed out before for saying 3 letter double you - tee - eff
yeah of course. did you not know
they conspiring in terms of giving too many resets, and now, hiding js_repl
ahhh i get it compation degrades over time , not just the code but the compaction potency itself
compaction in a new threqd reduces tokens by almost half , in a lagre thread it reduced by 10k the most
remeber cydia compactor
Really have no idea why they removed js_repl, it was really good for anything node/js etc...
cydia was so great
youd think wed have more jailbreaks with ai but apple is probably using a military grade ai
or a super computer type ting
@torpid trout It gave a persistent node backed js runtime with top level await, local imports, package imports, helper access, and reusable bindings the model could run js logic to call multiple tools, transform output, loop, filter etc... while staying inside one tool flow. (from my previous explenation about it). pretty potent stuff.
i feel like ad space in an app paywall is horrible for sales idk why people do that
yeah it is suggesting that or something like that rn, some mix between node script and using chrome headless
however, this all semms a total mess to me
wow making great progress now its looking like the triads page is complete !
wow
excited @plush harbor
new thread is finally matching the consistency of the old thread it just needed a pep talk lol
You can make a yakuza page next (jk I know its a music theory app lol)
lmfaooooooo
it will have a free browser version and a paid app version
Neat
atleast thats my plan idk if the brower cn handle switf code
I honestly dont know if it can or not
i want it to be music app of the year tbh
and honestly i think it can be
the goal is to sell it to apple for 1billion then take a 10% royalty lol
This might now be it
But, always worth it for a meme
Anyway, it added playwright mcp, seems to doodle away now with it. So, A(va)nger averted
looks like spongebobs cousin
Time for me to try codex I think, I came here because I want to switch from claude
claude is duce
etihs si edualc
Cant login to openai lol
Lol the little things if one look. One of them, the cup at the last panel xD
AGI is close...
can you smell the reset?
Is there a way to get higher context window in codex? Iam a pro 5x subscriber
wen
yeah... that one messed it up
But I liked the tealeaves consultation lol
pretty sure its limited in codex, need api for 1m
doesnt it allow 400k atleast? Iam capoped at 256k
256k is preatty low if you ask me
i noticed this when i used pi/opencode, gpt 5.4 had 400k there while codex had 270k
i have my autocompact set to 230k and it feels like infinite context with no degradation to me
You can use Oauth in opencode?
yes
Kinda scam if you ask me
its even semi-official, although TOS still forbid it
256k is with their margin. also 400k is with the ouput/reasoning that the model have as a play field which i belive is 120k.
where does TOS forbids it? have a link at hand ? curious to give it a read
Think Or Swim
would it be possible to code a program as interactable nodes that hold thier function within a container so say if i deleted one node the rest of the program wouldnt collapse ?
like leggo bricks
I was just researching removal of js_repl and poo. š
About poo : It's confirmed that the word is moderated but that the emoji "probably" should not be, and it was expected to work. So that will be looked at by mods.
About js_repl : It looks like that was removed, in part, because it was perceived (?) as only (?) being useful for Bazel testing. Removing js_repl resulted in removing Node from CI and a related Dockerfile dependency.
It looks like code_mode (still in development?) is replacing js_repl. It looks like they removed js_repl without introducing its replacement - generally bad form, but since it wasn't advertised as being a feature for us to plug into, it seems they were at liberty to remove it at their discretion.
While js_repl is gone, there is a node_repl/js or mcp__node_repl__js used with PlayWright - which kinda explains how they're still using PW without it - but that doesn't seem to be fully implemented either. š
I'm thinking, if there is a devloper use-case for keeping js_repl in Codex, make your case in a new ticket. Otherwise stop using the undocumented (I found none) functionality. (sorry)
There's a lot of vague guessing going on there but that's what my limited research turned up, short of creating a ticket to ask about it.
And now, on to real work...
Yeah but the method is dependent on the language/framework. You're using Python. Is that your target for the question?
And how is that related to Codex?
im using Codex and Xcode and Swift
ok what is this ?? 3 min just to move 3, 15mb files
its possible your thread is congested
Oh yeah, confused with another project.
when i started a new thread it was 50x faster
i just restarted the pc with no other apps open and its a new thread
And destroyed your life. š¤£
i wonder if you baked your file paths into an agent it would be faster
type shi lololol
its okay to cry , but its not okay to not look in the mirror an call your self a Bish after
lol and thats what i did i man'd up and powerd through it
0.133 just dropped
codex update ?
Goals are now enabled by default
this is mac right
yea mac supremacy
Something else to mess up your life, PIC.
ah shoot lol
crazy to think my project had 6435 lines of code
have they fixed usage consumption ? Im on plus and i burned through ~90% in around 1h. Definitely wasnt like this last week. Maybe im hallucinating
humans are amazing
Mobile app says to follow instructions on desktop but nothing appeared there yet
I don't see anything about the token consumption
https://github.com/openai/codex/releases/tag/rust-v0.133.0
my app is looking lovely
and it follow all the apple design principles
masterpiece
its murky like all TOS, but i would say it falls under the forbidding automatic/programmatic access because its third party software
so if they want to enforce it, the TOS allows it. so far they havent because its better marketing
aw hell nah
I know it might sound weird, but... Is there a way to make codex talk to chatgpt using my Pro subscription?
yes steipete/oracle
or browser use in the app
upgrade to pro, plus has almost nothing
wym ?
codex told me it's against the TOS to scrape data out of ChatGPT
yes it is
honestly i kinda like the limits it makes to live your life and think of new ideas
just zip your repo, upload it and prompt pro manually
it can even return a git patch
even if it wouldn t be they could add it any time when they would want to enforce it š
yea sure. i mean right now they are relaxed about all this, i hope it stays this way
So what was added? I just got a windows app update
one of my accounts got a "suspicious activity" warning (i guess because of oracle cli use), so it seems they are starting to be more strict
No, the usage still same with real API token usage. so it's not change, it's just calculate by real token usage.
But if you compare with API token cost, it's cheap than API.
