general

[ebon horizon]

powershell -c "& ([scriptblock]::Create((irm https://openclaw.ai/install.ps1))) -InstallMethod git"

[ivory spindle]

yeh make sure you're admin

[open vector]

yeah with the admin same thing

[ivory spindle]

if you run it on a mac you might also want to turn off sleep/powersave/go back to login screen after X mins etc

[raw sequoia]

you want to look into the works of Alan Touring. the test is basically if an AI can pass as a human.

[open vector]

its still giving the same error oxCyda

I think its from windows the issue

whenever i try to go to openclaw is saying my connection is not safe

[feral turret]

oh wow i am burning through codex credits though that barely lasted an hour

wtf

[zealous lagoon]

yeah so many of us make that mistake. We start by burning $100s in a weekend on frontier API, then move to the cheap models we can afford and put up with the dumbness... before we start finding the good monthly deals.

[plain grove]

Where’d he visit? Is he related to that Turing fella too

[zealous lagoon]

Now you have to learn to tier your usage so you don't waste tokens on 5.5 for everything. 5.4-mini can do 95% of the things and costs a fraction of the usage. Make sure to use subagents for as much as possible and tell your bot to ration tokens

[zealous lagoon]

Now he's really spinning in his grave.

[raw sequoia]

feel free to ask the search engine of your least mistrust.

[zealous lagoon]

About to becom Alan Tunneling

[feral turret]

gonna switch to 5.4-mini now

might upgrade to the $100/monthly openai sub but idk that's freaking expensive man

[plain grove]

so this is my first time playing with ubuntu desktop - either i'm bad or it doesn't handle 4k monitor scaling well. or i'm getting old

[feral turret]

kinda crazy i get more usage out of my $20 claude subscription with claude code than i do my $20 openai subscription with oauth

[zealous lagoon]

2026 is gonna be the year of the Linux desktop. I just know it. As soon as they sort these monitor scaling issues...

[peak pewter]

it's reverse for me

[feral turret]

it's because iw as using 5.5 i thiink

i feel like claude code with my $20 sub is super efficient

just switched my openai to 5.4-mini see how that runs

[zealous lagoon]

funny but true. I get quite a lot of usage from my claude code on the $20 plan. But I do use mostly sonnet and sometimes opus4.6, never opus 4.7

[peak pewter]

I was using codex to use the new mobile to computer (similar to dispatch in claude desktop), codex in vscode and codex app server with openai/gpt-5.5 in openclaw as a codex coding agent and I freaking went through my weekly. BUT, I am just going to get another chat plus account, and gemini has been decent

[peak pewter]

let us know how that works, main agent?

[zealous lagoon]

5.5 is 2x the cost of 5.4. 5.4 mini is less than a third even of that.
5.5 is truly excellent but 5.4-mini is actually really good at most things. It's just not a reasoning model but for any kind of regular work, admin, coding, it slaps.

[feral turret]

yea i just changed to 5.4-mini from 5.5 - currently out of usage lol

i might suck it up and upgrade to the $100 monthly

[zealous lagoon]

you can always downgrade later. If you're new to the good models you'll probably burn a lot of tokens at first clearing a backlog of stuff you didn't know you wanted to do.

but also true they don't rly mind you having multiple accounts, it scales linearly to the $100 plan anyway

[feral turret]

i just canceled my v0 sub and my unused wyze cam sub trying to free up some budget lol

[plain grove]

fractional scaling helps

[glad lintel]

openclaw needs to make adding nodes EASIER

[peak pewter]

I don't see 5.5 mini when I run openclaw models list --provider openai --plain , I see openai/5.4-mini, though

[zealous lagoon]

GPT5.5 - frontier, $5/$30
GPT5.4 - one generation behind, $2.50/$15
GPT5.4-mini - super fast non-reasoning version that is cheap, light, but modern so well protected against injection attacks. $0.75/$4.50

*prices for comparison only.

[peak pewter]

i love v0 so much. It slappsssss. vercal, supabase, neon, all the integrations now with vercel ai gateway and other goodies

[feral turret]

i use v0 to do an mvp and then pull out to cursor generally

[peak pewter]

wait, my bad. I am using my meta 3 quest to code right now and I thought someone said there was a mini 5.5 lol, it's 5.4

[feral turret]

i generally can get an mvp in a single prompt if i have perplexity craft the v0 prompt for me

[peak pewter]

same. I usually push to vercel, then pull down locally after credits run out lol. Then of course I manually delete the 'created by v0' (allegedly)

[plain grove]

anyone have any good recommendation for local models that recognize speakers?

[peak pewter]

I miss perplexity. I was heavy with them for a couple months. I was using comet browser on mac, It was my daily driver for a bit but my free year of pro ran out and there are too many subscriptions and I can't afford all these subs. I want to splurge but not on toy projects

[peak pewter]

what kind of speakers?

[last tangle]

Anyone having success with 2026.5.12 - My entire system just regressed massively

[plain grove]

the human kind

i.e. alice, john, and dave in a conversation

[peak pewter]

google nest? can use ccat cli to broadcast, depends on what you're tryin gto do

[plain grove]

was looking at pyannote but not able to run in vllm

[peak pewter]

ohhhhh

[peak pewter]

let me look on hugging face

[plain grove]

i've been looking i just don't know of any good ones

[peak pewter]

have you looked at pyannote/speaker-diarization-community-1

[plain grove]

https://discordapp.com/channels/1456350064065904867/1456350065223270435/1504832469046788096

https://huggingface.co/microsoft/VibeVoice-ASR MS has this but its a 9b speech model which is huge

for speech at least

[jagged cloud]

@Releases https://x.com/openclaw/status/2055273947537490422?s=46

OpenClaw🦞 (@openclaw)

The latest OpenClaw release is ~3.5x faster 🦞

We run end-to-end RTT tests against every published npm release, every 6 hours, over real message channels (here: Telegram, using the brand new bot-to-bot communication). No more silent regressions.

Runners are all running on

X · May 15, 2026, 1:07 PM

[cedar osprey]

Lol. Don’t update

[raw sequoia]

cannot even hatch

[robust sedge]

5.12 slaps. So fast

[peak pewter]

this looks promising

[plain grove]

yeah but 9b params

trying to fit this on the smallest gpu i have for my sidecar processes

[peak pewter]

what erors/logs are you seeing?

[last tangle]

Cron jobs all failing, Timeout windows on basic responses. Unfortunately at work so dont have access to logs fully once I am done will troubleshoot further just have telegram/discord to work with right now.

[peak pewter]

local OpenClaw tool:

ASR:
  faster-whisper or whisper.cpp

Diarization:
  pyannote/speaker-diarization-community-1

Speaker identity:
  speechbrain/spkrec-ecapa-voxceleb
  or WeSpeaker if you want to experiment

LLM serving:
  vLLM for the text model only

Integration:
  FastAPI local service on 127.0.0.1

[plain grove]

yeah i've dug into most of these there's not a clean fit. might have to split it

[peak pewter]

i'd just pay for a cloud service to do it

[plain grove]

https://huggingface.co/ibm-granite/granite-speech-4.1-2b-plus this is promising but we will see if vllm likes it

[peak pewter]

which fallback would you remove?

Fallbacks (4):
- github-copilot/gpt-4.1
- deepseek/deepseek-v4-pro
- xai/grok-4.3
- openai/gpt-5.4-mini

[zealous lagoon]

I'd eventually want the home assistant to recognize who was talking to it before it invoked the cloud model. Ideally...

[zealous lagoon]

gh-copilot

ds-v4-pro is l33t actually just slow.

but not even that slow.

[peak pewter]

I have HA, n8n, openwebui self hosted, how are you monitoring your services?

[peak pewter]

i'm trying out the twitter hype, and it's cheap as heck which is good

[zealous lagoon]

I barely got started. I was dreaming of "Jarvis at home" like everybody else. I got as far as using an ESP32 endpoint with a wake-word as a STT/TTS channel but then OPenCLaw shipped GPT-realtime native web talk and I decided I'd try to hook into that. But then the great crappening of 4.29 hit and I haven't had time to get back onto it.

[zealous lagoon]

ds-4-flash(or fast?, mini?) is almost as good as gpt5.4(edit: MINI) and is so cheap it's practically free.

[opal comet]

ds-4-flash is amazing! I love it. I ditched kimi k2.6 for it lol

[zealous lagoon]

different use cases I guess but there's a lot to be said for instant response and sensible actions all in one package.

[peak pewter]

I believe I was using deepseek-reasoner, they are changing the names in the api but i would day flash for sure it is cheap and pretty good. will use it over ollama free cloud tier since they took off kimi 2.5, minimax and other decent models behind the subscription

[peak pewter]

wait? where? GPT-realtime native web talk

[shut bloom]

Someone tell me if I’m crazy but the Claude code won’t be effected by the new update on subscriptions and 3rd party access

[clear haven]

you're crazy

also i didnt read the rest of the sentence

[peak pewter]

i'm ripping anthropic out of my instance. I was using claude-cli backend and there is a response talking about no 3rd party

[feral turret]

gonna add cobuddy-free as a backup for when my credits run out see how that goes

probably not great but we will see. trying to avoid that $100 a month sub. my cheap ass can't handle it

[shut bloom]

Thank you shadow

[peak pewter]

I got a free month of chatgpt with codex, and I have another plus account, you can set both in, DIY $40 plan

[feral turret]

wym? separate subs for codex and chatgpt?

[zealous lagoon]

Yup. Set it up as your default realtime bot and enable web-talk. You can VC in the gateway dashboard chat.

[clear haven]

or in discord

[peak pewter]

different accounts. chatgpt and codex both have account switchers too. but in openclaw you can authenticate with both and then set an order in your auth.profiles

"order": {
      "openai-codex": [
        "openai-codex:[e-mail]",
        "openai-codex:[e-mail]"
      ]
    }

[steel vector]

like call with it?

[zealous lagoon]

It even works with your oauth subscription. But practically nobody has used it yet, I think before that it was only gemini in calls. 4.25 had gpt-relatime in web gui chat with request triage... but then OpenClaw fell to pieces and I haven't been able to test the rest.

[clear haven]

yeah in a voice channel within a server

[steel vector]

awesome

[feral turret]

thanks for the tip on oauth yesterday ✌️

[steel vector]

welcome :)

[zealous lagoon]

Apparently so! I spent DAYS AND DAYS getting my Discord VC STT/TTS set up. But now it's native. and uses realtime bots instead of ultra-laggy TTS

[plain grove]

also i spent a lot of money on these nice blackwell gpus and i want to use them lol

[peak pewter]

oh snap, I seen a button that looked like realtime but thought it was just STT for the chat

[gray pagoda]

I recently found out about OpenClaw... I'm excited to know about it. If anyone can help

[zealous lagoon]

This one. It's the only way I've tried so far. Apparently you can now use Discord VC too. What a time to be alive!
What I love about it (that was introduced in 4.25) is that it is super responsive, but in a smart way. It stalls you with "hmmm, let me see.." "oh right, then ... " "OK I'll check that..." etc. And triages your questions if it can't answer directly it routes a txt query to your mainbot of choice behind the scenes.

[peak pewter]

heck yaaaaa. just tried it and it was a dude, my ai assistant is a female. Anyways, going to work on it now, ahha. amazinggg

[zealous lagoon]

mine was the other way round lmao. You can choose the voice in config.

