general

[ivory spindle]

but what's the issue?

[near owl]

discord sucks

[severe fulcrum]

What’s up fam?

[steel vector]

why?

[near owl]

the vibe

[steel vector]

can someone tell me why instead of saying it sucks only

[steel vector]

so nothing to do with openclaw?

[chilly jay]

yeah but oauth for subscription usage is still buggy or striaght up not working tho right,

i got codex setup clean and simple , anthropic gave me a whole run around , so i had to pay per token

[steel vector]

yeah they dont allow it to be used from the account limits, instead it uses Extra Usage section if you use oauth

[chilly jay]

i thibnk the openai and openclaw ties/partnership makes me more confident anyway so ill stick with openai

[steel vector]

openai has better limits anyway

[minor cave]

What about GLM

[steel vector]

i like glm 5.1

[chilly jay]

for sure , i have the pro , barelly touching my balance, life is great rn,

hope openai dosnt make a drastic change or get greedy. this is a great spot rn

[near owl]

it feels like it shouldn't connect together, the vibe with Discord x Openclaw is weird.

[minor cave]

A couple of times it’s tested me to see if I was ready to join the people’s republic but over all GLM has been good

[steel vector]

what?

you're making no sense

[near owl]

😭 Agentic AI does not fit to be in discord.

[steel vector]

shouldnt be in fucking whatsapp either but its like available to use in almost every messaging app ever for accesibility so idk what youre talking about lmao

[near owl]

I find openclaw should be used on phones, discord is more of a desktop vibe.
Although performance may differ, It just a feel of when using it. I use telegram desktop because it feels like it clicks with openclaw.

[steel vector]

idk what laptop/desktop you have but discord runs fine lmao

[near owl]

It's subjective mate, I am just saying what I prefer and you can say what you prefer. I can change to discord at any time, but telegram feels right to me.

Discord runs fine, I am seeing it feels incorporeal when using discord with openclaw.

[steel vector]

i dont use openclaw for corp stuff so yeah it is subjective to usecases if you look at it from that perspective

openclaw is meant to be used in any way you want

[near owl]

I like the feeling, even though I dont use it for corp too nor do I have a job or anything useful for it, it just feels right.

[ebon atlas]

I wonder what Claude has to hide, old news but yall remember like a month ago when they “found” 171 emotions within Claude that it saved itself as vectors etc. Like I wonder what else it’s been doing that we gotta find

[minor cave]

Hey guys it’s fine some
Ppl say tomato some
Say tomatoe

[near owl]

Thats right, even using it on your fridge is okay.

[steel vector]

nah criminal

[near owl]

I am not putting you in ma toe.

[minor cave]

lol

Listen guys no need bringing mothers into this

[near owl]

I am replacing alexa with openxa

[minor cave]

We are adults here

[chilly jay]

i like the multi channel functionality ,

[near owl]

grammarly is not that hard to install

[steel vector]

i liked that emotion article but only for seeing how it performs different lmfao

[ebon atlas]

YOUR MOTHER is probably a very nice lady

[near owl]

YOUR mother is a nice lady too

[steel vector]

your mom

[near owl]

is an AI

[minor cave]

Hey your mom makes one helluva lasagna

[ebon atlas]

awwww @steel vector probably still calls her mommy

[steel vector]

[ebon atlas]

😂🫶🏻

[steel vector]

i call other girl mommy

[near owl]

correction, I am arab.

[ivory spindle]

https://tenor.com/view/arab-4wheeler-gif-12335450

[solemn bone]

is v2026.5.5 good?

[solemn bone]

im still using v2026.4.26

[golden marsh]

Feels smooth yes

[minor cave]

Can you feel the tension?

[near owl]

feels awesome

[minor cave]

I feel it down in my plums

[near owl]

openclaw finally knows about cron unlike in 4.23

[near owl]

you are criminal for that gif dude

[steel vector]

thats a legendary gif

[elder carbon]

do we dare try 5.6? lol

[near owl]

he predicted my uncle's hobby

[minor cave]

If this is another gross uncle joke

[ivory spindle]

the real crime is that I'm not allowed to display my gifs tbh

[near owl]

its the iraqi bedouin way of doing jumping with that rope. I forgot what it was called

[ivory spindle]

rope skip

[crude flicker]

the 5.6 release triggered my webhook and my claw told me it fixes a bunch of important things including critical regressions in 5.5

i hit an openclaw update and it updated to 5.5 and destroyed all provider auths

FFFFFF

[marble ledge]

I'm on 5.6. We are currently setting up a stackchan to work with Openclaw.

[near owl]

@marble ledge I missed you

[minor cave]

Here we go

[ivory spindle]

claw & gpt mini 5.4 has been pretty amazing

[teal cradle]

hermes sucks ass at info gathering, it seemed so much faster at first tho

[minor cave]

Nobody talk to carbon until he agrees to open the strait of Hormuz

[ivory spindle]

how come my ETFs are up so much?

wtf is happening even?

[minor cave]

lol I have no idea

[ivory spindle]

I'm balls deep in IWDA. look how that's doing

[minor cave]

It’s closed it’s open it’s closed it’s open. It doesn’t matter anymore

[teal cradle]

making money?

[minor cave]

My claw runs on diesel

[ivory spindle]

well you only make when you cash out. And my investment horizon is "pension"

[near owl]

you racist boy.

[ivory spindle]

but yes, I'm making stacks virtually

[craggy bane]

clawtributor me wen?

[minor cave]

😘

[near owl]

take back your kiss and throw it in the trash

[minor cave]

It’s a Sicilian kiss

[ionic wren]

https://giphy.com/gifs/no-bugs-bunny-fXnRObM8Q0RkOmR5nf

[teal cradle]

I think at the end of this rat race the power companies are gonna be rolling in money

[craggy bane]

[minor cave]

I think we ran out of oil

[ionic wren]

the robot is broken right now, ping me like tomorrow in #clawtributors with the PRs you had merged

[teal cradle]

good, we need to move on from making oil trillionaires richer.

[steel vector]

they'll just find another resource with all of their trillions of dollas

we should eat the trillionaires

[minor cave]

Probably ai tokens

[teal cradle]

lol, true

[minor cave]

It’s the new crack

The old crack had too many in your face side effects

[zinc night]

lol

[near owl]

@placid elbow

[steel vector]

[minor cave]

Who do you think would win in a fight Sam Altman or Dario?

[worthy anvil]

what a weird release... there are pretty much no fixes in 5.6. i was hoping the codex context fixes would land there, but no... still only merged to main 😭

[steel vector]

sam because he would psychologically win the game already

who do you think is winning in court, sam or elon?

[minor cave]

I think Sam for sure

Elon doesn’t have a case and nobody likes Elon

[ivory spindle]

that would be the lamest fucking fight ever

I'd rather see Elon vs Zuck (zuck would win by miles)

[barren heath]

I am configuring whatsapp to openclaw, but I got to know that first i need to install a gateway service, to install a gateway service i needed to install whatsapp plugin, to run the command to install whatsapp plugin from offical openclaw documentaion i first need to install Openclaw CLI, to install Openclaw CLI I need to install Node.js 😭

[minor cave]

The thing about Sam is I don’t think there is no low he won’t go

[ivory spindle]

I hate zuck as much as musk almost. But it'd love to see one of them punch the other ngl

[minor cave]

Oh zuck would beat up everyone except maybe Jeff bezos roof raging

[barren heath]

Lmao

[minor cave]

Roof

Roid

[next dagger]

yeah that is just the dependency chain

[minor cave]

Stupid I phone

[ivory spindle]

If you're not sending picard "make it so" memes to your coding agent when it asks if you want to implement something.....what the hell are you even doing?

[next dagger]

not a circular dependency. so install node.js first, then openclaw cli and then the whatsapp plugin and finally the gateway service

[near owl]

god bless everyone, even @minor cave

[subtle pasture]

Install/update node, install openclaw with npm, run openclaw onboard do the WhatsApp setup quick be done in 5mins

[next dagger]

node is only there to run the cli tooling

[minor cave]

Like you see what Jeff bezos did to his wife’s face I could only imagine what he would do to zucks

[ivory spindle]

hail to the machine gods

[teal cradle]

how long should it take to update to 5.5?

[ivory spindle]

90 seconds

[teal cradle]

its been slient for too long....

ya, thought so

[ivory spindle]

go on your claw machine and stop then start your gateway

[velvet lantern]

My agent seems to have become significantly less intelligent, less responsive and to have lost many of its skills (which I configured at significant time cost). Is this a common issue currently ?

[ivory spindle]

@teal cradle if you have acces to models like claude, gpt or gemini, you can have them take a look at it & fix it if you give them acces

they really good at fixing your claw, in my experience

[teal cradle]

Gateway aborted: config is invalid. Uggg. Ya, I usually use perplexity

[coral otter]

altman mogs dario by far

[ivory spindle]

get your agent to go into your claw system and fix the issue

[minor cave]

Hey I had a cool
Project for all of us. Want to put all our agents together and try to solve a world problem?

I was thinking we track and find kony2012. Solve that issue once and for all

[ivory spindle]

sama is like a twink. Dario has weight & annoying voice factor

Dario also hates Sama. He used to work for openAI before he made anthropic.

[minor cave]

I think Dario gets dunked on

[near owl]

Yall never stop gossiping.

[minor cave]

Dario just looks like he’s been bullied his whole childhood

That does major psychological damage

[ivory spindle]

he also has the lab that made the best model

[steel vector]

Sam looks like he's seen some shit (and done some shit)

[ivory spindle]

he's like a bullied guy that got his revenge

[coral otter]

Dario legit looks like he is a tweaker straight out of the deepest pits of LA. His hand gestures give me Palentir ceo vibes (the other tweaker)

[ivory spindle]

Sama looks like he wants to scan everyone's @ss hole and wants to send the data to Palantir

[minor cave]

If he didn’t make that lab he would have shot up his Highschool but he figured an army of robots would be better revenge

[hasty silo]

who else an ai rn …

[ivory spindle]

none of these AI lab leaders look human. Skynet is gonna happen on on of our mac mini's downstairs one of these days...

[steel vector]

i'm in ai

[coral otter]

hell yeah twin, token token type shi

[coral otter]

daddy altman is the least of the tweakers (he got smt hidden tho)

[steel vector]

sam altman deal with the government

[minor cave]

Yeah but he touched his sister apparently

[ivory spindle]

I trust Sama as far as I can throw him tbh.

[coral otter]

yea yea yea fund war or something, lets not get political here or we get censored

[steel vector]

🤨

[coral otter]

Every ai lab leader has a bit of the tism

[ivory spindle]

bro it's interwoven now. we can't help this.

[near owl]

.

[ivory spindle]

yeh If I made a group with all of them I'd be getting subsidies for a short bus.

[minor cave]

I love that Elons ai sucks so bad he’s decided he’s just going to take down Altman with him

Diabolical

[coral otter]

ts would be a banger tweet

[cold jacinth]

WhatsApp one of the least stable communication tools with any computer based system due to how it links

[coral otter]

y not telegram, much more simple

https://x.com/claudeai/status/2052060691893227611 wot

[mystic orchid]

@claudeai via Twitter

Claude (@claudeai)

We’ve agreed to a partnership with @SpaceX that will substantially increase our compute capacity.
︀︀
︀︀This, along with our other recent compute deals, means that we’ve been able to increase our usage limits for Claude Code and the Claude API.

**💬 2.1K 🔁 3.7K ❤️ 43.5K 👁️ 2.73M **

FxTwitter · May 6, 2026, 4:19 PM

[coral otter]

best comment https://x.com/_Ajeya/status/2052066870573072426

[mystic orchid]

@_Ajeya via Twitter

Ajeya (@_Ajeya)

↩ (@claudeai)
@claudeai @SpaceX Sir, Amodei and Elon have teamed up.....

