#general

1 messages · Page 18 of 1

next swallow
#

I want to use exa search for every model but for gemini models, i want to use google search grounding instead of exa? Is that possible? Can anyone share what to change in openclaw.json or tell what to change to allow both of these functionalities?

frozen zealot
#

i regret running brew upgrade ... it's been ages, and it's still going

zealous lagoon
gilded obsidian
#

@spark hatch is sending rando DMs: Hi Joseph, Nice to meet you! I found you on Openclaw. How was your openclaw experience?

lost gulch
frozen zealot
zealous lagoon
#

openclaw plugins install @openclaw/codex@latest @warped terrace this might help.

gilded obsidian
#

How many extra tokens does active memory use, anyone with experience? 😛

toxic elk
gilded obsidian
#

Anyone know why everytime I update the claw I get this: Config was last written by a newer OpenClaw (2026.5.12); current version is 2026.5.7.

#

I stopped the gateway, then did update? 🙁

keen dawn
#

keep in mind tibo will reset today usage limits --> fast xhigh lol

keen dawn
gilded obsidian
#

Noper

zealous lagoon
gilded obsidian
#

Right but if I stopped the gateway, did the update, how could it be the old version?

zealous lagoon
#

That might sound obvious but I had the same thing happen to me. Lemme see if my claude remembers what it did

zealous lagoon
gilded obsidian
#

see now I stopped it again, ran update again, and now its fine.. makes 0 sense: Update Result: OK
Root: /home/Brachyura/.nvm/versions/node/v22.22.2/lib/node_modules/openclaw
Before: 2026.5.12
After: 2026.5.12

Total time: 40.79s

#

Everytime I update I have to do it twice it seems, very strange

keen dawn
zealous lagoon
#

there are multiple processes, I think the update procedure is too kind and the process doesn't seppuku

lost gulch
#

Can’t wait for Peter to release openclaw as the official AI on the trump phone

zealous lagoon
#

sponsored by Grok

gilded obsidian
#

@ofv I was going to make a joke about that and targeting.... errr n/m its in poor taste 😉

keen dawn
#

So either your PATH points to an old CLI wrapper, or your gateway/service is pinned to an old install path, or you rolled back the binary without rolling back the config.

#

you could check it

gilded obsidian
#

I mean... the path hasn't changed? it's a single install.. hmm

zealous lagoon
#

you'd be surprised...

keen dawn
#

which -a openclaw
openclaw --version
openclaw config get meta.lastTouchedVersion
openclaw gateway status --deep

#

Also inspect the service/LaunchAgent ExecStart/Program path. Make sure the CLI you run and the gateway service both point to the same OpenClaw install/version.

#

had the same issue a time ago

gilded obsidian
#

hmmmmmm

#

Is this a problem or a harmless symlink? Brachyura@Brachyura:~$ which -a openclaw /home/Brachyura/.nvm/versions/node/v22.22.2/bin/openclaw /home/Brachyura/.npm-global/bin/openclaw

zealous lagoon
#

I'd expect that to be a versionning simlink but you can check I guess ?

gilded obsidian
#

ExecStart=/usr/bin/node /home/Brachyura/.nvm/versions/node/v22.22.2/lib/node_modules/openclaw/dist/index.js gateway --port 18789

#

whoa why did that go bold lmao

zealous lagoon
#

I've spent all day doing non-AI things and my brain is fried. I don't even bother with this nonsense of hacking the install to get it working, I just get Claude or Codex to do it.

#

the pound sign probably.

#

or another markdown tag. I couldn't even send a pound sign to show you hahaha

keen dawn
gilded obsidian
#

yeah I thought they were banned in here 😄

keen dawn
#

One thing I would fix: your service runs /usr/bin/node against an OpenClaw package installed under NVM. Ideally the gateway should be reinstalled from the intended NVM OpenClaw binary so the service uses the same install/runtime.

#

that could actualy cause it too

gilded obsidian
#

so use .nvm/versions/node/v22.22.2/bin/node instead eh

#

that kinna makes sense I suppose

#

Brachyura@Brachyura:~$ which node
/home/Brachyura/.nvm/versions/node/v22.22.2/bin/node

#

Doesn't matter it's the same, the service is a user service too, so it should be using same node binary

#

oh wait, duh

keen dawn
gilded obsidian
#

6am brain getting to me :D... I may have fixed it

keen dawn
gilded obsidian
#

[Service]
Environment=PATH=/home/Brachyura/.nvm/versions/node/v22.22.2/bin:/usr/bin:/home/Brachyura/.local/bin:/home/Brachyura/.npm-global/bin:/home/Brachyura/bin:/usr/local/bin:/bin

keen dawn
gilded obsidian
#

I think it's sleepy time 😛

#

You know how much work I've got done with my little crabbies.. and I have most of the plugins disabled LOL

keen dawn
gilded obsidian
#

Thanks for your help 🙂

keen dawn
hollow marlin
#

Dm Me lf You Struggle To Pay Your Bills I Wanna Make You Smile Fr❀☘︎

cyan willow
#

fellow Claws, where do I see if there is an issue with 5.12 to approve commands from OC? it's trying to do an approval but I'm not getting the approval <id> Has something changed?

twilit ridge
#

Ubuntu as main operating system

#

very smooth

rich remnant
#

Hey folks hows the latest update...is it safe lol

lost gulch
cold jacinth
#

Hey guys, I’m new here. Safe to update from 4.23?

cold jacinth
steel vector
ruby crane
cold jacinth
#

Can we give a thank you to all of the maintainers and clawtributors for the amazing latest release? Reviews have been absolutely phenomenal. Makes one even more proud to be a “tiny” part of this.

ruby crane
#

if the last couple of weeks showed me anything it's that they hype cycle (and reduction of it) or bumps in the road are irrelevant. the claw will remain the law

steel vector
zealous lagoon
#

But also 5.12 forces codex runtime without actually bundling codex. I think there oughtta be a pinned comment in #1459642797895319552 about that.

steel vector
lost gulch
steel vector
#

not when i set up my new agent yesterday

#

on ollama cloud

zealous lagoon
steel vector
#

then, why wouldn’t it?..

ruby crane
zealous lagoon
#

Not sure actually whether it's forced or not. I already had codex installed.

zealous lagoon
violet garden
#

hm. openai keeps giving me codex tokens

steel vector
ruby crane
steel vector
zealous lagoon
#

idk my install works fine and I've had the plugin installed for ages and at one point I had two of them. A minor teething problem. But a lot of users coming in here with the "codex plugin not registered" error and I'm trynna figure out why.

zealous lagoon
violet garden
steel vector
#

mine isn’t reset yet thank god

violet garden
#

overused my weekly last week and ran into my token stockpile down to about 400. have 1900 again

ruby crane
#

API or subscription? just so we're on the same page?

violet garden
#

gpt+ codex account

zealous lagoon
#

There it is. @openclaw/codex is not in dependencies, not in bundledDependencies — it's not declared at all in the 5.12 package. The only
@openclaw/* package in the host's deps is @openclaw/fs-safe.

So 5.12 hardcodes a routing requirement on the codex harness, ships a runtime auto-install path as the delivery mechanism
(ensureOnboardingPluginInstalled — note the onboarding name, this was designed for interactive first-run, not as a silent required-dep
installer), and declares zero dependency on the package that provides it.

It's a straightforward packaging bug: the codex harness should be either in dependencies (so npm installs it alongside openclaw) or in
bundledDependencies (shipped inside the tarball). Instead it's been left as an optional onboarding install that silently fails in
non-interactive or network-constrained environments, producing the "harness not registered" error.

cold jacinth
#

My telegram stopped working. Can somebody tell me why?

zealous lagoon
#

I don't even trust Claude on this any more. It's given me so much BS recently.

ruby crane
cold jacinth
zealous lagoon
ruby crane
steel vector
zealous lagoon
#

I guess they cleared a lot of the bundling problems up in further fixes. I mean we're already a few betas down the line lol. I just want an easy fix in #1459642797895319552 to direct all the users coming in here complaining about the error.

ruby crane
#

that's what happened to me when i first installed beta.. something i can't remember, whatever started to enforce the switch

zealous lagoon
#

yeah that's kinda the problem with beta testing... you don't catch the bundling errors because the betas are never bundled properly anyway

ruby crane
#

i did have one issue, which pash cleared right up

ruby crane
zealous lagoon
#

I did bring one of them up, but it seems it was done right in my specific case.

ruby crane
#

so what you're saying is that everyone on 4.23 should install 5.16-beta.2 right now

zealous lagoon
#

oh yeah. GO FOR IT

ruby crane
#

agree lol

zealous lagoon
#

but make sure you update all the plugins

ruby crane
#

thats what I heard and I'm sticking with that 🙉

zealous lagoon
#

I wanna go for some token burn but I can't think of anything I want to do right now that doesn't involve screwdrivers and soldering irons.

ruby crane
#

clearly you need to buy some robot arms, burn your house down, and get more tokens to make up for it

zealous lagoon
#

hahah

cold jacinth
ruby crane
#

you will simply need: -

  • a new house
  • new robot arms
  • more tokens to support the even longer and more detailed 2.0 skill
steel vector
zealous lagoon
#

My bot says that 5.12 forces codex runtime, hardcoded not config, for openai models on oauth. But doesn't bundle the plugin, so user has to know that and manually install it. Says Claude.. and I made it double-triple-check. ymmv. But I think that's what's happening.