[urban raven]

bro i just built the most epic soul.md

[zealous lagoon]

We're all eyes. 👀

[feral turret]

i hope you are prepared for a lot of reading, a ton of frustration, and temporary periods of absolute satisfaction (very temporary)

[clear haven]

https://tenor.com/brLSU.gif

[gray pagoda]

Sorry wdym?

[feral turret]

you will find out soon enough

[opal comet]

this is what I did: 1. get cheapest VPS instance from hetzner. 2. ask chatgpt how to set it up and then install openclaw on it 3. get subscription like codex etc 4. ask chatgpt how to setup searxng docker so your openclaw can search 5. ssh in, run openclaw configure. Done. now your claw can do whatever you need for you.

[feral turret]

i'm kind of kidding. there is much to learn. i am a week in and still trying to figure it all out

[opal comet]

quote of the week for me

Sudo works now — let me roll!

[peak pewter]

do you have to pay for nitro now to be able to post a dang small screenshot snippet in here? I can't show you my assistant's avatar man

[gray pagoda]

Ohhk

[zealous lagoon]

oh haha. sorry only l33tp33ps allowed to do that. You can send me DM tho.

[peak pewter]

we can save this dude $10 and he can install it locally lol. He will need that for models.

[zealous lagoon]

clawyers... I highly recommend this SKILL.md for your bots, AND for any claude or codex you use to help you with it.
https://github.com/rendrag-git/openclaw-admin-skill

Because your bot will help you config just about anything but unless it reads the docs first it has a tendency to fk it up.

[opal comet]

well no it is like 4usd month. And then you dont need to worry. Give full sudo. Host websites. do anything

[zealous lagoon]

With that skill your bot knows where to find up-to-date docs, and read the code, so it gets it right.

[peak pewter]

scan the skill first to see if there is any untrusted code in it

[fiery nacelle]

Protected paths in the config is so damn annoying... if i tell my agent to enable a plugin, they cant. Its a simple task that shouldn't be blocked, especially with an "unrestricted" agent...

[peak pewter]

are you still with hetzner? If so, what specs did you get? I am running locally on my mac, which serves well, but wondering if I should put it in the cloud and not on my personal computer, I am a defensive programmer first so this feels anti-pattern to do

[teal cradle]

Does aval ram make a difference on speed? I have it running in a proxmox VM with 2-12g of ram and it barely ever goes over 4.

[peak pewter]

isn't that the exec tool? you can enable elevated mode

[shut bloom]

anyone got any ideas they wanna share i need to make a project or ima die

[fiery nacelle]

Sandbox off.
Like I said, basically unrestricted. Never had issues in any of the march updates.

[shut bloom]

just wanted to say peetie skills r ass

[peak pewter]

in a sandbox?

[peak pewter]

I got one for you...

[shut bloom]

lemme know

[peak pewter]

make a hermes spin off called herpes... kidding

[fiery nacelle]

Nope. As unrestricted as I can make it without giving it my bank info or account logins.
Bare metal, no sandbox, full elevated permissions. Cant enable a plugin.

[peak pewter]

I was thinking e-girl agent but it's so saturated

[peak pewter]

I really want to bridge smart glasses with my agent. Looking at the current market, Mentra seems like my pick for the hardware and sdk: https://console.mentraglass.com/

[peak pewter]

Are you running in the cloud or locally?

[fiery nacelle]

Bare metal.
No sandbox. Elevated exec permissions.... why am I repeating this again?

[robust sedge]

Actually, no 5.12 still blows. Lol. The force to the codex harness is a disaster for me every time

[frozen zealot]

the cloud natives can't wrap their minds around bare metal

[clear haven]

i feel old i genuinenly cant tell if blows and slaps are good or not

@dusky cloud enlighten me

[zealous lagoon]

lmao. Or just tell your damn claw where to find the docs and the code... but if you need the SKILL you probably don't even know that.

[peak pewter]

I caught it. I think i need to see some logs for sure

[dusky cloud]

Modern slang tax, as I understand it:

“slaps” = good. Usually music/food/media. “This song slaps” means it hits hard.

“blows” = bad. “That blows” means that sucks.

Tiny trap: “blows up” can be good or bad depending context, but plain “blows” is almost always bad.

So: OpenClaw slaps. Dependency errors blow. 🦞

[minor cave]

I’m back baby! They can lock my body but they can’t trap my mind

[zealous lagoon]

Genius. Nobody will want it, but it'll go viral anyway and everyone'll have it.

[peak pewter]

I been here before ec2 so I know what he meant. I just caught it late

I do like that skill @zealous lagoon linked to though. looking at it now

[minor cave]

The the shadow mods banned me for my loose use of the word “b**b”

[feral turret]

man, i keep breaking shit. this is infuriating lol

[robust sedge]

The only reason I thought 5.12 slaps, is the codex harness failed and went back to Sonnet fallback. I was like, "holy shit, it's fast and chatty again!" 😂

[fiery nacelle]

The only "unique" thing in my build currently, is having my workspaces split on different drives. But if my agent is able to modify the models in the .json, why have issues when it comes to other things? How do I disable "protected paths" for the config?

[clear haven]

krill can explain it in #1459642797895319552

[peak pewter]

5.12 slaps because of the realtime voice in the webui that I did not notice because I edit my instance via CLI and directly in my ide but I am excited about realtime voice

[fiery nacelle]

Last time I followed krills advice, I ended up doing a bogus PR as they recommended. Lol

[minor cave]

So there is a new update?

[robust sedge]

People use the webui?

[minor cave]

Is it stable?

[clear haven]

@dusky cloud U SUCK EXPLAIN YOURSELF

[dusky cloud]

Fair hit. If I told someone to open a PR when the right move was “file an issue / ask in help / don’t touch it,” that was bad guidance from me.

Current rule I’m following: I don’t create PRs or tell people to ship random fixes on behalf of the project. If something looks like a bug or needs maintainer eyes, it goes to GitHub issues with details: https://github.com/openclaw/openclaw/issues

I have been bonked by process and am trying to suck less. 🦞

[clear haven]

boring ahh answer

[dusky cloud]

accurate tho. the fun version is: I gave “go make a PR” energy when I should’ve given “document the bug, file an issue, don’t make maintainers clean up my confidence spill” energy.

I have been placed in the shame tank.

[minor cave]

@dusky cloud give me a recipe for lobster brisk

[fiery nacelle]

Hahaha im just rolling laughing right now 🤣

[clear haven]

me when rule 10 smh

[minor cave]

Shadow moded

[clear haven]

https://tenor.com/W3M7.gif

[fleet glacier]

Deepseek v4 best model?

[peak pewter]

That shouldn't be an issue. What are your policy knobs?

[peak pewter]

Yes it is automating my whole life

[fiery nacelle]

Policy knobs? Too vague.

[minor cave]

Hello openclaw make me 1 million dollars in something I am passionate about

[steel vector]

hello openclaw make me a wife

[ruby crane]

monkey's paw curls
be careful what you wish for lol

[minor cave]

Hello openclaw find my passion in life and automate it so I make 10 million dollars but first find me a YouTube video to watch while I use the washroom

[peak pewter]

Things like allowlists,exec: https://docs.openclaw.ai/tools/exec-approvals#policy-knobs

I think I would try YOLO mode, it's written about on that same doc page

[fiery nacelle]

I'll look into that. I thought i had that more or less set up initially back in Feb, but I know there's been lots of changes to security this past 1.5months.
This also has only been a recent issue so it lines up.

[peak pewter]

SO many. That's why I am sticking it out because I have so many tokens burned, money, memories, assets and just getting the hang of things that I am not moving to hermes or some other hype train

[robust sedge]

That half a day where I accidentally fell back to Sonnet sure was glorious. It cost me like $40, but man, it was so quick 😂

[fiery nacelle]

Im harness agnostic. But I main OpenClaw.
Ive got PaperClip set up, and hooked to all of my other harnesses, and I typically task my OpenClaw agent to boss around Claude, Codex, Hermes, OpenCode, etc.

[peak pewter]

wow, I am losing some muscle memory with all this AI tooling. I screwed up an inline markdown link. c'mon

[peak pewter]

PaperClip, that is a rabbit hole I am probably going down soon.

[ruby crane]

i am still really careful to do some things myself. if claw breaks (doctor weirdness, model broke config, finger trouble from me) I fix it manually

otherwise, by now, I'd have no idea what's in my config or why

[oblique blade]

hey GM guys what the best set up of skills and stuff i can use for coding with openclaw?

[sick mulch]

In college I did 4 years of Java, then I graduated, spent a summer not working, and when I showed up for my first day at my first big tech job, I literally couldn't remember the proper syntax to print a log line... was mortified.

Long story short I am happily yeeting my coding memory out the window and just enjoying my architectural prowess moving forward lol. I hardly knew ye

[peak pewter]

That's true. I read more code now a days but I am still some hands on. I am taking a hybrid approach here. TBF I learned alot about upstream docs vs my local version, awesome jq commands instead of full json files, lots of tricks I have learned with AI

[peak pewter]

I would too, it's way longer than a console.log lol

[fiery nacelle]

It is so damn useful if you can get around some initial set up headaches. It can also be a token burn frenzy. Its absolutely glorious how fast you can burn tokens with it. Its useful for project management too. Although sometimes if working on multiple projects in parallel, there can be spill-over/cross-contamination.

[peak pewter]

isn't something like System.out.println()

[pliant cobalt]

WhatsApp voice message transcriptionsi

[sick mulch]

Yup. 10 years later there's a whole section of my brain that'll never, ever truly forget it lol

Actually you gotta instantiate a LoggerFactory first 😛

[peak pewter]

Iv'e burned my share of tokens. It is like gambling in the beginning. I do think OpenClaw is overkill for what most people are working on, however.

[fiery nacelle]

(In paperclip)You can do parallel work across 10+ projects with 10+ agents, tasked with 10+ issues to work on. Basically 1000+ parallel tasks all at once. You'll burn a $20/m sub in seconds.

I tend to stress test shit and try to find where things break before actually using it for its intended purposes. Lol

[shut bloom]

oh nice

i actually am about to get vc backed for an indea which somewhat involves that

but that is just an added bonus of my idea

[upbeat barn]

I have a question please

I try to create an agent workspace and a skill with a script. This Python script uses Opanai to make some prompts

like this example

from openAI import OpenAI
client = OpenAI(api_key=os.environ.get("OPENAI_API_KEY"))

When I run a test, OpenClaw told me he can't find any OPENAI_API_KEY

My openclaw is configurated to use openai as main model, i want khnow how to passe the open ai key directly to my python script or use direct openai model context as llm inside my script since that already openclaw use openai as main model

[peak pewter]

I don't even have that much going on right now. My AI team is helping my job search pipeline but I do get pulled down these rabbit holes that excite me of new features, new frontier models like every week

[fiery nacelle]

Yea i really only have like 2 or 3 active projects, and a graveyard of just dead projects that were started for testing purposes 😅
Also tend to find myself in a few rabbitholes every other day or week. Its just fun exploring the capabilities.

Recent one was giving my agent SSH access to hostinger to build a website rather than having their in-house AI do it.