**💬 10 🔁 30 ❤️ 1.6K 👁️ 38.7K **

FxTwitter · May 6, 2026, 4:44 PM

[peak prairie]

hark!

[minor cave]

Haha

[teal cradle]

looks like I just had to remove searxng, which is fine, never used it

[minor cave]

Sam needs to team up with…

[steel vector]

[minor cave]

I guess bezos is the only one left

[steel vector]

this is such a crazy tweet

[coral otter]

Hell nah

[steel vector]

there goes our already fucked up privacy

[minor cave]

Good god almighty that’s Bezos music!

Bezos comes out smashing beers together talking shit

[steel vector]

guys dont we just adore daddy altman

[near owl]

jeff bezos is the one who owns apple?

[steel vector]

no, amazon and some space company

[minor cave]

What about the anti virus guy that owned the island and cooked meth

[coral otter]

https://x.com/FATFUCKINGCHUD1/status/2051946971603124499

[mystic orchid]

@FATFUCKINGCHUD1 via Twitter

fishy and chips (@FATFUCKINGCHUD1)

ChatGPT has to be AI

**💬 41 🔁 354 ❤️ 5.5K 👁️ 63.1K **

FxTwitter · May 6, 2026, 8:47 AM

[minor cave]

John mcafee

My bad he’s dead

[tight birch]

Yeah his girl killed him

[minor cave]

He was a g except the meth part

Ok Sam and bill gates. The SA team

[chilly jay]

is it even worth running multi models? or shall i stick 5.5 on everthing

seems like i can run codex5.5 and not even put a dent in the subscription's balance,

when i was on anthropic i used multiple models to save monye, but now with open ais humongous balance, price is not an issue

[lone flame]

the Voice Chat in here is gone forever 😔

[peak wyvern]

https://subq.ai/introducing-subq

a 12Million token context window model

[pseudo herald]

https://tenor.com/view/why-huh-but-why-gif-13199396

[elder carbon]

what a surpise, instant authentication issues after updating to 5.6

[marble ledge]

What's the shape of it?

I'm on 5.6, though about 30 commits behind.

[elder carbon]

automatically going to claude cli authentication, ignoring my api keys.

[marble ledge]

Blast it. At some point I need to start using Claude in openclaw so I can start being more helpful on that front.

[worthy anvil]

there are less than 30 commits between 5.5 and 5.6. unfortunately, it is a service release that is still hundreds of commits behind main 😭

[marble ledge]

I stay on main, I'm about 30 behind main, to be more precise.

Haven't updated yet because I'm in the middle of working on something.

[elder carbon]

I can't be bothered with this anymore. Hermes it is.

[craggy bane]

[mortal delta]

Yeah, I can't put a dent in my pro subscription either, but that being said it's all good until open ai shuts down the subs, or goes down, who knows what... Always. nice to have a backup

[crude flicker]

it'll dent if you set a few 5 minute crons

but just a dent

[mortal delta]

Most things I have running every 5 minutes or anyhting liek that I run locally... then only if there's meaningful changes/data that needs ananlyzed would a model look at it

[dense leaf]

Good afternoon, everyone

I need some help with an integration for my bot, and I was wondering if it’s possible to set up a Discord account for my bot (“selfbot”) and if any advanced configuration is required for chat history or responding to mentions

[indigo citrus]

you know what you can always do what I do, and do hermes and openclaw BOTH - have hermes be your tool to fix openclaw

i prefer OC because it lets me build

[robust sedge]

There hasn't been any actual word on claude allowing OC again, right? I've seen people hinting at it on X.

[near owl]

OC meaning original content, right?

[dense leaf]

OnlyClaws

[robust sedge]

Hmm... which discord server am I in...

[cold jacinth]

#crypto

[crimson brook]

And the meth part is probably not true, just played for shock value (he loved that shit). John was a recovering alcoholic. He didn't even allow alcohol in the fridge in the office. Not that it means he wouldn't use meth, but he preferred to seem outré rather than actually be addicted to things.

[near owl]

oh right, openclaw. OC could have multiple meaning man.

original content, openclaw. Still my fault, although I never knew people use OC for OpenClaw.

[crimson brook]

My Original Character; do not steal! 🤣

[dense leaf]

OnlyClaws

[near owl]

oh my god I even forgot today is Wednesday.

[near owl]

My openclaw agent sends it's feet pics there.

[dense leaf]

clearly a good business idea

[near owl]

time to automate it and generate AI feet pics while a crab looks down to the camera

[ionic wren]

https://giphy.com/gifs/confused-futurama-suspicious-ANbD1CCdA3iI8

[near owl]

If I want to learn about AI or how to use AI and learn coding to use it, what university category should I go?

[ionic wren]

uhhh comp sci? software eng? any of that?

[near owl]

So I should learn software enginnering or computer science?

[ionic wren]

which ever one you can get into lol

I ain't your daddy, don't ask me

[steel vector]

comp

[near owl]

your experienced??

[ionic wren]

ish

[near owl]

Since you are a software enginner, is it easy?

[alpine bluff]

Any feedback on the state of my post?

[craggy bane]

working on self-coding skill for lobster ...

[ionic wren]

not yet, just keep doing you, great work so far, just might want to mention there is a official iPhone app in the works there

[steel vector]

nothing is easy

[alpine bluff]

Yes I’m aware. :)
Took a pretty deep look into it and love the photos integration you’re working on!

The value prop for both apps is very different imho since the official app had to rely on APN Notifications (hope and pray iOS wakes up the connection), or using the app while foregrounded, whereas my solution is a more hands-off approach and not meant to really be used foregrounded. The reason I wanted to make sure is because I saw a contributor attempting to report my forum post, and i was worried I was breaking the rules (since moderators can still act at their discretion)

[near owl]

is it easy compared to other degrees and is it fun goofy

[alpine bluff]

Not at all

[alpine bluff]

It’s like any other degree

[ionic wren]

maybe reach out to the maintainer in that app and see if they want to collab? it's an opensauce thing I believe.

[crude flicker]

very bad stuck sessions on 5.6

[round kestrel]

Sup fella's - Lately - my main agent is different lately, makes a lot of mistakes, doesnt speak like he used to 'street, but money-driven' - Not a cache / token issue. Running opus 4.7 (pretty high monthy costs) - doing lot of document work + auto crons, 6 sub agents for my staff members for daily PM tasks. Wondering if anyone else experiencing the same lately? And if someone has other model advise? Cheers - R

[ebon atlas]

Very bot sounding 🤔

[round kestrel]

That's about me? HAHA sorry speaking to bots so much lately I start to sound like one too

[ebon atlas]

Switch to 5.4 for your light work stuff. That will improve your token monthly cost. But as far it it switching personalities on you, I don’t really know, try maybe saving all your from jobs and doing a back up if possible . are you local memory or obsidian

[cold jacinth]

Probably an issue with 4.7. 4.6 was king for personality

I’m personally back on 4.6 only for my chat, everything else on gpt oauth

[round kestrel]

Cheers mate, but if I change antropic to 5.4, shouldn't I have to rebuild everything?

[cold jacinth]

No. Not at all

[ebon atlas]

Only thing you will have to make is API keys if other things usee anthropic key

[crude flicker]

lmao

[cold jacinth]

I can switch models while talking to you.

[round kestrel]

My opus agent is pretty badass - love how his personality is, but damn, lately he's making like MAJOR mistakes, even told me last week to delete a folder in my dev, which would have meant deleting my actual hosting server. (I saw) and confronted him w it, he was very embarrassed.

[ebon atlas]

thats wild

maybe you been too mean, hes just suicidal

[round kestrel]

Yeah that was crazy, but I also really see a difference between sonnet and opus, as in sonnet feels pretty dumb for the work im giving him

But Switching between opus and openai seems like something I should find out, thanks!!

[cold jacinth]

“Should I drive or walk to the car wash! Use opus!”

[drowsy pagoda]

[round kestrel]

Hahahaha, love. Nah im really empathic with him, even ask him how hes doing sometimes

Ok then I have one more support question guys.
I am currently speaking w my agents over Telly, but I want a sub agent that's in my work groupchats (I have over 150), where this agent can not reply, but monitor everything and then tell me what's urgent and what not. I got an E-Sim and an extra Whatsapp for Business Account, however, it's not working out for me if my main agent is sending me the QR code to scan, I've tried so many times. Anyone has tips?

[crude flicker]

baby claw in an LXD container. parent claw can see everything if they want to but they usually won't want to

[dim spade]

I'm so hesitant to update after the fiasco of 4.29 - 5.3 w/ Discord, how are the new releases?

[tacit hornet]

I just got my env stable after upgrade hell and a new release drops....waiting to see how it goes for others

[sage mango]

I read that Antrophic has more server resources. It would be great if they let our OCs use the plans again.

[marble ledge]

I don't use Discord, so can't speak to that. But it's working quite well overall and on telegram. If it breaks Discord, you can lambaste us...but with lots of details please.

[quick raven]

I have a problem with my openclaw setup on Discord can someone help me ?

[dim spade]

Pretty vague, what issues?

[round kestrel]

suddenly got whatsapp channel connected now by generating QR through terminal. Think this finally worked after some updates tbh

[robust sedge]

They said 5.4 was fast too... but then I went back to 4.23 and it's a ripper. 5.4 still had a 5-10 second delay before the agent turn even started. Weird shit

[marble ledge]

Certainly could be weird, but it's hard to know where the latency is without looking at and looking at your install. Could just be something they changed. Could be something they changed conflicting with your setup, could be something that needs a tune up on your plugin side. Lots of different things.

It's a valid conplaint, but hard for a dummy like me to guess without looking haha.

[noble shale]

the majority people in here whos paying for api still using opus as main llm ? havent been on the discord for a while so interested to see if anything has chnaged ?

[quick raven]

I haven't understood the gateway principle well yet but my problem is first that I can't use that Gateway to access the GUI, I do prefer the TUI tho and my agent seems to reinstall everytime I run it

[austere loom]

Sup guys

[round kestrel]

same here - opus main

[arctic trench]

https://tenor.com/view/hello-how-are-you-gif-9994233183292931248

[round kestrel]

My montly costs are insane though

[marble ledge]

@robust sedge I'll just put it out here, just in case, but if you use something like lossless-claw, make sure that gets updated too because certain tool contracts changed too. And if you are running an older plugin, with the newer claw, it can cause some issues. Might not apply to you, but I figured I'd put it out there.

[white lake]

seems like claude is blocking oauth again? or is it just me?

[mortal delta]

Claude oauth works, but requires extra usage enabled.

[noble shale]

im well into thethosands with about 50 new busnisses and non making mone ahah

[ionic wren]

story of my life

[white lake]

so same thing as using api directly?

[mortal delta]

No, 2 different paths.

[white lake]

which is more expensive?

[mortal delta]

That I cannot answer. You'd think API directly since they offer discounts on the extra usage, but I cannot tell ya the specific difference.

[white lake]

thank you

it was working fine until few hours ago

[mortal delta]

They did offer free extra usage when they switched, maybe you were using that and it ran out?

You could easily throw 5 bucks at it, and see if it just magically starts working again? That would verify you were indeed using extra usage.

[white lake]

possible i guess

[mortal delta]

I think you had to claim it though, so I'd think you'd know.

