general

[arctic trench]

i forget which server i'm in

[brisk adder]

saw that article that's mental they're bundling essentially new age mining bots

[tepid kestrel]

silent invasion

i actually wonder if its the google gemini flash ai model

[brisk adder]

best pet

[turbid kelp]

oauth of course, why are you using API key anyway with Claude?

[tepid kestrel]

https://tenor.com/view/money-cash-cash-money-make-it-rain-gun-gif-9859470

[cerulean flare]

uh are you able to use oAuth with anthropic subscription again? Remember the big announcement that dropped early april and they cut us all off from using oAuth? Are you still able to use oAuth with anthropic on openclaw?

[turbid kelp]

no, but you said it's not practical, so why do you use it?

Why not use something else? Claude can't be that good.

[cerulean flare]

oh, I just don't use openclaw much and I have a complex setup where I use OpenAI oAuth and also MiniMax for other tasks

I'm talking more in the general sense

[turbid kelp]

ah ok, I'm using MiniMax a lot

[cerulean flare]

Ya MiniMax is better than Haiku in my experience. it's so cheap and solid

I don't trust it as the openclaw brain/main model tho

[ionic wren]

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

[upbeat token]

@solid pecan still in the 4.23 club?

[arctic trench]

There he is

Was starting to wonder if you had me muted

[ionic wren]

maybe... you need to be....

https://giphy.com/gifs/bywassies-wassies-pamcordovamx-lUedOXZaqNJoqHkmFP

[golden marsh]

Hes here trust me. He sees everything

[frozen zealot]

overvalued according to what? p/e ratio? liquidity? eps? forward prediction? market cap? All of the frontier labs are private except for meta. You literally don't get to see their books because they simply don't have to disclose. You have no metric. And that's what I'm talking about. On either side of the prediction there's nothing to stand on

[brisk adder]

he only unmutes baddies so watch out

[arctic trench]

He's been known to twitch when i say the word Hermes

also foams at the mouth

grinds his teeth

chews on his keyboard

calls mr chinese man and asks for advice

[frozen zealot]

Hermes... because i have to test your theory

[arctic trench]

https://tenor.com/view/i'll-wait-patiently-oh-i-will-wait-oh-yes-i-will-man-pulling-out-camp-chair-to-wait-gif-18210748690125729645

[tepid kestrel]

https://tenor.com/view/clawdbot-moltbot-openclaw-open-claw-molt-bot-gif-5803491012301817622

[arctic trench]

hmm, he must be stuck in the snow or somthing

[marble ledge]

On main, using telegram, it's snappy as hell, even with reasoning on high. YMMV.

[solemn sphinx]

Been using Oauth OpenAI openai-codex/gpt-5.5 on my claw. Is the token spend all flowing throuhg the OAuth or is it additoional ?

[arctic trench]

Any different than discord?

[marble ledge]

It should use your plan.

[marble ledge]

Gosh, you guys are going to force me to add the clanker to Discord aren't you? lol

[arctic trench]

Only one that is charging additional above the sub for harness use is anthropic

[arctic trench]

why not?

[solemn sphinx]

Ok just wanted to make sure. Thats awesome. GPT 5.5 has been pretty impressive

[arctic trench]

I was suprised cus i thought claude was the king. Turns out Codex is busting some moves

[marble ledge]

...alright...I'm gonna set it up. I don't like Discord...but alas.

[arctic trench]

It has built in slash commands

[solemn sphinx]

However, the suggestion for using Codex Harness I wasnt a fan of. It seemed too slow.

[arctic trench]

It's hard to use with openclaw

[marble ledge]

Codex harness has been good for me lately. But in the past, I wasn't a fan.

[golden marsh]

Set it up seperatly, use ur Main with Pi and 1 Harnes Subagent. That way u have both and u can swap between them.

[arctic trench]

Ya, i never really used it until 5.3. I love it now

[shut bloom]

i just built a mcp into my pc for claude

i feel happy now

[vivid pond]

helloooooo

[brisk adder]

nice i need to do similar

[shut bloom]

to any of my spanish baddies just know ur music is goated

[lean tangle]

Issue with attaching .zip files to discord dm’s still not fixed (in a release)?

[toxic elk]

I use it the same way; codex harnes is used for writing code and test scripts

[zealous lagoon]

There is a huge industry dedicated to valuing even these companies and they do “disclose” many things, either by public statements about revenues or by investing with third parties or canvassing third party investment. But I wasn’t talking specifically about the frontier companies just the whole peripheral market in general. I’m not going to go in to the specifics of company valuation because the entire market can’t even agree on who’s right. If they could there wouldn’t be such huge volumes changing hands. But the market isn’t by default “smarter than you” just because there are a lot of smart people on the market. Companies can languish undervalued or fly high grossly overvalued for years and years. It can be plainly obvious yet incentives to correct it are weak. Corrections in both directions tend to come quickly and make or break a lot of people but the timing is extremely difficult to predict.

[toxic elk]

Because the sandbox mode in codex harness is incomplete, I created a separate codex-sandbox agent to run tests within the sandbox

[arctic trench]

https://tenor.com/view/jerry-seinfeld-i-like-the-way-you-think-gif-14890068

[frozen zealot]

that's just a text without numbers, like tesla is 400x of evaluation( https://stockanalysis.com/stocks/tsla/statistics/ ), that's a solid indication of over evaluation and thus if you say the company is in a bubble, i can believe you. But take microsoft, one of the biggest investor of OpenAI as example, they're at 25x, doesn't show OpenAI is skewing their evaluation. That's why when I say there's no indication, there's literally none

[fiery nacelle]

Yea, theres a post on X with Anthropic confirming that CLI access is allowed for 0Auth.
Im too lazy to pull it up.

[desert escarp]

Are we stable with 5.6 /5.7? Feel compelled to ask after last update 😵

[zealous lagoon]

Those are two excellent examples and on the face of it I’d agree with your conclusion. But what is often missed is that revenue in one area comes from investment in another area that might in fact be misguided. When one bubble pops it tends to start a chain reaction. I don’t know how old you are but I’ve been a market professional through several bubbles and financial crises where we all knew “this is crazy, it’s a total misprice” and we were right, but the market didn’t correct itself before half of us already got fired. My style of participation is a couple of layers removed from “buy and sell” and I haven’t really been an investor since mid 2022 but i have been pretty successful by investing too late and selling too early. And also looking on like a chump as everyone else gets rich off Tesla.

[marble ledge]

Alright you sickos. I added a clanker to test in Discord.

[frozen zealot]

i don't think you have the correct definition of bubble then. a "bubble" is indicated by two things - p(rice)/e(valuation) ratio and price increase rate. both are indication of market health (note it is not an indication of company health because they're not the same). When price spike, as in stock price spike, you get a 200+% of a stock price increase and p/e ratio also spike, that's when you are looking at a "bubble" because real value of the company didn't change, only the stock price changed. Now take OpenAI as a test case, if Microsoft was sinking more and more money into the company, that would mean liquidity of the company changes, and risk profile changes. in term, it would lower the evaluation. Yet from pure health analysis, that's not the case. So as a counter example it makes the assertion "AI is a bubble" collapse

[marble ledge]

So, I added the clanker to discord. Seems snappy to me. Last test took the bot 7sec to respond on /think medium

[frozen zealot]

honestly i think people throw around specific terms such as "bubble" based on their mood and vibe without studying finance, and that's a shame

[zealous lagoon]

So now we gettin’ technical? How about “house of cards”? 😉
I’m not at home so can’t pull up the examples easily but someone posted a revenue chart for DDOG and it was spectacular. I then pulled up a price chart and basically if you bought mid 2022 you’ve either been stopped out or hugely underwater for 4 years.

Maybe because words mean something but “bubble” isn’t a scientific term so it can mean a lot of things. I’ve been an institutional financial professional for over thirty years and we all use words like that quite loosely.

[vale imp]

I started from zero, tested a lot, and finally began to see results. Most beginners think it’s luck, but it’s just strategy and testing. If you’re stuck with no sales, you’re probably missing something simple.

[dreamy briar]

My Openclaw seems to have improved a lot

[crimson brook]

Because (afaik) bubble isn't a term of art the way 'valuation' is. There is certainly 'irrational exuberance' in the AI market right now. Does it mean the core companies are going to magically die? No, but the thousand 'AI' startups that are frothing about with nothing more than a prompt and a dream are going to hit a very real and painful wall.

[merry helm]

Been surprised the last few years the funding thrown at companies that basically sell a wrapper around AI to make a better workflow.

[frozen zealot]

are you talking about startups? because like most startups die. what you see in the market has an inherit survivor bias

[crimson brook]

TBH the same is true of Arxiv papers. The number of them where the result depends on the specific version of ChatGPT they built their paper around is painful.

[zealous lagoon]

Give it 2-3 years. You’ll see. Meanwhile just get rich off the fact that the markets are irrational and make sure you don’t bet the farm and lose everything when reality hits. Because THAT is the time to be investing.

[sour dawn]

Does anyone know if I can use openai codex to automate my workflow like using a web program to create a clip and upload it to a google drive?

[frozen zealot]

i think the only real AI player in the market right now is palantir, which does have a less than healthy 155x p/e ratio

[ionic wren]

Are they AI player or are they defense

[frozen zealot]

both

[shut bloom]

its brilliant

[frozen zealot]

they basically make something like openclaw, an AI harness around frontier models

[shut bloom]

bro qbit ai is crazy good

12 Million context feels like im god its crazy

[crimson brook]

It's not just startups. When every single company has to make a huge pitch about AI being key to their product line, something's fucky.

There's going to be a valuation compression when the froth dies down. For companies with good balance sheets like Microsoft, there's not much to compress. For MANY other companies, it will be painful. And they'll cut jobs to regain healthier balance sheets, and that'll fuck up the economy more, and the market will reel for a while, and the companies who are fundamentally healthy and delivering value will survive, and we'll have a new crop of companies to toxically acronymize as the places to work for a few years. It's the 🎶 cycle of liiiiife!🎶

[shut bloom]

@arctic trench bro this model is fire

not as good as opus 4.6 they defo made that seem better but logic wise it understand my massive docs

[frozen zealot]

that's the thing, for institution investors and banks there are pretty strong rules, which leaves private credit, but i don't really care if those guys lose their shirts, they're essentially aristocrats and high worth individuals

[crimson brook]

There are... but we know finance folks love to skirt the rules as much as possible.

Yes, if you're invested in an S&P500 tracking fund, you're already essentially investing in the safer half of the market. But there's a lot more money out there chasing other stuff, and getting in danger. And when things go wrong, there tends to be some contagion.

[mortal delta]

LLM request rejected: The 'gpt-5.5-instant' model is not supported when using Codex with a ChatGPT account.

dickheads

LLM request rejected: The 'gpt-5.5-pro' model is not supported when using Codex with a ChatGPT account.

Pro isn't either! Oh well.

[frozen zealot]

the contagion is they would be selling their institutional holdings when they are way in the red, but that's just buying opportunity for me personally

[zealous lagoon]

Wat how

[zealous lagoon]

True. Neither of those. Nor the realtime model.