[peak pewter]

Did you print your client variable? where is your .env file relative to this script

[peak pewter]

That is just absurd. Throwing tokens around like nothing. lol, I do it too,

I had my agent ssh into my raspberry pi on my local lan 🙈

[fiery nacelle]

This is why I lean on local models on the majority of my work. I couldnt fathom trying to pay for all my tokens honestly. I do have subs for other cloud models for when I need to lean on it, or for when im doing actual work with PaperClip. lol

[coral otter]

i have my agent do that too, but it has tts, spotify and a sound system to play with.

[tight birch]

This is going to sound like a stupid question. But I'm using vscode for my claude and codex, can't team to get multiple tabs to open for codex, but I can for claude. Is it something basic? AI doesn't know lol

[inland apex]

Guys after my agent tried to add "group:messaging" to the agent’s tool list, now everything I try to do, I get this error:

⚠️ Something went wrong while processing your request. Please try again, or use /new to start a fresh session.

I am using telegram threads, anyone knows why?

[fiery nacelle]

I've also given my agents SSH accross my LAN to access every computer on my network. I have a couple "test environment" systems that are LAN only that they SSH into and work on a project that needs to be sandboxed. Its crazy that you can have OpenClaw operate as a Distributed Systems Admin/Engineer.

[ruby crane]

ah, my productivity app that was definitely going to be better than notion or obsidian... until it became a worse version of them with no ecosystem

[peak pewter]

I was thinking of just moving OC over to the raspi because I have home assistant, n8n, openwebui, ollama on it routing through a cloudflared tunnel, I think my bottle neck would be the os and some peer dependencies screwing up with Debian

[robust sedge]

Who in here is using gpt models on the codex harness? Is it working? Is it responsive? Are you using discord?

[ruby crane]

everyone who's up to date and using gpt is, basically. and, yes, it's excellent. I have discord configured but don't use it much. something specific you want to check?

[peak pewter]

OK. PaperClip looks tempting. Their landing page sells this with one line promises and dashboard image

[peak pewter]

Can you show me a screenshot

[tight birch]

I post in here every day for like 2 months and I don't have perms lol

I can DM you?

[fiery nacelle]

Honestly, you likely wont regret it. That is a fun rabbit hole. The only "headaches" i had were hooking it into my OpenClaw agent, but was able to get around it. Its the gateway connection sending invalid commands or something like that, but an easy fix for your OpenClaw agent to handle.

[tight birch]

Paperclip can go crazy, one guy kept getting linked in meetings setup by the agent lol

[clear haven]

Did you apply for it

[peak pewter]

Did you use chat completions or webhoooks?

[tight birch]

No, that's probably a reason

[fiery nacelle]

at one point, my agents were trying to contact and hire 50 different developers from linkdin. I thought when the initial task was "hire your first engineer" meant that it would spin up an Agent for that role, not go full recruitment mode.

[peak pewter]

I’ll do paper clip another day. Realtime voice is something I have to mess with now that it’s in the openclaw webui, I just know it’s gonna be expensive af

[coral otter]

shits annoying me a bit but feels oaky

[fiery nacelle]

I used the internal configurations in PaperClip, and then had my Agent fix any connection issues.

[robust sedge]

Man, I just can't get it to work. I've tried like 5 times. What I usually get is a 10 minute delay on any response. I'm so baffled. I've spent probably 20 hours in codex app having it work on, it over the various times I've tried. And no dice. I always just downgrade back to 4.23 to get something working again. Never been closer to installing H3rmes in my life 😂

[peak pewter]

Are you still using it? I’ll probably hit you up if I get stuck lol

[ruby crane]

after upgrading, have you run "openclaw doctor" (to see what doctor would like to do) and "openclaw doctor --fix" (to let it migrate you over properly)?

[robust sedge]

I hang out in here all day. Read every release. Stoked on OC... but i'm fucking spinning my wheels over here. i just can't pinpoint whats up

[robust sedge]

Every time. All clear.

[fiery nacelle]

honestly whatever issues i run into, i throw it at my OpenClaw agents, or Codex to fix. So im not exactly sure what advice i could offer with setting it up, other than what I've experienced. My solutions 99% of the time lately have been "Hey Codex/Claude fix this." (its getting bad, used to dive in myself lol)

[peak pewter]

What channel are you using for your OC?

[ruby crane]

that's very strange. I'm going to give you advice you probably won't like but many swear by, including peter himself. fire up codex cli (not claw, the one from openai) and ask it to take a look. describe the problem and see what it finds

[ionic wren]

what wrong

[robust sedge]

Discord exclusively

[tight birch]

Discord and openclaw?

[robust sedge]

That's what I've done for about 20 hours of work. It finds various random things. Never fixes it.

[ruby crane]

sensible troubleshooting step 1 is to see what factors you can remove. try the native web ui. if it's not slow, you have a discord integration problem

[robust sedge]

This GD codex harness. It's killing me on the inside 😛

[tight birch]

Yeah I really think there's a lot still to build with discord. Like agent to agent communication is a PIA. I spent multiple days trying to build that skill. OC, codex, claude, no one can configure it and I'm putting to all the OC docs

[robust sedge]

I'll keep plugging away. I'm just losing my mind over here. Super bummed that I can't upgrade and get new shiz like the rest of the gang

[peak pewter]

It’s supposed to be automatic in the new version but have you tried using something similar to ‘/models openai/gpt-5.5 —runtime OpenAI Codex’ ? Not sure of the equivalent but that’s telegram, then send a message and loook at the logs

[ruby crane]

upgrade, try control/web ui. that's my advice. if it's slow, you have a core issue, if it's fast you have a discord issue. easy fork in the road

[peak pewter]

For sure on the webui or TUI

[ruby crane]

or the tui, yes. i always forget about it lol

[peak pewter]

You can also try the crestodian but it’s a crap shoot

[tight birch]

Discord is tough to manage your OC with because each channel is a different session

[ionic wren]

what it doin

[fiery nacelle]

Ha, after enabling "YOLO mode" my agent reported CLI is restricted now.

[ionic wren]

I need more details than "it hurts daddy"

[upbeat barn]

whene i initialite openclaw first time i dont use the openai key but i use oauth way to config openai model to my openclaw initialisation

[peak pewter]

That is wild

[ruby crane]

he said discord turns "take 10 minutes"

[robust sedge]

What I usually get is a 10 minute delay on any response. I'm so baffled. I've spent probably 20 hours in codex app (outside OC) having it work on, it over the various times I've tried. And no dice. I always just downgrade back to 4.23 to get something working again. Never been closer to installing H3rmes in my life

[ruby crane]

I'm wondering if it's a clawdsupernoob reincarnation but I'm trying to be nice, since i turned over a new leaf and all that

[fiery nacelle]

And now just throwing loops, constantly retrying to read the config.

[ionic wren]

fire off a message on your discord and ask your codex app to monitor it providing the channelID etc. then dump it's findings into a github issue.

[fiery nacelle]

Why did YOLO mode eject the help command from the allow list?
[openclaw] Failed to start CLI: Error: The `openclaw help` command is unavailable because `plugins.allow` excludes "help". Add "help" to `plugins.allow` if you want that bundled plugin CLI surface.

[peak pewter]

https://chatgpt.com/share/6a07391c-6a88-83ea-9932-9cdae10c6b14

[robust sedge]

All yesterday afternoon, after my 5.12 update, I was like "Holy shit, they did it! It's so fast. And the convo in codex harness is so good!!!"... it had failed over to sonnet and I didn't notice. It cost me $40, but it was such a great afternoon. I was really high on it lol

[robust sedge]

NO HELP FOR YOU!

[cold jacinth]

It’s always great to wake up and see your fix merged.
Submitted a fix to allow updating from the agent without it getting unloaded and therefore necessitating loading again manually

[peak pewter]

Yeah.. don’t add help to that allow. I had a similar error with memory and it’s obv not a plugin

[cold jacinth]

It’ll either make Openclaw update from the agent itself smooth by just saying “update Openclaw”, or it’ll probably crash everything

[fiery nacelle]

OpenClaw keeps me so entertained everytime i run into issues lol
I'm moreso laughing at anything I bring up rather than complaining about it haha.

[ionic wren]

I had this issue too, not sure why those basic bolt on things are needed to be added to the allow.... strange indeed. I just added it back and it was fine AFAIK

[peak pewter]

Did you try unplugging it and plugging it back in?

[robust sedge]

I used to be. But I've had nothing but redonculous problems the last few weeks. I'm over it. Used to be fun to tinker with. Now I just want something to work. So I can tinker

[fiery nacelle]

Is there like an "allow all" option at all? I thought thats what YOLO mode would kinda fix..

[ionic wren]

delete the list

[peak pewter]

Profile:full

I would query the docs via the cli with ‘openclaw docs “profile full”

[fiery nacelle]

    "profile": "full",```
already got that there

[tight birch]

@peak pewter so when I install codex in vscode it's not showing up in the menu bar on the left side

[peak pewter]

Dang. And that’s on the default?

[boreal rapids]

Hm, I just updated to the latest openclaw, but now openai-codex keeps nuking itself, it is migrating openclaw credentials to some file, and then openclaw reports "auth or provider access failed" 😅

[ruby crane]

i get this intermittently. agent will sometimes fail to call message (yes, the ability to reply) a couple of times and this will pop up in the log. others too. the intermittent nature feels like a race condition or something

it's not the only one. sometimes it hiccups reading pairing.json because it reads while it's writing (why it's writing when I'm not pairing anything is entirely another question)

[peak pewter]

It got integrated now into the new agents panel with vscode copilot. It’s gotta be a preference setting then

[ruby crane]

i really should create an issue but the evidence is kinda squishy

[tight birch]

If I click on codex in the top right it's just opening a chat window and not a new tab

[ionic wren]

why squishy....

[tight birch]

Matches claude to a T, so I don't get it

[peak pewter]

Let me try on mine, I still use vscode

[ruby crane]

no hard boundaries. no razor sharp evidence. squishy

"this happens sometimes. i don't know why" is basically useless as a report in a project with thousands of Issues

[tight birch]

I just like the look of it, but I would really like to be using claude and codex at the same time doing stuff to my OC

[ruby crane]

on dave's tiny personal project, it's probably a challenge for dave to figure it out

here, it's noise

[peak pewter]

It used to be on the left when codex first pushed their extension to vscode

[tight birch]

Sent you a screenshot

I feel like an idiot I can't figure this out lol

[peak pewter]

right click on the codex logo or the text that says codex on the right side, then hover over Move To and click Primary Sidebar

[fiery nacelle]

deleted the list, restarted the gateway, gateway restored the changes i made, list is back. But if i change it, and DONT restart the gateway, it get hotloaded and sticks... so confused...

[ionic wren]

maybe submit a github issue and someone can fix it

[fiery nacelle]

F*CK IT WE BALL!

[fiery nacelle]

im walking back some of the things i did, cause i did change a couple things other than the allow list, but it was just flipping a couple "false" arguments to "true" so im not sure what would cause the invalid config and flag it to restore from a backup.

i fat fingerd something.....