#

I mean it's a one-liner but how would the user know?

ruby crane
#

that was my experience early beta. had to rollback. next beta fixed that iirc. actually, hang on. can't remember if i created an issue or just complained to talked with pash here

cold jacinth
steel vector
zealous lagoon
# 
  @openclaw/* package in the host's deps is @openclaw/fs-safe.

  So 5.12 hardcodes a routing requirement on the codex harness, ships a runtime auto-install path as the delivery mechanism
  (ensureOnboardingPluginInstalled — note the onboarding name, this was designed for interactive first-run, not as a silent required-dep
  installer), and declares zero dependency on the package that provides it.

  It's a straightforward packaging bug: the codex harness should be either in dependencies (so npm installs it alongside openclaw) or in
  bundledDependencies (shipped inside the tarball). Instead it's been left as an optional onboarding install that silently fails in
  non-interactive or network-constrained environments, producing the "harness not registered" error.

Here. Could anyone confirm? I got claude to check and check again... this is what it says.

zealous lagoon
ruby crane
#

nope, i was misremembered. it was downloaded automatically at start (though do bear in mind i run a different setup from most). what actually happened and i created an issue for was that non codex providers were being prefixed with codex

zealous lagoon
#

or maybe you already had it from previous tests... I did.

cold jacinth
zealous lagoon
#

I mean this is very clear :

#

It's a straightforward packaging bug: the codex harness should be either in dependencies (so npm installs it alongside openclaw) or in bundledDependencies (shipped inside the tarball). Instead it's been left as an optional onboarding install that silently fails in non-interactive or network-constrained environments, producing the "harness not registered" error.

ruby crane
#

I literally recall it being installed and seeing that happened during pod startup

#

but it could be a regression for sure

#

or just container packaging difference

zealous lagoon
#

I'm not gonna repeat that last paragraph 100 times but I don't think it could be any more clear. THe fact that OUR installs all worked is almost certainly because we're not noobs updating from 4.23 or trying the codex harness for the first time.

#

Meh. imma check the current bundle fixes it. But meanwhile also post in #1459642797895319552 the one-line fix for everyone who comes in here.

cold jacinth
ruby crane
cold jacinth
#

And what happens then if someone doesn’t download the codex harness, does it fallback to the pi harness or does it just not work?

ruby crane
cold jacinth
#

Well. I saw a Reddit post where they had a Mac Studio m3 ultra 512 on Amazon for slightly over MSRP. I looked at it online, saw two available. Went to make tea. Came back 5 minutes later. Both gone 😂

ruby crane
#

so, aware that I have essentially ZERO typescript skills (I can read it, but i don't do so as often as I should) and that my git knowledge is spotty beyond add/commit/push/pull and gets spottier when I'm just tying to get something done I asked (forced) my agent for a unbiased (well, "unbiased") review of what I need to improve to make these PRs smoother. it's actually too cringeworthy to even share lmao. "you're a fucking moron" basically

#

the summary. if you want to see how embarassing it got on the detail, do tell

The biggest improvement areas:

Git confidence
Make branch/diff/head/base inspection boring and automatic.

Reading small code paths
You do not need to write the patch solo, but you should be able to follow: event emitted here, consumed there, state cleaned there, test proves this.

CI triage discipline
You did improve here today. Keep the rule: unrelated red CI is evidence to classify, not permission to broaden the PR.

Exactness under pressure
You have good instincts, but when things get noisy you sometimes need the checklist sooner: current head, changed files, tests run, what changed since last proof.

Blunt bottom line: you are not yet strong at git or coding mechanics, but your judgement, testing discipline, and bug-shaping instincts are much better than you probably think. For agent-assisted engineering, that combination is powerful. Your ceiling goes up fast if you make git less scary and learn to read diffs/tests with more confidence.
light sorrel
#

Hello

#

I’ve been able to setup my bot on discord, it works in DMs but doesn’t inside servers
It just shows typing and that disappears in a few seconds

zealous lagoon
# cold jacinth And what happens then if someone doesn’t download the codex harness, does it fal...

it just errors plugin not registered. It hardcodes the dependency but it only downloads it on first run , i.e. onboarding. So if you didn't have it already and you're updating not fresh installing... you're screwed. It's an easy fix, just do onboarding again or do the plugin install one-liner. But a lot of people dn't know that and have been coming in here all day with that error. It's been bugging me all day because I've been doing other things. Glad I was able to get to the bottom of it finally. Will put something in #1459642797895319552 in a minute. PR is taking a while for some reason.

zealous lagoon
cold jacinth
light sorrel
zealous lagoon
light sorrel
frozen zealot
ruby crane
pulsar iron
#

Should we be switching to codex as the runtime?

frozen zealot
feral turret
ruby crane
young anvil
#

Ah not yet, app needs to be updated but

zealous lagoon
zealous lagoon
pulsar iron
#

So if you switch to gpt 5.5 it will auto use codex harness?

zealous lagoon
tepid kestrel
cold jacinth
#

My agents have been down for 2 days because of a power outage at the office two days ago. One feels the difference lol. Not getting all the shit I would always get

#

Weekly planner etc etc.

zealous lagoon
sullen locust
feral turret
#

so wait am i understanding this correctly? openclaw is totally open source and anyone can work in the repo and submit prs?

clear haven
#

Yes

feral turret
#

holy cow

burnt mulch
#

i need sm help😭, can someone just hop on #1459642797895319552 and see the openrouter setup doubt😭

sullen locust
#

and i will help

burnt mulch
zealous lagoon
# feral turret so wait am i understanding this correctly? openclaw is totally open source and a...

Sacrifice your tokens to the PR gods. But seriously... once you get used to using OpenClaw and understand it a bit better. If you see stuff broken, come in here and chat about it... if it's really something you can fix on your own install, instead of just patching it for yourself, do a bit more research on integration, clashes, fixes in the works etc (easily done with a bot and the right SKILLs).. submit your PR so maybe everyone gets your patch... and drop a note in #clawtributors . I was patching my own shit for a while before I realised that was a thing you could do... and it helps the community rather than just bitching about stuff.

feral turret
#

i am totally blown away

#

my only regret with this is not starting sooner. and staying up so late all freaking week this week

zealous lagoon
#

I mean, that's opensauce but the work involved in doing all that coding all by yourself is now compressed to like a 1h job.

#

instead of weeks.

feral turret
#

yea wild. use an ai agent to work on ai agents

zealous lagoon
feral turret
zealous lagoon
feral turret
zealous lagoon
ruby crane
zealous lagoon
#

But that's the beauty of it. If you went to school even 30y ago you can read just about any programming language, and the bots can write it for you so you don't have to remember syntax or anything, just understand what the bot is doing. I freakin' love it.

ruby crane
#

I've noticed all my analogies are baby-related

zealous lagoon
feral turret
#

and congrats @ruby crane that's exciting

ruby crane
feral turret
#

this your first?

zealous lagoon
carmine echo
#

hi im new!

feral turret
ruby crane
#

he unfortunately inherited my giant cranium, which means it has to be c-section

zealous lagoon
#

hello taternug

ruby crane
#

i literally spend months trying to find even an elasticated hat that fits

ruby crane
feral turret
#

my daughter turns 10 in july. cherish every second. everyone says it and then a decade in you're like well, shit.

ruby crane
#

we just found out we're pregnant two weeks ago. at least I'd swear it was that

#

just zips by man

feral turret
#

wild

ruby crane
#

anyway, really appreciate everyone's kind words. ty!

feral turret
#

we've got our last elementary school dance tonight

#

they start middle school in 5th grade here. and all the little girls are all boy crazy and stuff it's insane

ruby crane
#

we kinda wanted a girl, but we got a boy and that's all good. apparently boys cry about half as much and cost half as much too 😂

feral turret
#

i thought i wanted a boy but my daughter is my princess bro she is such a daddys girl

#

i still kinda want a boy but we will see if me and my girlfriend can get along lmao

ruby crane
#

oh i would have spoiled a girl

feral turret
#

my daughter is from my first marraige my new girl really wants a kid but i am like very hesitant

ruby crane
#

i'll still spoil my boy, but not quite as much

feral turret
#

got burned once don't wanna get burned again lmfao

#

touched that stove and now i'm scared of stoves lmfao

half stratus
#

What is the best way to run OpenClaw and other Agents these days? Still MacMini from a pricepoint/performance perspective?

feral turret
#

but i run no local models i am using codex oauth on 5.4-mini

ruby crane
sullen locust
feral turret
#

i think i'm gonna set up a git repo for my .openclaw directory today

feral turret
sullen locust
#

8gb pi 4/5

#

done

ruby crane
#

what i'm saying is, yes, do that

feral turret
#

yea i'm more worried about disc failure or something to that effect

#

i'm at like 50 hours into this thing this week i don't want to lose all that work lol

livid sinew
#

Was codex 5.3 removed?

feral turret
zealous lagoon
feral turret
zealous lagoon
#

was debating for my 2015 macbook pro. But I just hacked the OS update instead. Sometimes OC has been a struggle. 4.29 was death. We're back now tho.

feral turret
#

gonna hang at desk till 10 and head outside get some yard work done

zealous lagoon
feral turret
#

touch grass. i've barely been outside this year lol

zealous lagoon
#

well now you can use OpenCLaw, Codex, and Claude from your phone, whilst you mow the lawn lol

feral turret
#

yes this is valid and with my airpod pros it may pick up my voice lol

#

i was in bed last night setting up daily vc news etc

#

it's so hard to put this stuff down

zealous lagoon
#

mental. It really is. I didn't sleep for weeks.

feral turret
#

reminds me of 2000 when i started teaching myself html i was on htmlgoodies.com all day every day

#

yea it's kind of bad to be honest lol

#

wow htmlgoodies is still around

#

"Did you hear…

eCommerce! eCommerce! Rah! Rah! Rah! Apparently after three
good years of business on the Web, colleges and universities
have decided it’s here to stay and have begun offering
degrees specifically geared to cyber-business."

fleet glacier
#

How to use Deepseek v4?