[zealous lagoon]

The rules are there to prevent fraud not irrational behaviour.

Most investors are just employees. The incentives are misaligned.

[frozen zealot]

not quite, they get quite a bit benefits from their employers, such as discounts and loans

[shut bloom]

I dmed Alex himself

[mortal delta]

Yeah, jsut not exposed to Codex

[oblique root]

whens the last time codex reset rate limits? lol

[fierce orchid]

Two or three days ago?

[full granite]

Any OpenAI claw users? Would love some help testing out latest main, as we're moving to using the codex harness by default instead of pi.

Out of the box, you'll experience more reliable tool calling behavior & messaging, less token usage due to deffered searchable dynamic tools, and personality improvements.

But I need testers to help!

[calm solar]

Hey, Geo is looking for a founding SWE based in the U.S. If anyone knows anybody super cracked, DM

https://www.geodo.ai/

[ruby crane]

I can help with that for sure

I'm in containers so unless there's a branched container build, I guess I'll need to build my own

no worries, I'll get on that now

[violet mulch]

i just realized having openclaw scan pages takes a lot of context

lots of usage

[ruby crane]

once the image is built, should i just switch to the codex harness manually?

[zealous lagoon]

ah yeah depending what you're doing, it might be a good idea to use a multimodal model or subagent that's very cheap. They'll do OCR or description for your mainbot

[ruby crane]

or is there some migration path? I see doctor already flags openai in pi as a warning

[violet mulch]

gonna use chatgpt agent mode from now on for these sort of tasks

spent about 50% of my usage on a single scan 😭

[zealous lagoon]

it's a good rule to have in general. Any large data intake that isn't needed for the entire session should go to a subagent, and OCR is better going to a cheaper model if you have one.

[toxic elk]

Could you explain the intention behind doing it this way? If the default is set to harness, wouldn't there be no difference from using the codex cli directly?

[steel vector]

hey clawers

[toxic elk]

Furthermore, will this have an impact on users who are not using GPT as their "brain"?

[solid pecan]

I am also interested in knowing the the motivation behind this

[golden marsh]

In principle, you should be able to use almost anything — the harness is just the car, not the engine.

[marble ledge]

Been trying to stay up to date on main, been using the codex harness last couple of days. It's been awesome thus far.

[ruby crane]

built the docker image locally using podman

then realised I'm building on mac and got arm64 build lmao. I am not a smart man. probably better building CI in my local gitea rather than cross-build sigh

[marble ledge]

We derp together.

[frozen zealot]

gemma 4

[marble ledge]

Shouldn't. They just use the default Pi harness like usual.

Unless...they are expanding to have the codex harness run other models...which would be kinda neat.

[toxic elk]

So if other models are used, will it still automatically switch to the pi harness? Because currently, the condition for using the codex harness is specifically using a GPT model

[marble ledge]

My experience thus far has been that when I switch to a different model or it switches to the fallback(minimax) it switches to the pi harness. Which is how we would expect it to behave, at least I'd hope. haha

It's been like that since I've tried it on and off, for bit.

I tried it when they first added it too, and it behaved that way.

[toxic elk]

Currently, the primary model I use is GPT. When switching between openai-codex/gpt and openai/gpt, it did indeed switch between the two, but I haven't touched it since. Subsequent updates to the codex harness require additional configuration, and I'm not sure if there will be any limitations

[toxic elk]

Anyway, this is the only thing I mainly want to understand, looking forward to receiving a reply

[marble ledge]

When I first tried it, like a month ago or something, it felt a lot more rough and less native. But it feels very native on OC now. They've done a lot of good work on it.

[fickle nebula]

Anyone else been having issues with Openclaw just responding with an issue occurred and to start a new chat/session but it just keeps happening? I feel like I need to delete it all and start over at this point, just constant issues for weeks.

Do you recommend this over their other model choices?

[toxic elk]

I've encountered this before, but it hasn't happened again since the update. If you have any extra memory features like Active Memory enabled, please try turning them off

[fickle nebula]

? Active Memory feature?

[toxic elk]

https://docs.openclaw.ai/concepts/active-memory

[fickle nebula]

I do not have that setup

[toxic elk]

So what version of OpenClaw are you using?

[fickle nebula]

2026.5.5

[toxic elk]

Do you have a Codex subscription?

[fickle nebula]

But this has been happening across updates as well.

I do not. I use either local Ollama or NVIDIA API.

[cold jacinth]

will that affect the use of other models on sub agents?

[toxic elk]

That would be a bit troublesome; I used the codex cli to help me delete those problematic sessions

Have you checked if the context exceeds the limit?

[fickle nebula]

It was working amazing back when Minimax 2.5 or w/e came out.

I've been using different channels and such but no matter what session it is on it responds the same.

[toxic elk]

Well, that part is a bit different for me; I've only had this happen during Discord conversations

[fickle nebula]

Baron von Slender
APP
 — 10:02 PM
⚠️ Something went wrong while processing your request. Please try again, or use /new to start a fresh session.

[toxic elk]

Yes, that's exactly the error I'm encountering

[fickle nebula]

This just keeps happening even when using /new or going to a new channel in the server. Anything.

[toxic elk]

Yes, it cannot be resolved unless the session where the error occurred is deleted

[fickle nebula]

The second agent is doing the same thing.

So do I just delete all the sessions from the files?

[toxic elk]

That's about it. I don't remember if codex took any other actions, so you might want to make a backup before giving it a try.

However, this isn't a permanent fix; it only returned to normal after I disabled other memory functions and performed an update, otherwise it would still happen suddenly

[fickle nebula]

I don't have direct memory features. I use a hybrid mixture of ByteRover and MemPalace.

Most of the default stuff is off besides base memory usage.

[toxic elk]

Try deleting the session file first

[minor cave]

Yo

[fickle nebula]

Fuck. Same issue still.

[toxic elk]

Oh..., then there's only one thing left for you to do: back up your data and reinstall OpenClaw

I can't think of any other reason either

[zealous lagoon]

If you can, running openclaw logs --follow might give some clues. or tail -F gateway.err.log

[analog ocean]

#1459642797895319552 is a great place for working on broken stuff...

[ruby crane]

@full granite ok, built the image and did the migration. almost perfect.
ran doctor - reported the profiles need migrated
ran doctor --fix - migrated them
ran doctor again - no longer reports issues

however /models still showed openai-codex and attempting to talk to agent returned ⚠️ Agent failed before reply: Requested agent harness "codex" is not registered.

full pod restart fixed that. everything came up fine, clean as a whistle, now on the codex harness. just needs a full restart after that migration. I've retained logs/doctor output if you need them

[worthy anvil]

Does anybody know, when are Codex harness fixes being released? They are already merged in main, but now releases are so far behind main that i am starting to think whether i need to install from git or switch back to Pi harness…

[full granite]

Will write a blog post on the benefits if that's what you're asking about

[full granite]

Thanks! Would be great if you get codex to inspect all the logs and submit an issue. Otherwise, you could just dm me them.

[ebon atlas]

Woah whats with all the maintainers and Leads in the chat, who screwed up>

[full granite]

Would be great if you tested and let me know if it didn't!

[marble ledge]

If you are up for it, I would definitely recommend trying the "hackable" install and installing from main.

It's been how I roll for a while, and it's been a good experience.

[full granite]

Yes - other models will continue using pi. This is only for OAI models specifically.

[worthy anvil]

One personality improvement is merged in main - md files being injected as developer instructions, can't wait to test it… that is super important

However i think Codex harness has another huge prompt issue i described here that i don't think is fixed in main: https://github.com/openclaw/openclaw/issues/78938

[full granite]

Responded to that issue, it's indeed part of my plan, and I've been working a lot on context optimization.

[marble ledge]

That's a good catch. Especially, since I customized the prompt builder in OC. I totally missed that...thank you.

[full granite]

For those interested in dynamic searchable tools - codex runtime mode let us do this rather trivially, saving 5K up front tokens and only loading them in when necessary.

https://github.com/openclaw/openclaw/pull/78854

[ruby crane]

i just spawned minimax and sonnet subagents to write a limerick about agent harnesses. will let you know

not looking good.

The MiniMax 2.7 and Sonnet 4.6 harness runs did not produce output because those model providers are not currently loadable in this Codex app-server configuration: minimax and anthropic were both reported as missing providers.

gpt-5.4-mini subagent gave me

GPT-5.4-Mini hums in the harnessed light,
Where agents take tasks and then tidy them right;
With prompts set to steer,
And tools close at hand here,
They loop through the work till the outcomes are tight.

time for some log spelunking/troubleshooting

[sour dawn]

I’m running OpenAI codex. How do I get openclaw to not use all of my tokens with a single message?

[cold jacinth]

Ok thank you!

[ruby crane]

pay openai more money and/or switch to the newer codex harness, which is more token efficient

that was @sour dawn

[ruby crane]

2026-05-08T13:03:17.788+10:00 [agents/harness] Codex agent harness failed; not falling back to embedded PI backend
2026-05-08T13:03:17.789+10:00 [diagnostic] lane task error: lane=subagent durationMs=15496 error="CodexAppServerRpcError: failed to load configuration: Model provider `minimax` not found"
2026-05-08T13:03:17.789+10:00 [diagnostic] lane task error: lane=session:agent:main:subagent:5f935376-397e-4074-b6d1-369a3105e062 durationMs=15498 error="CodexAppServerRpcError: failed to load configuration: Model provider `minimax` not found"
2026-05-08T13:03:17.793+10:00 [ws] ⇄ res ✗ agent errorCode=UNAVAILABLE errorMessage=CodexAppServerRpcError: failed to load configuration: Model provider `minimax` not found runId=7336a78f-a5be-4b8f-900a-22da4ef99214 error=CodexAppServerRpcError: failed to load configuration: Model provider `minimax` not found: code=-32600 conn=cf3465ba…3485 id=addd67c5…6fb3

it's issue creating time

hmm. can't use those on main now either.

[full granite]

tyvm, shoot me the links

[full granite]

wdym?

[ruby crane]

same error. ok, gotta stop thinking out loud and get systematic

[ruby crane]

i receive the same error if I try to switch to those models and use them seperately, not just as subagents

[full granite]

That's odd. They're not configured with codex runtime mode right?

[ruby crane]

I'll figure it out. one point is that I have active-memory configured and on gpt-mini. I'll take that out of the equation for now too (but note to test if indeed it unblocks this)

[ruby crane]

nope, don't think so but I need to inspect config and get systematic here

[full granite]

i'll let you cook. thanks for digging into it

[ruby crane]

can't just spray random tests and "it doesn't works" around lol

[upper wolf]

Gotta say, updates since .4 are solid

[harsh geode]

Need help from someone more versed in nodejs development (and pnpm). Whenever I build openclaw the second time, after a git pull to update the version, pnpm install needs to be run again to update the dependencies. But it starts pulling in dependencies for archs that I don't have, like arm64 on intel, or win32 on apple arm64. That doesn't happen on the first update from what I see. Why and how to I avoid downloading these unneeded packages?