Example I paid $200 plan. If I switch to API, it's might cost $1600 while $200 plan cover for this price.
i noticed too if you hit your limit mid task it will complete tht tasl first
yeah, what @bright swift said was non-sequiter and unsupported. The codex-for-oss quote that he provided isn't related to TOS, and he conveniently left out the rest of the sentence:
yes the oss program supports oss work
two different parts of a sentence
learn to read
bro got ai induced rage lol
Is there a way to check specifically what eats up the token usage? Like a specific prompt, or the compounding of AI needing to re read everything said earlier to maintain context etc ?
a new codex update just came out, but is it even trust worthy
still not for mac though
link?
ai induced rage and discord induced stress you gotta chill bro
gndk - we're (the rest of us) are adults that can discuss business issues without drama. cool down bud...
all i'm saying is that the TOS are typical broad legalese
vs a tweet by altman or whatever
boutta go shower and eat se yall in 3 hours when my limit is up
OK, I'm trying to reconcile what you're saying with the text.
BTW, there's an EU-specific version of that: https://openai.com/policies/eu-terms-of-use/
this is about extracting data....
it refers specifically to trying to mine the model for it s training
not about you using codex programatically
it literally says output bro
No... Only see the "compact" context when go near 100%, so 256k context can costly more than you think. If you found suddenly context go 100% in short time... You should re-thinking why consume that much.
output is the llm response
you extract it programmatically by using a third party tool (pi/opencode)
its crystal clear
ok, i don t care enough to debate
anyway back to work š
same, have fun
so this update did nothing š im about to uninstall it and reinstall
Is there a way to fully disable its compacting? Or is this not possible. This would mean that context will probably have to go too. Is that possible also ?
Itās crazy codex solved stress induced baldness with the help of Ai , now weāre getting discord induced baldness
Thermodynamics is real
šš¤£.
what s the issue?
We didnāt solve a problem we merely shifted it
earlier today when i updated the one they release yeday i got that bug with the broken db on windows. had to ask codex to fix it by giving it the link of the issue from cli
after the last update my codex started being much slower and kept freezing after clicking out then back into it
You could try reinstalling the whole app but idk if that removes your thread and tokens
Is there a way to store tokens ? And restore a thread
No, compact is part of codex and it's require it.
Every compact require to re-send some detail data from the previously again and might take around 30k-40k.
@bright swift
From the TOS:
What you cannot do. You may not use our Services for any illegal, harmful, or abusive activity. For example, you may not:
... Automatically or programmatically extract data or Output (defined below). ...
... Interfere with or disrupt our Services, including circumvent any rate limits or restrictions ...
Your content. You may provide input to the Services (āInputā), and receive output from the Services based on the Input (āOutputā). Input and Output are collectively āContent.ā
Summary:
- "Output" is defined as responses that come from OpenAI tools.
- "Automatically or programmatically extract data or Output" means we are not allowed to scrape results from products like ChatGPT, or code into their pages, like for injecting prompts and extracting responses for the purpose of reselling services.
- There is ambiguity in the exact terms, but in-context they're talking about "You may not use our Services for any illegal, harmful, or abusive activity". This is not unusual in legal documents or common human conversations.
- These terms do not apply to legal activity. That one line should not have to read, again, "for unlawful activity".
- CONTEXT MATTERS!
Im not quite sure how it fully works though. Lets say I have done 10 total compactions. Would the first information before its 1st compaction have been compacted 9 extra times since our total is 10? That would essentially just make it lost information after that many compacts. Hopefully makes sense.
@lean lark
"store tokens"? no.
"restore a thread"?
I don't understand the questions.
And make a website to share threads and tokens
We can "resume" a thread programmatically.
And we can just go back to a thread easily from Task History.
Iām creating a new function for codex
You can "save" tokens, as in "not spend them". But a token is not something that is stored.
A thread forum
Where you can take someoneās thread
And input it into your codex
And then resume that exact thread
That's not how it works.
Every transaction in a session is the culmination of all of the applicable context. It's not just a copy/paste of transactions in-sequence.
If nothing else is a factor: After a compaction (maybe at other times) there is an encrypted block inserted into the session which contains context data. 1) you can't decrypt that, and 2) I think the key is specific to the account/environment that created it. So you can't actually rebuild a session like that.
Take a step back... What is the goal???? What is to be accomplished by such an effort?
ChatGPT says it's interesting and a "real idea" ... but right above that it also said it's not possible.
nope and not sure if the idea makes sense to download random session that could potentially do malicous stuff, eat all usage or get account banned for violations
what exactly does '/goal enabled by default' mean?
Like every message I send will be treated as if I used /goal?
Well, you haven't provided context. Now you're saying that it's given some credibility to whatever you told it, but you haven't shared what you told it.
OK, have fun, but you still haven't provided a reason for what you're doing.
If you want shared discussion with ChatGPT, we already have it in ChatGPT. And we can share/fork a ChatGPT session.
"/goal" previously had to be enabled explicitly in config.toml.
Trying to connect my ChatGPT app to the codex desktop app - missing the option on desktop for ācodex mobileā. Made sure Iām up to date also
Windows?
Yea
š
So I basically canāt use my iOS codex then as I just get this screen
When I hit chat itās the same as hitting connect
I mean the desktop app is on windows
Itās just the mobile connection isnāt there
Windows is Duce
Yeah, mobile to Windows is not yet supported.
Moving off is time consuming that I donāt have time for rn
The remaining question:
is there a way for me to use the iOS app now or am I screwed here
It's actually a 1 time thing though. It consumes exactly one unit of time to move off, and that's why it's 1000% always worth doing
My irl schedule doesnāt allow me the time rn point blank
iPhone to MacOS, yes.
Android to MacOS, yes.
swipe it off in the card view and reopen, if you meant the cant cancel issue
iPhone alone. Before this update I could use the app alone
I have restarted my whole phone Iāll screen record
Please be more specific Zenologia - iPhone to what?
then just reinstall as all data is server side anyways
Before this update, I could use it on iOS alone and it would sync with what was on the codex website I guess using their servers instead of my pc
OK, so you're saying you connected your iPhone to Codex App on Windows.
That was possible last week and only with by using a "remote-connection" option, which was unsupported and that hole was closed. So now we need to wait for any moble-to-windows connectivity.
FFS dude, you didn't say!
Drop it. Iām done
i think he means he got stuck on the codex onboarding phase that some others has as an issue too
but he said it was working and didn't clarify what it was working to. It actually wasn't working except as described. So if he had it working with "something" then it's possible that it might still work for him. But if he doesn't clarify the tooling then we can't help.