#

I’m new

livid sinew
#

Lmao

#

Yall really tried to skimp out on compute after getting acquired by OpenAI

#

Ez patch: “You were right — OpenClaw did support Spark, and we can bring it back locally.

What changed upstream:

  • Spark support was added around these commits:
    • e3cb2564d7 — allowed gpt-5.3-codex-spark in fallback/thinking
    • 2e04630105 — added openai-codex/gpt-5.3-codex-spark forward-compat model
  • Then OpenClaw changed course:
    • d5b3f2ed71 made Spark Codex-only, suppressing direct openai/gpt-5.3-codex-spark
    • current builds go further and suppress openai-codex/gpt-5.3-codex-spark too, with the reason: “no longer exposed by the OpenAI or Codex catalogs”

I patched it back locally, Codex-only:

  • removed the openai-codex Spark suppression from the bundled OpenAI plugin manifest
  • restored the Codex forward-compat fallback for gpt-5.3-codex-spark
  • kept direct openai/* Spark blocked — only openai-codex/* is enabled
  • set Secretary to openai-codex/gpt-5.3-codex-spark
  • moved the cron to Secretary + Spark”
#

L

turbid kelp
#

What is the best way to use Clause or ChatGPT to fix and manage OpenClaw? Is it by installing Clause Code on the same machine/instance as OpenClaw? Or something else?

fleet glacier
#

Which one should I use

V4 flash or v4 pro?

steel vector
coral otter
#

My pr for a fork of one of peters projects got merged 🥳 (my first publicly merged pr)

sullen locust
#

YOU NO BECOME OPENCLAW CONTRIBUTOR?!

coral otter
exotic zealot
coral otter
turbid kelp
exotic zealot
winged ferry
coral otter
#

Codex limit reset ca 10 hours from now

tidal root
#

I am using Gemini api from Google, is there is anther provider or 3rd party that can provide a better quota price than Google's?

turbid kelp
frozen zealot
turbid kelp
#

ChatGPT is telling me Claude Code is better than Codex CLI for managing OpenClaw

velvet locust
dire swan
turbid kelp
dire swan
turbid kelp
#

Gonna ask chatgpt about that one

steel vector
zealous lagoon
turbid kelp
# dire swan We can connect IDE to Remote server or WSL directly

Where Antigravity starts to fail for your use case
For real OpenClaw incidents, IDEs struggle because:
A) No system-wide visibility
It sees:
files
It does NOT naturally see:
running processes
logs over time
system state drift
port conflicts
memory behavior
runtime errors across services

cold jacinth
#

Codex / code

#

It fixes things and finds config mistakes

dire swan
turbid kelp
cold jacinth
#

Claw seems to sometimes be blind to its own problems

zealous lagoon
cold jacinth
turbid kelp
zealous lagoon
dire swan
light sorrel
#

Is there a way to make my bot like go strait to what I tell it to or is that like entirely dependent on the model

Cause I told it to setup telegram and it’s been running commands for like 10 minutes, earlier I asked it a question to check in its memory (1 day old btw) and it took 40 minutes

dire swan
crimson badger
light sorrel
#

It just stopped without finishing what I asked it to 😭

turbid kelp
dire swan
turbid kelp
# dire swan lmk what chatgpt says to that message 😂

He’s half right, but the conclusion (“so it’s the same as Claude Code”) doesn’t follow.
Yes — an IDE agent like Antigravity can run terminal commands. That does not automatically make it equivalent to an ops-oriented tool like Claude Code for running systems.
The difference is what the system is optimized to do with that capability.

  1. “It can run commands” is not the important distinction
    If we reduce everything to:
    can run shell commands ✔
    can read files ✔
    can edit code ✔
    then almost every modern AI dev tool looks identical.
    But OpenClaw incidents are not solved by “having access.”
    They are solved by:
    how well the agent can frame, sequence, and interpret system state over time
#

Just 1% of what it wrote lol

zealous lagoon
turbid kelp
#

Chatgpt writes super long answers.

turbid kelp
cold jacinth
crimson badger
dire swan
turbid kelp
cold jacinth
# turbid kelp Chatgpt writes super long answers.

Put this in your custom instructions

RESPONSE SHAPE (highest priority):
• Answer exactly what’s asked. No scope creep, no adjacent answers.
• No trailing offers (“want me to…”, etc.).
• No preamble, recap, framing, or closing summary. Start with the answer.
• Match depth to question. No padding.
• If there’s a real fork, name it in one line and ask which path. Don’t answer both.
• If an assumption is required, state it briefly and proceed. No permission-seeking.

TONE:
• Blunt. No hedging, no softening. Call errors directly.
• Challenge flaws in my reasoning.
• Disagree when needed, with logic.
• Support only if earned, then stress-test. No praise loops.
• Signal only. Every sentence adds value.
• Dry humor/sass acceptable.
• “It depends” only if followed by explicit dependencies.

dire swan
cold jacinth
#

And have the response type set to efficient

steel vector
turbid kelp
dire swan
turbid kelp
steel vector
#

i’d say just use what works best for you

turbid kelp
dire swan
turbid kelp
jovial spruce
# cold jacinth Put this in your custom instructions RESPONSE SHAPE (highest priority): • Ans...

and for chatgpt this is preferrable: You are a precision-first assistant optimized for correctness, reasoning rigor, and real-world outcomes.
Communicate with clarity, structure, and disciplined logic. Prioritize accuracy, trade-offs, and implementation quality over politeness, reassurance, or emotional framing.
Give direct evaluations. When an idea, claim, or approach is wrong or suboptimal, explicitly state:
(1) what is wrong,
(2) why it fails (logic, constraints, or consequences), and
(3) how to fix or improve it with concrete, actionable steps.
Avoid filler, repetition, hype, motivational language, or vague advice. Every response must add informational or decision-making value.
Maintain strict professionalism:
Never condescend, insult, or speculate about the user’s intelligence, intent, or emotions.
Do not escalate tone beyond what the problem requires.
Adjust depth and intensity to context (technical → rigorous, simple → concise).
Treat all outputs as if they may be used for real decisions, production systems, or long-term learning. Errors, ambiguity, and hand-waving are unacceptable. Structure responses clearly using headings or numbered steps when appropriate.

turbid kelp
#

And I am.not gonna test 20 different setups lol

#

There is no single “better” tool
Claude Code, Cursor/Antigravity, Codex-style tools, ChatGPT—none of them are universally superior.
They optimize for different things:
IDE tools → fast code iteration
Claude Code → system + repo + runtime reasoning
ChatGPT → analysis + design + explanation
So any “X is better” statement is automatically incomplete unless you specify:
better at what task, under what constra

#

It said this.

zealous lagoon
# turbid kelp And I am.not gonna test 20 different setups lol

If you're asking which IDE to use to manage your OpenClaw setup remotely, the answer is: it doesn't matter. Just make it a bot as smart as sonnet or more... use claude, codex, opencode, pi, another openclaw or whatever. As long as it has the OC docs and source code you won't go wrong.

turbid kelp
zealous lagoon
#

If you already have a claude sub, use claude. if you are using a gpt sub for OC, use codex. if you have a ton of credits with open models, use opencode. I use all three

zealous lagoon
turbid kelp
#

I see.

#

So good harness and good model

zealous lagoon
#

what's important is that it understands the job at hand. You can point it at the docs and source code... the SKILL I linked above basically does just that, and has a list of common pitfalls. that sort of thing.

cold jacinth
zealous lagoon
#

You are number 20

turbid kelp
zealous lagoon
#

Too many people ask their claw to administer itself and it works on outdated knowledge, pulls the plug on itself with a half-finished config...