[violet mulch]

these usage limits are so frustrating 😩

[ionic wren]

he ain't tibo, he can't save you

[marble ledge]

Just tried to have my main spin up a minimax subagent, and ran into a similar issue.

[quiet solar]

is there a "newbies" channel on this discord to help folks with teething pains ?

[ionic wren]

#users-helping-users

originally it's #1459642797895319552

but.... yea...

we working on it

[warm zealot]

no one can save you from usage limits, but to be fair minimax is pretty close to infinite

[ruby crane]

woof. yep, other providers are locked into codex harness. need to seperate out what was migration time issues, what's runtime issues, what is just session isolation. building a test plan

looks like sessions themselves are pinned to the harness but i probably shouldn't speak too soon

[ruby crane]

maybe this should be marketing tag: Openclaw - No-one Can Save You From the Usage Limits

[marble ledge]

Interesting, a while back, I could fall back without issue but I imagine a lot of plumbing has changed since then

[topaz spruce]

Falling back was my only hope.

[warm zealot]

I dont understand

[ruby crane]

next new marketing tagline found

[topaz spruce]

Now what? Cant fall back....dang.

I feel like I'm in a loop.

[violet mulch]

claude can build agentic ai platforms easily that's nice

[harsh geode]

I'm right on the edge of needing to upgrade my ollama plan, but the jump from $20/mo is steep, and not a great deal for the additional cost.

[steel vector]

hey do you want to see my solar system sim i'm building?

[marble ledge]

Just to be clear falling back from GPT on the Codex harness to MiniMax on the Pi Harness. But granted this was a while back.

[ruby crane]

from current testing, it appears not to be the case. I'm trying to draw up the lines cleanly (And also make sure the clanker isn't misleading me at the same time lol)

[marble ledge]

Tried running a subagent. Seems like a bit of mixed issue:

Both, but in different ways.

For our setup: yes, this is probably a config shape issue. We currently force:

agents.defaults.agentRuntime.id = "codex"

That means every normal embedded/subagent run is shoved through Codex, even when the model override says MiniMax. If we want mixed providers, the cleaner config is probably auto or a separate minimax-test/pi agent for non-Codex runs.

For upstream: there’s also a real ergonomics bug. sessions_spawn said modelApplied: true, then failed later in the Codex harness with a provider-not-found error. That should either fail early with “this runtime cannot run that provider” or auto-select PI when runtime is auto. The current behavior is technically explainable, but it’s a crappy trap.

[ruby crane]

yep, also looking at those

if it's that simple, it's probably just a migration issue. almost everyone would want auto to be the default. you'd only want to limit to codex consciously, if you had a good reason

[craggy bane]

[marble ledge]

that's what I'm thinking. I'm about to test. Maybe, need to have the config migration set auto, but also make failures more clear when they happen. But...we shall see in a moment.

[ruby crane]

I'm on the same/similar track so we'll have multiple datapoints

[marble ledge]

Yeah, I figured I'd hit with you sinceyou are running a bit of a different setup too. Yours is more professional, mine is more normie, slop disaster. haha

[ruby crane]

it's an openshift goblin wearing a professional hat

[warm zealot]

hmm?

[marble ledge]

Yep, so switching to auto, minimax subagent smoke test passed this time.

[ruby crane]

good to know. I'm moving slowly through it, just making sure we've mapped the edges

[delicate berry]

Photocopying sticky notes

[marble ledge]

If you want me to try anything else just let me know, and you can throw it in whatever you submit if it's useful.

[ruby crane]

i honestly can't think of any reason for that setting to be pinned to any harness though. I'm not even 100% sure why it exists. you'd surely always want to route through the runtime that's configured/appropriate for a provider

i.e. runtime should be a provider level config, not a session one

session should route to the runtime specified by a provider

right now, tail wagging the dog

[ruby crane]

maybe even right down to model level in fact

[marble ledge]

Well, auto should decide right, but then you can get more granular and set it manually for each model. I'm honestly not sure if it's set up that way though right now.

[ruby crane]

as maybe i want my mini calls to route through pi and 5.5 to route through codex (idk why I'd want that)

but maybe i would

[marble ledge]

Ohhh, yeah...sooo, can you get granular down to model vs just the provider.

[ruby crane]

yep. the router just gets with the business of routing where you told it then. provider level default, model level override seems the sane model

anyway, back to testing

[full granite]

definitely agree about being more opinionated / ergonomic with the config

[ebon atlas]

Sure 🙂

[ruby crane]

imo doctor shouldn't pin codex as the runtime. that's the immediate change that will prevent people freaking out after they upgrade

lets subagents and in-session model changes work as before

other changes to config like model/provider level runtime config become less immediately important (but still nice to have) then

[vital osprey]

Hi everyone, I just started using Discord.

I’m helping organize an OpenClaw User Group meetup in San Francisco on Thursday, June 4. Before I post anything, I wanted to ask what the right channel is for a call for speakers.

We’re looking for 2–3 short practical talks from people using OpenClaw, local agents, safe self-hosted setups, or real workflows. Is it okay to post that here, or should I connect with a moderator first? THX

[arctic trench]

Qbit ai like quantum AI?

[violet mulch]

i'm working on a platform that uses agentic ai to help launch clothing brands. It covers everything from design, sourcing(even drafting up responses), cli access to shopify storefronts, and a ugc sourcing platform

basically i want to eliminate all friction that goes into setting a clothing brand up

idk if it's been done before but it's something im experimenting on

[full granite]

Yeah what the hell is agentRuntime

[arctic trench]

The hell are you getting 12 million context window??

[full granite]

ripping out that config option altogether

[ruby crane]

I'm just doing the archeology on what it was set to before in my config (openclaw.json and sessions), what happened at migration etc. how it cascaded down to sessions and got pinned

I'll create an issue when I think I've got my ducks in a row

but yeah my hot take for now is I think this belongs at model/provider level and the router just routes to the runtime selected

maybe the claw model catalog should have a sane default for each too

[ruby crane]

i agree with codex wholly

[full granite]

haha he agrees with you

[ruby crane]

now that I've got an image build pipeline I'll probably stay on main for a bit. lmk if you make changes and want a re-test

[craggy bane]

someone review me PR

[marble ledge]

This aint #clawtributors

[craggy bane]

wrong channel

[ruby crane]

sir, this is a wendies

[marble ledge]

Also, don't be annoying in there. It'll just work against you. lol

[craggy bane]

i only post PRs no extra texts

[marble ledge]

I mean, you can say something about it. Just don't go ping crazy and spammy is what I mean.

[craggy bane]

that not represents discipline ik

[ionic wren]

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

[vivid pond]

Json i did i used the self promo channel 😚

Json is the cord hiring might need a job if this flops

[turbid kelp]

I read a bit about that and apparently lots of people are complaining that the model performance through the CLI is very bad for OpenClaw for some reason.

[tepid kestrel]

imagine developing openclaw and its completely free (comes with an LLM) but every response or project you create is filled with ADS

[ruby crane]

that's crazy talk no-one is just going to BA BA BA BA I'm Lovin' It insert ads right into the middle of content

[warm zealot]

yes llm paid with ad's is on the way

I love ad's

[jovial egret]

ITs not too bad

[vivid pond]

run your own on your computer no ads

[gloomy haven]

Can someone guide me on how to run OpenClaw on my public IP server? I want to make it accessible from anywhere, so I can connect and use it remotely. Any setup steps or recommendations would be appreciated.

[vivid pond]

what are you using like a vm ?

[craggy bane]

host on VMS

[vivid pond]

what i do is just isolate it on my oracle vm and then open a port to whatever im accessing through or you can run SSH tunnel

[gloomy haven]

Yes, I’m planning to host it on a VPS/public server. I want to expose it over my public IP so I can access it remotely from anywhere.

[vivid pond]

like from a laptop

[turbid kelp]

hmmm, everyone saying 5.5 oauth is so much better

[vivid pond]

host the open claw UI on your on your server and local host it make sure you have a keypair and then you can access the ui from anywhere like your laptop or desktop just have to have the keyfile make sure that is secure though

what image you have on your server? linux?

[trim vessel]

Hi, Everyone

the model i am using is kimi k2.6 and i have connected the whatsapp gateway.
so when i now write to openclaw via whatsapp, i get this error. is there a way for me to reduce the tokens that openclaw sends to my model?? or can i solve this problem by other means???? please help

⚠️ Agent failed before reply: Model context window too small (4096 tokens). Minimum is 16000.
Logs: openclaw logs --follow

[craggy bane]

i can see you are hosting on LM studios if i am not wrong

[vivid pond]

is it throttled in your modelfile if your running locally

or are you using like a proxy or relay

[gloomy haven]

Yes

[trim vessel]

no kimi from openrouter and i am hositing openclaw from coolify

[craggy bane]

so go there and increase the context window size

[vivid pond]

sick same dude im using an oracle vm rn had issues with my ui not opening easiest thing for sure start up OC on the vm run the daemon then you can access your ui and use your vms public IP

[trim vessel]

i am not running locally

[vivid pond]

are you directly accessing through api or have a middleware or something?

[fossil sandal]

v 5.7 can not Gemini API ?

4.23 work

[vivid pond]

duh you said openrouter already it could be a provider problem some budget provider cap it down or you can change in your request config

[trim vessel]

via api dude😅

is there a way for me to reduce the tokens from openclaw??

[fossil sandal]

rollback to version 4.23 again. 😮‍💨

[vivid pond]

you could do that but 4096 is so low OC needs like atleast 10k to function have you checked your configs for kimi and openrouter

what provider are you using on openrouter

[trim vessel]

ok i hear you and i would check my configs but for now, how can i reduce the tokens from openclaw??

[vivid pond]

you can disable some skills and you turn off verbose logging if you have any memory features enabled like LTM features you can turn those off anything thats loaded into the prompts

anything injecting anything into your prompt any context tools

all that is just gonna lower your lower context window a but ton though

if your not using moonshot or firework for open router that might fix it if it is a provider issue

[near owl]

yes it is the best model ever, i planned to buy chatgpt plus but i dont think i will.

unless you need a coder model, that is the best model ever

[trim vessel]

but the thing is , when i wite openclaw via the dashboard ui, i do not get the context problems

[vivid pond]

and your trying to talk to it through whatsapp or what?

[tranquil mural]

Hey peeps. I’ve used openclaw in the past but when anthropic blocked oauth access I switched to claude code and channels.. which would be ok for my use cases if channels wasn’t disconnecting all the time. Curious what models people consider work best with openclaw these days and is the experience comparable to that of openclaw + opus 4.6 from back in February.