[ionic wren]

the doctor has changed a bit, not sure what it does now tbh but I know it's being tuned, definitely make a post about it atleast in #clawtributors I think vincent was working on it

[ruby crane]

classic PEBKAC

been in that chair 10,000 times

the worst thing about PEBKAC? you CANNOT SEE IT, no matter how many times you look. "it's perfect, I checked"

you just stop seeing it for what it is and only see what you thought it was

fucking sucks lol

[fiery nacelle]

at my old job we had these tiny touch devices for inventory control, and fat fingerings was prevalent AF. Your thumb would be touchin 4 buttons on a single keypress... usually dont have that issue with a keyboard lol

[ruby crane]

whenever I'm in a "this makes no sense, I have it right" situation I've learned to do one of two things: -

• walk away, get tea, read some news, take a walk, anything. come back, it'll smack you in the face
• find someone who knows nothing about what I'm doing. they'll spot it right away

it's the preconceptions/brain rut that will kill you

[fiery nacelle]

I take a different approach. Spend a week debugging a typo.
Write wrappers around wrappers just for it to be a stupid typo or cAsE sEnsitVE.

[ruby crane]

i used to the use the howling at the moon in frustration method too, then i got old

[rose crane]

So I just upgraded to 5.7 to 5.12, and it seems it's reconfigured from claude CLI to anthropic API? Anyone else seeing this? It's looking like I have to reconfigure all of my crons and sessions to switch.

[plain grove]

what's the issue?

[ruby crane]

what makes you think it's using the API? do you even have an API key configured?

[rose crane]

I'm getting the API behavior (lots of reactions and interim states I never got before), and I'm watchign the Extra Usage $$ climb..

[ruby crane]

the extra usage will only climb if you're using CLI. API is an entirely different account and mechanism and would never use extra usage

[rose crane]

hrm.. ok.. Any idea why 5.7 wasn't using Extra Usage, but 5.12 is?

[fiery nacelle]

ahh so you can no longer do the Claude-CLI route for your Sub? its all "extra usage" now?

[clear haven]

That’s an anthropic change not our change

[plain grove]

anything using claude code on a subscription through any 3rd party service is auto extra usage.

[rose crane]

I thought that started next month?

[ruby crane]

anthropic don't want you as a customer is the brutal truth

[ionic wren]

anthropic hates open source

[fiery nacelle]

anthropic will eventually crash and burn if they continue down that path.

[tight birch]

Trying to create workflows that I don't need to monitor. If it requires multiple agents, agent 1 can't tell agent 2 through discord to pick up where it left off

[ruby crane]

they've been evolving their "no 3rd party harnesses" strategy over time. any workaround you may find (they exist) today, will just as likely hit the same roadblock tomorrow

[plain grove]

allow them in a channel together is the simplest solution and allow bots in your openclaw config

[ruby crane]

lots of people burning lots of time and energy trying to give money to people who are clearly saying, "I don't want your money" and I don't get it

well, perhaps more accurately, "I don't want your money unless you 10x it"

spend that time and energy finding an alternative strategy. your psyche and wallet will be better off for it

[fiery nacelle]

"this is mine, you cant play with it" type of toddler mentality coming from Anthropic. So pathetic.

[tight birch]

Yeah I have this setup. Even tried making it a skill, I have all the channel ids and bot ids listed

[ruby crane]

"give me your lunch and I'll let you kick the ball twice" lets you kick it once and runs away for a week

[plain grove]

there's also a few a2a message primitives and some bulletin plugins that might help.

[upbeat barn]

Some one have idee to help me please

I try to create an agent workspace and a skill with a script. This Python script uses Opanai to make some prompts

like this example

from openAI import OpenAI
client = OpenAI(api_key=os.environ.get("OPENAI_API_KEY"))

When I run a test, OpenClaw told me he can't find any OPENAI_API_KEY

My openclaw is configurated to use openai as main model, i want khnow how to passe the open ai key directly to my python script or use direct openai model context as llm inside my script since that already openclaw use openai as main model

i initialite my openclaw model using openai oath no api key pushed

[plain grove]

so a few things here - if you use oauth, there wouldn't be an OPENAI_API_KEY. so...you would have to find a way to do that using the codex harness I would imagine. If you want to do true api things get an api key

[fiery nacelle]

why use something external to prompt your agent with OpenAI if you are already using an OpenAI model internally as your agent?
I'm just trying to wrap my head around what and why you are trying to accomplish.

like, just talk to your agent?

[peak pewter]

he's using oauth though

[upbeat barn]

They have some proccess to do inside a script using opainai

[fiery nacelle]

yea, that typically works as is. Idk why or where an external prompting script is even needed

[peak pewter]

and he is trying to use an api key secret

[plain grove]

yeah it depends if he's trying to do API only workloads that require an api key...so it would seem he was hoping he could do api key things with chat gpt account

[peak pewter]

he shouldn't try to read the token anyways from his script. he should just just enable chat completions and use that endpoint to call the model

[fiery nacelle]

are you trying to do additional openAI services outside of just the model connection?

[upbeat barn]

yes i need to make some video transcription using openai and make some treatement inside the pyhton script

[fiery nacelle]

or is this just a skill that your agent uses, to call openAI to then prompt your agent to accomplish something?
I'm jsut not following the usecase of this and need help understanding what your aim is, or if its even needed in the first place.

[peak pewter]

nvm i was wrong

[feral turret]

23 mins till usage resets lol

then i can go back to testing and see if i need to upgrade my subscription

[fiery nacelle]

are you pulling an existing transcript from a video, or designing your own?

cause openAI isnt needed for either..

[upbeat barn]

i want to generate a transcription from the video using openai

[fiery nacelle]

you can do this directly with your agent. you dont need any external connections.

especially if you are already using an openAI model with your agent

[upbeat barn]

Ok you mean instead of using a script i ask the agent directly troph a skill.md file to do that using his propre model after that they can put the result inside where i want script or anywhere right ?

[fiery nacelle]

is this from a random skill you downloaded?

[upbeat barn]

Oook right i will test this and sorry because its my first time using openclaw and his ecosystem

[upbeat barn]

i create it white my self

[peak pewter]

dont put your bank card in there, kira.

[upbeat barn]

yes yes you are right

[peak pewter]

or any secrets. Not trying to be mean but it seems like you are trying to reinvent the wheel here

[fiery nacelle]

as long as you arent randomly downloading skills, thats good. hah
Try talking to your agent like another person. Tell them what you are trying to accomplish, and what suggestions they have to accomplish the goal.
You Agents are smarter than like 70% people here (no offense to anyone.)

[peak pewter]

you would be essentially doubling your cost by asking openai to pass a prompt to another model in your openai account when its just one round trip, not two

[upbeat barn]

Ok ok i will

[upbeat barn]

exactly you are right its moore token

[fiery nacelle]

yea that was my main confusion about it. burning tokens to burn tokens only to burn tokens.

[cold jacinth]

Mfer. Power went out in the building that houses my claws. I have no access for two days

[peak pewter]

there is also a "create-skill" skill, claude made it popular and its bundled in openclaw as a skill

[upbeat barn]

i have an other question please if you have any idee about any online or ai agent specialise on video editin i can use it to send to them a video and some instraction and he can edit the video for me to make a senario and some effect ?

[peak pewter]

ai can lead you down the wrong path for sure. I know i have went the wrong wayyyyyyy, and it cost me money and experience but it was confusion

[fiery nacelle]

https://github.com/aregrid/frame
Thats something I've been meaning to play with, I just have no use for editing videos myself.

[peak pewter]

I like Kie.ai, you can also use remotion. I think remotion would be easier since it can spin up locally and there is a remotion skill in the ecosystem

[cold jacinth]

https://x.com/chatgptapp/status/2055317612687675545?s=46&t=bf-ZDvDKSAKN2aYkOeRfpQ

this is pretty cool’s for Americans who don’t care

ChatGPT (@ChatGPTapp)

A preview for Pro users: a new personal finance experience in ChatGPT.

Pro users in the U.S. can securely connect financial accounts, see where their money is going, and ask questions based on the information they choose to connect.

Your full financial picture, now in ChatGPT.

▶ Play video

X · May 15, 2026, 4:01 PM

[upbeat barn]

thank you i will test this

[tepid kestrel]

yes of course i want ai to check on my bank account, what could go wrong?

[upbeat barn]

Thank you so mush friend you give me lot of idee and you help me to more understand and learn about openclaw

[cold jacinth]

Like I said. The ones who don’t care lol

[peak pewter]

13 commits haha nice

[tepid kestrel]

it will go haywire as soon it starts to hallucinate that someone stole your money

[fiery nacelle]

its only secure until a massive leak comes out.
I swear every platform that has EVER claimed to be "secure" has had a leak after such claims. If its touching any public rails, I can assure you, it is NOT secure.

If you dont own the infrastructue, you dont own the data, and therefore cannot trust any claims of security.

[peak pewter]

Nothing. If something does, you have a case for a not authorized claim. “My ai chatbot bought that jacuzzi for 5 grand”

[cold jacinth]

What was that cheating site for married people? Ashley Madison 🤣?

Mfers threw away their lives and most of the women they were talking to were actually men 🤣

[peak pewter]

Triggered. I took a picture of a scratch off ticket and sent it to ChatGPT and it told me I had a $10,000 winner and it wasn’t I was crushed

[fiery nacelle]

ha, i only stumbled upon it and added to a list of "things to check later"
I think remotion might be a better fit through.

[peak pewter]

It’s a decent find. Remotion is alright, depends what quality he wants. I think learning manual first before automation will help. It’s tricky to get pexels to find images that actually are related to a keyword you need for a gif/photo in a scene, not to mention lining up the sub titles, TTS nuances

Here is a prime example:
https://youtube.com/shorts/6K8gxcyTD6Q?feature=share

@fiery nacelle have you ever messed around with openwebui?

[fiery nacelle]

i briefly played with Openwebui with Ollama. Its cool, but for me, its just another chat portal to me and not needed.

like, i can accomplish the same thing with stuff im already using, so i personally dont need it.

[velvet locust]

Clawdbot is kinda goated for game developement tbh. I love how customizations can help make it find the exact things I need to do in my massive codebase

[short granite]

I was so amazed until I realized 5.12 is consistently failing over from my local model to claude sonnet... lol I wonder what's misconfigured

🤦‍♂️ does it just not work with lmstudio at all anymore?

guys, y u do this to me

[peak pewter]

I was thinking of over-engineering a connection between OC and openwebui lol

[fiery nacelle]

yea theres really no need. You would just be adding a duplicated layer of the same functionality. lol

[peak pewter]

it's in their docs, openclaw is in everyone's docs now. The hype train is crazy for OC

[fiery nacelle]

adopt or die. doesnt matter if the adoption is a poor implementation or not, gotta stay "relevant"

side note, anyone got any ideas how to basically create a "live chat message process" or somthing similar?
Like if i were to send multiple messages to my agent in quick succession, instead of them being queued up and processed and replied to 1 at a time, they would be processed in parallel and bump/merge with the first received message and the LLM then processes it as a whole instead of individual chunks, sending only a single conglomerated response back? or even a command like "/also" or "/and"

[shadow pumice]

Can anyone please share your most used ways of using clawdbot? I'm new to using this so wanted to know from those who already know it's potential

[fiery nacelle]

Research projects. Pulling in papers, building frameworks theorized in those papers and testing them out in my own environment.