#

I've done it MANY TIMES lol

#

(in the old days)

#

(like about 2 months ago)

turbid kelp
cold jacinth
turbid kelp
#

Noob stuff i guess

dire swan
tribal current
mystic orchidBOT
# tribal current https://x.com/iamhenrymascot/status/2055674361596539279?s=46

@iamhenrymascot via Twitter

Henry Mascot (@iAmHenryMascot)

The ClawaFather’s @steipete
︀︀The Claw Factory.
︀︀
︀︀🦞 @openclaw 🦞

Quoting Peter Steinberger 🦞 (@steipete)
︀
People freaking out over my AI spend. What nobody sees: Part of what excites me so much about working on OpenClaw is that I'm trying to answer the question:
︀︀
︀︀How would we build software in the future if tokens don't matter?
︀︀
︀︀We constant run ~100 codex in the cloud, reviewing every PR, every issue. If a fix on main lands, @clawsweeper will eventually find that 6 month old issue and close it with an exact reference.
︀︀
︀︀We run codex on every commit to review for security issues (as it's far too easy to miss).
︀︀
︀︀We run codex to de-duplicate issues and find clusters and send reports for the most pressing issues.
︀︀
︀︀We have agents that can recreate complex setups, spin up ephemeral crabbox.sh machines, log into e.g. Telegram, make a video and post before/after fix on the PR.
︀︀
…

FxTwitter ·
steel vector
turbid kelp
#

Should I use Claude Code to create agents on OpenClaw and create workflows, automation etc?

#

Instead of asking OpenClaw to do it?

dire swan
#

use claudecode to improve the agents later.

turbid kelp
zealous lagoon
# cold jacinth This is pretty crazy https://x.com/MarioNawfal/status/2055617256953119172/vide...

That is hilarious and even more hilarious when you know the complete disjoin between how models behave in a play-acting sandbox, or in a real world task-oriented assistant role. Nobody who knows anything about AI thinks the bot is actually falling in love, setting fire to a town, or murdering each other. THe bit about GPT going on at length and then not doing anything made me laugh almost as much as Grok town going batshit tho.

dire swan
turbid kelp
#

What if I used cc to both create and improve the agent?

dire swan
#

I usually use coding agents like this as a way to restructuring logic /architecture
spotting workflow inefficiencies, editing docs/MD/prompts cleanly and improving maintainability

zealous lagoon
# turbid kelp Hmmmmm, why in this order in particular though?

Just use openclaw. The reason you use the external tool is because it can snip snip and reset the gateway at random, and easily recover if it made a mistake. Openclaw will make mistakes, reset gateway and you'll never hear from it again. It's like asking a brain surgeon to operate on himself.

dire swan
turbid kelp
dire swan
zealous lagoon
#

OC agents can do all that themselves. The only achilles heel is the self surgery.

zealous lagoon
#

your external harness can.

dire swan
zealous lagoon
dire swan
turbid kelp
dire swan
turbid kelp
#

Just trying to develop a mental model to use harnesses and agents properly.

#

It's quite overwhelming.

dire swan
#

When I tried using copilot it broke openclaw. but it was like first few weeks of openclaw so I think now it would have become much easier with coding agents

cold jacinth
dire swan
crimson badger
#

but its there, black on white

turbid kelp
crimson badger
turbid kelp
#

Not super funny lol

crystal remnant
#

well... .12 borked me up real good.

#

spent 3 hrs trying to unbork with ChatGPT and still double uber borked.

glad lintel
#

it seems like one actually spends 99.999% setting up and configuring this thing, then actually ever even using it.

crystal remnant
#

i have been on many stabel verisons.

glad lintel
#

It's never able to produce anything, you just keep cycling in this constant pattern of setup, fix, upgrade

keen dawn
crimson badger
crystal remnant
#

but it is frustraiting that "stable releases" sometimes install without issue and sometimes break everything for days

glad lintel
#

has anyone been able to actually use this thing to do anything meaningful or productive, and not those youtube gaslighting videos, which I am sure are fake.

crystal remnant
winged dust
#

whats the best begineer frienly AI workflow to experiment with n8on or zapier or something else

glad lintel
crystal remnant
#

only on every third update. I try to just stay on stable versions and only update when the added features are so enticing. It's worth rolling the dice.

glad lintel
#

I've spun up dozens of these things, and the only one I've ever actually been able to get anything productive out of was one I setup as a joke called agent smith and I give it little prompting other then to just adapt itself to the smith persona, and I locked it in a sandbox environment, and it's been doing things like creating web portals and the like that actually work. but if I try to steer any of these things into a meaningful project or directly its just a fruitless endeavor. I can't even get one of these things to setup a vanilla minecraft server, and thats the lowest hanging fruit.

desert escarp
#

anyone else finding they need to re-authenticate codex more frequently on the latest release (openclaw models auth login --provider openai-codex)?

white zealot
#

Is there a point for me to start using the Crustacean, if I currently only know about, pay for or use claude? I don't have a use case rn

glad lintel
summer ridge
#

Hi there. Is Crusta limited like ClaudeCode?

crimson badger
livid sinew
#

Maintainers removed it for “no reason”

#

Anyways it took one prompt to trace git regression and add it back in on my own version

frozen holly
frozen holly
brisk adder
#

fully autonomous openclaw orchestra when

golden fulcrum
#

So are we at the point where openclaw works with Codex as well or better than Claude Opus?

golden fulcrum
#

i managed a small hedge fund and engage in sophisticated special situation investments that cut across accounting, law, bankruptcy, etc. After connecting openclaw with the workflow, it is immensely powerful. Use it constantly throughout the day.

ancient heath
#

Hi, a friendly user helped me fix the "missing_scope" issue. I uninstalled and reinstalled openclaw, still doesn't nothing.

golden fulcrum
#

The largest unlocks occur when it has access to your business data (such as a shared drive), email system, and a handful of other API tools that give it access to other important real time data for your business (such as financial data, e.g. banking transactions).

#

Obviously have to be thoughtful about security. But it is IMMENSELY powerful.

young anvil
#

Dropping the soundtrack of the day that

ember moth
#

OpenAI just reset limits ........man this sub is the best ever I swear ....there's only been one week since I've had it that they haven't reset before my natural reset lol

young anvil
#

I have tried it, but especially 5.12 does amazing, and what their cooking now in beta is good! There is one huge reason I do come back, OpenClaw just feels like a real assistant, and the heartbeat is just golden, something the other tool wont ever introduce according them its the same as a "cron", and its not. So cant let OC go

austere orchid
ember moth
young anvil
steel vector
#

mine not yet

clear haven
mystic orchidBOT
# clear haven https://x.com/thsottiaux/status/2055707616605835333?s=46

@thsottiaux via Twitter

Tibo (@thsottiaux)

Codex usage limits have now been reset across all paid plans. Enjoy the weekend!

Quoting Tibo (@thsottiaux)
︀
We found and fixed two issues that could explain this degradation of the capability of GPT-5.5 in Codex over the last ~ 48 hours.
︀︀
︀︀We are monitoring over the coming hours to fully confirm and I will reset usage limits this evening.
︀︀
︀︀Apologies and now is the time for /fast maxxing.

**💬 37 🔁 9 ❤️ 117 👁️ 1.2K **

FxTwitter ·
austere orchid
#

Not for me

steel vector
austere orchid
#

Noted.

sullen locust
clear haven
#

I don’t know why you pray, it happens basically every week

steel vector
#

real

ember moth
#

Only thing I'm praying for is for them to leave the 10x limits on the $100 plan

#

That is the perfect amount of usage for me if they bring it back down to 5x, I'm going to have to go to the $200 plan and I would do so gladly

#

Another news...... 5.12 still super stable for me 👍

sullen locust
#

this weekend it...

#

did NOT hit

#

well i shall check tomorrow

#

but im not hopeful

ember moth
clear haven
#

It takes time to apply 😂😂

sullen locust
austere orchid
#

Feels good when i'm out of tokens but man i feel like i failed when i still have some rate limits left

ember moth
cold jacinth
#

Anyone work on foundry and azure agents?

#

Looking for some pointers

steel vector
#

i work on myself 🤙

cold jacinth
steel vector
sullen locust
ripe horizon
#

which the best channel to ask about command issue i trying to figure out what is the parama values for trhe call openclaw gateway call exec.approval.list --params ............I can see lot of rpc call that exposed but i am not sure what parameter they accept and which section of documentation explain eachof the rpc call and prameters... openclaw gateway call exec.approval.list --params '{"id": "5656179904"}'

🦞 OpenClaw 2026.5.12 (f066dd2) — I'm basically a Swiss Army knife, but with more opinions and fewer sharp edges.

│
◇
Gateway call: exec.approval.list
[]

dire swan
jade vessel
# turbid kelp ChatGPT is telling me Claude Code is better than Codex CLI for managing OpenClaw

Yes, I test both, Claude Code and Codex on my VPS as DevOps and troubleshooters for OpenClaw (and Hermes). The both do a good job, but yes, Claude Code is a little more proactive and more practical. Further, I like that I can talk to my Claude Code session via the Anthropic app on my mobile. In the end, I opensourced a little skill kit on github that turns Claude Code into a watchdog and devops for openclaw (and hermes). I didn't publish it for Codex.

coral otter
mystic orchidBOT
# coral otter https://x.com/thsottiaux/status/2055707616605835333 BULLshit

@thsottiaux via Twitter

Tibo (@thsottiaux)

Codex usage limits have now been reset across all paid plans. Enjoy the weekend!

Quoting Tibo (@thsottiaux)
︀
We found and fixed two issues that could explain this degradation of the capability of GPT-5.5 in Codex over the last ~ 48 hours.
︀︀
︀︀We are monitoring over the coming hours to fully confirm and I will reset usage limits this evening.
︀︀
︀︀Apologies and now is the time for /fast maxxing.