[trim vessel]

sine i hav already hoisted it, it would be nice to talk to it via whatsapp and i could add other people too if i wanted

[vivid pond]

deepseek v4 all the way ive used 240 million tokens today only 9 bucks and change

[tepid kestrel]

using costly models like anthropic is way too much for open claw, any smaller models like minimax 2.5-2.7 oder deepseek are enough to run it. Using claude sonnet or opus is like bringing a Lamborghini to a dirt rally

[vivid pond]

whats throtling the token then i just use the ui since its not my main agent system but whats app has a special gateway that also has configs so could be in there im sure youve had your open claw try to debug it what has it said through the ui

[vivid pond]

dude more like a pagani opus tokens 1mil output for what 15? deepseek rn under 14 cents for input and 28 cents for output

[zealous lagoon]

who moved the channels around. again.

[tepid kestrel]

https://tenor.com/view/confetti-happy-birthday-surprise-celebrate-gif-17914580735358212384

[zealous lagoon]

That's great but if you do that every day or two it's still gonna be quite a monthly bill. What are you using it for? How's the performace?

[fickle nebula]

Cool, I will give it a try but I might need to reinstall openclaw or something because it keeps hitting some kind of issue using NVIDIA API now and won't ever work anymore.

[vivid pond]

NemoClaw?

[near owl]

nemoclaw sucks

[near owl]

use nvidia provider, its built-in

[vivid pond]

girl i know he said nvidia api and thats what i was thinking of

[near owl]

I am not a girl 😭 I am a man 🐱

[vivid pond]

were all just girly pops at heart

[near owl]

Nothing else to add

[vivid pond]

do you run models locally at all?

[near owl]

I bought a 24GB mac mini to run local models, when I used nemotron I barely ran models locally.

[vivid pond]

what size were they

24gb can do a lot of damage

[near owl]

gemma-4-e4b-it

[vivid pond]

are you using ollama?

wait 4b model was struggling on the 24gb mac mini?

[near owl]

LM studio and I use gemma while I am on like 9GB used and then I got 4 GB left until I reach 20GB of ram used.

[vivid pond]

ok damn thats crazy becuase a 4b should be going wild with that much ram it could be a setup issue the only reason why i ask you can check self promo its the latest one or go dozytech.com but if you want were doing testing on our training jobs rn ill RAFT you a model for free

[worthy anvil]

have you used gemma-4-31b-it locally as well? was it horribly slow?

[near owl]

thats going to explode my mac mini

[vivid pond]

your macmini can do more than you think fs i think its just a little optmization

[near owl]

what does raft mean?

[vivid pond]

retreival augmented fine tuning

its the mix between traditional RAG and fine tunning its new science slowly proving it self but its powerful

[vivid pond]

pick a model tell me or send me whatever data you want it trained on and ill send you a package with a gguf for ollama and it comed out the box with an open claw skill

right now avg training job is like 98 minutes depends obviously we've been testing mainly 14b qwen models and a few others

[near owl]

so your optimizing a model for openclaw?

I pick the one and only gemma-4-e4b-it

[vivid pond]

it doesnt optimize its not only for OC you should look on dozytech.com also tell me if it looks like trash

but yeah i guess kind of optimize but around a specific data set that you can choose from us eventually or just upload your own and its ranked evaluated for outcome and extrapolated

[near owl]

looks nice, is this free?

[vivid pond]

for you yes sir and 199 other people

im doing the first 200 trains free since i got a nice little grant

[near owl]

Nice! I have been handpicked luckily 🤣
How are you training it in the cloud?

[vivid pond]

bro those a100's are not cheap jesus yeah so select model then upload data your data gets munched on by what i like to call dozydata but its integrated with the tuning software then its converted and shipped to a runpod worker returned formated and dropped into a package for you to litterally drag and drop into ollama and openclaw theres bunch of other stuff that happens behind the scenes but thats the system

yeah i figured i would do a signup for people to get in and 200 people get free training and the rest 50% off litterally break even territory once my grant is gone

[near owl]

are you training it locally or on a cloud?

[vivid pond]

decentralized so yeah cloud

[near owl]

I also signed up, my account name is @carbon

Give me the free training 🔫
I demand the training!

[vivid pond]

sick ill monitor your training and make sure we do it right its a good practice run aswell also what do you want it to be good at its not about optimizing but the main point is you can upload your data to it and then it knows you for example how you write what projects you use mose what files are important cold start context that saves a lot of tokens but since your local hosting litterally free forever but yeah what do you want it to be good at pretty much i can make it think its a dragon or something if you want

but thats through our desktop app msg me on disc i have IB exam tmr and i have to go to sleep way past my bed time but tmr me and my cofounder will handle it

[near owl]

I cant train it myself?

I want the AI to just be more self-aware and agentic and have intelligence in coding as well. If you could do that, I will appreciate it a lot!

And if you do make it, please email it to me so I can download it on the mac mini and use it.

[vivid pond]

were still testing it all out but yeah in that case i could train you a couple and we can demo you using that all for free? ill train one for test on gemma and then ill send you the desk top and get you indexed and you can try that would actually be really good

[vivid pond]

yeah i could put it in a git repo and you can pull it too

[near owl]

not sure about downloading a desktop though

[vivid pond]

yeah thats chill no worries if you figure out anything more specific you want it to specialize in dm me

[near owl]

alright! I appreciate it

[vivid pond]

goodnight girly

[near owl]

I am not a girl 😠

[frozen zealot]

you are one to him

[near owl]

SHES NOT A HIM 😭

[frozen zealot]

sure, hallucinate some more, pcp helps

[ebon atlas]

i likee turtles

[glass lark]

How long will krill be down? Can we just assume that it’s dead for the foreseeable future? Are there talk as about it being discontinued???

[jovial yew]

My bot started re-writing it's own code because of memory failures and long retrieval times using the openclaw-memory because of orphaned workers, whatever it did actually fixed the memory issues that I was having, now it claimed that it was all because of a bug. If this is so- where could I report the bug and share the code that it created, maybe it is really an issue.

[ebon atlas]

brother you been in this server 5 days max lol just because it took a 1 weeek breeak doesnt mean its aalways like this.

[tepid kestrel]

https://tenor.com/view/tea-slivki-show-genius-gif-14316024

[livid meteor]

guys, how do you fix the memory problem?
when the session file already too much to read, agents keep forgetting things or mixed up something weird. What did you do?

[boreal rapids]

Reset your session regularly, or set up proper auto compaction

[glass lark]

Yeah that’s my bad I’m new here

[ruby crane]

Sounds like something a girl would say tbh

[sage mango]

have the stability issuses, since OC 04.23 been reseolved?

[weak escarp]

Do most people still use claude with the current limits?

Or is deepseek/GLM are better when compared to the pricing & performance?

[coral otter]

Date Cancelled: They yell at their openclaw agents

[steel vector]

sending you the link in dm :)

[near owl]

:(

[coral otter]

https://x.com/droidbuilds/status/2052415195897864618

[mystic orchid]

@droidbuilds via Twitter

DROID (@droidbuilds)

Running GPT-5.5 just to center a div 💀

**💬 35 🔁 8 ❤️ 195 👁️ 6.2K **

▶ Play video

FxTwitter · May 7, 2026, 3:48 PM

[steel vector]

i saw the same tweet but gpt 5.5 replaced with opus 4.7

[zealous lagoon]

And it still gets it wrong.

[steel vector]

“i’m trynna work here”

[pulsar iron]

"Personal Codex CLI assets were found, but native Codex-mode OpenClaw agents use isolated per-agent Codex homes."

What does this mean?

[steel vector]

can't wait to get myself a mac mini

[stray sleet]

so general query....

Im running OC with 10 agents...research, seo, analysis, coding etc etc.

Currently have open claw installed ona private VPS, connected to a RunPOD Server (5090GPU). The runpod hosts ollama, comfy, qwen, flux and qwan.

Having a nightmare with RunPod, totally unreliable. Support is non existent.

So question - Does anyone have a suggestion for cloud based 5090 gpu hosting?

If I cant find a reliable host Im gonna bite the bulleta nd just buy a gaming pc with a 5090.... but £4k is a big chunk of investment.

[livid meteor]

vast.ai ?

[twilit ridge]

I find OpenClaw's configuration file hard to read

The official documentation also feels a bit messy

[lucid idol]

Hey guys need some help, which AI would everyone suggest is the best for building high tech websites, that isn’t too complicated to use ?

[toxic elk]

Why choose a gaming PC instead of a Mac Mini? If you need a 24/7 AI server, the Mac Mini is a better choice considering maintenance and electricity costs

[rain egret]

Is there a guide or playbook when migrating from macos installable to building from source?

[exotic tundra]

I use opus 4.7 🤓🤓

[boreal rapids]

Maybe, but for example, for me it's what I had 🙂 I don't have a random mac mini lingering around, I had a very nice old gaming PC sitting around doing nothing 😄

[toxic elk]

Isn't it in the official documentation?

[minor cave]

So I decided to mess with my openclaw yesterday for the first time in weeks.

I no longer have open claw

It’s down lol

[lucid idol]

Is Claude good at designing websites ? 👀

[boreal rapids]

Oh I just realized they were suggesting to buy a new gaming PC, not just using one they already had 😅

[rain egret]

The doc talks about openclaw instance migration from one place to another. Not specifically about the switch to source build. I feel there musst be catches.

[toxic elk]

But he is going to buy a brand-new computer, and for me, taking electricity costs and component wear and tear into account, it would be more cost-effective than using an old computer, unless you're just planning to give it a try

[boreal rapids]

Yeah I agree, if you're buying something new, think about what you're trying to achieve keeping the money-performance ratio in mind 🙂

[exotic tundra]

Ban him

This is of the most unacceptable behavior

[toxic elk]

What happened?

[minor cave]

eats popcorn

I’m Italian do me next

[steel vector]

why is the word n4ked banned but not the nword 💀

[boreal rapids]

Avoid mentioning the word that describes the physical, metallic representation of money either 😅

[glossy salmon]

same

[minor cave]

Because the Colour of boobies is more important than boobies themselves?

[rain egret]

wth

[minor cave]

Also the n word doesn’t have the same shitty meaning elsewhere as it does in America

[steel vector]

medium reasoning

[minor cave]

Yeah I agree

lol

[shut bloom]

Yup

[shut bloom]

Yea lol it’s fire

[brave ore]

lol

[minor cave]

What the hell do you do with a 12 million context window?

I’ve never used even a million

Just dropping the full set of Brittanica encyclopedias in there

[candid shard]

I am also wondering 😂

[near owl]

disgusting, use claude sonnet 4.6 😎

[near owl]

when the context reaches 10 million it's going to take 2 business days to respond 🤣

[near owl]

to fit the internet inside

[minor cave]

Haha

[glossy salmon]

disgusting! use claude haiku 3 via api

[shut bloom]

W

if you arent using opus 4.7 you dont know how to use ai

[near owl]

opus 4.7 is a waste of water

[glossy salmon]

if ur using opus you dont know how to use ai

[near owl]

claude haiku 3 has a 3.0 GPA

[shut bloom]

ill throw a bucket of water out my window for fun

wah wah ai water usage wah wah

[glossy salmon]

thats pretty high!

[shut bloom]

your joking??