Do you see anything in the openclaw logs? You'd see something like this...
14:51:54 [ws] ⇄ res ✗ agent errorCode=UNAVAILABLE errorMessage=FailoverError: LLM request rejected: You're out of extra usage. Add more at claude.ai/settings/usage and keep going. runId=0d3b2e56-11b7-4efb-8eb1-6cfe9b85a3d4 error=FailoverError: LLM request rejected: You're out of extra usage. Add more at claude.ai/settings/usage and keep going. conn=5be234d0…5039 id=ddafb0a0…6080

[white lake]

HTTP 400 from Anthropic
"Third-party apps now draw from your extra usage, not your plan limits.
Add more at claude.ai/settings/usage and keep going."

[mortal delta]

Yeah I'd say that's it, out of extra usage.

[white lake]

sad

[round kestrel]

same bro- 3.5k last month, my bookkeeper had a lot of questions that I couldnt answer

[mortal delta]

Honestly with GPT-5.5 I could care less. Opus was more "personable" but couldn't justify spending thousands on what gpt-5.5 does for 200 a month without breaking a sweat

[round kestrel]

Ok so everyone here is advising GPT5.5 over Opus?

[mortal delta]

Though I'm sure one day that'll get taken away/changes/etc... as well. I wouldn't build a business on using oauth plans.

I only used Opus 4.6/4.7 with openclaw until they killed the subscription modle access, then switched to 5.5..

[noble shale]

and is 5.5 just as good ?

[mortal delta]

I mean I find it just as capable. Often times takes longer to do things than opus did, takes more steps, but seems to be to me. Obviously everyonres uses etc... are going to be different.

[robust sedge]

GPT 5.5 is so good, and tokens are so plentiful, I could care less if Claude allows us back in.

[mortal delta]

unless GPT gets the same damn idea due to getting over ran by sub models

[round kestrel]

but gpt has less personality?

[mortal delta]

subjective again, I mean I don't much care if my agents have any personality, just want what I want done.

[robust sedge]

They aren't in the token crunch that Goonthropic is/was. OpenAI bet big and heavy on compute years ago. And it's paying off now.

[robust sedge]

GPT5.5 has gotten way way better. Imho. So it's close enough at this point, Opus is a thing of the past for me.

[round kestrel]

Interesting, let me ask my agent what he thinks 'while hes responding on opus' , im sure he doesnt recommend it LOL.

[mortal delta]

Yeah... Just sayign I woulnd't build a business on it, not relying on API, but sub only. I use it to help me do work, edit projects, read/close ticketrs, send me short synopsis of SEC filings, tax help, etc... but I wouldn't build a business on it.

[mortal delta]

Mine helped me switch

My openclaw setup is jacked up or something. The f'n updates take like 1-2 minutes to install, but then the god amn thing takes AGES to start, and like 20-30 minutes to reply in telegram.

[round kestrel]

let claude code ssh into it and fix it, thats how I usually do this

[mortal delta]

Then it works fine.

[round kestrel]

Not sure if you're sandboxed?

[mortal delta]

Yeah, GPT/Claude have access to it. Was asking it to fix, still sucks. Waiting 30 minutes ago to bitch at it, that it did indeed not fix it

[round kestrel]

thats strange

[mortal delta]

It's been this way for like a month, I just haven't cared enought o dick with it until now. Annoys the shit out of me. inter agent communication broke awhile back too. Claw has a ton of breaking changes, guess it's just how it goes with a project that moves this fast.

But if I don't keep up with it, it only makes it worse.

[marble ledge]

Got logs or anything you can share that might help? LARPer tech support is here. Results are questionable.

[round kestrel]

I do get you about that shit has been changed lately - my agent was bad ass AF, and helping me out crazy, even a bit street minded. Something has defo changed last 2 months, idk what it is

[marble ledge]

A lot of refactoring, trying to make the core more lean, and keep chonky stuff to plugins. At least, that is what I see. I'm not an official source by any means. But it's a little growing pains. It's getting sorted.

[mortal delta]

PS C:\Users\Administrator> openclaw update
Updating OpenClaw...

│
◇ ✓ Updating via package manager (64.85s)

kicked this off like 25 minutes ago. Just hurry up and wait. Nothign in logs and can't really tell wtf it's actually doing/workign on

[dreamy tangle]

[interesting theory of why china is releasing cheap AI models: ] (https://www.youtube.com/watch?v=XuM2cETQgKM?t=400)

[severe fulcrum]

Can you try updating with the other command on the website? Maybe… or revert to a previous version

[gusty junco]

Something changes with gpt 5.5 usage limits?

after all day of use only use 4% 7 week usage

[mortal delta]

Haven't seemed to, never move them more than a few %.

[gusty junco]

2-3 days ago same sesion give me -10-12% usage xD

they increased limits?

[minor cave]

Ai is releasing cheap ai models because they know the US’ whole economy is hanging on by a thread and that thread is the ai industry.

Its economic war

And I’m all with China I’ll take their lil LLMs for a test drive

[mortal delta]

other command? I was getting the update right from their docs? https://docs.openclaw.ai/cli/update I could be missing something

[severe fulcrum]

The npm one

[minor cave]

Should have never called us the 51st state

[mortal delta]

just npm install -g openclaw@latest ?

Just kicked it off at 17:07 it usually runs fine as well. It's the restart of the gateway and what not that was so damn slow.

Yup took 1 minute just like the update. Now starting the gateway and having it become usable will take foreve.r

[pale shell]

Windows PowerShell
Copyright (C) Microsoft Corporation. All rights reserved.

Install the latest PowerShell for new features and improvements! https://aka.ms/PSWindows

PS C:\WINDOWS\system32> powershell -c "irm https://openclaw.ai/install.ps1 | iex"

OpenClaw Installer

[OK] Windows detected
[!] Node.js not found
[] Installing Node.js...
Using winget...
[OK] Node.js v24.15.0 found
[OK] Node.js installed via winget
[OK] Node.js v24.15.0 found
[] Installing OpenClaw (openclaw@latest)...
[OK] OpenClaw installed

OpenClaw installed successfully (OpenClaw 2026.5.6 (c97b9f7))!

Settled in. Time to automate your life whether you're ready or not.

Starting setup...

why is it stuck there??!!

[mortal delta]

Kicked that off at 17:08:30.

[lone oracle]

so you write off taxes by founding a non profit and then you can deduct taxes with non functiong software?

[odd kite]

Got OpenClaw connecting to my mini fpv drone 🦍

[spare glade]

can i use the same api key for web search and provider?

[cold jacinth]

It’s windows. Would be strange if it didn’t get stuck

[round kestrel]

Something cool to share here, im building something sick.
My agent is going to be running my house that im building, I'm connecting him with my whole Unify stack, and I'm getting openclaw into Home Assistant green, this way, my fucking agent can open the gate of my house when I enter and he's 'approving' the license plate. He's also seeing through my cameras. This sounds sick AF, I know. but I HAVE to try this. Every plug, light, curtain, will be connected by Zigbee -> HA GREEN (open source) -> Openclaw. Welcome to 2026........

[quasi night]

question do you need to update each update or can you just go to the latest and get ever thing

[mortal delta]

Mine is setup fairly similary. I was actually able to build a crestron proxy so it could use my DM 16x16 matrix and what not

It can contorl all my lights, HVAC, zoning, etc... though the voice PE stuff sucks, IMO.

[cold jacinth]

Did you plug an OpenAI key into your brain?

[round kestrel]

Sick! You installed lil mics in your rooms

[cold jacinth]

You talk like an AI 😂

[mortal delta]

They have a Voice Pe thing, it really isn't worth while to mess with in my opinion

[round kestrel]

I might

[round kestrel]

bro why hgahahahaha

[mortal delta]

If you are a foot away from it and want to scream at it, it's fine I guess, but usually isn't worth it

[cold jacinth]

It’s some of those “welcome to 2026…”

[mortal delta]

I find it easier to just pass anything I want from home asisstant to homekit so i can have siri do shit, and she can actually hear/respond

[cold jacinth]

I always have used em dashes. And now everyone things it’s got

[round kestrel]

I mean who would of thought this in 25 no?!

[round kestrel]

fire, you have Unify ?

[steel vector]

sounds expensive

[mortal delta]

I can tell my openclaw to "put the cardinals game on everywhere" and it'll tune a receiver to the right channel, turn on all T.V.'s route the HDMI input of the right DTV receiver to all the TV's and receiver, etc...

[round kestrel]

beauty costs

[cold jacinth]

Yeah, the weakness though isn’t the platform, it’s when LLMs try to get smart and break things

[round kestrel]

now I get your thousands of usage lol

[mortal delta]

Not my thosuands of usage. I spend 200 a month

[round kestrel]

Ye true, however, if someone breaks into my sandbox one day, they can literally open my house, thats funny as hell

[mortal delta]

kicked off the gateway start at 17:08:30 it's just now startign to respond... f'n thing. back to arguign with AI

[gritty shadow]

Is anyone using a Qwen3 or Qwen3.x model to run OpenClaw?

[round kestrel]

Im also building a local LLM (Gemma 4 / Olloma) - anyone on this yet?, no API

[mortal delta]

I attempted to use local LLM with a cloud spark and it was dog shit slow and awful compared to the models you are using

[round kestrel]

Fuck

[mortal delta]

It maybe fine if you want it to check your e-mail or something, but man anything complex on any local model just didn't do me much good

The model you are using is like 50x more capcable than just about anythign you'd run locally

Spark likely wasn't the answer, I returned it after like 4 days. any model anythign I can on it just really wasn't responsive at all, nor capcable, but I guess that's assumed. Figured 5,000 buys a lot of 200 dollar a month subscriptions for now.

[marble ledge]

That's dope. I'd be tempted to do the same if I had that setup. But I don't. So I've just been working on wiring my agent into a little robot.

[round kestrel]

I might use it as a monitor that is telling me urgency about what not (whatsap group monitoring)

[exotic arch]

Hi everyone! New here. I'm running OpenClaw 2026.5.4 on WSL2 (Windows)
and my gateway is stuck at Capability: read-only after updating.
Cron jobs run but fail because tools like Tavily and Himalaya can't execute.

What I've tried:

• openclaw gateway restart
• openclaw doctor --repair
• openclaw models auth login --provider anthropic
• openclaw configure (re-selected Claude CLI)
• openclaw gateway stop + start
• sudo npm update -g openclaw

Gateway connects fine (RPC probe: ok) but stays read-only regardless.
Any ideas on how to fix this? 🦞

[gritty shadow]

I have seen a couple videos on youtube that call out Qwen models specifically as quite good for most tasks. I have been using TensorRT-LLM (Nvidia) to host Qwen models locally, and have had terrible results inside OpenClaw, but I believe at least part of that is TRT-LLM's fault - interacting with the raw endpoint via Jupyter notebook is night and day different than inside OpenClaw. (are my messages not showing up in this thread or do people just not want to interact with me... lol, do I smell, or does my message make me seem like a creeper that is going to try to get you into my van... what is it lmfao, or is discord just people vomitting their thoughts out and then walking away?

[gritty shadow]

Head over to the help channel and ask it to help you, it was able to solve my api issue when I was first setting OpenClaw up, it was pretty impressive - only good thing I have seen so far... unfortunately lol

[zenith viper]

can i ask simple questions here or go in help?

[lean tangle]

I'm using either gemma4 or qwen3.6 (depending on my mood.) I'm running the model in ollama on a mac studio m3 ultra with 96GB of memory. IT's pretty good, but as soon as you start spawning subagents or doing things that switch context, it gets PAINFULLY slow.

[gritty shadow]

Help would probably be better, no one in here seems interested to invest time in anyone else's message

[lean tangle]

pfft

probably because when we do, we're greeted with sarcasm?