Nah, Zenologia is saying they used to be able to use Cloud/Web codex, but now it wants them to use local development only
Jesus Christ someone finally got it
Thought I was talking to a brick wall
Maybe need to use web browser as stopgap
But the screenshot is trying to connect to desktop. It says so. Not Web. š¢
Because I CANT GET TO THE WEB PORTION ANYMORE
OK, sorry, I'll back up...
I donāt want long explanations. Is there a fix or no? Thatās all I want
Nothing more
Not a stop gap. A fix
and known issue is that if you abort the onboarding you get soft locked... but ill just mute since it was vague
You didn't say web, you did say "yah" to Windows, the screen says desktop, I was going on the information available. If iPhone app connects to cloud, I didn't know that. I use Android and this was only published a few days ago.
https://chatgpt.com/codex/cloud
in ios web browser, codex cloud available. In native app, i guess they removed it
And I go to my browser for Codex Cloud all the time too. That's not related to the device OS, it's a web page.
Did they just limit goals to 4k characters?
( Another announcement of 26.519 https://developers.openai.com/codex/changelog)
The video shows my email nvm. But yea I canāt get to the web portion on the app anymore
Iām just stuck on the connect to desktop BS
Kirkās dik dik š„¹
It seems many of the GUI textboxes that OpenAI offers in tooling are weirdly limited far below the actual limits of the fields. I dunno if 4k is a real limit like from CLI, but obviously it is in the GUI.
Does it work with Linux? I can dual boot for the time being I guess.
Annoying as hell tho
Zen - just use your browser. š«£
On my phone is problematic.
TLDR my eyesight is trash and I have to use the iOS accessibility settings as it is but it doesnāt carry over to safari
Ahh, got it.
Like without my glasses or contacts Iām over the ālegally blindā threshold
My eyes go blury from too much phone time too.
(I need a doctors note if I have to do a vision exam for my drivers license renewal if that gives you any idea)
Hey if anyone else here feels the pain of maintaining different AGENTS.md files across different OS, Local/Remote, sandboxes, systems, lemme know ... I have a utility for building files from component documents. It's a poor man's contract management system in a thousand verbose lines of code. š
I'm also alpha-testing a 100 line AGENTS.md directive block to make sure the assistant knows where it can and cannot make file changes. It's a soft sandbox intended to supplement OS controls.
This might be a silly question, but is there any way to set those limits? This basically just took away using goals from me.
I gotchoo bro
Settings > Apps > Safari > Page Zoom && refresh page
Use in-app reporting and create a GitHub issue. Make your voice heard.
Man they really flopped with all the vague posting
Will try, thanks
Who is they and where are they posting?
Codex update today?
2 hours ago, yeah. Apparently they fixed the problem where the agent stays focused and gets the job done or something
adding the translation from Chinese to English
Codex has had a really good run for a while now, wonder if tech debt is starting to take effect
That Moment When : I express paranoia about Codex self-authorizing insecure file system updates, and ChatGPT says "Your paranoia is justified - the weak point is self-authorization." 
Sometimes I find myself thinking about that, which is why I spend more time working on my app's security than it does.
This happens any time you try to make a super app. Instead of being just for developers, Codex is gonna be for everyone, so they focus on all the wrong features.
They're also riding the endorphin wave of pushing out updates every couple of days ... quality is suffering.
Hello, im new in this of codex, i want to build a good and big app/appweb, what model you recomend me? i have GPT Plus
did you solve your problem with caps and breaking lines?
OpenClaude Mimo v2.5
Yes, I made a system to detect speech bubbles.
I prefer to use the Codex App, but like what model, i have this options with different intelligences
Find another hobby if all you got is $20 to spend on models.
itll be a struggle to create a good big app with only GPT Plus
I have more money, but i like to speend, what i want, but im not a full coder, 24/7, i work when i have free time.
Good luck...
Thanks
youll run into usage limit real quick
I come from Claude, i think is better gpt Āæno?
Almost in limits
openais $20 sub is much better than claudes yes but still it wont be enough unless you are one very patient person
Im patient, really patient. i have more work to do so i have so much time to wait
i mean you can get started and build the barebones of the app but if you want to actually make it good and polish/refactor/optimize and finalize the app, itll be a struggle with GPT Plus
Someone need to make a pov of none developer use coding agents. Cuz im curious how they used it without all the prior knowledge of good software practices, deployment etc
I know, but if i need more, probably i use credits or put in a Pro plan for some months.
donāt use credit, itās the fastest way to burn through your funds
But thanks for the recomedations
I like your dog picture anyways
thank you š
You could probably try Plus. If you use 5.4-mini and stay away from xhigh you might get good consistant usage. If it turns out you need the big brain model or wanna do multiple agents then the natural choice is Pro (higher 5hr limit, often overlooked, more important than credits)
Im not planing to do it in a month, i have like a year for do it, so idk if Plus is okay if im pacient
I'm a seasoned professional developer and I'm satisfied with my ChatGPT Plus sub and the token quota for all development. I use 5.5 Medium with Standard speed for actual work, 5.5 Low/Plan for discussing challenges. I never use Fast and rarely High. I never use XHigh - if my code requires a PhD to understand it then I've done something wrong. I've only run out of 5 hour tokens once and that was because I was testing the limits. Use solid prompting skills and AGENTS.md directives. You should be fine with Plus.
Thanks Captain. i use GPT Plus time ago but never test Codex
I have seen a number of folks here talking about dev'ing games with codex
Is anyone of you open to a quick brainstorm (pm, I think it would be offtopic for channel)
Basically, I created a thing, and its a monster, and it proobably is unrealistic, I hit some hard issues that not even codex can reliably solve
As a hint, it revolves around "real 1:1 world map vs hexagon based map"
I understand the challenge but can't actually help with that, I'm new with that too. I'd like to sit in on a thread if you get one going.