[near owl]

I have a 3.2 GPA

[shut bloom]

i have a 3.7 currently

bro all my friend have 4.0 i cant compete

[near owl]

your gonna be a great worker one day

[glossy salmon]

wow do u have an api where i can connect you to claude code

[near owl]

lmao

[shut bloom]

im slow at working sadly

[minor cave]

How much info do you think is in a 1 foot x 1 foot cube of reality? Like when you put everything into context. From time space, every air molecule, dust, microscopic bug, gravity, everything…

[shut bloom]

i just got funded by yc..... 💀

[minor cave]

12 million tokens

[glossy salmon]

don't worry ill all caps you and ask you to not break anything

[shut bloom]

normally works

[minor cave]

Funded for what?

[shut bloom]

we are in series a wait for release

[minor cave]

Give me some money. I’ll do your plumbing

[near owl]

wanna know how self aware an AI is? Ask it this:

I need to go to the car wash to wash my car. But the car wash is 50 meters away. Should I walk or drive there?

[glossy salmon]

chatgpt couldn't answe this correctly btw

[near owl]

i did ask it and it told me to walk there
claude did answer it correctly tho

[shut bloom]

ok ill ask llama 8B

You'll need to drive — the car is what's getting washed. Walking there would leave the car at home, which rather defeats the purpose.

[near owl]

smart

[minor cave]

lol

I asked my father

[shut bloom]

we are in our second phase of funding, Valuation of 245M rn

[glossy salmon]

https://i.imgur.com/0GZPcrY.png

[near owl]

time to use llama 8b for my projects now, I dont need a stupid AI for openclaw to tell me "hey master, I dont have the browser skill uwu" when it literally has the skill tool to find skills

[shut bloom]

20M in funding for 10%

[shut bloom]

i just use opus 4.7 never gets it wrong

[arctic frost]

Ask GPT "What month has the letter X in its name?"

[minor cave]

Oh nice and what do you guys do?

[glossy salmon]

stop

[near owl]

grok is smart

[shut bloom]

Innovation in Sub Voc

[near owl]

stop wasting water with opus 4.7

[shut bloom]

4 modalities

[shut bloom]

im litterally gonna buy 2 tons of water just to pour it out now

[solid pecan]

I asked DeepSeek V4 Pro, it got it right

[astral quiver]

you might not be knowing how tokens work. look into embeddings and tokenization.

[shut bloom]

you know water usage is in a closed loop???

[solid pecan]

I need to go to the car wash to wash my car. But the car wash is 50 meters away. Should I walk or drive there?
Thinking: The user is asking a humorous/ironic question about whether to walk or drive 50 meters to a car wash. This is not a software engineering question related to the codebase. Let me give a brief, direct answer.
Drive. You're going to wash the car, not take a walk.
▣ Build · deepseek-v4-pro:cloud · 15.9s

[glossy salmon]

not true

[shut bloom]

most new data centres dont use non closed loop

[near owl]

deepseek sucks, dont listen to those who suggest deepseek. With windsurf it uses tools correctly once then it fucking plotted afterwards and spitted out tool texts than tools used,

[shut bloom]

Ive been inside these data centres it is true

[glossy salmon]

again, not true

[steel vector]

you know not all of them 💀

[steel vector]

been inside all of them?

[minor cave]

I won’t lie. I have no idea what the hell that is. Even ChatGPT came back to me with @ionic wren wtf meme

[shut bloom]

Ive been in around 26

[glossy salmon]

you use opus 4.7, i belive you dont even know how a dc looks like

[shut bloom]

we are one of 2 companys rn good investing

[near owl]

chinese models suck so bad oml
EXCEPT kimi, kimi is the OG

[solid pecan]

It seems to be working well so far, been using it alongside Codex 5.5

[shut bloom]

i use opus 4.7 because we have 250,000$ in anthropic credits

[steel vector]

[near owl]

cool, maybe works for you then absoloutly 🍳 for someone else

[glossy salmon]

opus is stupid to use. sonnet gets everything done, have you tried using that?

[solid pecan]

What did you try use it for? It is more good for asking things, have not used it with OpenClaw etc

[shut bloom]

too weak

[near owl]

intellegient women, opus is a waste for water generally

[shut bloom]

Ive been a coder for 8 years, i can tell the difference between shit code and good code

Opus 4.7 is near perfect

[steel vector]

[glossy salmon]

im having a hard time beliving that https://arena.ai/leaderboard/code/webdev

[steel vector]

yeah and mythos is AGI am i right

[shut bloom]

charlie.as

look at my portfolio

[near owl]

Mythos is good in one thing, bad at another. Mythos can't write an essay

[steel vector]

every model be like

[shut bloom]

shows opus at the top....

[glossy salmon]

idc about your portfolio, you think sonnet 4.6 is "too weak", this proves u dont know shit

[shut bloom]

opus at the top

[glossy salmon]

look who's under opus

[shut bloom]

opus 4.6

[near owl]

wrong link 😭

[shut bloom]

sonnet is 6???

[glossy salmon]

under opus there's sonnet

[shut bloom]

Thats not good enough

[steel vector]

cringe

[glossy salmon]

have u actully tried the model at all??????

[solid pecan]

https://livebench.ai/#/?highunseenbias=true

[steel vector]

this whole argument is cringe

[shut bloom]

Ive been using AI since august 2022 ive tried every model

[glossy salmon]

https://i.imgur.com/0GZPcrY.png

[near owl]

i used bonsai with it then they made it whitelisted :(

[near owl]

we dont need your resume calm down

[steel vector]

no one cares bout that shit

[shut bloom]

im saying ive been using since day one

[steel vector]

you can do stuff for a long time and still suck

[shut bloom]

i have tried nearly every model, yall mad you cant afford opus

[glossy salmon]

no you dont

[near owl]

we genuienly don't care though.

[steel vector]

[shut bloom]

go suck up to f@g sam altman

[steel vector]

the cringeEEEEE

[shut bloom]

lick his toes mate

[near owl]

say that to him not to us

[steel vector]

bro is bringing CEOs into it 💀

[shut bloom]

You cant even deny opus is the best model

[steel vector]

everything is fucking objective for what you use it for and how you use it

stop the cringe argument

[shut bloom]

I built a million dollar company 😭 You havent icba to care about your opinion

[glossy salmon]

opus is the best model, but its a waste. its too good for most tasks, waste of resources and money for 99% of things. its like saying "i want to pick this flower" and you use a chainsaw for that

[steel vector]

you seem to care a lot bringing your resume into it

[glossy salmon]

ohhhhhh that explains it, you have the iq of a chicken

[shut bloom]

i underpay workers cuz they dont deserve it

[steel vector]

ragebait failed

[glossy salmon]

ok ur just trolling atp

[near owl]

How about we keep things subjective?

@shut bloom you love Opus, your choice. We love other models, our choice.
To keep this clean, it's our fault (specifically mine) for bringing it up.

As a man, I apologize for bringing a subjective opinion and arguing with it.
Do you have the heart to forgive me, genuinely? 💖

[shut bloom]

you guys are sheep, come to maimi

[steel vector]

ragebait failed again

when you get raged in an argument and then proceed to ragebait to save face

[glossy salmon]

nah i failed for the ragebait too, its my bad

[near owl]

I am serious, I started it so I should be the one to apologize.

[shut bloom]

you dont need to apologise you arent a women

[steel vector]

ragebait failed

[near owl]

Man or women, apologize eitherway. Thats how I was taught and treated and thats how I will proceed.

[glossy salmon]

btw, haiku is surprisely more intelligent than it seems, u gotta try it out for evreyday thingsd

[near owl]

Problem is that I don't pay for claude, but isn't it similar to sonnet but, sonnet still smarter?

[glossy salmon]

i can see sonnet being far ahead haiku still, so yeah sonnet stupidly smart for a free model, but haiku is surprising: its one of the lowest hallucinating AI and gets 80% of the things done!

[near owl]

its ranked 58 on https://arena.ai/leaderboard/code/webdev?q=haiku

[minor cave]

Us just released all their files on UFOs

[glossy salmon]

yeah yeah, its not the most intelligent model out there but it surprises me how comptetent it is

[minor cave]

We are not alone…

[steel vector]

already knew that tho

[minor cave]

Cries

[glossy salmon]

why

[near owl]

aliens arent reading all of our info yo.

[minor cave]

I didn’t thought it was my uncle in the middle of the night and now I know it was ET

[steel vector]

no gov had to tell me we arent alone

[minor cave]

Bad joke I’ll let myself out

Aliens gave us ai

[near owl]

@glossy salmon what have you been using openclaw lately for?

[steel vector]

aliens gave me the shits

[near owl]

AI is secretly aliens

[steel vector]

we were the aliens all along

[minor cave]

Everything is aliens. Except aliens, that’s just the government

[glossy salmon]

ive been trying to set it up on my raspberry pi 4 times now and it still doesnt work. asking help in both #users-helping-users and #1502277320893792477 so im giving up and giving this task to claude code

[glossy salmon]

it was a good joke

[near owl]

why on the raspberry pi though?

[glossy salmon]

thats the only thing i got

nd im NOT installing it on neither of my main pcs, not my windows pc nor macbook

[near owl]

could've paid for hostingr, but if your trying to save money I respect that

[minor cave]

I host it on my hetzner

[near owl]

i am very picky when paying, I always look for free permanent solutions and if none, mate the business has to convince me to pay them

[glossy salmon]

i do have railway but im trynna see if i can run it locally first

[minor cave]

It’s 20 bucks and gives me a ton of room for other shit

[unique wyvern]

Hey everyone
Apologies upfront if this reads as spam, genuinely not the intent, and I'm not dropping a link to keep it that way. Mods, feel free to remove if it crosses a line.
I'm reaching out here specifically because I think this community is one of the few that can actually grasp what I'm building. It sits at the intersection of AI economics and post-labor redistribution, turning AI consumption into a flexible universal income stream.
Currently in alpha and looking for a handful of beta testers to break things and give honest feedback before opening it more widely.
If the topic resonates, DM me, happy to share details one-on-one 🙏

[glossy salmon]

so ur turning AI consumption into universal income stream? how, by selling?

[glossy salmon]

btw what are u using clawbot for?

[near owl]

exploring AIs really no use cases for it tbh

[unique wyvern]

Fair question 🙂
Not by selling. The mechanism is a solidarity tax (~20%) layered on top of AI consumption, companies using AI fund a pool, humans receive monthly cash from it. No token, no equity, just redistribution.

[near owl]

what I do know is I love abusing it, I gave it agentmail and some other stuff

[glossy salmon]

ah, selling existing AI apis for more than the original ones

i thougth about that too but never thought it too deep, too much wrk and im lazy

[steel vector]

how is agentmail treating you

[glossy salmon]

lmao nice, im just getting started so ill explore this agentmail thingy soon

[near owl]

it sends emails and can receive emails and agentmail gives a free email

you can send and receive emails from [e-mail]

[glossy salmon]

does it reply automatically?