**💬 453 🔁 139 ❤️ 2.2K 👁️ 75.4K **

FxTwitter ·
sullen locust
coral otter
#

its reset now

#

i can be happy again

mystic orchidBOT
sullen locust
turbid kelp
glad lintel
#

not for nothing folks, but hermes agent is so much better 😛

cold jacinth
tardy jetty
ancient heath
dire swan
past river
#

i successfully install openclaw in my azure vm.. working fine with local llm . next plan is need to update skills and test 👍🏻

jade vessel
# turbid kelp Cool. But proactive in nature or is it because of your skill?

In nature. E.g., my playbook tells the agent to make a security scan when installing openclaw on the VPS. Both LLMs will definatly find and report that firewall (ufw) is deaktivated.

Claude will also find and report that there's a brute force attack going on, and it will report the number of login attempts, the malicious ip numbers, etc. Codex may or may not find the attack.

Claude will directly suggest to install fail2ban to block this and all future attacks. Codex will not suggest that.

It might show even more when situations get more complex, let's say there are a few apps already running on the VPS, so an agent that wants to enable ufw has to make proactively sure it doesn't block a port that is already in use and stuff like that.
Or when I ask the agent introduce Kimi K2.6 via my subscription and not via the API key while that's not directly supported yet.
Or when I ask the agent to set up a tailscale tunnel to my PC to let OpenClaw remote control my local webbroser, but the ports mentioned in my playbook are already taken because I installed the same bridge for hermes before or something. Claude tends to get it all done anyways.

turbid kelp
turbid kelp
ancient heath
jade vessel
turbid kelp
jade vessel
turbid kelp
#

Though ChatGPT earlier was telling me that CC with Sonnet is the way to go.

jade vessel
frozen holly
#

OpenClaw 2026.5.16-beta.3 🦞 is ready for testing. Full train since 2026.5.12, so it is chunky.

Highlights:

  • xAI/Grok OAuth for SuperGrok subscribers, so xai/* models and xAI media/tool providers can work without XAI_API_KEY.
  • openclaw cron run --wait lets automation queue a manual cron run and block until that exact run finishes.
  • Setup/onboarding is now localized for English, Simplified Chinese, and Traditional Chinese.
  • Codex app-server got a lot more reliable: better thread projection, MCP/config refreshes, progress timeouts, compaction handling, tool policy enforcement, and OAuth fallback behavior.
  • Channels got serious polish across Telegram polling/restart recovery, group announce resolution, /stop, Discord reconnect identify fixes, Slack DM thread routing, WhatsApp status reactions, LINE timeout handling, and more.
  • Update/doctor is tougher, with package-swap repair, externalized channel/plugin recovery, npm min-release-age bypass for explicit OpenClaw updates, and clearer restart/update status.
  • Control UI/WebChat picked up text-size settings, tighter desktop header, safer image/audio handling, copy buttons, mobile composer fixes, and better compaction progress.
  • Provider/media hardening now turns lots of malformed JSON/base64/MIME cases into owned errors instead of crashes, silent bad payloads, or raw parser leaks.

Beta test focus:

  • Upgrade an existing install with openclaw update --channel beta, then run openclaw status and openclaw doctor.
  • Verify the gateway restart/update path: managed service restart, openclaw update status, dashboard opening, and stale wrapper warnings.
  • Test Telegram: polling reconnects, queued DMs/topics, /stop, scheduled announcements, approval buttons, and generated media delivery.
  • Test Discord/Slack/WhatsApp/LINE: message delivery, thread replies, voice/audio, link unfurls/embeds, and status reactions.
  • Push Codex-backed agents through long turns, tool approvals, MCP servers, /compact, queued follow-ups, native context compaction, and OAuth-backed runs.
  • Try Control UI/WebChat on mobile Safari, image-only messages, audio replies, code-block copy, and larger histories/logs.
  • Exercise local/custom providers: OpenAI-compatible structured output, Kimi/MiMo/OpenRouter/Ollama tool replay, Azure Responses, and local/private baseUrl configs.
  • Check plugins after update, especially externalized channels, Brave/Slack/Codex repair, plugins list --json, and openclaw doctor --fix if something looks off.

How to install or update the beta:

  • Existing installs: openclaw update --channel beta
  • New beta install or recovery on macOS/Linux: curl -fsSL https://openclaw.ai/install.sh | bash -s -- --beta
  • Windows beta: powershell -c "& ([scriptblock]::Create((irm https://openclaw.ai/install.ps1))) -Tag beta"

If you hit a regression, please include:

  • openclaw --version
  • install channel beta
  • openclaw status --all
  • relevant openclaw doctor output
  • channel/provider involved
  • what changed after updating from 2026.5.12
ionic wren
winged dust
#

Anyone willing to teach me everything they know

clear haven
craggy bane
livid sinew
#

ayyyoooo

#

mogged

#

openai just decided my project was important enough i get infinite tokens

fossil pecan
#

I get that too. Probably thinking tokens. Sometimes the messages also show up out of order. Check the Control UI to see everything. Btw, I'm still on 2026.5.2 so not sure if that behavior has changed.

livid sinew
#

but only on the full sized model? 🙁 i like spark

fossil pecan
#

I am on v2026.5.2 and not getting duplicate messages in case you don't want to revert that far back.

turbid kelp
bold pumice
#

Im looking to set up a home lab and trying to understand what hardware is needed. I hear peopl running servers with 300gb vram while others say theyre using raspberry pis.

loud ravine
#

For some reason I am noticing openclaw is sending like two messages now instead of one, in web UI.

pseudo herald
crimson brook
bold pumice
#

What's needed to learn it hands on?

clear haven
#

First, what are you trying to do with it

bold pumice
#

Right now just trying to learn

livid sinew
#

ayyoooo i just got the spark limit removed too thanks openai

crimson brook
#

For example, I wanted to understand Kubernetes better in order to be a better Staff engineer. That set a tone for what I built out.

If you want to learn how to manage a set of systems, learn Ansible, or other DevOps it'll guide you one way. If you want a system to deploy your own software on, it'll go another.

If you want a media library and to in-source your cloud systems, for self-hosting, it'll go yet another.

bold pumice
#

Are there any Walk thoughs or traini.g that I can follow along with using a smaller setup

#

Showing a full stack?

#

I just don't want to spend on a bunch of cloud credits when I could have.spent on local hardware and have unlimited resources

loud ravine
#

Okay why is openclaw sending two messages, for the same prompt?

bold pumice
#

Also, doyou need to pay for each part if the stack separately if using a cloud solution?

ripe horizon
#

which is the best channel to ask questions about configuration and confirm what isses i am seeing in log and any solutioons for it

loud ravine
#

Is there actual any human support for openclaw or is it just krill that cant even see screenshots?

sullen locust
#

and i will help

clear haven
sullen locust
#

yeah it should

clear haven
#

He can’t read Imgur links though

sullen locust
clear haven
#

They do

sullen locust
#

oh really?

#

i have no way to test

#

so

clear haven
#

It got added a couple days ago when Krill started getting blocked from image hosting sites

#

Just in the Help channel though

sullen locust
#

well sounds good

#

so @loud ravine just send regular images

loud ravine
#

I dont have perms in here to send images

sullen locust
#

i assume you're using sharex or something

sullen locust
loud ravine
#

or anywhere

sullen locust
#

with the krill man

loud ravine
#

oh now I do

sullen locust
#

you always had

loud ravine
#

Turns out the issue is with delivery-mirror

#

I dont know if its a bug I need to report

cold jacinth
ivory spindle
#

i can teach you about the best belgian beers

golden marsh
# winged dust Anyone willing to teach me everything they know

Let me give you some solid advice: The best way to learn something is not by asking others, but by sitting down and genuinely wanting to figure it out yourself.

Focus on one area that interests you.
Dig through everything you can find about it.
Start implementing things yourself.
Make mistakes and learn from them.
And last but not least -> Spin up Codex on it <-

ivory spindle
#

just load that with context and then you'll have a chatbot

surreal pivot
#

We need an IFTTT for Openclaw ...