5.5 Low/Plan
MY MAN
How can I check the number of consumed tokens in the app? Doing that in the CLI is trivial, but I can't seem to find a way to do it in the app
Current status https://photos.yoonect.com/share/bTmywx2Sr9aymyc42tnGcdly23B7cvhGuYlMkTbx0JJTJZP0CSW_H8NDoMAxczEMFv8
And, while it looks like my idea, it is just impossible to get performant path previews, reliable unit spawning onto walkable paths, and my ships learned to walk over tiny peninsulas š¢
GPT keeps going places trying things that do not work, and the only other option I see is "civ 6 style map" (hexagons), which basically blows one of my cornerstone ideas: real world map.
So, thats all I will say here, the rest is too offtopic.
I'm curious how you do that trivially with the CLI. I wrote this to make sense of the data:
https://github.com/CaptainStarbuck/codex-usage
Only for tell, but in spain the names of the Intelligence are diferent, we have this, idk if is same in other lenguages
The 4000 character limit on /goal prompts is a massive L
Yeah, same settings.
For Web's what model and intelligence do you recomend me?
thread?
Same as #codex-discussions message
5.5 medium, ok thanks ::)
what are goals
How's my favorite group of Codex users today?
dont ask
Bro it's definitely possible to do all of those things. Tell it to bisect the 2D space of non-water bodies to create a hit-testable bounds, draw the ship path naively, sample the path every n unit length, and if any of the points are within the bounds of a non-water bounding circle, find the angle from the center of the body to the failed sample point and bezier curve it in the opposite direction.
baldness is at an all time high today
As I said: trivial
code induced baldness and discord induced stress
Hey all, it seems there is a significant increase in Codex issues, but so far it's all anecdotal and whining. Nothing here is actionable or effectual. I suggest we use the normally useless #1070006915414900886 channel to at least centralize reports and move people to contribute details onto single issues. Otherwise when someone says "OK, what's wrong? Everyone else seems to like it." We can point to something substantive, not just random threads of angst.
Claude?
Impressive. I feel like I use it a decent bit and never come close
What is exacly? Im bad remebering UI's
Is that the plus or pro count?
Apple , Codex , Xcode , Iphone , Mac is the only way to live
I have so much catching up to do
Thanks! Claude looks the same today too
also
right, but current app uses OSM data from remote server for landpaths
That alone is a killer, it takes some seconds sometimes to draw a land path
And selhosting the OSM is like downloading BitCoin core in full and processing it lol (possible, but retardedly crazy, and almost non-sellable to other users even for free lol)
The water movement aspect is just one of the two major issues I was (or am) hitting
I have to add that I am not only lazy but also no "editor", it took me 2 weeks to create the map overlays and masks in an app called ... forgot the name:
Remembered: QGIS
Might be that I am on Pro that I never run out of codex use
Sonnet Only? What is that? Time ago i have a pro suscription and dont appear something like that
costs too much
we need a mid tier plan for 40$
I understand that lol
I'm tryna use some credits. Might actually make it to 0 this week if they don't reset!
Oh, I see, you specifically want counts in the app. Can't help.
BTW, for those of us who work in different environments, the stats we get in one environment will be different than another, and we only get aggregates from the OpenAI server, including their website.
So for example, my code-usage app reports quotas which are aggregate from the server, but specific transactional usage only comes from the current system, so it's not the full picture.
I dunno if anyone cares but it's a factor that I might address later.
If you're getting your own stats from somewhere, you might want to consider this too.
how do credits work what is 20$ worth of credit get you ?
is it a month worth of usage ?
If that's a web app, can spread the load across web workers and use WASM for parallelize it š
and do they rollover ?
I keep sayin bro - burn extra tokens on someone's FOSS project. Do some good for humanity.
hell nah save them credits this aint a charity
I AM š 2 projects right now. MLX and Rinf
lets hate on that. I got 5 % left, and 5 days till reset, and 10 % on my claude 5 days till reset, using composer and kimi to stay alive. I was betting on codex reset today or tmr. hope it comes lol
we cant have it too easy we need struggle
A lot of FOSS dont want AI submissions
how do credits work ?
do they rollover
is 20$ a month worth of usage like the basic plan ?
I should know, but I don't. Sorry š
my basic plan is 20$ a month
Credits last 12 months after purchase
ahh
thats dope
but
is 20$ of credits worth a month of activity like my basic plan
Yeah, that's why I'm not trying to sit on 4k credits right now
GO ROB GO. Send out all of the agents
I do wonder if some people would prefer codex credits over API credits for discord dev challenges
I forgot the link to that page
It's always so annoying to find
Can you send it?
Bruh
Markdown and links often get purged on post lol
https://chatgpt.com/codex/cloud/settings/analytics
Lol when I pasted, I had text selected, it turned into markdown and got automodded
that would honestly be a better idea for most people. I don't know a ton of people that use a sub AND API
Thanks
Will see if there is a codex credit grant api system...
Same
I'm getting close to 20%
You're about to have people lock in for the challenges if so
Fine with me lol
Does anyone know how to do this? #codex-discussions message
( Nice to see ya stopping by @robust lintel )
someone mak this thier pfp
Bro the peaches were gas
It was your imagination. There is no Bau here.
aye yo pause
These aren't the droids you're looking for.
oh shoot bro is the top dawg
what are goals
i need to know
im locked outa codex
going through withdrawals
woah
thats cool
I gotchoo bro
thats stright wild
Protein trend out of control
shoulda been meat cup
thats all wrong
zorgmeat industries
lmfaoaoaoaoa
make that your pfp
anyone compared 5.5 to composer 2.5? thouughts?
it would be dope if you could sacrafice your chat gpt limits and send them to codex
like a codex only plan
whats composser 2.5?
done
legend
they aren't shared under general usage atm?
i dont think so
You don't want to go down that route, because then your codex limits would also apply to GPT, which is what Claude does
aye yeo
Is GPT 5.5 being extremely annoying for anyone else? (just stops midway, uses hotfixes instead of actual fixes, doesn't check it's work, etc?)
nope
what thinking level are you using?
Medium
How to see the context after today's Codex App update?
Did they entiurely remove the ability to see it?
generally i use xhigh, but not had any issues with medium either tbh the couple of times i used it
your website is a bit confusing @ember spire
The cincle for the context usage is gone
what's confusing about it?
its like i wanna read "Own your Voice " but you have it tabed below
instead i just read Own Your
own your what ?
i think voice ai infra makes sense on its own line
am i buggin people ?