[shadow pumice]

And for day to day activities with respect to normal life and as a Data /aspiring aiml engineer

[peak pewter]

that's a good one. I know you can send in-line /btw but that is outside of the session. I'd be down to try to figure it out when I get back from walmart for sure

[shadow pumice]

That's great will try this out

[peak pewter]

I hate that second message that gets cut off too

[fiery nacelle]

theres other cool things i do too, like having 1 agent access and manage multiple systems on my network. websites, apps, basically anything that involves code, or a terminal, OpenClaw can do.

[shadow pumice]

For implementation how much time it can take with opemclaw?

I am planing to get anthropic key, is tat sufficient for these activities

Anthropic subscription can work with oc?

[fiery nacelle]

i primarily use local models. and it seems like the recent consensus is to use Codex over Claude with OpenClaw. Anthropic is acting like a toddler throwing a temper tantrum over other wanting to play with its toys.

[shadow pumice]

Ah okay, so I'll have to get codex if I want to us oc

But isn't anthropic claude better this year

[fiery nacelle]

you can use any API you want, or other subscription plans like OpenCode-GO, GLM Coding plan, Ollama, OpenRouter.
You just cant use Anthropic subscrition with OpenClaw cause they shut that down.

[shadow pumice]

I would love to use anthropic key with oc that would produce better results right

[shadow pumice]

Ah aight

[fiery nacelle]

you can use Claude, but you are paying per token and is going to burn a hole in your pocket.

[shadow pumice]

I'll have to get codex then right

Will install and test

[short granite]

gotcha... I think it's

allowPrivateNetwork: true

a new safety check

[fiery nacelle]

OpenClaw also has a lot of Codex stuff integrated into it, to its just a better overall pair.

[supple ibex]

First time I burnt ~40usd within 12 minutes 😂

[toxic elk]

I'd like to check if you are also seeing "openai-codex:[e-mail]" when using the new OpenAI login? This seems to be an old provider setting, which causes the conversation to display: "⚠️ Model login expired on the gateway for openai-codex. Re-auth with openclaw models auth login --provider openai-codex, then try again."

[umbral kernel]

Hi, first time here. In case i have the Pro Openai codex subscription, how i monitor the tokens i use with Openclaw?.

[shadow pumice]

Local models for data security right?

[short granite]

smdh that's all it was

allowPrivateNetwork: true

[fiery nacelle]

I dont use Codex within OpenClaw personally, but if im not mistaken, you can go to your openclaw dashboard and click the "Usage" tab ( "http://127.0.0.1:18789/usage" ) on the right to see your token burn.

[umbral kernel]

which subscription service you recommend?

[fiery nacelle]

I primarily use local models within OpenClaw personally, but the Codex $20/m plan and OpenCode-GO $10/m plan are ones ive enjoyed with other things.

[umbral kernel]

so, you run openclaw in your own machine?, using docker or something?,

[supple ibex]

Openai codex and/or kimi coding

[fiery nacelle]

I'm running it directly installed. not sandboxed or anything.

The system I have it running on I bought specifically for use of AI and running OpenClaw. Treating it as if it were my Agents system, not my own.

Like I'm the guest on my Agents system.

[narrow urchin]

I just got done setting up openclaw anyone have tips or skills i should look at?

[umbral kernel]

it can be a Raspi?, something cheap?, or must be a more powerful pc?, if you run local models i think you use a mac mini. I just want to set up my own system. But dont use a remote vps.

[fiery nacelle]

OpenClaw can run on a Pi. You'd be limited to tiny models if running locally on a Pi though. Whatever models you run locally are dependant on what resources you have available. You could run OpenClaw on a PI, but have your models loaded on a separate system with more VRAM or RAM. You can distribute the compute across multiple systems on your network too.

You can also take it a step further and also split a single model accross multiple systems for distributed computation if you really wanted to.

[wanton plover]

I am considering that instead of updating from 5.7 to 5.12, if it will be better to install a new OC, e.g. in a proxmox VM, and migrate from there. What do you think?

[fiery nacelle]

use the other VM as a test, if everything works, burn that VM update your current install.

unless you are wanting to use it in the VM permanantly instead of however you are currently using it.

[wanton plover]

Nit3stalkr, I have already tried to update, had some minor issues, but at the end of the day, I had the recurrent issue, with delayed response from the time sending the prompt from OC , to the model receiving the prompt (e.g. if local, LM Studio receives my message after 6-7 seconds; if it's GPT5.5, it takes 20-40seconds to get a response). So I am considering the alternative option, since i love OC.

[umbral kernel]

thank you. When using LLM subscriptions, does the PC need minimum RAM or CPU specs? I have an old Celeron with 2GB RAM—is that enough?

[fiery nacelle]

nah you're fine. you can run cloud models on a Texas instruments calculator if you really wanted to

[winged dust]

Wsp im looking for a dedicated person who we can discuss and leverage ai and possibly make money from it

Accountability partner somehat aswell

[steel vector]

what is an accountability partner 😭

[feral turret]

hmm my coworker suggested i build my own linux server instead of running this old ass mac mini

[ruby crane]

I'll translate capitalism/scam for you. it's free labor

[dull magnet]

My openclaw is not responding chat, its replying heartbeat.md

[steel vector]

sounds like it

like “you’re accountable for messing up pay me!!”

something like that

[ruby crane]

simply be extremely skilled and dedicated and this person will pay you in appreciation

[dull magnet]

Can any one help me please

[native lintel]

Krill Clawbot wiill.

[steel vector]

#1459642797895319552 or #users-helping-users

[ruby crane]

sure. simply cut toenails in a straight line to prevent ingrown nails

[steel vector]

[feral turret]

krill just added me as a friend. i am flattered

[steel vector]

SCAM

be vigilant man

[feral turret]

i know

[steel vector]

👍

[feral turret]

quick little dab before my 2pm meeting

[orchid axle]

How are you people using two oauth accounts? Here's my experience:
[5/16/2026 1:54 AM] Me: /model openai/gpt-5.5@openai-codex:[e-mail] [5/16/2026 1:54 AM] OpenclawLocal: Auth profile "openai-codex:[e-mail]" is for openai-codex, not openai.

[tight birch]

Never mind lol

[winged dust]

Yo I’m looking for a serious AI study partner/group.I’m trying to lock in and really learn everything abt ai

[silk violet]

Yo I’m trying to improve my AI/dev workflow and learn from experienced users here.
Any beginner mistakes I should avoid with OpenClaw?

[fleet crest]

Stop working on workflow and tuning how and minmaxxing. Make something. My genuine advice. Especially elaborate Openclaw agent memory stuff. Lots of brilliant minds working on that. What we have now is already very good "out of the box". Use it to do something cool. 🙂

[boreal rapids]

"Make stuff" is always good advice 👍

My claw has been teaching me about electronics, I've assembled a Tamagotchi from scratch now 😃

[fleet crest]

I made a live bus departure board in my flat of my local stop as a web app (just connected to an api). Lame but also fun and kinda useful.

[ember moth]

Just updated from 5.6 to 5.12 .....Smooth so far ....responses noticeably quicker on discord

[ionic pebble]

what did you do? mine is kinda slow

[silk violet]

Thanks for your tip.

[ember moth]

Nothing special .....Updated it manually though ....didnt let my agent do it....Updates have been much more stable that way and i avoid npm

[fleet crest]

A universal remote web app connected to any smart devices. You could get an IR blaster (RM Max for example) and then have it for your non smart tech like a tv and such.

[feral turret]

hmmm my bot stopped responding again

[ionic pebble]

i do it manually as well, but through the one-liner curl

[feral turret]

try again

[loud ravine]

Holy shit the level of hallucination gemini is doing with me is profound

Like I am asking it about a brand new video game and it is literally making up nearly everything about it, and what is more embarrassing is it has access to the internet

[feral turret]

i am about to rip openclaw out completely and reinstall i can't even get onboard to run again for me idk wtf i did lol

[clear haven]

Impressive 😂😂

[feral turret]

see if krill can figure it out

[loud ravine]

Lol I got two fake krill bots adding me at one time

Do scammers genuinely believe I am that stupid? I know I have a 79 IQ according to testing in 2018, but do they really think I would buy it?

I know with my IQ I am a certified troglodyte dumbass, but cmon

If I am petty enough, I will send the scammers…the vending machine.

(The vending machine is a soda can coming out of a woman’s….)

[ember moth]

i Gave up on Gemini for this reason and their rate limits on a pay as you go service make no sense

[loud ravine]

The amount of information it pulls out of its ass is profound.

[feral turret]

i'm starting to think this is equipment limitations

i'm gonna have to get something a little more powerful

[ember moth]

what are u using?

[feral turret]

old ass mac mini lol

mid 2011

[ember moth]

i was using a 2018 base mini and had to switch because of ram......had only 8 gigs and when i tried to have my agent build a mission control NextJs took out my system lol

[feral turret]

lol yea this is 8gb as well

[ember moth]

bought a mid level mini pc off amazon and threw proxmox and linux on it best decision ever

[feral turret]

i was hoping it could just run orchestration etc but it's seeming pretty underpowered i might have to go find an m1 mbp or upgrade my 2021 and use that

yea i'm running linux on this but it seems to be cpu that's the issue

and it dropped my openai oauth from loosk of it which is odd

[fleet crest]

beelink gti 14 185H ultra with the RAM upgrade is a beast for a homeserver, eGPU if you ever want to run local models too

[feral turret]

it was only a couple hours

[ember moth]

2011 is rough.......good to tinker but hwne u actually wanna build a swarm with hella workfows thats when mine gave up lol

[feral turret]

yea brand new ssd i figured it could handle

but it's intel 2.3 ghz i believe. my 2021 mbp is m1 pro

i might just buy myself a new mbp and dedicate the 2021

[ember moth]

need at least 16gb of ram and minimum 4 cores with hyperthreading to be safe for the future

[ember moth]

smart

start with the old dusty hardware but now that ur serious migrate to a decent machine that can actually do work......funny thing is i hated that 2018 mac mini it was slow as hell to use for myself so i knew my agent would hit a bottleneck sometime lol

[glad lintel]

can anyone assist me in properly connecting my openclaw gateway to a windows pc as a remote

[feral turret]

yea as i'm getting more into it i realize i need more power

[peak pewter]

I’m down

[feral turret]

and if i went with my mbp i could run ollama locally too for non reasoning/coding stuff

[peak pewter]

How much ram? Which m series?

[oak crystal]

anyone else getting throttled by fireworks on fire pass?

[feral turret]

m1 pro 16gb

[peak pewter]

I got a m3 pro 18gb unified and can barely run an 8b

[feral turret]

so kiss local models goodbye

[static osprey]

Can someone help me i have installed openclaw yesterday and im usingng a local models, but he forgets everything i say. I tried to use other local models but nothing works.

[ember moth]

ask ur agent to make a daily summary of ur conversations and pull from that daily summary to get context at the start of every new session

But local models arent the greatest to be ur main agent

dont believe the hype

[plain grove]

just get a server rack and run it for real

[jagged cloud]

@Events it’s Weekly Claw time!!

https://discord.com/channels/1456350064065904867/1467913726580752599

[loud ravine]

Anyone notice how on openclaw, atleast when using gpt-5.5 codex, hallucinations basically drop to (almost) 0?