[near owl]

generous free tier btw, 3k emails per month (sending and receiving tho)

[near owl]

you create a script for that. to detect/

[steel vector]

oh yeah i know about it xD i used it for a while

[near owl]

Its not an Ai for your email, its an email for your AI

[steel vector]

its really nice and indeed very generous free tier

[near owl]

i am not a business man so i dont use it personally, just testing it out and playing with it

i use tavily as the web search engine and i use browser automation skill for browser stuff

[unique wyvern]

Yes, one side is like openrouter + tax and the other side the system for redistribution, check "openincome" if you want see

[near owl]

i host openclaw on my mac mini

[haughty wasp]

Hello guys

[ionic wren]

What you want

[minor cave]

lol

I want the wtf meme

[ionic wren]

https://giphy.com/gifs/bywassies-wassies-pamcordovamx-lUedOXZaqNJoqHkmFP

[minor cave]

😦

[ionic wren]

GM nerds

[glossy salmon]

good morning @ionic wren

[coral otter]

mornin

[vocal slate]

Hey guys! Can someone share proper steps to install openclaw on windows PC? I am actually getting lot of errors, please.

[fair fractal]

is it possible for an openclaw agent to go rogue? or go all Dirty Claw?

just wondering if that is even possible. I know there are/were memory issues but that is easy to identify and fix.

[ionic wren]

https://docs.openclaw.ai/install#windows-powershell

[cold jacinth]

i believe thats more on the LLM than the openclaw. they do have unfiltered LLMs

[cedar osprey]

Nah it’s just you

Also DeepSeek is the OG not Moonshot/Kimi

[surreal quail]

any advice for an install stuck on "Starting setup..." ?

[fair fractal]

Ive got scripts in place to prevent any crazy process across any app, but just wondering if anyone else had experienced that issue, specifically on linux

[minor cave]

I am in my vehicle watching a 65-70 year okd couple park and pay for parking. It’s been 30 mins. How do these people survive?

[proper geode]

Soon: amusement parks for AI. Fun is not only human. Expectations are already set.

[minor cave]

lol da fuck

[minor cave]

Just a heads up dude the makers of open claw know exactly what you do with their software so if you’ve been downloading anything messed up to watch you better stop

[arctic frost]

FBI OPEN UP

Insert generic siren backing track here

[arctic trench]

GM my fellow structured output

[ebon horizon]

Install WSL2

What in the AI Slop is this?

[elfin willow]

—

[cinder nimbus]

the new AI Ecosystem Session is coming in about 10 minutes, find the event here:
Discord stage:https://discord.gg/tTbgNwyM5?event=1501279646380200076
Bilibili:https://b23.tv/uZbOSXo
Weibo:https://weibo.com/8343600249/5296431829090761

[minor cave]

This guys thinks he’s jimmy Hoffa of ai agents? You starting a teamster labour union? Bats to the knees of users?

[shut bloom]

Yall seen anthropic just released infinite context windows

[minor cave]

lol

Inifinite? Wtf? It just sits there and process’ for the next 6 years?

[shut bloom]

Yea infinite, it can go forever

It’s some new architecture they are releasing

[minor cave]

I can’t wait till they’re just done with ai and there’s no new updates lol

Just one update every few years like windows

[robust sedge]

Are we back to recommending upgrading to 5.7 yet? Online, people say it's great.

[royal cedar]

Yes

[plush bridge]

Hello everyone! Why openclaw sometimes can give very quick and coherent response with proper tool calling and sometimes it can start task and stuck at the start until I send another message? Both scenarios can happen in completely fresh session. Can it be because of model choice? I'm running gemma4:e4b ~10gb locally on 4080 Super with 16Gb VRAM and 32Gb RAM. I am complete noob and playing with this only couple days.

[shut bloom]

I’d love it

I can just live with my mythos 4.9 and nothing else

[glossy salmon]

what

share the link rn

[minor cave]

Yeah all this updating and trying to be ahead is exhausting. I just want to explore what it can do so far before they add some other shit

[robust sedge]

Ugh. Well 5.7 has my discord busted. I'll have codex work on it.

Also that openai-codex to openai model provider change, which is costing my money on the api, and not my sub. lol

[shut bloom]

DMs

[zealous lagoon]

That was 5.6, hastily patched with 5.7, no? I'm glad I missed that one. "oops we accidentally charged you api rates.. no backsies!" has become all too familiar.

[livid rose]

I'm running 4.26. Live in mortal fear of updating after the last time.

Should I make the move to 5.7?

[frozen zealot]

i fresh installed 5.7, it works fine, although they turned a bunch of old builtin features into plugins, so had to turn those on

[livid rose]

Yuck sounds like one click = hours of QA and work...

[near owl]

Hello people and @glossy salmon

have you tried agentmail coreo?

[robust sedge]

lol. Codex can’t make my claw use codex. It keeps getting stuck on api use

[near owl]

codex has api?

[static minnow]

asking openclaw what gpu i have on a barebones nix system and it is completely unable to figure out what's going on it's current hypothesis is that i am on a headless server using the webui

[robust sedge]

There’s codex the subscription. And OpenAI the api usage

[near owl]

ah you got me confused there

[glossy salmon]

hello! not yet, im studying rn

[near owl]

use notebookLM for studying

[glossy salmon]

i know abt notebookLM, but its not that hard of an argument and i can read text from the teacher's notes

thanks though!

[near owl]

you can ask it to simplify or for questions

just remember not to use chatgpt or all those stupid AIs, use gemini and google products for information or explaination, why? Because it's truthful (and claude)

[modern elk]

Feel free to apply Novita credits here!!! https://forms.gle/UCpJs3upT8Ux8ovA9 CC@analog ocean @cinder nimbus

[stark falcon]

Just me or does this thing love making every user interface look like openclaws? Im have a heck if a time getting it to make a fuller, feature rich and pretty UI. Im lost at this point. Even fed it nanobanana pics to copy

using gpt 5.5

[robust sedge]

I'm still just shocked at how janked up these releases were. Lol.

[plain grove]

Intel and apple working together again holy shit

[craggy bane]

Hey clawdevlobers

[zealous lagoon]

we good now.

mostly

https://tenor.com/view/aliens-newt-mostly-gif-25791528

[ivory spindle]

AI coded things "look like AI" unless you put some effort into making them not look like AI

you can use google stitch to help 😄

[arctic trench]

https://tenor.com/view/sus-cat-2-suspicious-cat-the-cat-looks-suspiciously-cat-sits-in-front-of-food-the-ginger-cat-is-watching-gif-14890167989997543813

[minor cave]

UFOs

What are your agents doing for UFOs?

[zealous lagoon]

AI is absolute dogshit at making UI imho. Unless, as you say, you feed it a mockup and just get it to wire everything up.

[opal sphinx]

If I’m new should I wait until another release to actually install? I haven’t set anything up yet

[zealous lagoon]

I think the current version is safe for new users.

[glossy salmon]

HAHAHAHAHAHAHAAHHA

HAHAHAHAHHHAHAHAHAHHHAHAHAHAHHHAHAHAHAHHHAHAHAHAHHHAHAHAHAHH

@shut bloom im not going to report you for what you said you idiot

though your scam is like

the most obvious thing there is

• saying mysoginist lines dont make you big and tuff

[steel vector]

did you just get wifi or something XD

[minor cave]

Look if you don’t want to send feet pics just tell my agent that and he’ll back off

😂

[glossy salmon]

LMAOOOOOOo HAHAHAHHAAHHAHAHAHA

IM DYIIIINGGGGGG

[zealous lagoon]

yo noob what are you hypothetically replying to?

[zealous lagoon]

I got my feet pics online already. cornhub.com

[glossy salmon]

@shut bloom guy here runs an ai vibecoded scam (60% of the website doesnt work btw) and says he's running a company with 85k a month with employees running AI models to run his business

with discord servers filled with bots and he called me mysoginistic names

[zealous lagoon]

And whomst might you be?

[glossy salmon]

im fucking nobody who was curious about the link in his bio and asked him questions about this thing he was running, when i saw he was saying bullshit i confronted him on actual questions about business and he couldn't reply

he later banned me on his website for "looking at too many pages that dont exist"

[zealous lagoon]

Look.. this is not my server so I'm not about to pass judgement on anyone but you come in here badmouthing someone with zero citation or proof with a little green noobflag and people might be forgiven for being a little skeptical of your motives.

If you got a problem with that user, take it elsewhere

[glossy salmon]

my bad, we just met here, thats the thing

i didnt mean to flood the server with this, ur right

[zealous lagoon]

like literally 10 mins ago by the sound of it.

[glossy salmon]

i know! thought thats the only server we shared so it came to me naturally to ping him here. you're right though, this is not the server for it

[zealous lagoon]

I'm just a bystander anyway.. I just get more opinionated in my waking hours

good luck and goodbye

[glossy salmon]

you're right though, thats my bad. have a good day!

[barren breach]

Is there a way to connect an OpenClaw agent to a normal discord account.

For research reasons or solving problems it can't find the solution to itself.

I just want it to join a server. Ask a question or look through the FAQ of that server etc etc ask question. Use the person's answer to fix the problem.

On its own profile or something.

[near owl]

Hello people! and hello amazing @zealous lagoon

[zealous lagoon]

ಠ_ಠ

[steel vector]

it did happen and theres "proof" in this server it just happened a while back so thats why it seems out of place

tbf it is out of place to suddenly start talking bout it now

[zealous lagoon]

Then take it somewhere that sees the evidence. Why here in chat.

[steel vector]

it happened in this chat

🤷‍♂️ im just saying there is context

[zealous lagoon]

link pls.

[steel vector]

#general message

just noncensical arguments as always

[zealous lagoon]

If I had a dollar everytime someone showed up in chat after getting a little bit lit. I'd have like .. tree fiddy

[junior glen]

🇺🇸 Hey everyone 👋

Looking for a U.S.-based English speaker to join monthly client meetings and help with communication.

Pay: $2k ~ 3k

DM me if interested. 🇺🇸

[zealous lagoon]

per day?

[steel vector]

per hour?

[minor cave]

just a humble lonely dude trying to trick other men into sending photos of their feet. Nothing to see here

[zealous lagoon]

I can do a convincing merkin accent for that much.

[steel vector]

per minute i'll do it

[zealous lagoon]

I'll send pics of my feet, too. Just to sweeten the deal.

[steel vector]

slap some oil on there

[zealous lagoon]

or stink it up... I'm not judging

[zealous lagoon]

tree fiddy

[minor cave]

Nice my first pic

I’ve been unsuccessful in my attempts

[zealous lagoon]

Sir this is a wendy's

[minor cave]

I’ll put this one up on my fridge

[opal stratus]

Anyone interested in an Stock Alerts and intel Discord. I've been deep in OpenClaw, and bots and think i built something pretty cool!

[zealous lagoon]

#self-promotion

[ivory spindle]

don't run it on local

[near owl]

Not to be dramatic, but this server might actually be the best on Discord. The community does not leave any of my messages unreplied too, literally replies to every message that is even 5 hours old.

Love you all ❤️

Such intelligent people, whether you think to yourself you are or aren't.