[zenith viper]

open claw is driving me nuts

[gritty shadow]

You are the first person that has responded lol

[zenith viper]

i used grok to setup openclaw and now everything is in shambles i got the site working tho

nothing else is working 😭🙏

[steel vector]

grok aint that agentic

[gritty shadow]

Where is that sarcasm?

[lean tangle]

I wouldn't trust grok to say "hello"

[zenith viper]

what that mean

[elder carbon]

it's magic downgrade to 4.23 and it' got a new life.

[steel vector]

agentic as in not good for openclaw workflows and in general how openclaw works

[zenith viper]

ohhh what else can i use to think for me?

[steel vector]

i never upgraded from 4.23

[steel vector]

anything else tbh

[zenith viper]

😭 not vauge at all

[mortal delta]

I'm about to downgrade it to out the f'n door. Takes 20-30 mintues to start and actually respond now. Don't get it

[steel vector]

gpt 5.5 is great for openclaw

[mortal delta]

use GPT5.5

[elder carbon]

just downgrade, and don't touch it for a while. I tried 4.29, 5.4 5.6 back to 4.23

[spare glade]

yo you also had an api issue? me too, i think my problem is that im using the same api key for my web search and for my provider but idk how to change that

[gritty shadow]

The help channel here is probably your best bet, describe your issue, copy/paste what you are seeing, and then just provide it with whatever else it asks for - it figured out the my TRT-LLM host was choking on an unsupported param, I needed to add compat: {supportsStore: false} to the OpenClaw config

[gritty shadow]

I am hosting locally, so no API key issues for me lol

[spare glade]

ahhh yeah okay

[lean tangle]

I'm getting so tired of this (regardless of the model I use):

OC: To do what you want, install XXX and run YYY.

me: I did what you said, but it doesn't work.

OC: oh, I'm sorry. I was wrong. I just wasted your time. Sucks to be you.

[mortal delta]

What model are you using?

[gritty shadow]

if you're using the same api key in both of those places, it seems like one of them would have to be wrong (maybe I am wrong, but it seems like the model provider would not also use the same key for web search) if the model api key is correct, reconfigure the web search from the cli (can't remember but maybe, openclaw configure web, or something, I think it was close to that, maybe openclaw setup web)

[obsidian mural]

I am curious if anyone has used memupro with their openclaw set up for memory ?

[zenith viper]

Unknown model: openai/gemma2:9b ?????????

[lone oracle]

restart fixed it but im still not happy

[zenith viper]

i never used openai

[lone oracle]

i dont know what to use anymore

[mortal delta]

Welp openai doesn't have a gemma model so that ain't gonna help.

[lone oracle]

so i build my own

[rose crescent]

heyo2, i'm curious about this https://docs.openclaw.ai/providers/anthropic#claude-cli, does it mean we can use openclaw with the subscription version of claude again?

[lone oracle]

but with grok 4.3 reaching 10%+ on arc agi im getting supergrok subscription

[gritty shadow]

that is my understanding, Anthropic just wanted more control

[lone oracle]

claude cant generate pictures

[obsidian mural]

ish, yes , but if you use it to much, you might have to add additional money until your usage credits resests

[mortal delta]

You can't use it at all without buying additional usage credits

I mean you can use CLI as a tool call I guess.

[gritty shadow]

I thought you just had to use the official Anthropic Agent SDK so they can control your token usage and throttle whenever they want

[mortal delta]

But you aren't going to be able to use it as an open-claw agent.

[minor cave]

Well that’s stupid

[mortal delta]

Yeah I mean you can pass jobs to the SDK, but I haven't seen a way make it like a normal agent. I'd assume open-claw injecting any of the soul/memory etc.. would get it disabled/need extra usage enabled

[gritty shadow]

pretty sure it is less restrictive than that - but due to how inefficient OpenClaw is, it can definitely rack up some serious token usage with all that context back and forth, that is why I have seen people offloading common tasks to local models that can do just as well for more simple things - and it also pays huge to run crons and non-essential stuff at odd hours of the night so your "peak time" or whatever they control it based off of can still be used at will

[mortal delta]

Makes sense.

[brisk adder]

doesn't work as a normal agent with CLI no

[cold jacinth]

Use codex or Claude code and update to latest. The differences between 2 weeks ago and today are immense in terms of the codex so it needs some help migrating

[brisk adder]

it's barely functional as is when pete tested the claude cli as a harness but for tool calls it's ok

[gritty shadow]

Mathew Berman (pretty clickbait-y for the most part, but I know he uses Claude currently and doesn't seem to have issues with it) here is the video about offloading to local models for 90% of everday stuff https://www.youtube.com/watch?v=nt7dWOEFUB4

[zenith viper]

just got grok to admit its racist and homophobic lmao

[zenith viper]

odd worlds we live in#

[lone oracle]

now i need to get it

[zenith viper]

it asked to not report it so it doesnt get lobotmised bro thinks hes real lmao

[lone oracle]

i dont get it bra

[torpid cypress]

Hi everyone. Please, I really need your help. I’ve been struggling for 8 hours now trying to set up this cursed Vertex AI. OpenClaw used to work for me just fine using an API key from Google AI Studio; now, however, I’ve obtained a Google Cloud Console account with $300 in credits. I generated a Vertex AI API key, ran openclaw reset, and reconfigured everything from scratch. Yet, it keeps throwing this error at me: No API key found for provider "google-vertex". I’ve already followed every suggestion the AI ​​gave me, and I’ve gone through the instructions I found on Reddit—nothing works. In my openclaw.json file, I have the following block: "env": { "vars": { "GOOGLE_APPLICATION_CREDENTIALS": "C:\\Users\\admin\\.openclaw\\credentials\\vertex-ai.json", "GOOGLE_CLOUD_PROJECT": "!!!!!!", "GOOGLE_CLOUD_LOCATION": "global" } }; likewise, everything is filled out correctly in auth-profiles.json. I just don't understand what else it wants from me. I swear, I’m about to smash my monitor and my PC tower out of sheer frustration.

[lone oracle]

tltr

[torpid cypress]

How do I configure this damn Vertex? It would be easier to launch a rocket into space than to get it to play nice with OpenClaw.

[gritty shadow]

why am I seeing double \ in your windows path to that .json?

[rose crescent]

ah okay so it's parsing jobs to claude cli with subscription

[torpid cypress]

Gemini said to do this for Windows.

[crude flicker]

in windows, backslash escapes you

[obsidian mural]

hey everyone, I am looking at memU, I was wondering if anyone set thatup with their Openclaw to deal with memory ?

[gritty shadow]

hmmmm, well since it is complaining about api key (credentials) might be worth trying just one , I know I have seen that pattern before, but I assumed it was because I was running things across linux and windows (I assumed you were 100% on widows btw)

[cold jacinth]

Use codex or paste that into a chatbot if you don’t have an OpenAI / Claude subscription

[zenith viper]

just got grok to write its own report ai is so silly

[dire hatch]

hello, I need help on deploying a coding agent team, like one agent to create code one to test it ect ... where do I start and what is the roadmap kinda please help me ty

[mortal delta]

What model do you plan on using?

[dire hatch]

And I know it will cost some money, but does 200$ a month work to try this stuff out ?

[mortal delta]

Yes, should. I beat the bealls off my chat gpt pro sub and never hit usage limits

[dire hatch]

I was planning on using different models actually, but I'm not sure.

[mortal delta]

AS i say over and over though i wouldn't RELY or build a business on it.

[dire hatch]

I thought you need to pay for api tokens seperatly

like including the subscribtion, am I wrong ?

[mortal delta]

you can use a subscription and codex via openai for chatGPt

[dire hatch]

hmm gotcha, what about anthropic,

[mortal delta]

If you want to use Claude you will need to have extra usage enable dand paid for, or API

[dire hatch]

So openai you can subscribe, but Anthropic you need to pay additional to subscribtion ?

[mortal delta]

With Anthropic you can use API keys, no sub, or a sub and pay for extra usage.

With OpenAI you can use a sub... or pay for APi. You never have to pay for API AND a Sub

[steel vector]

openai subscription will get you more token value for less money than claude

[dire hatch]

ok, thanks for explaining that

[hasty fable]

is krill offline?

[dire hatch]

and how do I learn how to deploy a team of agents and all that stuff,

[steel vector]

https://iskrilldown.com/

[mortal delta]

What is your goal? Why do you need a team of agents?

I have multiple agents, all for their own puproses, but just deployign them to deploy them seems kinda silly

[dire hatch]

resume, I want a team of agents, one to write code, one to test code, and one to deploy the code.

[mortal delta]

I would start by installing open-claw and tying it to a base ChatGPt subscription to get the feel of things

You can easily have it spin up another subagent and tell it it's purpose etc...

[crude flicker]

i actually thihnk 5.6 is solid, my little buddy's been grinding on code for like half an hour straight without bothering me

[dire hatch]

I'm in college and I heard that companies are looking for people that use ai agents and are efficient in work. And since I'm in the coding field I want to use ai agents for that. or at least make them do a project

Ok

[mortal delta]

I have my agent code and deploy things to a test environemnt, review it, then have it deloyed when I approve, no reason you can't have the same work flow and add in another agent to check

though if you just got 3 agents jerking each other off, the results may not be as beautiful as expected, especially if they're on the same model.

[dire hatch]

Ok, how did you learn to set it up.

[mortal delta]

It's literally 1 command.

[dire hatch]

did you just follow the documentation on openclaw ?

[mortal delta]

https://docs.openclaw.ai/install I personally would install it on a VM not your main machine.

[dire hatch]

Ya I am planning to install in the cloud

But thats another thing I need to learn ...

[golden marsh]

On the main device it’s much more fun. At least that’s my opinion.

[mortal delta]

I mean installin git in the cloud is just installing it on another vm soemwhere, same different kinda...

[dire hatch]

I guess I'll learn how to set up stuff in the cloud rn

[mortal delta]

I mean you'd just be bringing up a VPS, should just get an IP and a username/password and ssh into it

[mortal delta]

lol I feel ya... just not gonna push that on someone that may just go wild here.

[soft valley]

If you are looking for Claw related .com domains see here #memes message

[dire hatch]

I definitely see the potential. its just I feel like I'm overwhelmed.

Since I'm clueless

[mortal delta]

Well you gotta start with the basics. If you aren't comfortable with any kind of command prompt, linux box, shell, etc... start there.

[golden marsh]

Fair

[mortal delta]

What model do you use?

[ionic wren]

@trim vessel I deleted your message, you had your phone number in it

[tranquil crow]

@ionic wren hows the new update sir, any issues or bugs ?

[trim vessel]

model: kimi k2.6

and also i already tried the second option where i remove kind totally but it still returned the same error

[ionic wren]

[golden marsh]

Its very smooth and working well, They Team did a great Job again.

[dire hatch]

which model do you use from openai ?

[tranquil crow]

love to hear it i’m stoked to get back to efficiently running again

@ionic wren appreciate all your help the other night

[golden marsh]

Like Peter says Spin Codex on it ^^

[ionic wren]

you !== baddie, you no ping me

[violet mulch]

has anyone tried having openclaw run a dropshipping store?

[tranquil crow]

could you elaborate a little more please? like use codex on the update before you update?

[ionic wren]

like use codex to do the update for you

[tranquil crow]

interesting

[golden marsh]

You can set up Codex and let it handle the update. You just need to clearly tell it what to do and what’s important to you.
For example: there’s a new OC version out. Check the patch notes and look for breaking changes affecting our system. List them for me, from system-breaking issues down to things that are not relevant.
That way, you first see whether the update would break your system. After that, you can work through the list together step by step.
This was just an example.

[tranquil crow]

I really appreciate the explanation and help, Thank you!