Yeah, I had 1 agent increasing stack size from 8MB to 8192MB to troubleshoot a stack overflow instead of fixing the infinite recursion lol it was crazy
not me, thx for the feedback
there'll be another redesign when i'm next accepting clients anyway
Cursors beast. Doing pretty well on benchmarks
lol, no š
idk what that is
yea its too easy to misunderstand imo
i build custom voice ai infrastructure for people who've outgrown managed platforms š
shouldnt it be
ok, i'll keep that in mind when i redo the messaging, thanks
i build custom Ai voice infrustructures ?
Tokens and Quotas
or it woueld be "I Build Custom Voiced Ai Infrustructures "
cant be voice itd have to be Voiced
well the offer is ownership of your voice ai infrastructure
lolol
it's a pretty specific niche
yeah you may not be my top of funnel target audience š
people who come to me generally have some experience with voice ai and want to take it to the next level
mmm
the technical content of the site presmes some familiarity
can i redesign the wbesite mock up ?
and if you like it you can use it
you make complete programs or just the voice ?
i'm not looking for a new website at this time but if you want to whip something up feel free to msg me and i'll check it out
latest project is a full saas platform for law firms with integrated voice agents
so yeah i also build full apps (this one is a two-app monorepo)
i bet i could be of great help with the ui
btw the site is more than a landing page, there's a backend for observability on the voice demo
feel free to reach out on linkedin š
i saw it looks very nice
how risky is it to give codex full access when building an app?
it will report you to the FBI
thats okay Im Canadian
Canadian fbi then
RCMP they dont scare me just a bunch of dudes on horses
Dont forget the hats
In the beginning, probably fine. When the app starts to get big, or when it starts getting confused, it can take drastic measures that might result in data loss. You wanna work towards making sandbox work well
Does the entire machine count as being sandboxed enough?
Did they hide the current context on today's Codex App update?
I mean yeah if it's in a VM or a dedicated server then that's different for sure. Maybe just keep em away from prod
Yeahhhh. I gave codex all of it, including the NAS
I gave exactly 1 agent prod access and I watch its every move when it's running. The rest are heavily sandboxed
Sandboxing is a skill. I'm finalizing a set of directives to ensure the assistant stays in its lane and doesn't mess up stuff even when it's in an authorized scope.
There's a good set of sandbox guidelines for Windows on the OpenAI site.
https://developers.openai.com/codex/concepts/sandboxing
https://openai.com/index/building-codex-windows-sandbox/
https://openai.com/index/running-codex-safely/
https://developers.openai.com/codex/agent-approvals-security
Codex can open in a pre-defined sandbox mode. I haven't tried that.
So i basically use codex and claude at this point to just run dedicated agents to do the work. Some local, some claude, and some codex depending on the jobs. I just treat my main machine as dev, and they dont get to touch prod.
Also, windows sandboxing is a joke in general
Codex said
why is suddenly everyone a yoghurt in here?!
I have it set up so all my big project agents work with workspace-write + networking, I maintain a server that runs privileged commands, so as they experience sandbox issues I add privileged commands to the escape hatch. They have full access to local docker for all their destructive testing. The production agent deploys to staging first, and once it's stable it points the domain to the staging server (never actually touches production, only performs DNS swap)
He's a meat yoghurt though
this is beautiful.
I actually just got codex and claude set up on a server with 3x 6000 pro q, and a 5000 and 4000. Local models for days too
Tracking Codex Issues
Niiiice! I have only messed around with local inference. Was considering getting a Cerberas API account and wiring gpt-oss-120b into codex over Cerberas. It's surprisingly cheap for the speeds they offer
found that model too dumb to even use in a voice ai pipeline but ymmv
120b is cheap, but if you have the option Kimi K 2.6 and GLM 5.1 are fantastic models to throw behind either the claude or codex harness
Cerberas has a pretty limited selection, and gpt-oss-120b appears to be the best one they have that shares a vocab with GPTs
I think it's like 3000 tok/s lol it's insane.
anyone tried out grok and compared to 5.3 / 5.4 ?
that's actually nuts for the speed, and if it's the full model and not quant at all, then it will work a lot better. I'm used to a quant version of 120b
Yeah I think it's MXFP4. Since OpenAI only released a quant and not (BF|FP)16 there's no way to unquant and change the bit level
I still think it's funny that 120b still calls itself 4o specifically
Sorry Robert
GPT-5.3-Codex-Spark is powered by Cerebras (that's why it's like 1000 tok/s) so slightly relevant š¤”
SMH - I'm just being a grump. Ignore me. š“
No. You're here. Cheer up captain grumpybuck
Codex the CLI/Harness has a lotta good bones. Nice, strong foundation. I wanna plug other models into it haha can't help it
No codex app for linux? Thats really dissapointing
I just finished shoving Claude into the Codex app itself, only for them to remove claude -p and the agent SDK from plan usage next month
I just want to use Codex app with AWS Billing.
download the CLI, and have it rebuild the macOS version for linux for you
I tried that and it wont build the linux desktop one
I should just port it to linux and throw it on github like I did with the windows one before they released it
I'm pretty sure you can if you change the default model provider in config.toml. The desktop app uses the defaults, and you can replace the whole default provider with AWS
Error: Required Linux Chrome plugin auto-install patch failed: could not enable bundled Chrome auto-install at Object.applyLinuxChromePluginAutoInstallPatch [as apply] (/home/laz/codex-desktop-linux/scripts/patches/chrome-plugin.js:62:11)
And it says useless stuff like check github issues, check this check that
Throw up the recipe š that way they can't getcha for copyright or whatever
I'm unusually distracted over the last couple days.
I put my main app on hold to create an app template generator from it to support other upcoming apps that are built on the same scaffold.
Working on that I realized my Agents directives were behaving differently in different environments, so I need to standardize them.
That led to writing this utility that builds AGENTS.md files so that I don't need to do it all manually.
That led to wanting to beef up the sandbox safety, because inherent in everything I've seen and done so far, "if the assistant determines that it needs to do something, then it becomes authorized to do so, whatever that happens to be, including wiping out something important".
That led further down this rabbit hole of getting the wording exactly right, that "The assistant's determination that an operation is useful, necessary, relevant, implied, cleanup-related, or within task scope does not authorize a destructive operation." ... along with a new section that defines what "authorized" means, and what a "destructive operation" is.