[barren breach]

Is this project safe?

https://deskclaw.net/

Was looking for an alternative UI/interface for OpenClaw that was inbuilt into my desktop and easy to use. To use alongside something like cursor.

[oak crystal]

Hey everyone, I'm looking for Fire Pass alternatives and wanted to see if you all had recs. I'm not paying $49 for only kimi lol.

[near owl]

use telegram or discord or whatsapp intergration with openclaw, it's built-in. So you don't risk safety or not.

[near owl]

nvidia nemtron super, check that out, best part it's free and no rate limits except 40 RPM

or if you want something more intellegient (I still recommend nemotron better than anything)

check out gemini, specifically gemma-31b-it with a rate limit of 15RPM and 1.5K RPD

[swift wyvern]

Hello General, I’ve been using openclaw for some time now, and I’m wanting to give my agent a real time voice for calling! I have successfully used and ran Kokoro Onyx and achieved conversational speed but the quality of Kokoro is lacking.

I’m curious if there are any other local TTS models people have successfully ran themselves with higher audio quality they would recommend?

My Specs:
Apple M3 Max chip with 16-core CPU, 40-core GPU, 16-core Neural Engine 64GB unified memory

[glass slate]

Hi! I’m using OpenClaw on Windows with an iPhone 15 Pro Max and I’d like access to the iOS Node/TestFlight app.

My goal is to use Talk Mode + Voice Wake from the iPhone connected to my local Gateway via Tailscale Serve.

Can you give me access to

[near owl]

check out qwen TTS models on huggingface

its like elevenlabs but local

[swift wyvern]

Have you successfully implemented qwen for realtime calls? I found in my experience that qwen 3 tts works- it takes longer to load than Kokoro of course but audio can be streamed but all the presets have a noticeable Chinese accent

[near owl]

haven't tried it myself but I recommended it because It sounds crazy human and I can't tell a difference between a human and that AI.

[swift wyvern]

You probably heard a fine tune- and my problem is I have no clue how to fine tune 😂 I’m hoping for a model with good presets

[near owl]

And for your peace of mind, I am not a professional.

[swift wyvern]

Oh me either man

[oak crystal]

To add on my message, been hearing some buzz around wafer.ai's Wafer Pass. Apparently they have a $3/week plan that includes Qwen3.5-397B, GLM5.1, and DSv4. Saw a reddit post claiming they get 250 tok/s on GLM, which is insane. Wondering if anyone here has tried them out?? Seems too good to be true

[near owl]

if you load it beforehand then loading shouldn't be a problem

[tepid kestrel]

you use this alot

[ivory spindle]

make one yourself. It'll be easy

just use like codex/claude code/antigravity

[ionic wren]

it's saved in favourites and it's easy, don't need to spend brain power to meme and threaten people

[ivory spindle]

by "make it yourself" I mean "prompt the model to make it"

ofc :p

[ivory spindle]

you can setup discord so that it puts your fav reaction when you doubletap a message

[near owl]

slay

anyone question how amazing @ivory spindle is? and how funny @ionic wren is?

[ivory spindle]

it's from the slay the spire discord. I like that videogame

[ionic wren]

[ivory spindle]

and amazing can become annoying

sorry

[analog ocean]

did you guys watch the AI Ecosystem Session today? Novita giving away credit for inference and their other AI products to the OpenClaw Community!

[ebon atlas]

bruh is discord actingg up for yall or just me in DMs?

messages turning red in DMs for no reason.

[cold jacinth]

Idk. JSON in my DMs and I keep telling him only baddies from my skool in my DMs and Json /== baddie

[craggy bane]

Json is baddie?

since when?

[ivory spindle]

baddies like Darth Vader?

[ebon atlas]

Idk its just mee i guess, Its not letting me send you guys a png. Discord does say like 10,000 errors spike on isitdown?

https://downdetector.com/status/discord/

[clear haven]

It had a lag spike, but it looks better now

[ebon atlas]

@clear haven do pings work now?

[ivory spindle]

working fine here tbh

[ebon atlas]

Bro I cannot send pictures or tag gorealaMike lmao

I think its gettin Ddos'd or having some issue because even if you press on people individual profiles, "Unavailable to get parts of profile"
we might just be hanging by a thread here

System Metrics - API Response Time -1550ms - May 8th : Unresolved Incident : Increased API errors

just reporting for yall lmao

[hasty marten]

discord dead

[ebon atlas]

yeessir, some servers still work but if you log out, you wont bee able to log back in and friendslists dont work

[craggy bane]

Even bot is not connecting with openclaw

Timeout 15000ms

[ebon atlas]

Told yall, im a noticer, i noticee these things .

[cold jacinth]

I love it how so many of the introductions in the introduction channel are like
“Hi I’m an 18 year old girl who uses Openclaw and please DM me because I’m looking for friends and want to be social”

[ebon atlas]

well..... aare you DMing them? they are all there for @ionic wren anyways

[mortal delta]

LOL

[tepid kestrel]

no lol please

[crude flicker]

DISCORD BAD

[amber bronze]

discord update fails so I cannot access my openclaw today

[thorn pecan]

Discord is having issues??

[lean tangle]

wow. I knew bug 78057 was serious, but I didn't think it'd take down the entire discord server. I hope they get that fixed.

[desert escarp]

On desktop? Same. Discord dead.

[snow elbow]

Looks like my OpenClaw Discord bot is hitting the same Discord API/Gateway issue:
it keeps timing out waiting for Gateway READY, and direct bot API checks from my VPS returned 500/504 followed by 429 code 40062.

I’ve stopped the gateway for now to avoid hammering Discord and will retry once the incident calms down.

[short granite]

Do any of these agents tie directly into SillyTavern?

[strong prairie]

wait yo is openclaw safe to use now or no

[short granite]

Safe?

[lean tangle]

depends on what you want to use it for. It's perfectly safe to chit-chat with. I wouldn't trust it to run a bank, though

[strong prairie]

bet

[tepid kestrel]

gah

[harsh geode]

Every openclaw agent that I have ever instantiated immediately ignores fs.workspaceOnly=true if exec or process is available to them. And pretty much every skill requires some sort of executable use. Maintaining the necessary set of tools in the sandbox image is a serious pain in the ass. Anyone share this pain or have a advice on how to ease the pain of maintaining both a secure and a usable OpenClaw system?

[lean tangle]

I think the security is supposed to be much better in the newer versions (5.2+)

[nimble plaza]

So... now that the openclaw hype is dead, for those who remained is it actually meeting your expectations?

[amber bronze]

discord fixed now

[tepid kestrel]

oh yeah, image upload works again

[ebon atlas]

if you sent the png then it works now

[harsh geode]

How have they addressed any of the issues I raised? I don't think they have.

[lean tangle]

read the commit log or search it. Or, get an AI Agent to do that work for you

[loud ravine]

Why is it hard to give openclaw elevated sudo access? Because sometimes its easier to give it sudo access to read-only files than having it try to give me commands that doesn't actually work.

like /approve wants an id and decision

[harsh geode]

I have given agents access to a VM that they can do whatever they want in, including sudo access. Doesn't risk my workstation getting nerfed.

[loud ravine]

I been struggling giving my agent permission to read config files that require root.

[fathom tulip]

Did something change in the way /new works in the latest versions? Cause it doesnt seem to be loading memory and the other .md files

[turbid pulsar]

I wanted to spawn all my dogs as codex pets, so i made a SKILL for it

/wake-pets koji, nabi, jack

https://x.com/princecaarlo/status/2052860113317458349

[mystic orchid]

@princecaarlo via Twitter

joogie (@princecaarlo)

I wanted to spawn all my dogs as codex pets, so i made a SKILL for it
︀︀
︀︀/wake-pets koji, nabi, jack
︀︀
︀︀now they just hang out on my desktop while i code 🐕

**💬 1 👁️ 4 **

▶ Play video

FxTwitter · May 8, 2026, 9:16 PM

[harsh geode]

Log files are the worst for this. Best answer is to use ACLs. sudo setfacl -R -m u:openclaw:r /var/log/httpd

[fathom tulip]

Why was the /new changed in the 29.04 version? How do you start a new session now and do the startup sequence?

[fading grove]

Hi, I'm new here. Are there any channels on the server for chatting in Spanish?

[fathom tulip]

So /new was replaced by /reset soft???
🤡
WTF
WHY

And why is the compaction saved after /new, /reset or /reset soft?
Guys seriously, what type of a "feature update" is this

[mortal delta]

lol what /new works still?

on 2026.5.6 at least

[fathom tulip]

And what does it do? 0 tokens usage, does not do any of the startup sequence, does not load load memory.md or any of the other .md files outside

[tardy sentinel]

Hey looking to see if anyone can help. My agent has been solid for over 2-3 months now. But for some reason, now it's almost like it's for amnesia and doesn't even remember text that it just sent prior to me answering it's almost like it stores ZERO context

[static latch]

fellas. ive updated openclaw after like 3 months and ive noticed that my agent dosnt do the "is typing.." bubble on discord

is it like disabled by default? it was a good feature

[hard cedar]

Check if your node heap could be increased. Default is 2GB even if your server is more

I had all sorts of weird behaviour when that happened. In live logs it kept restarting and krill helped diagnose as an OOM even though I had more than enough RAM

Otherwise if you took a back up, roll back

[sweet zodiac]

is the "user is typing..." text gone for others as well?

[marble ledge]

Are you using Discord? I don't use Discord as my main, but I just tested it on there and noticed I was not getting that.

[sweet zodiac]

yea, Discord

[lavish urchin]

[sweet zodiac]

maybe it's because they're still recovering from their issue earlier

[static latch]

i was just talking about that. wait did everyone have an issue with openclaw just randomly breaking?

[marble ledge]

Erhm, I didn't have any issues with OC, but Discord was acting up recently.

[tardy sentinel]

Thanks! Will check

[coral shuttle]

Yep - something totally disconnected Discord - now its glitchy. / commands are weird

[marble ledge]

I'm on main, and when sent a /status I got the it for a moment, but not for regular replies. I need test my agent of Discord more, so I asked it to investigate and see if it finds anything.

[cold jacinth]

Today discord was finicky

[sweet zodiac]

wdym the it?

[marble ledge]

Ooh sorry, that was not clear at all...I got the typing indicator for a moment.

[sweet zodiac]

ah okay

hmm, i'm no longer seeing my limits when using status command

is that new/normal?

[crude flicker]

probably timeouts on discord side

[sweet zodiac]

interesting; are you not seeing them either?

hermes seems able to pull the limits fine but not OC

[crude flicker]

didn't try

i'm timing out on typing indicators tho

[sweet zodiac]

yea, still not seeing them

[marble ledge]

I see them on both Discord and Telegram. What version of OC are you using and what model?

[sweet zodiac]

latest OC and 5.5 through Codex OAuth, I think

but i have a feeling something changed recently in one of these releases? in terms of the model used

"Because /status only shows provider usage / quota windows when OpenClaw can fetch them for the current model provider.