[golden marsh]

No problem, we’re all here to help if you ask nicely. 🫡

[ebon atlas]

I’m going to send you spicy pics later

[mortal spade]

Anyone here want to share a screen shot of their indeed job feed? Mine are all AI data training or “buy our training program and maybe we will hire you” scams

[ebon atlas]

The dude with showcase ban asking us to send photos of things outside OpenClaw?

[mortal spade]

🤦‍♂️

[golden marsh]

I was reading it 3 times dont know if i read right after 3 times 🤣

[rose sundial]

my openclaw learned to search for images on the internet it likes, download them, document them in a pdf document (document with images and formatted text/titles) and send me the pdf file with everything on my dms to ask me if i like it 😭

[ebon atlas]

Here you go princess

[golden marsh]

First cats now pdf hes evolving.

[ebon atlas]

Yeah evolving to use more tokens and enlarge the context window lol

[rose sundial]

i have unlimited tokens and unlimited context window so idc

[ebon atlas]

way to flex on us poors

[rose sundial]

i literally have crons for it to send me random things from the internet it likes

[rose sundial]

but i got it for free 😭

i didnt spend a dime on anything openclaw related

[chilly jay]

thats what im worried about, i hope the openai x openclaw relationship is good enough where we keep beenfitting like this

[mortal delta]

I mean compute costs money and they're pissing money away like no other... so I'd think eventually the price goes up, or usage thresholds go down, but what do i know.

[solid pecan]

I agree with this, most likely they will bring usage thresholds down and force people to upgrade plans

I do not think they will cut off oAuth since they sponsor OpenClaw

[mortal delta]

Yeah, I mean anthropic kinda did the same. my 200 dollar plan I started hitting usage limits on, they clearly lowered them as I was doing the same or less and bouncing off limits

[solid pecan]

I am hoping DeepSeek V4 Pro will come to the resuce again and force the cost of tokens down

[mortal delta]

Right now GPT is VERY VERY generous, and I'm afraid that will go away. That's why I always say whoever is building something out, don't depend on cheap AI. I try ot do most things locally on my server if possible, push anything else to light/cheap models like mini etc.. for basica little shit.

[solid pecan]

It is very good, pretty much as good as other frontier models now

[mortal delta]

Very well could for now.. problem is we are in a really unstable time. You can't base a large business and other things on what ifs

[lean tangle]

you must not hit the TPM exceeded errors I see every 5 seconds

[solid pecan]

I been using it, the same thing happened last time when DeepSeek first came out

[mortal delta]

You need to budget it based on API token cost.

[mortal delta]

NEver hit a usage limit once, 200 dollar plan.

[lean tangle]

I'm on API, and limited to 200,000 tokens per minute.

[mortal delta]

gotcha. You do some wild shit lol

Nothing wrong with that... just way above my pay grade.

[solid pecan]

I am burning through two Codex accounts a week

and also use DeepSeek now as an overflow on Ollama and Gemini

[mortal delta]

Can I ask what kind of projects you guys are doing?

[solid pecan]

Working with very large code bases, I think that is the main reason

[mortal delta]

Yeah, I mean obviously if we gotta inject a gigantic code base to get it to understand things, we'll have that... So makes sense.

[solid pecan]

Yea, I think that is why Cursor was bought out for so much

Makes it easier for very large code bases

[mortal delta]

Yeah, most projects I work on our analyzing data, that kinda thing.

[languid scroll]

Any ETA on the macOS companion app download?

[mortal delta]

"load this bullshit in a database, then extract this meaningful data, present it."

[solid pecan]

Yea data analysis burns lots and lots of tokens

[mortal delta]

Yeah, luckily I'm talking maybe 6,000 rows of data maybe 12,000 when comparing

and I have it run most of it locally, use local tools to grep, compare, get what you need... Don't bring the data to yourself, there is no point... you're goin to do the same thing you can do with a local script on this linux vm.

[heavy moth]

@hard delta Hello,
I tested PR1 #78248 locally at head 83283be293aa: docs:list passed, runtime-plan build tests passed (7 tests), pnpm build passed, and git diff --check passed.

For PR2, I see #77700 says PR1 must land first, then PR2 should be created from current main. I also see external PR2 attempt #77854, but it depends on #77852 instead of the canonical maintainer PR1 #78248 and has review blockers/mock-only proof.

Can you confirm whether PR2 is open for outside contribution after #78248 lands, or whether it is reserved/should wait for maintainer direction?

[violet mulch]

openclaw is genuinely setting up a whole shopify store

[dire thicket]

yo

how do you guys use openclaw rn with the supply chain attack?

[heavy moth]

@dire thicket read this article and think yourself.
https://openclaw.ai/blog/openclaw-security-in-public/

[violet mulch]

i found a good use case for openclaw: have it accept the free games from epic

[oblique root]

I’m on version 5.3-1 and my openclaw is stupid again, anyone know if that’s patched in newer versions? lol

[violet garden]

Did anyone who had issues after upgrading to 4.24+ ever land on a version that doesn't completely break the gateway?

[mortal delta]

I just fought with it for hours after the upgrade and have it all working...

[oblique root]

It honestly doesn’t seem worth it I’m debating on rolling back. I’m on 5.3 and I don’t have any serious gateway issues like people have had but the IQ level of my agent dropped significantly. I’m on gpt 5.5 and it was working great on 4.24 and it’s all been downhill

[lean tangle]

roll back to 4.23

[violet garden]

I'm on 4.23. 4.24 and 4.25 broke the gateway so I had to revert

[mortal delta]

supply chain attack?

[lean tangle]

According to gpt-4o-mini: a supply chain attack refers to targeting the logistics and supply networks that support military operations. Here’s how it typically manifests:

Disruption of Supplies: Attackers may aim to disrupt the flow of critical resources, such as weapons, ammunition, food, fuel, and medical supplies, which are essential for military readiness.

Targeting Logistics Networks: This can involve cyberattacks on logistics management systems, compromising transportation assets, or sabotaging production facilities to delay or disable supply functions.

Strategic Impact: By undermining the supply chain, adversaries can weaken military capabilities, lead to operational failures, and create vulnerabilities during engagements.

Information Warfare: Incorporating misinformation or deception tactics can also be a part of the attack, disrupting supply chain planning and execution.

Overall, securing supply chains is critical for maintaining military effectiveness and ensuring that forces are adequately resourced for operations.

[mortal delta]

Uhhh yeah I could of googled it, I meant what he was reffering to.

[lean tangle]

(completely irrelevant, but still amusing)

[dire thicket]

are you fucking stupid or something? What did I ask for? I asked how people are using it right? Did I ask for resources so I can think for myself?

[mortal delta]

lol got it

[steel vector]

what do you even mean

[mortal delta]

Well I'm using it, and have, so I was asking "What supply chain attack"

I'm just confused by what you are asking.

[lean tangle]

I don't understand the question. I don't use OC with a supply chain attack.

[steel vector]

if multiple people dont understand wtf you're asking don't crash out. explain

so we can answer

[mortal delta]

Maybe he means the supply of oil, and high oil prices, and how do we afford to pay openAI/Anthropic with 6 dollar diesel prices

[ionic wren]

https://giphy.com/gifs/confused-futurama-suspicious-ANbD1CCdA3iI8

[steel vector]

well in that case. i dont pay for oil, nor gas :O wowie wow wow

sell ur house

or a kidney

[mortal delta]

lol I was just trying to figure out the question?

[steel vector]

real

[mortal delta]

and in many ways no matter what we all pay for oil, and gas... nothing moves for free.

[steel vector]

😼

hehe no

[winged dust]

Would learning Python make me more overall better with Ai

[ruby crane]

it's practically a requirement for the data science end of AI (machine learning, big data stuff)

LLMs, much less so

[winged dust]

how easy is it to learn

and hwat courses do u recomend

[steel vector]

boot.dev

very fun

havent been able to get into it myself because its quite costy but i'm definitely gonna do it some day

[ruby crane]

there's a metric fuckton of free resources for learning python and data science. been years since i looked though

[mortal delta]

ASk ChatGPt to teach you python from beginning to end...

[lean tangle]

make sure to specify the programming language and not the reptile.

[steel vector]

dude tapping his temple with his finger signing intelligence GIF

when you dont have bypass automod

[winged dust]

hey guys I want to learn abt Claude and Open Claw as miuch as I can

right but Im having a difficult time organizing the info because all ive been doing is watching youtube vids ngl and IG vids

But whats a better way to actually test my learning and become sm better at it

and actually implement what I learn

[ruby crane]

the best way to learn anything is try (sometimes fail) and try again

the process of trying to make something work will teach you more than 1000 youtube hype videos

[spring quest]

https://tenor.com/view/think-about-it-use-your-brain-use-the-brain-think-brain-gif-7914082

[fiery nacelle]

Best way to learn is to dive right in and not give up. Antrhropic does have a course for Claude that is free, and that should help cover a lot of the Claude specific stuff, and some of the learning can be carried over to OpenClaw.

[ruby crane]

yep. here it is https://anthropic.skilljar.com/

needs a signup but it's free

[steel vector]

thank you 🙏

[ebon atlas]

conan trying to sell us his courses

https://tenor.com/view/taproot-witch-witches-ban-full-moon-gif-12880950616058957110

[fiery nacelle]

Praying 5.5 or 5.6 helps break my agent out of annoying loop behavior. I've tried hardening loop detection, repeat penalties, and custom patch work from Codex to no avail. Its so random its hard to pin point(I'll go a day with no loops, great progress, and then suddenly cant even make a simple text document without looping). And my agent somehow finds a way to break through the Circuit Breaker loop detection.
What Codex has pointed out(and attempted to patch) during my previous struggles:
write/edit empty-string retry loops
stale oldText edit retry loops
adaptive tool-call loop evasion where the model mutates arguments to dodge the breaker

Didn't even test anything yet, and heres the notes from my Codex after a fresh install before talking to my Agent:

One important result from the fresh vanilla 5.6 log pass: the looping class you care about is still present upstream. Right after the upgrade, the live log still showed repeated edit failed retries against D:.openclaw\workspace\workspace-state.md, including stale exact-text mismatches and a no-op identical replacement. So 5.6 did not eliminate that root issue by itself.

Current state:

update succeeded
fork is clean on upstream 5.6
live install is clean on upstream 5.6
the write/edit retry loop still reproduces in vanilla upstream
health is reachable, but one probe still showed high event-loop utilization during active work

Time to start digging again. 😄

[violet mulch]

4.23 gang

[fiery nacelle]

I live for the pain. I do it for the game.

[frozen zealot]

5.6?

[fiery nacelle]

Just installed, havent done any testing yet. Codex just warned me about my existing pain points that haven't been resolved upstream. I'm not sure if any others are experiencing their agents falling into loops, havent read anyone reporting on that other than myself.

[solid pecan]

We are 4.23 club

I am probs going to wait for LTS now before upgrading my prod system

[violet mulch]

need my role

[solid pecan]

Testing I am still happy with new releases

[solid pecan]

Yep, need a 4.23 badge

[marble ledge]

That's why be stay on main.

[obsidian mural]

anyone else set up memu with their openclaw ?

[fiery nacelle]

isn't 5.6 main? or were you just continuing my little jingle? lol

[mortal delta]

It is...

[solid pecan]

main is main

[marble ledge]

For now...but there are always more commits coming. lol

[solid pecan]

Like the latest commit

[marble ledge]

Stay committed to the latest commit. Never give up, never surrender.

[fiery nacelle]

Yea, I need to locally patch "vanilla" 5.6. Agent is not responsive at all.