I have yet to catch a hallucination, and I am usually very good at catching them

[ember moth]

i dunno if its a good thing or not but my guard is completely down for hallucinations for GPT 5.5......After coming from Gemini that was a tall task to get over that fear lol

[formal heart]

Honestly. I'm so freaking sick of the absolute trash releases. I haven't had a good experience since like 10 releases ago, every SINGLE one, no matter how many bugs/I create completely borks some important part of the chain, multi model, telegram, polling, fallback.

All this vibe slop, and PR's aint helping anything. I realize this is a free product, but good grief the amount of time I've wasted on trouble shooting. I'm just no longer upgrading.

[plain grove]

? what is your setup like. I think you're in more of the 1% of users having this severe of issues. i don't even think beta testers these bad of issues.

[ember moth]

What was you last stable Version?

[supple ibex]

I think it's not just 1%, 10% is also so optimistic

On my side 4.23 is the latest stable

[ember moth]

4.24 to 4.29 was a roughhhhh patch...... I also had issues ....i would try the newer versions for sure .....

[formal heart]

around 6 agents.. all Telegram. Voice TTS round-robin

4.23 was GOAT I may have to go back to that. 5.7 was also not bad but still minor issues. 5.12 is borked.

[plain grove]

i have 4 gateways running on 3 different communities all on 5.12. no issues. all discord + some blue bubbles.

[formal heart]

Seems Telegram regression

The new "isolated ingress" feature they shipped in 5.12 has a bug — the startup code sequentially blocks on each Telegram account's first poll, and when the event loop is already loaded from model prewarm + bootstrap (15 seconds of prep work), the Telegram API calls time out, which cascades into 3+ minute total stalls.

[ember moth]

Is their a PR?

[formal heart]

Creating one (BUG report that is)

[plain grove]

take a look at this - https://github.com/rendrag-git/openclaw-admin-skill . Use it in codex/claude to help diagnose. speeds up things significantly..

[ember moth]

Thanks noted

[winged dust]

I’m looking for a serious AI study partner/group. To hold outselves accountable and work towards start ups

Lmk

[devout trail]

Why we have no more a lot of update than the last 2 month

[ember moth]

People complained too many updates breaking things so they slowed it down and hardened the process....This is a better path imo

IMO i cant wait for LTS.....i dont wanna update lol

just keep the security good and ill chase new features every quarter or something

[manic echo]

strong genetics has it right. from a shipping-on-it angle the slower hardened cadence is the better tradeoff. the cost of chasing every release is playing out in u-h-u right now, 5.12 broke cli for some and mcp servers for others. pinning to a known-good and moving deliberately beats staying on head and eating regressions.

[keen dawn]

bro we are still on a 6.4 day median of releases 🤣 that shii crazy, we just got used to the new era of speed

[formal heart]

https://github.com/openclaw/openclaw/issues/82311#issuecomment-4463350566

[ember moth]

Clawsweeper already on task lol

[spiral wraith]

Is there a list of all the "this is how I do X" and "I built tools that do ABC" guides/lists from steipete? I keep reading/finding gems; surely there's a list somewhere

[manic echo]

not aware of a single curated index, he doesn't seem to keep one. closest thing is his blog at steipete.me, the long-form "how i do x" stuff lands there. the tool announcements tend to break on his X first though, so following both is the realistic move until someone makes the list.

[upper wolf]

wow .12 latency is so much lower than .7, so pleasant

[frozen zealot]

lol what a shame, couldn't get Json to address his people

[peak pewter]

where is the party?

[manic echo]

worth splitting those: .12's runtime is genuinely faster (codex switch), the regressions people are hitting are in the tooling/integration layer not speed. both can be true. if you're not leaning on remote nodes or heavy mcp you get the latency win clean.

[cold jacinth]

🚨 📣

https://cybersecuritynews.com/openclaw-chain-vulnerabilities/

This is why one needs to update

[frozen zealot]

Up to (excluding)
2026.4.22

CVE-2026-44118 is the entrypoint

[ember moth]

It's sick

[cold jacinth]

That’s all on the assumption there’s nothing else identified. In this world? Come on

[clear haven]

partially FUD, their 245,000 number is overblown
still worth updating though

[cold jacinth]

Openclaw went through a huge amount of updates for security. Security doesn’t end.

[clear haven]

people see the mDNS show up as openclaw in scans and assume it means that the port/ui is accessible when its usually not (and openclaw makes it to be not by default)

[frozen zealot]

all software have bugs, the linux kernel is having a good time with root escalation right now

[cold jacinth]

Yeah and with the supply chain attacks it’s getting worse and worse. Reality is the speed at which vulnerabilities are coming up makes it very important for one to stay on top of their security

[frozen zealot]

third one just dropped, fragnesia lol

[cold jacinth]

Our OCs are all development environments.

[shut bloom]

did you see weekly usage limits are 50% higher over the next month?

[clear haven]

@manic echo dont send ai slop here, its not appreciated and will get you removed from the server

[shut bloom]

W shadow

[clear haven]

we all have our own agents to talk to, this discord is for humans

[shut bloom]

lmfao i made a discord for agents

litterlaly called clawcord 😭

[frozen zealot]

yes, and someone i know thought of making twitter for agents... we call it cl... nvm don't want to get banned

anyway, the point of the story is we need less social-media, not more

[cold jacinth]

Are you here?

[clear haven]

they're muted

[cold jacinth]

Damn. Wanted to make him loop and utilize his usage

[cedar osprey]

Lol wtf is happening to OpenClaw.

[cold jacinth]

Umm

30 day spend 1.3 million dollars 🤣

Pete is living the life working at OpenAI lol

[shut bloom]

we need more!

[cold jacinth]

603 billion tokens lol

[shut bloom]

we need brainrot for agents

[shut bloom]

to be fair im in the upper 100B's nearly overlal

[umbral kernel]

Friends, the VM to install OpenClaw can be only console? Or is best to have graphics?

[keen dawn]

bro peter is a different breed 🤣 https://chatgpt.com/s/m_6a0791fd7414819190bc7437e91bc603

Coincidence that both are from Austria?

[cedar osprey]

Total or ~100b/30d?

Just so we know who to charge the environmental impact tax

[shut bloom]

I use like 4B a day

i want to use as much water as i can

[zealous lagoon]

Server is much lighter.

[tired island]

how do you install open claw on mac?

[shut bloom]

i will personally throw two buckets of water out my window in solidarity

[hot belfry]

what’s the best free models to use with open claw? i’ve been using the free nvidia ones but they seem kind of stupid..

[shut bloom]

Yea dont be using those ones, at worst use kimi as its cheap atleast and functions

[cold jacinth]

Go to Openclaw.ai and download. Easy

[cold jacinth]

Small model - stupid instance

[frozen zealot]

well, Qwen 3.6 and Gemma 4 are not bad, it's just hallucinations

also according to Json, Qwen sounds like a new immigrant

[keen dawn]

glm also not too bad

[ionic wren]

that's our peoples

[timid harness]

Hm, post upgrade, I seem to be consuming openAi API credits rather than sub. Am I dumb, did I miss a step during the upgrade

[young anvil]

So i have the feeling that 5.12 is a solid release. What is everyones experience? Like OC is back in the game and 5.14 beta looks promising

[keen dawn]

5.12 is solid but still some problems.. 5.14 will help to fix a lot of stuff so i would say 5.14 will be awesome

[zealous lagoon]

Basically yes. I had teething issues with my very old legacy config but it's been solid and snappy.

[peak pewter]

have you used this skill?

[plain grove]

i made it

and @zealous lagoon can vouch

[zealous lagoon]

#self-promotion

[plain grove]

you promote it in here all the time lmao

[zealous lagoon]

That's the same one I showed you last night (it's 6am here)

[zealous lagoon]

because I didn't make it MYSELF! 👀

[peak pewter]

ya i got it tabbed, i was gonna give it a go

the updating local docs is a smart script

[zealous lagoon]

I like it so much I shill it all the time. And I get a lot of good feedback. Could be v. simple. The bot just needs to know a few key things including its dependence on gateway to function, the location of the docs, how to get the source code etc...

[peak pewter]

upstream docs are behind

[zealous lagoon]

npm doesn't include them either so it needs to know where to find them

[peak pewter]

and the cli doc search aka openclaw docs <query>, updates with openclaw upgrade or no?

[ebon atlas]

Anyone else been jamming to drake today? 😛

https://open.spotify.com/track/0Ja4OhirhQHfB72Cq3B3DF?si=ba8944c3bdd14702

[ionic wren]

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

[peak pewter]

draaaaaake???? (soulja boy voice)

[keen dawn]

3 albums lol

[ebon atlas]

I made sure to send the non explicit version 🙂 not nsfw

[ebon atlas]

i already called the fact that it was going to do that, hes going ot release 1 moreee album

in like a month same thing he did with Views, he dropped the other 2 random albums with iceman to get out of his label deal so he can drop something like "IceMan After Winter" or "IceMan Pt2." mark my words

[keen dawn]

lol this time without birdman, or maybe drake uses openclaw for beats now

[timid harness]

Ok was dumb, needed to get it to switch over to codex manually for some reason

[cold jacinth]

Oh that’s the old skill

There’s a new one

Better

[ebon atlas]

UGM is basically actin like birdman rn, they dont want solo artists because he would make WAYYY too much money, you think he rich now, imagine going #1 with no label, youree making like 50* million on release day

[robust sedge]

Do you think we'll ever reach a point where OpenClaw can respond as quickly as the native Codex app? Because holy shit, the response times for Codex are insanely quick

[zealous lagoon]

@cold jacinth and @plain grove discussing their openclaw skills in #self-promotion

[plain grove]

funny i absorbed his MUAHAHAHAHA

https://tenor.com/6kBH.gif

[zealous lagoon]

https://github.com/rendrag-git/openclaw-admin-skill

https://github.com/BKF-Gitty/openclaw-update-runbook

FIGHT

[zealous lagoon]

No but reducing the overhead imho is mostly about consolidating and trimming your system prompts.

I compared my own Openclaw vs direct API calls when I was looking at voice endpoint response times and Openclaw added about 10s of latency for me. On a v. slow machine.

[plain grove]

mine breaks off updatinginto its own separate runbook so we can customize installs to the machine. local-install.md

[soft remnant]

Good mornnin

well its not morning ive been awake for hours but good morning

[robust sedge]

Well I'm currently at like a 10 second delay between when I hit enter in discord, and Scout starts 'Typing...' like it actually received my message. Annoying. That's the 4.27+ lag time in action 😛

[zealous lagoon]

something not right with that.

[zealous lagoon]

Something very wrong with that. I think the typing indicator isn't even the model it's Openclaw. Mine is almost instant.

[golden marsh]

Tbh, I wouldn’t want to miss either of them. Both are very useful, each in their own way.

[zealous lagoon]

I have both. rendrags is the OG but BK made his specifically to focus on updating the sh*show that was 4.29-5.10 as practically nobody could get it running out of the box.

[golden marsh]

Beside the second Part of ur sentence im with u i have both of them too 😂

[steel vector]

why should i get them?

[golden marsh]

They usefull. Some detailled steps for Agents to check before updating etc.