This stuff to "do everything for us and save us time" has become the only thing I've been working on for days.
I'll snap out of it. I'll get back to myself, get out of chat, push another product into production, then move on to the next. šŖ
looks like a perms issue?
Haha yes, this is the rabbit hole. I feel it man
They should have a codex rabbit hole support group channel on here š
I'm not sure
I dont see why this isn't just a flatpak
And it says useless stuff like check github issues, check this check that
https://github.com/openai/codex/issues?q=is%3Aissue state%3Aopen chrome-plugin.js
I built an entire scaled agent platform on -p, still very annoyed
Claude found it: Root cause: Upstream Codex changed their code structure and the regex no longer matches. The codex-desktop-linux patch is outdated.
I'm not looking at those issues, I'm just saying the error handler did kindly direct you to maybe find a solution. That's the world we live in...
I swapped most of mine with codex --exec
I want to try and build something on codex's app-server API
Which I dont need the desktop app for, but, still want to check out the whole desktop and mobile stuff
app-server is different from mobile, which I believe relies on "remote".
Am I wrong anyone?
There's so much fun stuff to do with Codex, and the app is so much better right now than it was at release.
Like building more agents with sub agents.
boyle is wild
Plus is fine for hobby level stuff
Its a shame mobile doesnt work with the headless remote-control
hater
Yeah, it wasn't supported but we got into it anyway, then they closed it on the way to supporting it. That happens.
I patched out the throw, lets see if I can install the linux app now
"Linux App"?
kk
Codex* because they dont wanna support linux for some reason
Yo im getting budget limited in codex app for my goals
or Windows
any ideas?
Mac being the "best" target currently is a bit wild
widows is horrible though
and wit mac its only one porcessor and os to target
windows has no vertical integration
all macs use the same processors now
windows spreads thierself too thin imo
same with android
I think Linux, aside from severe software ecosystem fragmentation, works just as well as mac for development
nice, they fixed the crash bug when right-clicking the top window edge on macos
As developers we know the challenges. As consumers we want our vendors to address the challenges - or stop advertising tooling for platforms they aren't actually supporting yet.
I use this bad boy with my mac
that was impressively quick
I see a Starbuck cameo
Busted!
nice jacket
(Shameless) Everyone is welcome to visit my recent threads:
https://discord.com/channels/974519864045756446/1507113548931666020
https://discord.com/channels/974519864045756446/1507120682247389326
i want to get realistic guitar sounds in my app
time to go into logic and take the cleanes guitar they have and play each note 1 by one
then edxport them all as audio
luckly a chord is only the same notes played at the same time
Get ChatGPT to find links and Codex to incorporate them.
Be Human, Be Lazy.
so it should be pretty easy
Find MIDI samples for some keyboard?
you can find files, no?
Look up free instrument packs for DAWs, you'll have more luck
ive been making beats for 12 years i dont mind it lol
i need a control group i can have digitally
I just googled it and there are a ton of hits. I've got guitar sounds in keyboards that are over 30 years old.
What about connecting codex directly with the VSTs
send a link ill check it out
that would be wild
idk if it can even do that
Im working on an mcp server for FL studio, ill see what I can dig up
If you can code it or do it in a browser, then Codex can do it too. Sad but true.
whats an mpc cerver
MCP is about to change your life
( Oh boy, he's in for a whole new world... )
So MCP is the model context protocol. It's basically how you can connect things that normally dont connect to AI, and connect it. https://developers.openai.com/codex/mcp
Google "ai mcp"
brother brother
make one for logic
wait i can make one
lol
fl studio seems smart because its mostly a cell / clicked based daw
I use them for connecting claude and codex to comfyui, blender, fl studio, godot, unreal, and basically anything else I'd need for a full pipeline
It was. Ableton might work better. Fl studio doesn't expose a whole lot through the available APi
I dont mess with logic simply because im.....a windows user.
Booo
horrible
I miss my hammer. Stupid server boost icon
can you show us an example of your mpc workflow
Once I get to my computer, sure
like what can it do ?
I still cant get decent clips or screenshots with remote
cause windows is bunz
Oh, so it can do what you make it to do. The fl studio one is still being built, but I can show you an example of comfyui generating an image and turning it into a mesh then putting inside a game engine, but it will take a bit. Im doing it on a 4080 lol
Be very careful about MCP. It's widely accepted but can be highly insecure unless you control the entire pipeline. Don't trust remote service providers who offer MCP Servers or Tools. When you get code for a Server or Tool, read the code and understand it before using it - it's random code "off the internet" and that should tell you how insecure it can be.
Outside of that it's great. š
wow
this costs api coins ?
or is an MPC not an API
in the litteral sense
mCP
MCP PCM
So MCP is local if you build it local. You can also build one for web. A lot of online providers will have an MCP these days for AI to connect with. The only thing it uses is my plan limits because the work itself is done on the Codex app or the VS Code extension, it loads up it's own built in MCP, connects to the other ones I start up, and then can route between them. I do use my own MCP routing though. That's not built in
wow
So disappointing I cant connect my phone to the PC, its just stuck at "waiting for desktop" and nothing happens there...
Yes, I said they closed the open connection and it doesn't work anymore.
So now ... we wait ...
They closed it? Meaning?
question why will my computer heat up when using codex ?
if its run on the internet
what processing is being done on my computer itself ?
They announced Codex Mobile on Thursday.
The website says "Windows coming soon" ... as in "not supported yet, not ready yet, no workey".
Someone published in X : "hey, you can use a remote-connection".
We tried it. Yay, workey!
But that code is still in development. So a day or so later that specific remote-conection was closed.
They are working on the real connection, the official one. Until then, there is no official or unofficial solution for Mobile>Windows ... or Mobile>Linux I don't think.
Codex still need resources to actually work on your computer. The reasoning is online, the work is local.
i see
I tried to see if I can run the CLI remote-control, that doesn't even start
So codex can start up the MCPs directly itself. Then I basically outline the job to codex, and it will send the job to comfyui for me.
I kinda knew you were off doing that. What can I say, dude. No workey!
Whats the intended use for remote-control?
It's documented in developer.openai.com
I was actually today years old when I learned that the CODEX APP PULLS IN COMFYUI IMAGES INTO THE SESSION. Nice
Awwww no new model today? I was kinda hoping for a new cerebras one
why is it so hard for openai to look at the codex usage data and go "oh my god this is blank for everyone? This is a bug! Must fix!"