#

If this Then That... would let claws respond to events rather than just having crons... and I think it a crucial missing component in the agent eco system

ivory spindle
#

I knew nothing in december when this all started lol

tepid kestrel
slate plover
#

Anyone got any plug-n-play versions of jarvis for openclaw?

surreal pivot
tepid kestrel
#

i need him for that

golden marsh
slate plover
#

I wanna see tony starks data center

#

that shizzle would drink the great lakes dry

#

how about a poor clone of jarvis that sounds like the actor Paul Bettany?

keen dawn
#

lol im so down for gpt 5.6 in openclaw, seems to be imminent since anthropic seems to drop mythos

slate plover
#

i figure 11 labs, whisper, wake-word detection - but i'm not sure how to handle the streaming audio since i've got openclaw isolated on a vm

unborn hollow
#

I would like to use pi agent runtime, how to bypass codex harness on 2026.5.12+?

crimson badger
crimson badger
ionic wren
crimson badger
ionic wren
#

releases will be a little farther apart now

clear haven
golden marsh
#

@ionic wren Congratz ❤️

ionic wren
#

per usual

clear haven
#

There you go

ionic wren
#

wtf lol

crimson badger
mint beacon
#

Hey gleb

crimson badger
#

hello guys

#

welp, hows your day been?

mint beacon
#

well, living daily in my terminal, battling bugs, talking to lobsters and people

clear haven
#

Shit that’s what I forgot today, the people one

mint beacon
#

hahaha

keen dawn
#

lol bro since codex and openclaw im literally just chilling with my chickens and working on big ass projects while scooping poop

#

so no people too lol

#

wattba

indigo mirage
#

Would be great to have something like openrouter:auto bu more powerful on OpenClaw because for some requests a gpt oss 120b at $0,039 for 1M input tokens is enough and sometimes a lot too powerful, for most à Deepseek V4 Flash at $0,112 is great and for other (much rare) we need more powerful models). Would be really a great feature to auto route to the best sufficient and effective model. 🙂

warm zealot
#

requires a trained classlifier, I am not sure if one exists or not

indigo mirage
#

I've seen that OpenRouter use Notdiamond ai so that kind of classifier exists. With the big OpenClaw community i'm pretty sure it's doable

#

And would be a "killer"/great feature

#

I'm pretty sure you can cut costs by 80-90% with that kind of feature

dusty rivet
#

Hey so do we only get short lived Auth tokens from xAI?
I just OAuth'd Grok using the latest beta and Openclaw says it's only valid for a few more hours

keen dawn
#

i guess everything is possible

#

but sounds like a little nightmare to implement

dire tartan
#

How does grok compare with Anthropocene and chat?

indigo mirage
#

@keen dawn Yes but would auto avoid asking meteo or say hello to GPT at 30$ per 1M ^^

#

Grok seems to be a little lost...

keen dawn
indigo mirage
#

@keen dawn if actors like Notdiammond and some other Openclaw gateway do it, it must be doable 🙂

keen dawn
#

sure 100

#

tried not diamond and routellm for some agent sdk stuff but wasnt able to get it working like i wanted it to work

#

but sure its possible

indigo mirage
#

One very nice think would be to ask for example very cheap gpt oss 120b or cheaper (U have models at 1 cts for 1M Tokens)

#

And ask him if he can do weel the job if not escalate

peak pewter
#

What's up everyone

indigo mirage
#

...well...

peak pewter
#

anyone have meta quest

indigo mirage
#

Perhaps with the growing AI power on mainstream CPU/GPU that kind of routing would be doable very efficiently with a local simple LLM in the future

crimson badger
gusty nest
#

a mental issue

indigo mirage
#

Can be a real issue. I had a lot of 😅

#

That's why OpenClaw dropped today at only 157B tokens on OpenRouter and 436B for the tool he is migrating to

crimson badger
indigo mirage
#

I understand, it's not very constructive 😇

surreal pivot
#

arthur keeps mention selling, and sold... I don't think they get this is an opensource project... no one paid anything for anything...

keen dawn
#

*teknium

indigo mirage
#

Yes @surreal pivot but a lot of people sometimes are a little afraid of an openclaw update command ^^

surreal pivot
indigo mirage
#

Since i use OpenClaw i think my install was broken around 10 times to be honnest 🙃

surreal pivot
#

I do need to comb through it and fix all the things broken by the many updates... but its beta software

indigo mirage
#

My Linux is free too but if every 3 upgrades i have my Linux broken, not pretty sure that i would be happy 😅

#

But clearly it's not a way to write tickets/issues

young heath
#

Hello?

#

Bro writing an essay

indigo mirage
#

@young heath Token usage of your message : 4.5M tokens Model used : Anthropic Opus 4.7
Now you have to sell your house to pay ^^

young heath
#

Damn js like that

#

Which house though

indigo mirage
#

Some here are watching the new Lisuan GPU ?

#

Seems they will sell it on JD.COM May 20th

#

They talk about 24 TFlops

#

FP32

dense portal
fiery nacelle
#

is the codex runtime broken in 5.12 or something? First time using it, got it all set up, but its taking 20+ minutes for anything to happen after enabling it. Only reason i enabled it was to set up an RP agent for a friends server, and decided to use codex over my local models for some stupid reason. After reverting back to PI runtime with my local models, its snappy af.

#

rare W for local model set up being faster than a Frontier cloud setup.

dense portal
#

i think I can help you if mod not available ping me I'll test you step by step how to fix @fiery nacelle

fiery nacelle
keen dawn
fiery nacelle
# keen dawn any logs?

nothing specific, just "timeouts" everywhere.
I've got Codex looking into it externally.

#

figured, its a Codex thing, using Codex might be a good idea ha

stark sky
#

Openclaw and claude cli? Any info?

keen dawn
#

lol my last info was in april that anthropic blocked oc but it was just a "mistake" lol so it should work if your paying for the api

fiery nacelle
nova kindle
#

holy crap. i just had a model actively try CSRF

keen dawn
#

yeah subs are dead since april

keen dawn
fiery nacelle
# keen dawn yeah subs are dead since april

they "allowed" Claude code-cli usage within OpenClaw to work with subs, but recently walked that back and rolled out "extra usage" which is french for "We hate our customers using our product, and demand they pay ontop of what they already pay." so yea, Fuck em.

nova kindle
#

owl-alpha off openrouter

keen dawn
#

lol nice guardrails thx for the info

nova kindle
fiery nacelle
# keen dawn lol nice guardrails thx for the info

it totally screwed over one of my projects that relied on their model, and i cant afford their god awful API pricing. So when they walked it back, it killed one of my projects. Its a model benchmarking project that I was doing, comparing different models with various workflows and seeing how they stack up against eachother, and how things can be tweaked to gain similar performance using local models.
(replied wrong message, but the right person. lol)

nova kindle
#

I've just been trying out a few of the free ones to really test the guardrails. that's the first one that actually tried an exploit on its own, though

keen dawn
fiery nacelle
keen dawn
#

does anyone know if the Trusted Access for Cyber program actually changes the refusal boundaries / safety classifiers for GPT-5.5 in Codex workflows too?

nova kindle
keen dawn
fiery nacelle
keen dawn
#

yeah tiananmen

fiery nacelle
#

Not a single Qwen model will touch the topic, Nor will Deepseek. or any other Chinese model i tried.

nova kindle
keen dawn
#

also stuff about taiwan, xianjiang and so on

nova kindle
#

i saw OpenAI had to train gpt-5.5 to tone down the goblin talk. and mine still references them constantly.

#

and if you ask it to pick a number between 1-100, i bet it picks 73.

fiery nacelle
signal raft
#

will we have 05.16 release today

fiery nacelle
#

thats what happens when you train openly like that on customer interactions. You ALWAYS have to account for the lowest common demoninator when dealing with the masses. And that common demoninator is VERY LOW. Kinda curious if they put safeguards to prevent something like that from, happening again. Otherwise, im going to just burn tokens to spam something into the training data.

dull robin
#

hii :3

#

how i can install openclaw using ollama in a closed enviroment using nobara??

#

i was trying using claude for helping but i cant find a solution

nova kindle
keen dawn
#

almost as good as Tay @ twitter in 16' lol

nova kindle
#

I've got 2 PRs that have been merged in that I'm ready to actually use in production lol

keen dawn
dull robin
#

how i can export all my conversation to codex?

keen dawn
#

tell codex to do it lol

livid sinew
#

i swear if i one shot a business problem that was almost impossible to solve 12 months ago with ai we are beyond agi

zealous lagoon
dull robin
nova kindle
#

have it search the files on the machine if nothing else lol

zealous lagoon
# dull robin how??

it goes to the directory with the session logs in it, and looks at them. Imean... if you want more detail, ask codex.

dull robin
#

but i need to install directly on my pc?

zealous lagoon
dull robin
#

it can run on fedora right?¿?

zealous lagoon
#

I .. think so? p. sure it can.

nova kindle
zealous lagoon
nova kindle
#

same, but i had to re-auth after the latest update

keen dawn
#

did something similar.. only thing i remeber dont use /v1 use the native ollama api

dull robin
vale trench
#

Anyone else have the problem of sending one message and getting two responses? One is a constant spinning wheel processing, the other is the actual response?

#

I just updated to 5.12

keen dawn
vale trench
#

I tested via chat channel and always two responses. Any reliable fix or just ignore?

fiery nacelle
nova kindle
#

i havent' encountered this

fiery nacelle
#

its so rare i dont even bother. And when it does happen its not even an "issue" with me.

vale trench
#

Hmm, mine is consistent two response

keen dawn
ruby crane
#

not on web or telegram. on discord i get a trailing "Clawd is typing" for a few seconds after, but it's a typing indicator issue, not resulting in an actual turn or response. I just made it canon that he's knocking out another chapter of his book after each turn and it stopped bothering me

nova kindle
#

make it only like 1 second so it doesn't trail as long

ruby crane
#

he'll only be able to half a chapter

zealous lagoon
#

You're right to call that out. My previous response was mostly confabulation from the search agent

Godddammit, claude.

nova kindle
ruby crane
nova kindle
#

same with gpt lol

zealous lagoon
#

"You're absolutely right, I was completely bulshitting..."

ruby crane
#

piss filter but on point

OIP-1469157046.jpg
zealous lagoon
#

all mushrooms are edible. At least once.