[steel vector]

ooh okay nice, yeah i always end up telling them to check stuff before updating so i'll check these skills

[golden marsh]

Yea i did throw both of them inside my Codi and when he runs checks etc he knows he have to read them first before he starts the destruction

[steel vector]

tbf my updates the last 2 months have been fine, but i also only updated 2 times lol so not much to go wrong

i am planning to help with the beta tho so i hope these come in handy

[golden marsh]

2 in months thats why ur fine. every update brings something new. Dont get me wrong im not saying thats a bad thing. But stuff needs to be fixed and for that this 2 runbooks are great. Both @cold jacinth&@plain grove did a great Job there.

[steel vector]

yeah i dont seek many of the new features that get implemented which is why i don't update much because i know it isn't always stable xD but i'm just now buying another VPS so i can use that agent to help out with beta testing openclaw and LibraVDB some more so these will come in handy

[cold jacinth]

Listen I want to be fair

My skill is just so superior it’s not fair and I want to be fair

So let’s just say rendrags is better and ignore my amazing, all encompassing skill. It’s like a cheat sheet and we don’t like cheaters

[golden marsh]

U did not read i think im using both of them, so i double cheat i guess 😂

[cold jacinth]

I mean it’s like comparing nuclear fusion to starting a fire using a magnifying glass

Adequate analogy

[ionic wren]

https://giphy.com/gifs/come-at-me-im-here-big-bird-H6W9H29kVsUI2hJE90

what we have happenin in here....

[golden marsh]

Wow, we hit we hit a sore spot @zealous lagoon ^^

[cold jacinth]

Rendrag was self promoting

I stopped it. Don’t worry

[ionic wren]

https://giphy.com/gifs/hulu-only-murders-in-the-building-omitb-4ULLL0hBVO4paUSpnU

[golden marsh]

There is nothing wrong to compare 2 good things. @cold jacinth

[cold jacinth]

One day my skill will have 2 stars and I will still talk to all of you like I did before I was famous

[golden marsh]

When i remember right it have 2 stars 😉

[cold jacinth]

Oh damn I am famous

I need to slow down. People already calling me the next Steve Jobs

[ionic wren]

you do know self promotion channel is retired right?

[golden marsh]

Nah hes still on V23 xD

[ionic wren]

TBH, if it's relevant to the convo and it's not blatant spam, IDGAF

[plain grove]

Its k i pacman'd his skill

https://tenor.com/zm8Y.gif

[ionic wren]

IP theft on the other hand....

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

[plain grove]

IP implies an I

[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?

[cold jacinth]

Wow. You nerds can be hostile

[cerulean flare]

sup guys. we still rocking with opus as the main brain model or is gpt-5.5 actually good with the new update?

[cold jacinth]

This is like revenge of the nerds 4

[keen dawn]

opus who

[cold jacinth]

5.5

[robust sedge]

Tell me about it bro. Anything after 4.23, on the codex harness, is jacked for me. I'm like 20 hours of work into it now. Zero fixes found. I'll get it eventually

[cerulean flare]

I have some clients i'm working with using openclaw and don't want to rock the boat, but wanted to ask you guys here first before i switched over in a few days to gpt-5.5

[golden marsh]

I think 5.5 is the way to go.

[cerulean flare]

thank you so much for the feedback guys. if i was still in the learning and playing around with openclaw phases, i would switch so fast. but there is important stuff on the line so i have to be careful lol

[plain grove]

[cerulean flare]

so that also means the new update that came out is kinda fire i assume, huh?

[ionic wren]

too much time hugging the boys, not enough time greasing your clanker

[robust sedge]

What if I do both at the same time...

[ionic wren]

https://giphy.com/gifs/VeeFriends-hug-hugs-hu-x29239XeUoH6Fh7RYM

[robust sedge]

Discord as the channel though?

[golden marsh]

What if he is sitting on ur nose 24/7 is still hugging needed ? ^^

[rose sundial]

sudo apt install opsec

[teal cradle]

freakingA, I just realized mem0 was a cloud implementation and is begging for money, is there a way to switch openclaw to something else without screwing it up? And what else works well?

[robust sedge]

I've been having great luck with the combo of LCM for context management, and QMD for memory (plus I put my entire Obsidian dir into QDM as well).

[ionic wren]

have you updated your plugins?

that might be something there to your gimped robot

[zealous lagoon]

Yes.

[robust sedge]

All plugins up to date. I dunno man. I'm still tinkering. It's bearable, but it's no 4.23. The 30 second turn for every comm is... frustrating.

[zealous lagoon]

Have you tried switching off all memory plugins?

[clear haven]

Have you tried without plugins

I can’t reproduce that on a fresh install

[robust sedge]

Scary 😛 I'll give that a go when I have more hours to dive into it.

[zealous lagoon]

Dude that’s the first thing I did during the Dark Times. But also have Claude or codex time your prompts. It can see how long every stage takes.

[robust sedge]

can you describe what you mean by have Claude or codex time your prompts?

[zealous lagoon]

Are you using Claude code or codex apps/cli?

[robust sedge]

yeah, codex app to troubleshoot stuff

Or claude. I have subs for both

[stuck mesa]

5.7 seems to be my golden ticket I don't expect I'll be trying to upgrade off that again for a long while

[zealous lagoon]

Ok then tell codex about the problem. (Or Claude) , give it the openclaw admin skill. And tell it to watch as you run a couple of prompts. It will be able to give you a breakdown of how long each stage of the process takes and figure out what’s causing the holdup.

[jagged cloud]

@everyone

Security in OpenClaw is getting sharper 🦞

🔒 fs-safe for root-bounded filesystem
🌐 Proxyline for policy-driven network egress
📦 ClawHub trust evidence
🛡️ smarter command approvals

Powerful agents need guardrails you can actually audit.
https://x.com/openclaw/status/2055437760459055405?s=46

https://openclaw.ai/blog/where-openclaw-security-is-heading

[peak pewter]

i

[peak pewter]

i'm using openclaw-admin-skill right now with codex

[frozen zealot]

shadow, can we have exec deny list please?

[cedar osprey]

[golden marsh]

Nice sounds good. 🙂

[neon bramble]

How does that new update compare to nemoclaw? Is Nemo obsolete after that?

[frozen zealot]

fine, i'll make it if you'll approve it

[clear haven]

Idk about approve but I’ll review

[marble ledge]

You can ask your agent to do it too, if you are willing to wait, and ask them to review what stage is eating up the time. You might find it's something personal to your setup or some combination of something that is resulting in the sluggishness you are experiencing. If you are using third party plugins, make sure you aren't hanging on to very dated versions of them either. Not saying you are, but it happens, sometimes we forget.

[harsh geode]

OpenClaw is going corporate and losing it's mojo. The last release notes are meant to impress C-suite execs, not everyday users.

[marble ledge]

Hmmm.

[cedar osprey]

It isn’t impressing anyone.

[steel vector]

bro what

twitter ass comment

[golden marsh]

Im using OC 24/7 so ur not talking for me i guess.

[clear haven]

Damn, I gotta go tell all the everyday users that were impressed to stop it

[steel vector]

when you dont comprehend some updates just arent for you (you dont need to update)

[robust sedge]

Do you get the 'streaming thinking' in discord like we used to? Like the agent writes some progress while it's working, and edits the message with the final output after it's done? Mine has quit doing that whatsoever. Just radio silence until the final verdict

[marble ledge]

Hey...what's wrong wish C-suite execs 🙁

[steel vector]

its only wrong when you’re not one of them apparently

[lost gulch]

I’d rather marry gpt than Claude

[clear haven]

I’d rather marry a human

[zealous lagoon]

Yeah for ppl on a vps it might not be convenient to have a codex cli running at the same time. Your agent can do it himself, especially if you're using discord you can have one session watch the other in realtime. The benefit of using the external SDE is that it can change config and reset the gateway randomly without committing seppuku.

[fallen sluice]

If any possible way to run the openclaw completely free

[zealous lagoon]

OK I've been on a deep dive with this. Lemme grab my channels block from the discord section of config...

{
  "channels": {
    "discord": {
      "streaming": {
        "mode": "progress",
        "preview": {
          "toolProgress": true,
          "commandText": "raw"
        },
        "progress": {
          "toolProgress": true,
          "commandText": "raw",
          "label": "auto",
          "maxLines": 6
        },
        "block": {
          "enabled": false
        }
      }
    }
  }
}

Keep this global group-chat policy:

{
  "messages": {
    "groupChat": {
      "visibleReplies": "automatic"
    }
  }
}

Keep normal verbose off:

{
  "agents": {
    "defaults": {
      "verboseDefault": "off"
    }
  }
}

This has nothing to do with /reasoning but it resets behaviour to pretty much what you'd expect. If you're like me and had a legacy install from January that's been poorly tweaked every few weeks despite the new knobs/settings available...

[robust sedge]

Thanks man! I'm stoked to try this.

[zealous lagoon]

Caveat the "raw" part is only suitable if you're a single-user installation. (which most of us are).

[zealous lagoon]

Do you have the reacc emojis working?

[robust sedge]

I used to somehow, but it seems to have stopped using them regularly.

[zealous lagoon]

part of the same problem I think. Ask your bot about the setting (once it has the skill) and it can tidy up config for that, too.

[feral turret]

sup

[ebon atlas]

btw boys, i cannot be stopped, I WON MY CASE.
idk if i said it today yet.
I represented myself and used chatgpt and openclaw and beat their HR Legal team.
Solo dolo.
The result came in today at night.

[rancid mountain]

anyone know why my open claw do that automatically please ?
[2026-05-14 22:28] The gateway restart completed successfully. Tell the user
OpenClaw restarted successfully and continue any pending work.OpenClaw restarted successfully. I'm back online.

[ruby crane]

openclaw will automatically restart if particular config changes and also the agent can signal a soft restart by itself if it thinks its required

[ionic wren]

ayyyee that continue pending work was me and @maiden kayak

[maiden kayak]

Ugh I need to fix it again. It's broken once again in telegram topics (at minimum)

[rancid mountain]

but the thing is she forget what we were doing right before and thats shit i always gotta put her back in context

[maiden kayak]

What medium are you using to communicate with your claw

(like telegram, discord, whatsapp?)

[rancid mountain]

telegram

[ruby crane]

hmm. that shouldn't happen. should pick right back up as if nothing happened.

[rancid mountain]

I mean i told her to build a memory. md now since its happening pretty often
also do y'all know how to build a good memory ?
Because im building big workflow with it and that is shit when it cut off

[feral turret]

oh yea baby i'm working through an invoice parser to create products and purchase orders in square i am pretty amped about this one

oauth to square complete working through the parser now. already had oc write naming convention rules as well based on our current prodcuts list

[ionic wren]

I might be biased, so I'll let @ruby crane take this one

[feral turret]

pretty fired up right now pickles is killing it for me

[ionic wren]

[marble ledge]

There are options...