I don't even use cloud, why do I need to figure out how to use cloud just to lodge a bug report "Confirmation whether cloud or delegated tasks produce non-zero usage values, since the current report appears to mention only local tasks"
If you don't use it, it would be nice if you reported it. But if you don't use it and would struggle to show data, don't feel to report it. Let someone who cares do it.
That's "priority". Community/user priority is one thing. Developer/vendor priority is different.
This is what they want off me to even look at it:
Workspace / Org ID or Codex workspace identifier
HTTP status code and x-request-id header from the usage API call in the Network request
UTC timestamp of the usage API call
Top-level shape of the API response body, with no tokens or personal information included
HAR file or screen recording showing the Network request and response for the usage API
Confirmation whether cloud or delegated tasks produce non-zero usage values, since the current report appears to mention only local tasks
wouldn't even know where to get some of those. Workspace identifier?
Provide what you can. Just a sec.
like ... its a chart. It worked until 3 days ago. Now it doesn't work. This is way beyond what I'd normally get in a bug report as a dev already
Shall we go to a thread?
sure
Yeah, but at least they fixed Codex so when you use /goal the agent works endlessly now, but all tool calls require human approval even with full access š„ø
ahh i only need 1 sample per pitch
as the low E string 6th fret is the same pitch as the open A string
magnificent
Better be or it's out of tune! š
If you're talking EBGDAE tuning, that's the 5th fret (second dot on fretboard) and if you want it to sound really good you wanna sample the whole string, especially when it switches from brass wound to plain strings
I am fairly certain that OpenAI has switched on quantization on GPT-5.5 in the last few days
The model no longer has any context awareness. Every prompt I am sending is basically starting from scratch and needs to be mega multi-paragraph stating every single detail. It has lost all capacity to follow intent
So not only are you a SWE, but a musician too?
I definitely dabbled š
we really do need a chat for talking meta stuff about codex vs projects being made with codex. First it was andrew, now its pic13, and that's just in the last week
I'll bug bau to organize some codex spaces
Iāve used all my token, I call upon my lord sir Altman, pls refresh my token limit
Wrong god
Tibo pls refresh
If anyone itās lord Tibo of the tokens
who's gonna pull the meme trigger?
"Dear lord baby Tibo, we thank you for this bountiful reset"
different meme, but there you go
Tibo will refresh your token + nerfing the model to oblivion
GPT image gen 2
I could make a slop graphic tees business out of this
Anyone can now. It's a completely open industry. Print to an iron-on transfer, iron on to a cheap t-shirt. Profit!
You can even take it a step further and upload the design to your storefront, and when an order is made, it's directly sent to the print on demand services for the product that gets shipped directly from print to customer. Why bother with inventory
ill do that soon but for now im just using the same pitch class method
I just went AFK, thought of that, and came back to my desk to type it. Rui beat me to it. (If you've seen the movie, you know the quote.)
5th Fret A is the same Tone as Open A
same tone and octive
@boreal holly
yea ive been studying theroy for the past 2 years so i thought id make myself a cool app
I dunno dude, there's something about a little bit of a fret buzz.
OpenAI should disable text input in this channel so we can only chat with images š
holy embed fail
you play ?
anyone els lmfaoooo
i have the m5 pro and from time to time it starts doing fire bending lmfao
guitar, bass, drums, piano, violin, viola, cello, contrabass
I said earlier, I used to play out, decades ago....
you play them all ?
mmhm
showoff
maybe yall will want to demo my app at some point
I'm a jack of all trades and a master of fun, or none
i mean if you play guitar you can play bass
bassplayer here anywayz
if you play guitar you can play violin
now wood winds are a whole other monster
i never have but i bet i couuld play violin well its just a fretless guitar
with a long as pick lol
when 5.6
not diminishing your gretness though i can play it but i wont be good at it
This one does require a lot of practice still. I started with the Violin, but the finger placement is one of the hardest parts. Once you have that, go nuts
wrong channel for that question.
wrong question for any channel.
glad the way i learned guitar was through theroy that applys to all instruments
stacked 3rds are life
intervals 1 b2 2 b3 3 4 b5 5 b6 6 b7 7 1
You'll use your 5.5 and like it, or go without
1 3 5 7 9 11 13
modes are just the major scale played from degrees of the root note
things like that
The funny part is, I'm actually god awful with theory. I play by ear, and can barely read sheet music anymore
other apps exist they just arent as comprehensive as mine
then my app may help you : )
for a small fee of your soul
i like it but frontend is miserable
That's your problem right there. do the $100 option of both, and then you just win
claude is duce
codex users jam session when
im down
we can only play programed music though
i love ai but i dont like suno lol
i would use suo for drum sounds though
or single sounds i can then transpose
but not for full melodies
Dang, I only play guitar and sing. always wanted to do piano!
piano is easiest to learn on
Piano is actually the hardest for me. My left hand doesn't want to behave
Why companies arenāt paying for coding agent
Idk why but piano for me is a struggle. With guitar it's like operating a clutch and gas/brake pedal. With piano it's like operating two sets of both
this is ai though
lol.......
having made music for 12 years its so easy to spot AI audio
sup all. i searched but dont see anything. is computer use and mobile use for codex on windows still just 'in the works'? do we have any clue about those and newer features? im getting serious fomo
its an uncanny valley thing
im convinced rui is a bot now lol
Really? I couldnāt spot it. The tung tung tung sahur song to me sounds real
mobile yes. computer use, not sure. We can control Chrome on Windows but not sure about the rest, haven't tried. 
the genre I usually listen to is like dark house/techno/electronica and its already so heavily synthetic that AI songs sneak in there unnoticed. I tend to pick them because the lyrics are a bit weird or something is off and turns out its only the bad ones I'm picking - the good suno ones slip right on through my filter. So don't discount it
bro said my left hand doesn't behave lmfao send an ai song
nahhhh he took us for fools
Here. I threw this together in FLS. No AI. just the start of some baby making beats
lolololol
Well I'm probably gonna regret this lol but when I was in college I was in a class that required us to make some kind of media about our journey to entering college (e.g. a short story, poem, song) and had 1 week to do it. https://youtu.be/B1ygwSMsZ10
ok
great work brother your past self would look up to you you finally made an app
full points for rhyming stackoverflow
bro went from piano god to piano rob
"Until I wrote an app" Is about to become the new codex catchphrase
codex does the code too fast and then I block on graphics/content š
"I remember the day I stopped writing code..."