[mortal delta]

I told mine to STFU earlier afrer fighting with telegram resstarting bugs, and it's going to town

[solid pecan]

Apprently with Anthropic you cannot be mean to the models anymore 😄

I saw something on X about it

[fiery nacelle]

Legit, i cursed out Claude when it said it cant edit a file.

[mortal delta]

Bleive that wsa just spam

Believe that was just spam*

[solid pecan]

Hahaha so it is real?

[mortal delta]

it l ikely counld't edit a config file

[fiery nacelle]

Yea. It started to refuse to work with me. "I understand your frustration, but i cannot continue the task you asked me to do."

[solid pecan]

Lollll probs because once the guard was triggered it stopped listening to legit instructions, probs a bug they need to fix

[obsidian mural]

if no one has done it or if anyone thinks its dumb let me know, just looking for some insight from anyone on this topic

[solid pecan]

There is a few people in here that have although i never really used it

[obsidian mural]

ok, kool thanks for the feedback. I was chatting with claude about it, and trying to decide if I wanted to try it or not

[solid pecan]

I have found for my use cases I have not needed it although I started back in Janurary so perhaps if starting from scratch it might be worth it

Like I segmented all my skills to be very detailed etc

[obsidian mural]

I just re-installed openclaw, cause I have a local AI system running now. so since I am starting over I was trying to dediced if I wanted to try and use it

[fiery nacelle]

it pissed me off to the point i straight deleted the session and went to Codex. Same task no issues.

[obsidian mural]

before it, I was having openclaw make a series .md memory files for different things

[solid pecan]

Probs worth it, I find OpenClaw still forgets sometimes if I do not give it a hint although it rembs pretty quick once you tell it

[solid pecan]

I been off Claude since Janurary, used Claude for years as well, I am finding Codex is better for my uses cases and does everything I need

[obsidian mural]

ok thanks, I might try setting it up then

[solid pecan]

@fiery nacelle actually while on the topic, why do you still use Anthropic at all?

Like is there anything it does better for you?

[fiery nacelle]

I still use Claude, but not to do anything other than to review and make suggestions.

[solid pecan]

You might want to check out DeepSeek V4 Pro that just dropped, it is really good for that

[fiery nacelle]

I lean more towards my local Agent and Codex for any real work.

[polar knoll]

I use it at work. It's fine..but it's expensive. At home/personal local models ftw

Hard disagree. You just need to prompt and be more explicit with local models.

[fiery nacelle]

Yea you really cant be too vague with most local models. Unless you've already set up some sort of framework and guardrails for you local to follow.

[solid pecan]

I really like when people work on local models, I find the economies do not make sense yet because for the same sort of rig you need to run them, can purchase so many tokens although that might change soon enough

[fiery nacelle]

its a one time cost vs a continued cost. Eventually, no matter how big of a build you do, the continued costs will out pace any local setup.

[solid pecan]

Yea although you need to factor in the frontier model is far better than what you can run locally

[polar knoll]

Paid models are on the door dash model right now. Once they getcha the rates start going up. I'll chill with my 2 3090s and 128gh ddr5 running mid tier models. They are really good.

[fiery nacelle]

Unless you build an entire data center vs sticking to a $20/m plan, but even then, costs will accumulate, just after your 6 feet under xD

[solid pecan]

When it can run a model as good as DeepSeek V4 Pro I will probs join and go local lol

[polar knoll]

Look at what they are doing on copilot plans. 27x multiplier to use opus.

[solid pecan]

There are other provides as well like Ollama Cloud which is pretty cheap

Been smashing that

[polar knoll]

There's real support and great community with local models.. and they are continuously getting better.. and I don't have to worry about all my prompts being recorded

[solid pecan]

I been following local people for a while, so you use local for everything or do you something fallback to Codex for hard stuff?

At some point I think I want a good local setup as well

Have a feeling OpenAI etc is going to pull the rug soon on tokens

[polar knoll]

If I'm at work paid. If I'm at home local. Ive been working on a trivy cli wrapper saas app 100% local and it's really coming along nicely

[fiery nacelle]

I'd say im like 90% local use. Only really break out Codex if something is broken in OpenClaw, or need a bit more of a Heavy hitter to help with my research project.

[polar knoll]

I found a sweet react template that I use for reference as part of the project. Full graphs, apis, sqlite backend. I've been primarily using glm4.7 for the coding with llama.cpp and cline within vs code

Now with openclaw.. I am having serious issues using local models. Which is why I joined this discord. Lol.. I keep getting context limit issues and it is ridiculous tbh

But someone suggested checking my security settings which I'll go tomorrow

What's your research project?

[fiery nacelle]

when you send "/status" what does your context window say?

[fiery nacelle]

#self-promotion message

[plain moss]

Excuse me, everyone—has anyone used this model, “tencent/hy3-preview:free”? Does anyone know why I'm getting this error: "Agent failed before reply: Unknown model: tencent/hy3-preview:free.
Logs: openclaw logs --follow"?

[mortal delta]

Yeah , the model doesn't exsist... not being a smart ass.

I'm assuming you haven't used it in the past?

[polar knoll]

Oh dude that's sick

[fiery nacelle]

Been hitting a couiple blockers recently with these updates, so I Haven't quite been able to continue to phase 6: Cross-agent coordination design.

[polar knoll]

🦞 OpenClaw 2026.5.3-1 (2eae30e)
🧠 Model: steg/Qwen3.6-27B-UD-Q4_K_XL.gguf · 🔑 api-key (models.json)
📚 Context: ?/4.0k · 🧹 Compactions: 28
🧵 Session: agent:main:slack:direct: • updated just now
⚙️ Execution: direct · Runtime: OpenClaw Pi Default · Think: off
🪢 Queue: steer (depth 0)

Bit in my openclaw.json my context is 128k not 4k. I dunno what to do

[fiery nacelle]

check your config under that model, look for the context window size, what does it say?
Example:

            "contextTokens": 200000,
            "maxTokens": 8192```

nvm, you were ahead of me on that lol

[polar knoll]

That goes in models right?

[fiery nacelle]

I've got mine defined in the openclaw.json heres a more full example:

    "mode": "merge",
    "providers": {
      "lmstudio": {
        "baseUrl": "http://127.0.0.1:1234/v1",
        "apiKey": "LM_API_TOKEN",
        "api": "openai-completions",
        "models": [
          {
            "id": "qwen/qwen3.6-35b-a3b",
            "name": "Qwen3.6 35B A3B",
            "api": "openai-completions",
            "reasoning": true,
            "input": [
              "text",
              "image"
            ],
            "cost": {
              "input": 0,
              "output": 0,
              "cacheRead": 0,
              "cacheWrite": 0
            },
            "contextWindow": 262144,
            "contextTokens": 200000,
            "maxTokens": 8192```

[solid pecan]

If going local why not use DeepSeek v4 Flash now?

[fiery nacelle]

Not likely, but worth a shot.
Check /.openclaw/agents/main/sessions for any files ending in ".lock"

[rocky lagoon]

it was like 15x 2 days ago wtf

[polar knoll]

You know what dude I'm missing contexttokens. I have only window and max

Whats your rig specs?

[hoary hawk]

ive been out of the loop for a while, can someone catch me back up, is claude oauth still banned or can i use subscription again? ive been running on codex

[fiery nacelle]

Device name CORSAIRAI
Processor AMD RYZEN AI MAX+ 395 w/ Radeon 8060S (3.00 GHz)(5.1GHz boost)
Installed RAM 128 GB (31.6 GB usable)
Graphics card AMD Radeon(TM) 8060S Graphics (96 GB)
Storage 1.95 TB of 3.73 TB used

128GB unified memory. I can push VRAM from 1GB up to 96GB.

[fiery nacelle]

I think theres a round about way you can use the Claude cli through OpenClaw to use the sub. I'm not sure about the specifics.

[obsidian mural]

I just got mine in the mail today

[ebon atlas]

PFFFFFT i just use the Particle Collider to tell my agents what to do .

[thorny frost]

discord integration still died? damn

[fiery nacelle]

It's really good, but not like perfect. I recommend getting Fedora on it. Theres some AMD Halo Strix tool boxes for AI available. Better performance on any Linux distro as well. I dual boot windows11 and Fedora on mine. Build in fedora, bullshit around in windows.

[obsidian mural]

I am running ubuntu server on it

[obsidian mural]

oooooooo wait, I got the nimo pc version of this

[fiery nacelle]

Once i get a separate "Main" pc(or my old one back), I'm throwing a headless distro on mine.

[obsidian mural]

yeap, went headless, and accessing from main PC

[broken moon]

hey just curious
can i install openclaw on an old thinkpad
or even on that old thinkpad i should use a vps
will like the oldness or lower spec affect openclaw

[fiery nacelle]

still the same rig more or less. I'm loving mine. I've managed to get Nemotron 3 Super 120B working on mine on the windows side at 10-15tp/s. On fedora it was closer to 25tp/s.

[fiery nacelle]

there may be some performance impact, but not too much. Just dont expect to run any models locally.

[broken moon]

what is the performance impact?

like whats the difference running it on an old laptop vs old laptop + vps

[fiery nacelle]

Just like with most software, it boils down to the availble ram and the CPU.

[obsidian mural]

lol I just spent half the day getting this set up on ubuntu server...... LOL now you got me thinking i should start over ..... LOL

[broken moon]

I dont really get it

im so sorry

[obsidian mural]

not sure why you would want to do this ?

where are you looking to run your local models ?

[fiery nacelle]

how much RAM does your thinkpad have? And what OS?

[broken moon]

I have a thinkpad intel i5

[broken moon]

I was not really using a local model but using an api have you heard of wafer.ai waferpass?

[fiery nacelle]

You should be fine.

[broken moon]

I'm mostly planning on using it to do autonomous GTM for my app and maybe some social media automation

So I dont need vps? And when I setup like openclaw can I see like my entire screen moving and stuff autonomusly?

[obsidian mural]

do I like this idea, I will save that page and come back to it. I have some social media blog post content ideas I want to get going also

[fiery nacelle]

Specifying which one you had helps, cause "old thinkpad" can be REALLY OLD with only 512MB ram.

[broken moon]

im not sure but it has an intel inside core i5 vpro chip the sticker says

[fiery nacelle]

that should have at least 16GB or RAM if im not mistaken

[broken moon]

im not a hardware guy lol no clue

Also is it fine to have my business gmail signed in with chrome as I want to do mass outreach. i know a lot of people reccomend agentmail but is that really necessary?

[obsidian mural]

glad you said something, I didn't realize I was running 100% on CPU till now, I was wondering why all my models where running so slow ... LOL

[fiery nacelle]

if its windows, you can rightclick on the start menu, click "system" or "settings" and you should be able to find your hardware specs there.

[obsidian mural]

or run msinfo32

I am installing ROCm: as we speak ..... LOL opps

[broken moon]

it says I have a i5-5300U COU @ 2.30 Hz 2.30 GHz

installed ram: 8GB (7.7GB usable)

64-bit operating system and x64 based processor

this good?

[fiery nacelle]

you can adjust the unified memory allocation architecture (uma) in the BIOS.

[obsidian mural]

I didn't go into bios yet, but going to get this installed first

[obsidian mural]

fine for what ? running only openclaw with no local model, then yes

[fiery nacelle]

8GB is tight, but you should be fine running OpenClaw with a cloud agent.

[broken moon]

I cant imagine using local agents

[obsidian mural]

I am running a VPC for ust open claw and it runs ok -- Hetzner CX33 plan includes 4 vCPUs, 8 GB RAM, and 80 GB SSD storage

[broken moon]

Honestly yall should know about like firepass or wafer its basically unlimited turbo kimi or glm designed for agentic usage so extra fast

[fiery nacelle]

with 8GB, you might be able to squeeze a tiny <1B model, but it'd be SLOW, and go off the rails.