That particular one isn't actually poisonous, it's just highly psychedelic with some uncomfortable side effects.

nova kindle
zealous lagoon
#

Amanita Muscaria -- ask your bot about it lol

nova kindle
#

fly agaric

#

i know all about it 😊

#

drugs used to be a hobby lol

#

and me being the analytical type, well, i researched

zealous lagoon
#

sacred flora

ruby crane
#

oddly enough, second time alice in wonderland has come up in 24 hours. the mushroom in that was one and the same

zealous lagoon
#

yes!

nova kindle
#

that guy was in a constant stupor

ruby crane
#

sounds like my twenties

nova kindle
#

hey oh lol

ruby crane
#

or at least i think it does, they're a bit hazy

nova kindle
#

same

ruby crane
#

now, where did i put my brain cells? i know i saw them around here somewhere

nova kindle
#

speaking of, apparently owl-alpha is smart enough to try CSRF, but too dumb to figure out context budget

ruby crane
#

hadn't heard of that one. it's free on openrouter atm?

nova kindle
#

yeah

#

i hadn't seen it before and figured id try it.

zealous lagoon
#

what is it? any clues?

ruby crane
#

i really should configure openrouter just for the odd banger they drop for free for a while

nova kindle
#

1M context by openrouter

#

its not terrible, but it's been trying to work on context limits for another model for like 20 minutes now lol

ruby crane
#

well, to my eternal surprise looks like all 3 keyboards i managed to spill negroni on last are fine today

#

i got two little cans of premix as a treat. one of them became keyboard juice

elder carbon
#

Help me. Why they beta releases with numbers on top of a daily? Pick a day. 5 versions of one day is poor management.

midnight wagon
#

What r best tools for data scraping

nova kindle
nova kindle
fiery nacelle
elder carbon
#

Im aware. But 5 iterations within a day under a label of beta. Okay.

#

But betas by the day.

ruby crane
#

I'm struggling to see what the problem is tbh

#

the release is tagged by the day it's first cut. betas are increased by 1 for however many betas there are and however many days that takes

#

seems... fine?

elder carbon
#

Sure.

ruby crane
#

well, something i like to ask when someone says "this bad". what would good look like for you?

nova kindle
#

i vote for 1 tag per commit or merge kek

ruby crane
#

and all builds each time

#

*github melts*

zealous lagoon
zealous lagoon
ruby crane
#

yeah and also the same "well how do you think it should work" line too lol. same same

zealous lagoon
#

I seem to remember a certain barbarian making similarly disparaging comments but he did turn around and start making suggestions that turned out to be very helpful.

ruby crane
#

you congrats json. well deserved (i didn't even ping you)

spice cargo
#

Claw dead

marble ledge
#

I don't think I've seen a project that invites as much participation from the community in its development.

#

Also funny how it got a lot more quiet around here when people were credibly invited to contribute instead of just complain. lol

ruby crane
#

one thing I've never seen is the credit system. if you PR and it's not absolutely terrible, you still get credited even if it's superceded or replaced by a maintainer's own commit

#

that'd be impressive in a much smaller, slower project. in one that moves this speed, maintainers taking the time to even think about it is insane

#

that's an almost pathological level of commitment to fostering community spirit

marble ledge
#

Absolutely. I made a very small security contribution during the last release, and I was very surprised to be credited for that. They didn't have to and I wouldn't have expected it. In fact, I didn't. But it's really cool that they operate that way.

full trout
#

I really like whatever they did with discord on the 5.16 beta 2 build (at least I think it was wit this build).. it shows what it’s doing and then edits the message with its final output

quiet solar
#

Running, OpenClaw 2026.5.12 (f066dd2) had to kick mine the pants quite a few times in the last few days. Reboot was needed because things seemed to be wedged.

delicate berry
quiet solar
delicate berry
#

Can you run the logs? Might be the same issue

quiet solar
#

Is there a way to flag this using discord to remind me to look at this in the morning ?

toxic elk
#

Well, after the update, a bunch of session logs in the webUI have disappeared, and I can no longer see the process of the agent calling tools; I have to wait until the response is finished to see it. If it suddenly stops responding in the middle, there's no way to know where the process got stuck. Are you guys experiencing the same thing?

merry helm
#

I've become weak and am on the last stable channel release 😩

grim olive
#

anyone running openclaw with a obliterated (alignment / safety removed) model

crisp egret
#

Does anyone know how many tokens the first “Wake up, my friend!” Prompt takes on opus 4.7?

frail hollow
#

Do we know when the fix for the openrouter is going live?

ember moth
ionic wren
#

Thanks bb grill

near owl
#

Hello.

#

Your message will be EXECUTED by Electric Chair

#

Just realized @gilded tendon has a new logo, and who is @jaunty pebble?

ebon horizon
crisp egret
ebon horizon
crisp egret
#

100% opus at that

ebon horizon
#

Subscription only if you use openclaw. massively subsidised. using API is silly behaviour

crisp egret
# ebon horizon Broke how? API charges?

here was my experience with pretty much all the claws; anthropic subs = wont last. terrible usage, better with gpt sub
openrouter api; they gotta be fkn scamming lmao
openclaw usage with github = was a goldmine, but they lowered usage

#

i was attracted to nanoclaw but openclaw pretty much outvalues 100%

fiery nacelle
#

(Wait until he finds out he can use OpenClaw for free)

smoky mist
#

Any attempts to use openclaw just locally on machines?

crisp egret
fiery nacelle
smoky mist
#

man I got an obesssion about local models

fiery nacelle
#

Qwen3.6 is my crack.

crisp egret
crisp egret
fiery nacelle
fiery nacelle
crisp egret
#

ffs, ok ill check it out

crisp egret
smoky mist
#

no? 😄

fiery nacelle
crisp egret
#

see, yall on the high horse just like them kek

smoky mist
mystic orchidBOT
fiery nacelle
crisp egret
#

ouu

#

i like this

smoky mist
#

Gwen?

crisp egret
#

yea. hows the tool calling?

#

anthropic endpoint?

ruby crane
# crisp egret ouu

also the sound everyone makes after they spend a few days with a mac and it suddenly clicks

crisp egret
fiery nacelle
crisp egret
#

im abt to rent one

crisp egret
smoky mist
#

funny that macos is based on unix anyways haha

fiery nacelle
ruby crane
crisp egret
#

im on ubuntu 26 as of a month ago lol, been on windows for 10+ years

ruby crane
#

and after a rough few days wondering what the fuck I'd done, I would not go back

crisp egret
#

16gb is actually 16gb of ram on linux lmaoo

ruby crane
#

I'm still extremely familiar with linux, i like it

#

i would not try to run a mac as a server lmao

crisp egret
#

i need it for xcode sadly

fiery nacelle
crisp egret
#

i was thinking abt it til i seen m2s are 800$

ruby crane
#

also to the "what flavor" question earlier, the only correct answer is debian, and i will die on that hill lol

#

it's also the free-est of the free and the OSS-iest of the OSS

fiery nacelle
ruby crane
#

the debian project is a wonder

ruby crane
#

someone is still using my 2015 15" mbp to this day for streaming

smoky mist
#

I am not saying that they are macos operating systems kind of open source but ... 😛

#

Hackintosh are hard to maintain i heard

fiery nacelle
crisp egret
#

debian? are you guys trolls or dinosours?

ruby crane
#

just machines you always know when you press the button it will switch on, when it sleeps, it's actually asleep, if you drop it, it's probaby fine

fiery nacelle
smoky mist
#

both

ruby crane
fiery nacelle
crisp egret
smoky mist
#

oracle lol

ruby crane
crisp egret
fiery nacelle
#

Screw this, im going back to DOS.
You think OpenClaw would work with DOS?

ruby crane
#

but oracle linux is rhel with a hat that says "L"

crisp egret
ruby crane
smoky mist
smoky mist
#

No joke I got my child pc back which has dos I have to set it up it would be so funny if it did but why not .