i can do your ui for you
literally a 1 chord example with a half step in there. I never said I was a Piano god though. Just regular God
true true
my latest project I'm doing with gemini + codex
anyone having issues with codex telling you add your phone number\
Rob. I'm not poking fun of you, but I love this, and I'll be memeing the hell out of it in clips soon
i can do an UI i work in illustrator and photoshop and blender i can make any ui elements
This has some Timberlake vibes
dont steal my song though lmfaooo
i made this 3 days ago in logic
ive been on an electronic kick
Yep, cheesy as heck, I knew it was gonna get memed š knock yourself out!
i usually make hip hop style stuff
I miss logic. I wish windows had a DAW that was at least comparable
none are
my workflow seems to be - get codex to make the basic layout in its horrible beige defaults, then spend some time staring at other people's websites, start trying to distill something out of that, throw half done stuff at gemini, get it to mash things together, go back to css, back to gemini, repeat for several hours, then tell codex not to touch my css
No sir. That was a bop. Memes are recinded. Only respect
those days are over im a ui god
context?
yeah working with chrome is great. the plugin for general computer use doesnt exist yet tho :/
curious why mobile is lagging behind as much as it is. youd think most of it is system agnostic
I'm used to being provided with the designs someone else paid multiple thousands of dollars and just turning that into a site
be a graphic designer? Last one I worked with was charging $1500 a day
mobile should connect to services with the same interface but the issue is server-side (PC). They haven't finished those components.
that's aud but still, that's a lot. AI is still really bad at original designs but you can just throw a design at codex and it just implements it
my budget is nil, I'm still trying to decide if I'm retired, unemployed or just poor. I can't afford to pay anyone else for tech
half empty or half full
i told codex to change the font of this one letter
im so happy it did it correctly lol
1 piece of code was all it needed to do
magic
thats my fake sine wave logo lol
i told it to use a 90 degree S as a sine wave
and a 90 degree Z as a saw wave
masterpiece
I have no idea how apps use fonts, on web you'd get google material or fontawesome or an icon font
explain
i just learned about project resources too
wow
you can store so much stuff in that section of your app so when you send it off it has the things it needs
magic
those are all things you can google, can't post links here. I got codex to add google material for all my icon buttons, I used to have an icon font but didn't have enough icons in it for the openai audio
there's websites where you can make your own from their sets. Google material is probaby the most standard one out there though
hello folks, when transitioning to codex from Copilot CLI, how do i re-use my many instruction files under ".github/instructions/*.md" .... any method to make those available to codex vs. just telling it to read and follow them?
I usually use inkscape, and turns out the various AI bots can generate SVGs. You could probably ask codex to make a svg to insert there
all good id rather make it myself
This is horrible. For the first time as I'm writing security-related instructions I thought "what if the assistant is malicious for some reason? That's a loophole for destructive behavior." I mean, I'm coding in the plot for a sci-fi movie here. š±
when I migrated from Claude Code I simply asked Codex to explain what options there are to convert the memory, subagents, and custom commands. Then chose a path and had it execute. Worked a treat and I was productive again in under an hour.
I did that just opening Codex in my main project and telling it where the important stuff lives.
where will AI be in 5 years i wonder
Codex 16.8
wow
it will be able to program your dna sequence
5 years ago, could you have imagined we'd be here?
hell no
me neither
I abandoned a 10+ year old project because making the content was just too hard and too timeconsuming. Now? I can get AI to do it
crazy times to be alive
cant wait for real robots
irobot type vibes
honestly the first time i watched chappie i was ready to trade my body for a robots body lol
does mean I have a very real world anchor on what I get for $20ish a month from openai that used to cost way way more before
what would you have your robot do?
help me with anything i nedded like cleaning , cooking , making a space ship
I'll say again, I really wish Isaac Asimov were alive to see just this start that we have now.
i love the direction humanity going , its going to cost us the planet but we can start elswhere fresh on a new planet
its messed up but no stoping the order of energy
I wonder what other technological breakthroughs we might have in the next 5 years. Quantum seems to be evolving rapidly.
crispr
genome editing
will be the future
then immortality
for a fee
the more money the longer you get to live
dream ads
type shi
biological immortality, like lobsters, or digital immortality? š
in 5 years?
oh hell nah
I've been stuck home sick for 4 weeks now, I'll take immunity
maybe like 20 years from now
Another decades old Math proof was solved the other day, that'll be a more frequent event.
Renaissance only affected Europe. This is more like the discovery of fire.
paying for art is in the best interest of industry not humanity imo
we operate society through industry but that wil change
robots should take all the jobs and we get to run around naked again foraging for berries
lmfaoooo
but this time well have social media š
how long before you can talk to your dog in English? š¤
as someone with a berry patch, those things have thorns. Just sayin
but i think thiers a conflict of intrest somone would convince all the ants to fight with them on earth
lmfaoooo
sharknado type vibes
brb gotta go grab my food
Iā¤ļøHumanity
That's going too far.
I am not a fan of humanity right now. There are too many of them, they're generally lazy, they make bad decisions for themselves. The world would be better with fewer of them with higher quality. Like code modules.
You spelled AI wrong
lol
Bro tryna repopulate the earth
depopulate
Uh dangerous topic
Yeah, going afk, gnite/gday all.
where did it go?!?!
i thought it was only me
maybe related: something else happened to me on chatgpt web - #ai-discussions message
finally got real guitar sounds in my app
honestly i really think i outdid myself here lol
I like the Affinity tools, especially Designer! The export persona makes it really easy to export designs in a bunch o different sizes
Adobe , Apple , Sony , Codex , Fender , Water > Everything Els
anyone els usage data get deleted ?
I noticed it was gone as well, i almost stopped looking at it at this point though. Compaction is seamless.
at least will still have /status
i use it all the time...
i never run a plan under 60% context remaining, always compacting first
compaction degrades over time
How so?
why?
I'm not superstitious, just a little stitious š