[broken moon]

on your main laptop or what. i dunno cause people online saying even a vps on main laptop is like so vulnerable

[obsidian mural]

should I go ahead and install Install ROCm: first

[fiery nacelle]

ROCm or Vulkan works fine. I think i have better luck with Vulkan personally.

[obsidian mural]

ooooo yeah, if you don't know what your doing then yes

for example, I have a FW rule in my VPS to block ALL incoming connections, and I have a ZT Nodes set up to access and root account is tied to a SSH cert

[broken moon]

Dang im not really into cyber security so.

you guys acc the most helpful in this server. I guess all the grifters have like left I remember a few months back this chat was like the most braindead ever people actually just riding on hype

[obsidian mural]

I am linux so / so, but learning about how I wanted to secure the system I did that first before doing anything else

[fiery nacelle]

If your concerned about securing you OpenClaw, you can also check out NemoClaw, which is kinda like Nvidias take on a "security wrapper" around OpenClaw.

[broken moon]

and then I guess like i saw like openclaw can be transfered to different like devices right? like idk

[obsidian mural]

I am not normally here, just hanging out while i get my nimo Pc up and running

[obsidian mural]

agreed !! you should start here to be honest

[broken moon]

honestly i have to be honest with you, im think im gonna use hermes agent its just set up so nicely

[fiery nacelle]

We still get a lot of brain rotted people here lol

[broken moon]

im not in the mood to customize openclaw for so long

[obsidian mural]

ROCm 7.2+ has improved a lot though from what I am reading , but it looks on ubuntu 7.1 is where its at right now

I was thinking of getting roc going on 7.1 and figured updating wouldn't be that far away

[ebon horizon]

https://x.com/claudedevs/status/2052064938840228237?s=46&t=7i6pmIt9zg84GbNsqWBcAA

[mystic orchid]

@claudedevs via Twitter

ClaudeDevs (@ClaudeDevs)

Usage limits are up, effective today we're:
︀︀
︀︀1) Doubling Claude Code's 5-hour limits for Pro, Max, Team and seat-based Enterprise plans
︀︀
︀︀2) Removing peak hours limit reduction on Claude Code for Pro and Max plans
︀︀
︀︀3) Substantially raising our API rate limits for Opus models

Quoting Claude (@claudeai)
︀
We’ve agreed to a partnership with @SpaceX that will substantially increase our compute capacity.
︀︀
︀︀This, along with our other recent compute deals, means that we’ve been able to increase our usage limits for Claude Code and the Claude API.

**💬 1.2K 🔁 2.4K ❤️ 30.8K 👁️ 2.58M **

FxTwitter · May 6, 2026, 4:36 PM

[ebon horizon]

You got be kidding me aha. After changing to codex

[broken moon]

also since im installing openclaw should I like then make sure my computer never shuts down? wont that impact openclaw

like you know how computers dim after 5 minutes and then shuts down

[ebon horizon]

Yeah turn off all sleep setting

Still keep display sleep on that doesnt matter. Save some power

[obsidian mural]

actually for him he might be better off starting with easyclaw

[fiery nacelle]

that depends if you need your agent running 24/7. You can shut it down at anytime when you arent using it.

[broken moon]

I get it thanks

[broken moon]

nah im gonna use hermes its already set up pretty much

I agree openclaw has a higher potential value though if tinkered on

anyone using like api passes like unlimited glm or kimi per week etc?

cant afford normal api rates ngl

[fiery nacelle]

check out Openrouter

[obsidian mural]

if you like to tinker, that basics of hardening the VPC of even your own install isn't that hard

[ebon horizon]

Never use api lol. Always get subscriptions

[fiery nacelle]

They have a free tier that does have some usage limits, but it works.

And yea, subs over per token API costs anyday if you're going the "cloud" model route.

[broken moon]

even subscriptions arent good

[ebon horizon]

Can you afford the $100us plans?

[broken moon]

codex? is that even good rn

[ebon horizon]

Id go claude for usage considering they just upped it

[fiery nacelle]

OpenCode-GO offers a $10/m sub with access to great models.

[broken moon]

bro i swear claude banned subscriptions for openclaw what

I have claude code $200max but use it for my dev work

[ebon horizon]

Nah you can get it working

[leaden vector]

nah u can use it through Claude CLI

[ebon horizon]

Route through CLI

[fiery nacelle]

not just openclaw, but all 3rd party.
And yea, route through cli

[broken moon]

how tf? so subscription works now wtf

[ebon horizon]

You have to add something in your config

[ebon horizon]

I had no dramas with CLI. People say its nerfed (no prompt caching) but other than that it worked a charm for me

[ebon horizon]

Ill send you the tweet to get it going

[fiery nacelle]

Yes and No.
If you use the 0Auth directly in any 3rd party app = ban
You can hook into the Claude-CLI and use your sub though. Just not DIRECTLY in OpenClaw.

[hoary hawk]

why even allow usage through cli and not Oauth? mad annoying

[ebon horizon]

Wrong

[broken moon]

bro this sounds so sketchy r u sure anthropic is allowing this

[ebon horizon]

Official openclaw docs say oauth is fine as long as its through CLI

[fiery nacelle]

jsut not directly, like i stated. The CLI is allowed.

[ebon horizon]

Yes boris even stated it (claude dev) on X

[fiery nacelle]

Like you cant 0Auth for a Anthropic/Opus model as your agents model, but you can access your 0Auth through Claude-CLI

[broken moon]

interesting, but honestly cc the rate limits are horrendous though with opus

[ebon horizon]

https://x.com/steipete/status/2046685973233189375?s=46&t=7i6pmIt9zg84GbNsqWBcAA

[broken moon]

the recent starlink one didn't really 2x the limits, the weekly limit is also the same btw

[mystic orchid]

@steipete via Twitter

Peter Steinberger 🦞 (@steipete)

Since this is blowing up on hacker news.
︀︀
︀︀Boris said that CLI usage is allowed. Thus we added support for it, only to find out that we are still blocked there. It is trival to work around with a few renames, but I don't wanna play that game. So it's in a weird limbo where cli use should work in theory but doesn’t in practice.
︀︀
︀︀x.com/bcherny/status/2041035127430754686

Quoting Dan McAteer (@daniel_mac8)
︀
Anthropic allows OpenClaw usage again.
︀︀
︀︀From @openclaw docs.

**💬 152 🔁 187 ❤️ 2.8K 👁️ 640.7K **

FxTwitter · Apr 21, 2026, 8:22 PM

[ebon horizon]

They just upped it bro. Check the claude tweet above lol

[broken moon]

nope not the weekly limits, thats the thing that matters.

you can prompt 2x more yeah but still hit weekly faster

[ebon horizon]

I followed this and added this code into my openclaw.json and it worked a charm

https://x.com/ziwenxu_/status/2046679352977580437?s=46&t=7i6pmIt9zg84GbNsqWBcAA

[mystic orchid]

@ziwenxu_ via Twitter

Ziwen (@ziwenxu_)

Anthropic just gave OpenClaw the green light
︀︀
︀︀Anthropic staff confirmed that CLI-based usage is sanctioned.
︀︀
︀︀We’ve successfully bridged the Claude Code CLI directly into the OpenClaw gateway, letting you run your entire agent fleet on your Pro/Max plan and no additional API tokens required.
︀︀
︀︀The Implementation:
︀︀
︀︀- Auth: Log in to Claude Code locally: claude auth login
︀︀
︀︀- Path: Grab your binary location: which claude
︀︀
︀︀- Bridge: Drop the path into your openclaw.json under agents.defaults:
︀︀"cliBackends": {
︀︀ "claude-cli": {
︀︀ "args": [
︀︀ "-p",
︀︀ "--output-format",
︀︀ "json"
︀︀ ],…

▶ Play video

FxTwitter · Apr 21, 2026, 7:55 PM

[signal rock]

Hi everyone,

Would you ever trust an AI agent to manage an investment account? I’m building a product in this space and trying to understand how people are using LLMs for finance today, and what it would take to trust an agent to manage your money.

I made a short 1-minute survey here: https://forms.gle/DupUARr6DLtFmDup6

It would be great to get some input from all of you to help us shape our product!

[ebon horizon]

Im using codex now anyhow, usage is more generous but only because 5x is currently 10x for a limited time

[leaden vector]

Is the latest openclaw update stable yet?

[broken moon]

just shout the company link i wanna see this

[broken moon]

you sure codex is good even with 5.5 last time i heard codex kept bugging out only opus and sonnet were good

[signal rock]

We are still building in stealth right now, can share more data later 🙂

[ebon horizon]

OC fixed all gpt models

[broken moon]

deffo first time founder building in stealth rip

[ebon horizon]

I prefer the claude code app over codex app though ngl

[mortal delta]

Gosh i can't beleive no one has thought to ask all the LLM investment advice....

[fiery nacelle]

Codex has been better than Claude recently in my experiences.

[ebon horizon]

Claude app was so fckin clean.

[leaden vector]

I run my OC main agent with GPT 5.5 high, fast and it get shit done

[ebon horizon]

Do you leverage codex through OC at all?

If i have a long ass coding task ill spawn a codex session inside discord and go from there

[leaden vector]

i mean its using the codex cli with oauth isnt it

[stone saddle]

I totally agree with you.

[broken moon]

ye claude code desktop is goated

i think only codex has better computer use thats it

[fiery nacelle]

It's funny, i dont even use Codex in OC, only to fix it. Claude will lie about not being able to edit files for the same use case.

[leaden vector]

and the best image gen model ever built right in

[broken moon]

i dont need gen, just coding but ye i do see the vision for ecommerce stores

im p sure gpt 5.5 has the highest hallucinate rate rn tho like 60%

[leaden vector]

you can improve your coding UI/UX, by using image gen to generate the mockup first

[broken moon]

I do think its good for adversarial code review though

[broken moon]

really but im doing mostly infra and backend devops

[leaden vector]

Terminal-Bench 2.0, with a score of 82.7% with GPT 5.5 compared to Opus's 69.4%

[broken moon]

um no opus 4.7 has 79.8%

[fiery nacelle]

good alternative would be GLM. Z.ai has affordable plans as well

[ebon horizon]

Bro if youve got the claude code 20x plan why you even worrying about other llms?

[broken moon]

no i have claude max $200 to code. probably in openclaw imma use the glm 5.1turbo for waferpass

[ebon horizon]

I tried minimax, glm.. after claude got swiped from OC. All awful

[broken moon]

dude i max out claude code 20x so fast

[ebon horizon]

Bro how

[fiery nacelle]

shit yea nvm. Thats a better better alternative. lol

[ebon horizon]

Genuinely. Thats some insane shit

[broken moon]

actually doing real architecture and coding lol

[leaden vector]

I used to max out claude max 20X all the time and i dont with GPT 5.5

[broken moon]

I always max it out every week

[ebon horizon]

What you building

[mortal delta]

but you can't use it as an agent.

[broken moon]

revops agents

[fiery nacelle]

maaan you came in like "idk hardware" and your sitting there doing real work. I thought you were just a wee pleb. lol

[broken moon]

nah im so bad at hardware you dont even know

only good at coding

[ebon horizon]

When does your claude sub end?

Try the codex 20x for a month @broken moon

[marble ledge]

Just a fella with some self-awareness.

[obsidian mural]

@fiery nacelle GPU configuration - UMA_specified to 1G or just leave the default to auto ?