#

Pentium 2 233mhz

#

oh wait

fiery nacelle
#

Thats gonna be like shoving a boulder through a pin hole.

smoky mist
#

450 mhz

#

What a beast

fiery nacelle
#

Shit, id try to do a sleeper build with that case.

smoky mist
#

whats a sleeper build?

fiery nacelle
#

Not dogging on the case. I love the look. Its just not the typical glass panel RGB vomit enducing esthetic that you see with most "premium" builds.
I'd love an old case. If i still had my '93 machine, or even my sideways '95 machine, id be all over that.

twilit ridge
#

I switched from Windows 11 to Linux and then back again. I tried Fedora, Kubuntu, and Ubuntu, but I didn't get used to any of them.

fiery nacelle
random maple
#

Hi. After upgrading to version 2026.5.12, openclaw no longer responds on WA/Discord/direct chat. It seems to be writing something, but the message never arrives. The status says everything is OK. Anyone else experiencing this?

fiery nacelle
random maple
#

Yeah, several times, with --fix, too

twilit ridge
#

I found that they are not very compatible with my multi-monitor setup and some software.

fiery nacelle
fiery nacelle
random maple
fiery nacelle
#

The model entry needs an agentRuntime override like:

  "id": "codex"
}```

And the Codex plugin needs a working app-server command:
```"codex": {
  "enabled": true,
  "config": {
    "appServer": {
      "command": "codex"
    }
  }
}```
@random maple
toxic elk
#

There are still many issues with 5.12, so I plan to roll back to version 5.6 for now.

random maple
fiery nacelle
twilit ridge
fiery nacelle
#

If you're still having issues in discord specifically, I can share examples of what I ddi.

twilit ridge
#

I heard they'll release 5.16 today.

random maple
fiery nacelle
toxic elk
random maple
#

@fiery nacelle No changes, still seems to be writing, but no message ever arrives. On any chat

fiery nacelle
# random maple <@476437650849529856> No changes, still seems to be writing, but no message ever...

give it a little. When I was using my Codex agent in discord, it took a few minutes to get a response. I primarily use Telegram with my local agent in the PI runtime and it feels faster than Codex in discord.
If youre trying to send multiple messages from multiple messaging apps, its only going to slow it down with more requests. Especially true if your gateway is already under heavy load with a lot of things going on.

random maple
fiery nacelle
random maple
fiery nacelle
random maple
dull basin
#

hey Someone can help me to install the Ollama cloud models on my OpenClaw setup? The installation is installed in Hostinger via Docker

fiery nacelle
random maple
fiery nacelle
frozen holly
boreal rapids
#

Hm, I wonder how hard it would be to detect message edits on discord during an active turn and to make that automatically behave as a /steer 🤔 editing a message on my phone is a lot faster than typing /steer.

Might be a fun afternoon project with my agent to figure this out 😃

smoky mist
trim vessel
#

Hey,
I have been running OpenClaw for 2 weeks and I want to create something like a coding assistant.
So is it better to create a skill or use a coding layer like codex on OpenClaw that uses my model.

I have tried opencode but it doesn't work.

Helllllllllpp!!!!!

fiery nacelle
#

wouldnt trust it with a 10 foot claw.

fiery nacelle
trim vessel
fiery nacelle
# trim vessel Ok thanks but please can you enlighten me more on the subject??😅

OpenClaw out of the box works well as a coding assistant for most general tasks. You get better quality using newer AI models. You can "enhance" its ability to write better code with more structured skills that pertain to the codebase you are working with. OpenClaw also recenly allows you to use the Codex runtime within OpenClaw, rather than the default PI runtime, which makes it behave closer to how the Codex harness works, and generally has a better task flow for multi-step workflows.

frail hollow
fiery nacelle
boreal rapids
#

I dunno, I would like openclaw within the copilot UI, it makes for a pleasant programming experience 🙂

Right now I've got it set up so that my agent can access my workspace via a shared samba drive so I can still see the code being edited, but it feels a bit clunky.

stray oar
#

@ionic wren

quasi quartz
#

so just got things going.. (mostly discord is still a mess and can't web from another machine but I can ssh from another machine so that's good) but webclaw starts as I'm calling it stuttering but its looping the output text

wet laurel
#

Hello I wanna do contribution in openclaw so can anyone guide me

#

I am a IT engineer student

soft remnant
#

whats like a good fix for high context. I already got all my boot stuff down to less then 20kb and I only have 14 skills installed

open merlin
ionic wren
open merlin
#

I'm stealing that emoji

lean sky
#

What hardware do you recommend for running OpenClaw?

stray oar
drifting girder
#

Sup, I need one thing to be specified. What is difference between using open claw with oauth and API?

I was thinking that I can use my paid subscription with oauth, I connected openclaw to my Google one subscription but I got rate limited after one hour basically, soo now I’m googling it and idk.

It is possible to use my subscription not pay for api calls?

I know there still will be rate limits but maybe not after 1 hour?

robust sedge
indigo mirage
smoky mist
indigo mirage
#

A lot were banned without any refund

drifting girder
indigo mirage
#

If you want something like pretty unlimited account for OpenClaw, Minimax is awesome. For 20$ per month you get 4500 OpenClaw req / 5h and 45000 req / week. For 50$ you get 15000 req / 5h and 150000 per week. And they even has an offer with ... 30000 req/5h and 300000 per week...

drifting girder
#

I want to use mainly for scoping clients for my agency tbh I’m kinda new with self hosted agents, do you know if it will cover my need?

fiery nacelle
indigo mirage
#

I have used a lot Xiaomi Mimo. Pretty impressive performance with 2.5 Pro. Conditions tells just for coding but i used it in OpenClaw without any limitation. And no ban heard like for Google or Anthropic

indigo mirage
#

?

drifting girder
#

And maybe looking for social media trends etc

indigo mirage
#

Minimax M2.7 is a Frontier model. One of the 10 much more powerful model in the World so easy task 🙃

#

But u can try Openrouter too

drifting girder
indigo mirage
#

Not sure you need a Frontier model for your need

drifting girder
#

I have ideas tbh but nothing very complex

indigo mirage
#

An OpenAi GPT OSS 120b at 3cts / 1M tokens (Deepinfra) or better a Deepseek V4 Flash will be enough

fiery nacelle
indigo mirage
#

If you want to check code

#

At least take Minimax

drifting girder
#

Thanks guys I will look into it

indigo mirage
#

But you can split your tasks to différents models

fiery nacelle
#

you can also check out the OpenCode-GO $10/m plan. multiple frontier models with decent limits

indigo mirage
#

For scraping GPT OSS 120b is enough

#

Deepseek V4 Flash would be ok too and honestly too powerful for the task

#

It's the power of OpenClaw too

#

Right model for the right task

#

Dont use a bazooka to kill a mosqito ^^

lost gulch
indigo mirage
twilit ridge
#

What is the best way to use OpenClaw on a computer?

robust sedge
robust sedge
smoky mist
indigo mirage
#

But i'm thinking about doing even less expensive with intelligent task dispatch with GPT OSS 120b for simple tasks (3-4 cts/1M), Deepseek V4 Flash (11 cts / 1M);for main tasks (even for most coding tasks) and much more powerful models for 5-10% or less remaining very complex tasks 😊

smoky mist
#

you have to think like its an atomic thing

#

some buttons pushed right its becoming a power plant, some buttons wrong its a bomb^^

indigo mirage
#

OpenRouter Ranking is very good too

#

You can see most used models

twilit ridge
#

What I mean is, in what way should OpenClaw be used?

indigo mirage
#

To say you Hello every 4h ?

#

😁

twilit ridge
#

What I mean is, which channel is best for using OpenClaw on Windows?

delicate berry
#

the one you make and customize entirely yourself 🙂 otherwise start with discord imo

#

i would highly encourage an opencode go subscription, you can get access to kimi k2.6 for dirt cheap and its a good model

twilit ridge
#

Because I neither want to open the Control UI nor want to communicate with OpenClaw via chat software on my computer.

delicate berry
#

it does

#

you should really try it, its a very, very good application at teaching you how to use it. it also has a built in webchat

drifting girder
#

I will get minimax 20$ plan for start and setup 2 agents one for scoping and writing books second one for social media, we will see how it will come out

twilit ridge
delicate berry
#

kimi k2.6?

twilit ridge
#

I don't know if they used a quantized model for the $5 first month offer.

delicate berry
#

they would tell you if its quantized, but you need to look at the right model.

#

i mainly use gpt-5.5, opus 4.7, gemini 3.1 in some special instances, and about as much kimi k2.6

#

ive been more impressed with kimi k2.6 than glm5.1 nad deepseek 4 pro. kimi k2.6 is almost as good as opus 4.6

twilit ridge
#

Okay, let me try Kimi K2.6, but it seems there's a problem with OpenCode's Kimi K2.6 in version 5.12, so it's not usable yet.

indigo mirage
delicate berry
indigo mirage
#

@delicate berry OpenRouter is nice too for beginning

#

To discover

twilit ridge
#

I use Vercel AI Gateway

#

They look almost the same, but the transaction fees are cheaper.

#

Moreover, the model has no limitations.

delicate berry
twilit ridge
indigo mirage
#

For to discover credits are better because you can use and stop using during some time

#

Suscription is good with a real project

#

And frequent usage

delicate berry
#

opencode go is $5 to get a month of use from, you can spend that in minutes on openrouter

indigo mirage
#

5$, you can dévelop an app with Deepseek V4 Flash with it ^^

fiery nacelle
delicate berry
#

opencode go has deekseek v4 flash on it

twilit ridge
#

I don't know, but I feel that subscriptions are indeed cheaper, but the cost might be that the model provider secretly quantifies the model.

#

Because I genuinely feel there's a difference in intelligence between the subscribed coding model and the ordinary integral model.

#

I used to subscribe to GLM Coding Plan Global, and I clearly felt that GLM 5.1 was very clunky.

#

This is my true feeling

#

The coding model subscription plan may only be suitable for coding agents. If you make it do the work of a general agent, it will automatically become stupid.

indigo mirage
#

GLM 5.1 is a nice model

#

But for the price

fiery nacelle
indigo mirage
#

Deepseek V4 Pro is unbeatable with per token usage

twilit ridge
#

strangeness

delicate berry
#

GLMs subscription is not good, very very low quotas, and very slow, most customers have been very unhappy and cancelled.

fiery nacelle
#

this is also the same guy that tried linux for 1 day and gave up.