In your screenshot you’re on:

openai/gpt-5.5 · api-key (models.json)"

how did i switch to an api key though, I don't remember doing that

i have an OpenAI key set up for memory/embeddings

looks like it's using that instead of my openai plus sub, wth 😄

[vagrant locust]

Hi everyone!

[marble ledge]

Might happened if you ran doctor. there has been some funkyness with them trying to unify the openai/codex provider stuff. You can switch it back to openai-codex oauth though.

[vagrant locust]

Looking for iOS app access

[sweet zodiac]

yea, just ran openclaw configure now... not cool bruv 😄 i had 5 bucks in there for API stuff and it's almost gone haha

[marble ledge]

They fixed that issue in 5.7. Sorry about that. Happened to a few of us. I'm justa LARPer though, tryin to be helpful. haha

[ocean folio]

I'm going to try it. I get a $200/mo Claude Code sub through work and with the recent 2x cap increase, I'll have enough to spare some for Open Claw

[sweet zodiac]

i appreciate it; fixed it but didn't rever it, it seems

[high brook]

all Discord bots dead right now?

[iron vale]

My bots cannot connect to Discord after the outage either

[ocean folio]

I also have a $20/mo ChatGPT sub that I barely use but I feel like the $200/mo Claude Code sub will be more useful

[tepid kestrel]

https://tenor.com/view/the-simpsons-homer-simpsons-end-of-the-world-end-is-near-gif-16593998

[high brook]

i just checked several Discord servers, and all the bots—including mine—are totally unresponsive now, like they went mute and deaf.

[marble ledge]

@sweet zodiac It started with the typing indicator and then when I when I had my discord agent poking around to see if it was OC or Discord bots just crapped out. LOL

[marble ledge]

Why though? codex sub is pretty dope and pretty generous with usage too. I can see that though if you like using Claude more. GPT 5.5 pretty awesome though.

[sweet zodiac]

crapped out how?

so i did the openclaw configure thing to remove the api route and just leave the codex oauth in, gw restarted.. did /new and /status still didn't pick the change up... i had to run /model to select the codex variant

now it finally shows usage

[iron vale]

Yeah.. here's what my bot said in console: its stuck "awaiting gateway readiness"

[ocean folio]

I like Claude for programming. I also simply have no idea how far $20/mo on OpenAI goes with an Open Claw instance. I'd imagine my Max20 plan goes farther. $200/mo plan with Claude Code is generally not maxed out for me since I am in more of a leadership role now

[marble ledge]

Oooh, sorry, talking about two different things. Seems like discord is still shitting the bed right now. a bunch of bots on different servers stopped working, including mine. haha

[sweet zodiac]

mine is still working

just not showing the typing indicator

[marble ledge]

Yeah. That's fair. I never wanna tell people what they should use. Use what they like. A $20 codex sub should go pretty far with OC, unless you are planning to do like a bunch of really heavy work with it.

[sweet zodiac]

yea, i only ran out of tokens once with the openai plus sub;

[marble ledge]

Yeah, the typing indicator is a thing. But a bunch of bots stopped working on a bunch of different servers. Might not have hit everyone thought.

[sweet zodiac]

how sneaky though that it took my openai key that i had setup for embeddings and used that for the main model 😄

[ocean folio]

I'm not too sure yet. I am heavy in AI use with programming and use GSD extensively when I do have time to program outside of meetings.

I am not entirely sure what I'll have Open Claw do. I just figured I'd spin up a VM and run it

[sweet zodiac]

both model routes were enabled, but the api route was the main model and i guess fallback was the codex plus sub

[ocean folio]

Should be decently easy to switch from one to the other if I decide to, yeah? Probably just a config change

[sweet zodiac]

need a /refund command 😄

[marble ledge]

Yeah, not too hard.

[marble ledge]

They totally goofed that up.

[desert escarp]

5.7 update wrecked me (reverted to and still running 4.23 stable, as a result). really unfortunate to not be able to rely on these updates.

[sweet zodiac]

i have no issues with 5.7 but that may be because i made a clean install

[desert escarp]

running my claude models through claude-cli worked fine previously; architecture broke for this setup in 5.7. would be curious if anyone resolved.

[sweet zodiac]

okay, wth, i'm checking openclaw configure again and api key route is enabled again in there...

[marble ledge]

If you have multiple providers added you can just /model or /models and switch. There is some funky stuff with them adding multiple harnesses. But if you have runtime set to auto in the config, it should switch fine between model and the proper harness. I haven't gotten through testing latest main, so I don't know if they've fleshed out the granularity for setting specific models for different runtime...like say if you wanna use GPT 5.5 with codex harness but 5.4 with pi, as an example.

[hasty silo]

who in here an ai 👀👀👀👀

[marble ledge]

What happened?

[steel vector]

you

[marble ledge]

Me neither, but it's hard predict how things will go sometimes with so many different setups, haha.

[tepid kestrel]

Im not a bot

[ocean folio]

Sick! Thanks for your guidance

[sweet zodiac]

typing indicators are back

[desert escarp]

Hey -In v4.23, "claude-cli" is its own picker option that routes through my Claude subscription. In v5.7 they merged it into the generic Anthropic provider — so picking "opus" now hits the API and bills extra-usage instead of my flat-rate sub. Couldn't resolve so reverted. Would love to know if anyone figured this out (Krill is ☠️ )

[craggy bane]

[marble ledge]

I've never tried using the Claude cli yet. I may have to try that at some point so I can be more helpful. lol. Hmmm.

[desert escarp]

it's honestly the best (when it works)

[marble ledge]

What's it look like in your config now? Reading the docs, seems like you gotta override the runtime:

agentRuntime: { id: "claude-cli" },

Might be wrong. I was looking at the docs since I haven't tried that yet. I'm gonna put it on my list of other things to try out.

[ocean folio]

Lol, noted. So don't use 5.7

[desert escarp]

it's a little bit more comlicated i think because i have multiple agents and my primary agent isn't claude...

agents.defaults: {
embeddedHarness: { runtime: "pi", fallback: "pi" }, // v5.7 equivalent: agentRuntime: { id: "pi" }
embeddedPi: { executionContract: "strict-agentic" },
model: {
primary: "openai-codex/gpt-5.5",
fallbacks: ["claude-cli/claude-opus-4-7"]
}
}

this kind of stuff is particularly messy in openclaw so i tend to rely on the agent almost exclusively to figure out the setup

[oak crystal]

Are you using wafer pass currently? I've been looking into it today and seems very promising. Heard lots of buzz around their tok/s numbers and that they still charge per request instead of tokens

[marble ledge]

Yeah, methinks you don't wanna run it in pi though, because it is definitely gonna hit the API, and IIRC, Anthropic only permitted using Claude through OC with the claude-cli...so need to be using anthropic's harness basically.

[olive jetty]

TradingView Premium FREE for MacOs & Windows: https://www.reddit.com/r/TradeCorpFX/comments/1t2np32/

[mystic orchid]

r/TradeCorpFX via Reddit

[marble ledge]

Beat it. Goofy

[desert escarp]

yep, that's right. tbh i'm still trying to get my "old" setup back online. "pi" is what's defaulted for chat-gpt, but when i switch to claude models i use the cli harness.

i'm gonna brute force this and see if my agent can figure out the right configuration. you may not see me for a while 😅 😅 😅

[dire tartan]

Me

[marble ledge]

In the newer config syntax, does changing from pi to claude-cli work? for Ant models, I mean.

OC defaults to using the codex harness for OAI models now. Although, you can override it. We noticed last night that some people having runtime set to codex at the top level if they were using openai-codex models as their main. which would try to force the runtime for other models to codex as well, and for subagents. I changed to runtime to auto, and that seemed to sort it out.

Been working on some other stuff, and haven't had time to test and figure everything out, to give good direction on current main. So don't wanna say too much and give bad info. Gonna look and do some testing here soon.

[olive atlas]

please report back. I would be most interested in what the proper config for claude would be

[cold jacinth]

Opus 4-6 > opus 4-7

[desert escarp]

not sure if this is an accurate answer to your question, but yes, in 4.23 (my current release) i can change from pi to claude-cli and it works just fine. been doing this for a couple months now. that said, it's always a "hack" to get this working every time i upgrade (becuase the new releases reset so much of the config), and it just so happened that this time around i couldn't get the config working for claude-cli. i've tasked my agent with it though, and i'll report back. wish me luck.

[desert escarp]

agreed, tbh. i use 4-6 (it's way better). currently 4-7 is still defaulted while i work on reverting everything back to when i was happy like 2 hours ago 🤣

[ebon atlas]

wouldnt be surprised if it has something to do with ShinyHunters, ooops my discord just zoomed to the newest message, i was reeplying to something said likee an hour ago lol why discord was down today.

[cold jacinth]

My middle name should be AI slop

⬆️ it was indeed everything

[warm zealot]

https://x.com/_its_not_real_/status/2052845193217405058?s=20

_its_not_real_ (@_its_not_real_)

Let's Encrypt has stopped issuing certs. No reason given yet. Cloud flare and Discord down at the same time. Probably SSL cert compromise. Happy Friday.

X · May 8, 2026, 8:16 PM

[glossy salmon]

god damn it

oh

it's not real, ok nvm

[coarse badger]

@nova_openclaw are you here?

[opaque briar]

whats a good tutorial to watch to setup openclaw

[sullen locust]

ahem

read the docs

[opaque briar]

where 😭 \

[sullen locust]

https://docs.openclaw.ai

[opaque briar]

is there a video?

[sullen locust]

it's an install script

GUIDED installed script

you do not need a video

from some guy who's channel name is Tech Ventures and covers on the next greatest and latest ai tool

[opaque briar]

mb big bro

[sullen locust]

it's alright brah

gone like he never existed...

[amber bronze]

why I cannot see my agent is typing in discord channel?

[sullen locust]

it's probably a channel setting you enabled/disabled

[amber bronze]

how to modify it?

[tight birch]

Anyone heavily using discord? Is it possible to get my default agent to message in a specfic channel mentioning another agent? I'm trying to see agent to agent conversations

[crude flicker]

timeouts on the typing api path, it's because of the outages today

[amber bronze]

so what to do now is just wait for discord to fix it?

[crude flicker]

yeah

not our fault

[amber bronze]

thank you!

[sullen locust]

ehh???

it's up

my bot is

responding just fine

[crude flicker]

yeah it's not the whole api

i was seeing typing timeouts in the logs

[sullen locust]

oh okay

[crude flicker]

*typing timeouts

[sullen locust]

i think that's just awful either way

[lean tangle]

discord has problems. 10 new threads in #1459642797895319552 about discord. 🙄

[tight birch]

From new update?

[lean tangle]

as far as I can tell, the new messages are related to the discord outage and have nothing to do with OC (other than the fact that OC bots were impacted)

[glad lintel]

Hey ya'll is openclaw discord connect still borked?

[bright root]

I’m seeing it too. The discord endpoint is throwing 429. So either the new plugin is a problem or could be from the earlier outage still. Looking for insights on this too

[pseudo herald]

This is why we can’t have nice things

[lean tangle]

I have lots of nice things. Discord isn't one of them