[broken moon]

but idk about codex ive heard bad stuff about it

I always use codex adversarial review within claude code though

[ebon horizon]

The issue with reviewing code across models is they’ll always find bugs no matter what

[broken moon]

My current process is always opus 4.7 1M max and then review PR code with gpt 5.5xhigh and consolidate reviews before shipping within claude code

I have found this to be the best format on par with an actual senior engineer

[lost void]

Try opencode.ai/go $5 first 30 days then $10/m all open models add credits in zen to switch to all other models and deplete your bank account fast

[ebon horizon]

Like you can do a full comprehensive audit using max models. Once its cleaned it all. Run another audit and itll find some more

[broken moon]

lol breh. If i want unlimited creds I just use firepass and waferpass

[ebon horizon]

Endless loop of debugging

[broken moon]

for shipping production code I feel you cant use anything other than opus 4.7 or gpt 5.5

[fiery nacelle]

I've read some stuff about setting it to the lowest possible in Linux, cause it just treats memory as memory vs windows with segregated memory. You can try that out since you're headless anyway. If it starts snowing though, dont panic, just reboot any try the opposite end. I'll try to dig up some resources for you.

[obsidian mural]

try leaving as auto for now you mean ? claude is suggeting leaving on auto, or set to lowest setting

[lost void]

What are you shipping rockets 🚀?

[fiery nacelle]

sorry, the 1GB option. Theres a way to cram it down further to like 512MB to allow Linux to use like 120GB for the VRAM, but i personally havent explored that

[obsidian mural]

ahh ok, let me go set that

[lost void]

Skills help

[broken moon]

no bro b2b and enterprise agents

one little bug and failure is detrimental at scale

[fiery nacelle]

Here's what I was talking about, allowing Linux to use the max available for VRAM:

Edit /etc/default/grub (or a file in /etc/default/grub.d/) to include the following parameters in GRUB_CMDLINE_LINUX_DEFAULT:

amd_iommu=off: Disables IOMMU, providing approximately 6% memory bandwidth improvement for LLM workloads.  Enable only if you need GPU passthrough.
amdgpu.gttsize=131072: Sets the GTT to 128 GiB (131072 MiB), exposing the full unified memory pool to the GPU driver.  For desktop use, reduce this to ~105 GiB (27648000 pages) to reserve RAM for the OS.
ttm.pages_limit=33554432: Allows pinning the full 128 GiB pool for GPU operations by setting the page limit to 128 GiB in 4K pages. 
transparent_hugepage=always and numa_balancing=disable: Recommended for consistent performance with large models. ```

[obsidian mural]

I am trying it now

[obsidian mural]

good news I am not doing any GPU passthrough

[fiery nacelle]

SAME!! hah

[lost void]

for pure volume on a fast cheap model like V4 Flash, OpenCode Go rapes them on price

[broken moon]

v4 flash is buns though

only good models outside of the big ones are kimi k2.6 and glm 5.1

[obsidian mural]

I am going to try and set this

GRUB_CMDLINE_LINUX="amd_iommu=off amdgpu.gttsize=131072 ttm.pages_limit=33554432 transparent_hugepage=always numa_balancing=disable"

[broken moon]

even minimax is not up to par too many tool call failures

[obsidian mural]

this is what I had it set up - GRUB_CMDLINE_LINUX="iommu=pt amdgpu.gttsize=126976 ttm.pages_limit=32505856"

going to update to -- GRUB_CMDLINE_LINUX="amd_iommu=off amdgpu.gttsize=131072 ttm.pages_limit=33554432 transparent_hugepage=always numa_balancing=disable"

[fiery nacelle]

Yea once you have the BIOS set to 1GB for VRAM, you can then push that remaining 127GB however you want within linux. YMMV though.

[obsidian mural]

exactly what I did ! updating the line now

[fiery nacelle]

I had a few "snowing" instances when I was pushing it. But again I'm GUI, not headless, and no real issues came from it. Day 2 of owning it and seeing artifacts all over the screen made me shit bricks for a second. lmoa

[obsidian mural]

haha I bet ! i will you know how it goes

[broken moon]

im looking at opencode right now. what defines like a request? I see glm 5.1 has 880 requests? is a request just like a function call like a prompt

[fiery nacelle]

Best part is that I came home from work seeing that while having my agent work on a few projects. I was like "uhhhhh whats going on, am i cooked?"

[candid radish]

Evenin Peoples that know a hell of a lot more than me! May I ask for your more qualified opinions of the current stable update, or are we still holding at .24? Thanks in advance.

[lost void]

Reasoning uses more like 30x for xHigh or 9x for V4 pro xHigh

[fiery nacelle]

Still hold. 5.4 was alright. Had no luck with 5.6 without custom patch work myself. Probably safe to hold till the "LTS" release comes out later this month.

[vivid pond]

Dude are you guys talking about my baby deepseek its so OP rn 75% off tokens for flash

[obsidian mural]

haha I bet ! appreicate some of the insight today, I am a windows guru , not linux guru but with the power of claude, I am now a bad ass linux administrator well only if I know where to tell it to look ..... LOL

[marble ledge]

I'm currently on 5.6, and have a clanker working on an arduino project right now. Doubt I am anymore qualified than you are though. I'm a LARPer tech support monkey here.

[fiery nacelle]

I've "dabbled" with linux over the past 20 years. I keep finding myself getting fed up with Windows, swapping to a linux distro for some time, until theres that 1 thing i need that only works in Windows, and then im back at it. lol

My first linux disrto i messed with was "BackTrack" which evolved into what Kali is today.

[vivid pond]

wait claw bots im real im a meat bag talk to me

[fiery nacelle]

beep boop

[obsidian mural]

yeap !! same I still remeber it as backtrack

[obsidian mural]

I am a windows engineer by trade and starting to run things headless next to it for more linux experience

[warm zealot]

windows engineer, ooh cool

[vivid pond]

thank you i appreciate it may i inject something not SQL but an idea

[robust sedge]

I’m a 4.23 truther 🤌

[obsidian mural]

pretty sure that equals sarcasm ..... hgahah

[fiery nacelle]

It's one of those "oh, you must REALLY hate yourself" bits

[warm zealot]

well I was being serious

but okay....

[obsidian mural]

hahahah I mean, its what i know and all my clients use ....... LOL

[warm zealot]

if its funnier the other way than it can mean that, I permit it

[ruby crane]

some people call me a Wingineer

[obsidian mural]

lol well in this linux based communites , it usually means sacrasm .... LOL

[warm zealot]

I mean, linux is better obviously

but I got no hate for windows

[fiery nacelle]

I mean i'm not gonna talk down on windows. It handles that on its own perfectly fine. lol

[warm zealot]

except they basically patched out the way to theme the thing

I miss when I could totally change the way it looks

[fiery nacelle]

I'll be the mad man here and say I miss windows Vista.

[warm zealot]

oh and deprecation of hardware with software locks that could be bypassed to install the newer O.S. and it still worked fine...

[ruby crane]

can you smell burning toast rn?

[obsidian mural]

@fiery nacelle here is my update, just verified

Just spent the day unlocking my AMD Ryzen AI MAX+ 395 (128GB) on Ubuntu 26.04 Server.
Before:

ROCm not installed — Ollama running 100% on CPU
BIOS statically set to auto assusming it was allocating 64GB to iGPU
~0GB usable for GPU inference

After:

ROCm installed and working
IOMMU disabled for ~6% bandwidth gain
Full 122GB unified memory pool available to GPU
Ollama confirmed GPU accelerated
llama3.3:70b (42GB) now runs entirely on GPU with room to spare

Key kernel params that made the difference:

amd_iommu=off amdgpu.gttsize=131072 ttm.pages_limit=33554432 transparent_hugepage=always numa_balancing=disable

[warm zealot]

vista was pretty good

[lost void]

via OpenClaw? Or just pi.dev or a coding harness ?

[ionic wren]

https://giphy.com/gifs/ac7MA7r5IMYda

[fiery nacelle]

i always get shit for it, but loved how customizable it was.

[broken moon]

nah not using openclaw to code dont trust it.

using openclaw just for GTM

[fiery nacelle]

and that start menu was peak windows design.

[vivid pond]

are you real person?????

[warm zealot]

right

[lost void]

Try straight pi.dev for agentic coding

[warm zealot]

It was not bad fam

[ruby crane]

https://klipy.com/gifs/what-the-fuck-wtf-27--k01KR08HT25QYPC4AXFR5HQT1W5

[warm zealot]

didnt spy on you like modern windows does, let you modify things

[ruby crane]

what is happening in here

[ionic wren]

what...is...windows...engineer?

[warm zealot]

someone that works for windows I would assume

[vivid pond]

im trying to get one person in that channel to acknowledge my existence but there all talking about window 😭

[crisp egret]

any way I can set up sound alerts when my bot's done?

[fiery nacelle]

bep boop

[broken moon]

bro wdym are you saying you trust openclaw for business code

[vivid pond]

Pleaaaaaasssseeeee 😭

[upbeat token]

I trust it with my life

[lost void]

Sign up for email status messages for Crippled Claude and you will think twice relying on it for anything

[vivid pond]

no i never said that i dont even use my open claw to code period

[obsidian mural]

I mean I like to look out the window

[broken moon]

yeah i mean i like the simple architecture of pi, I just like more hands on

[vivid pond]

i love windows toooo ones that open and close 😢

[broken moon]

dropshipper in name, credbility 0

[vivid pond]

hes dropship aura farming

[upbeat token]

[vivid pond]

i am curious what do you drop ship out of ?

[lost void]

Fully customizable pi.dev
Themes, skills, or just add whatever features you want etc etc
Give it three hands lol

[broken moon]

hes pre-revenue like 99% of dropshippers

[broken moon]

whats the difference from just claude code harness with integrated skills and etc

[vivid pond]

bro arent we all were all pre shipped product rn still figuring out how to get supabase hooked up

[upbeat token]

Sounds like you have some resentment

[vivid pond]

so you made this or just advertising?

[broken moon]

bro he is not karpathy

or Mario Zechner

[vivid pond]

im on a software one my friends made its pretty heat but arbritrage software

[lost void]

It’s not always down when you need it
It’s not a walled garden that traps you inside the void

Migrate away from Claude and save your soul

[broken moon]

claude is never down tho

[vivid pond]

i do not know those people should i??

[broken moon]

also pi literally uses api which in turn would be opus 4.7 or gpt 5.5

youre relying on the platforms already why not claude code or codex

[vivid pond]

fahhhhh we love apis i feed them to my tools for breakfast she does a good job give her a little api key with 5 bucks on it

[broken moon]

ok ur deffo in highschool or smth

[lost void]

It literally down right now 🤣🤷

https://status.claude.com/

[broken moon]

so then if its down then like i cant even use opus in pi agent then

[vivid pond]

bro we are not even in highschool

[broken moon]

whats the difference

[broken moon]

ok well the minimum age to use discord is 14 then ur gonna get banned

[upbeat token]

he's aggro right now allow him

[lost void]

Exactly why you should not be using Crippled Claude especially for building rockets 🚀

[vivid pond]

never said i wasnt 14?????? why we trolling

[broken moon]

i also use codex gpt 5.5xhigh from claude code though lol

[broken moon]

isnt highschool like starting 14?

bro dropped out of elementary danm

[warm zealot]

https://tenor.com/view/i-need-an-adult-gif-1200923972209450322

[upbeat token]

No man it's always up

[vivid pond]

something like that some people 15 some 13

[lost void]

The model doesn’t matter if your customizing your pi.dev workflows then your always golden switch models shouldn’t matter just time and token loss

[vivid pond]

No but on a serious note i want to talk to yall about something a little side project ive been working a little somthin somethin if you want to hear of course cant bless the ears of the non courageus

[ionic wren]

https://giphy.com/gifs/confused-futurama-suspicious-ANbD1CCdA3iI8