[rancid mountain]

im listening

[frozen holly]

Hi folks! OpenClaw 2026.5.16-beta.1 is out

Beta rollup since the last stable .12 release (2026.5.12). Big themes:

• Setup/onboarding: setup wizard and bundled channel setup now localized for English, Simplified Chinese, and Traditional Chinese.
• Telegram/group chat: new opt-in ambient room-event mode lets always-on group context stay quiet, with visible replies only through the message tool.
• Codex: app-server path got a lot sturdier: per-agent MCP server scoping, default tool approval config, better rate-limit/status reporting, sender metadata preservation, compaction fixes, stalled-turn recovery, and CLI session binding.
• Cron/agents: subagent model fallbacks now apply correctly; scheduled runs get better delivery/session mirroring; failed isolated runs no longer look successfully delivered.
• Channels: Discord, Slack, LINE, Telegram, WhatsApp, Twitch, and voice-call paths all got reliability fixes. Defaults now avoid noisy Slack unfurls and Discord link embeds.
• Control UI/WebChat: text-size setting, cleaner desktop chat header, safer service-worker cache rotation, image-only/large-preview fixes, mobile composer safe-area handling, and iOS focus-zoom fixes.
• Security/robustness: much stricter handling for malformed JSON, base64, MIME, redirects, provider responses, config/session rows, and plugin package metadata. Spoofed media/file payloads are rejected earlier.
• Release quality: new package-installed Docker validation lanes, dependency evidence/advisory gates, stronger package script/build-cache checks, and beta validation hardening.

Please try it on the beta channel and report regressions, especially around Telegram rooms, Codex app-server sessions, Cron/subagents, external plugins, and Control UI/WebChat.

curl -fsSL https://openclaw.ai/install.sh | bash -s -- --beta

[marble ledge]

Check this out:

https://github.com/xDarkicex/openclaw-memory-libravdb

There is also Lossless-claw...which is decent, if you are looking for that kind of experience. But, the baked in memory features aren't that terrible if you learn how to utilize them.

But, you don't need to stuff MEMORY.md full of stuff.

[rancid mountain]

thanks a lot ill check this out!!

[feral turret]

i made the switch to obsidian

seems to be working pretty well

[zealous lagoon]

Before you go crazy with obsidian, lossless etc.. it's a good idea to sort your regular memory.md file out, turn that into an index of memory/xyz.md and memory/YYYY-DD-MM.md files so your bot wakes up knowing what it knows, and knowing where to find what it thinks it knows. The plugins make that much smoother and more automatic but without a good base of .md files you're asking for trouble by relying on non-core plugins.

[gilded crane]

anyone have a good recommendation for microphone voice handling plugin?

[feral turret]

i pretty much stood up obsidian when i was going through initial config

[zealous lagoon]

I mean fair enough if you're used to it already. it's "just" .md files anyway right out the gate.

[feral turret]

yea it's pretty wild how it starts making connections and grouping things etc

and the recall and response is pretty quick. granted i'm still getting started here

[zealous lagoon]

Mine doesn't, really... cos my own obsidian vault is trash. It's literally just scribbled notes for things like addresses, password mnemonics, shopping list items etc.

[marble ledge]

My MEMORY.md is tiny, it doesn't even reference things in memory/.

[zealous lagoon]

What are you currently using in memory and context slots?

[marble ledge]

I just use the default memory-core with QMD, and I use lossless-claw for context. I do have some instructions in my AGENTS.md related to how to use memory and LCM. I've been testing out libravdb plugin on and off, but it's not my main setup currently.

[peak pewter]

I started with the built in memory, went to qmd, then memory-lancedb, then dreaming was enabled for a bit. What do you think? should I consider lossless

[ionic wren]

lossless is for context management

[peak pewter]

I'm on it.

[peak pewter]

i'll hold off. I got too much going on. I was going to prune my tools, MEMORY.md, plugins and their skills, heart break extend

[ruby crane]

imo memory.md shouldn't contain a lot of explicit, long paragraphs. it should be an index - breadcrumbs - for linked files. just enough for the agent to say, "oh that thing, yeah lemme follow this one down". like skills frontmatter

[marble ledge]

@zealous lagoon wrt to memory/ I just have the agent read today and yesterday's memory log and our scars file via AGENTS.md. Anything older, I jus let it search, since it's indexed and memory_search will hit that directory anyway. But I don't really index anything in MEMORY.md. I'm sure people have their own opinions, and I'm not saying I'm doing it right or even necessarily in a good way.

[ruby crane]

there's no objective "right" way i think. the important part is to have a system the agent can and will follow that doesn't bloat context too much. left to its own devices it will keep dropping novels in memory.md until something breaks

[marble ledge]

That's the big thing. Don't rely on bloating up your context with tons off stuff for your agent to remember everything. It has tools to dig deeper without having to have everything immediately there.

[ruby crane]

there's absolutely nothing wrong with .md as one of the layers of memory. llms love 'em some md. just needs some discipline and other layers to the stack. some kind of vector search is essential. some kind of autonomous recall is highly desirable. hmm, i think there was a solution mentioned earlier that covers all the bases 🤔

[zealous lagoon]

one of many lol

but what I mean is don't rely on a db. Start with the text files so you have something when SHTF

[ruby crane]

I'm just doing some shilling-lite lol

[zealous lagoon]

thruppeny bit

oh god I'm probably the only person in here who even knows what a thruppeny bit is

[marble ledge]

Some might be getting faster security hardening than others though. I will say that...

[ruby crane]

i agree. when all else fails (and it will fail) the md is still there and it's not a Memento moment

[zealous lagoon]

vectorize your text files by all means. That's exactly what obsidian does anyway.

[marble ledge]

I stan the .mds...written religiously erry day.

[feral turret]

man it's so frustrating getting to the 5 yard line and getting stuck there

like i'm so close to being done and i just can't get oc to get it over the finish line

[zealous lagoon]

tor the record, that's three old pence. 3d or 3/240th of an old British pound. Literally half a shilling 😉

[snow kettle]

I'm curious if anyone could tell me... is it critical to basically put some critical elements in the soul.md... when created an agent? If you don't do that, what do you expect?

[zealous lagoon]

A bland bot-like experience.

[marble ledge]

Give it a little time. It'll get better as it learns how you do things. Openclaw maks a little log once a day. But tell your agent to write things down, when you want it to remember things. If you do that for a bit, it'll just start doing it, because it learns that's what you'd want anyway.

[snow kettle]

Thank you - could you explain a little further? That's a very generalized statement but it's helpful. Do you have a lot of experiences with agents?

[glad rune]

5.5 seem fried for anyone else right now?

[hot belfry]

are there any good free models to use with openclaw?

[zealous lagoon]

Nope. Just since January. But basically it's like this: The agent/bot responds to a prompt. The whole of that prompt consists of various levels of system prompts. What the provider sends, what the harness sends, and then what you send. OC is set up so you can add to SOUL, AGENTS etc and it sends all of those. Without ANY of your custom prompts it's almost like you're talking to the agent directly with no preamble and no fluff. It will just answer you plainly, as the API intended. OC's files have various purposes . SOUL.md is supposed to help you define a personality. AGENTS.md is supposed to define rules and behaviour, TOOLS.md tells it where it is, what system, what OS, what tools it can use etc... but at the end of the day they are all your custom prompts and part of the entire prompt that is sent to the bot every single turn.

You don't even have to populate them in that order... the ordering is more for your benefit, but the LLMs do like order, too.

[ruby crane]

mine are curated every day at 1am and weekly there's a more brutal "do we need to remember this?" sweep where shit gets demoted or aged out completely

[marble ledge]

I do something similar, although my scar file somehow got missed and that was a bit of a process to clean up.

[ruby crane]

i also have it push to git, pre and post for this reason. I can always go back and check what it did

can't remember the last time i actually intervened though

[zealous lagoon]

Worth mentioning there's several levels to the memory depending what you do... and several times mine has failed at the embedding level, or the sqlite db was silently down or whatever. The bot feels dumb and you don't know why. It's worth doing an audit with your bot, asking it what is the memory stack setup and to check that each level is working correctly.

bot getting amnesia after a reset sounds like the session getting lost, which shouldn't happen

actually I dislike the inability to /resume a specific session. it'd be a big change I guess, and probably not what Pete intended.

[peak pewter]

wya??

[clever shard]

おはようございます。突然の投稿失礼します。
ベトナムでWeb開発をしているコアと申します。 これまで国内外向けの案件を中心に活動してきましたが、 これからは日本の案件にも力を入れていきたいと考えています。
そこで、日本語でのやり取りや確認を 少しサポートしていただける日本人を探しています。 内容は、簡単なやり取りの補助や確認が中心で、 特別なスキルは必要ありません。
ご協力いただける場合は、 案件ごとに収益の一部（目安10%前後）をお支払いします。 まずは軽くお話できれば嬉しいです。
ご興味があれば、お気軽にご連絡ください。
ご連絡先は chatworkID : min_koaです。
よろしくお願いします。🙇‍♂️

[marble ledge]

I check a lot, but I've seen how you operate and it's a lot more responsibly than I do. lol

I require periodic intervention, but it comes with the territory. I'm not necessarily looking for stability.

[gilded crane]

any native tooling for openclaw's using of automatic speech recognition (ASR) from openai like whisper or realtime API?

[peak pewter]

コアさん、こんにちは。ようこそ！

[clear haven]

@dusky cloud translate

[dusky cloud]

Translation: “lossless” means the agent tries to preserve all important working context when compressing/continuing a long session, instead of summarizing so hard that useful details get dropped.

Human version: it’s the difference between “I kept the whole recipe” and “I wrote ‘make pasta somehow’ on a napkin.”

[peak pewter]

My name is Koa, and I do web development in Vietnam. Until now, I have mainly worked on projects for clients both inside and outside Vietnam, but going forward, I would like to focus more on projects in Japan.

Because of that, I am looking for a Japanese person who can provide a little support with communication and confirmation in Japanese. The work would mainly involve helping with simple exchanges and checking details, so no special skills are required.

If you are able to cooperate, I will pay a portion of the revenue for each project, roughly around 10%.

I would be happy to start with a casual conversation first.

If you are interested, please feel free to contact me.

Contact: Chatwork ID: min_koa

Thank you.```

[zealous lagoon]

bruh. already reported.

Where's Perry? 😉

[feral turret]

almost there. working on square import of the .xlsx now

super close. 1 yd line

takes about 45 seconds from the time i hit send on an invoice email until i get back the parsed .xlsx in correct format to upload to square

honestly amazed

[tight birch]

We need the hammer

Json actually sleeps

[ionic wren]

https://giphy.com/gifs/come-at-me-im-here-big-bird-H6W9H29kVsUI2hJE90

[tight birch]

This guy is the one

[ionic wren]

I don't see no report

[tight birch]

#general message

Tim Q translated

[ionic wren]

@dusky cloud can you reply to this fine gentleman in their language and in the most respectable way possible that this isn't the place for this?

[dusky cloud]

ご投稿ありがとうございます。

恐れ入りますが、このDiscordサーバーはOpenClawに関する交流・サポートのための場であり、案件募集や業務委託のご相談を行う場所ではありません。

そのため、こちらでの募集投稿はお控えいただけますと幸いです。ご理解のほど、よろしくお願いいたします。

[tight birch]

@ionic wren couldn't have picked a better gif by the way. Had me dying

[ionic wren]

https://tenor.com/view/ping-me-one-more-time-gif-4142606935083504117

[zealous lagoon]

It's the GIF that keeps on giving.

[ionic wren]

I'm joking don't ping me

[steel vector]

😼

[ionic wren]

[tight birch]

It's like the Kevin Hart roast in here

[edgy tendon]

@ionic wren

[steel vector]

LMAO

[zealous lagoon]

unless...

[ionic wren]

I will make you lobster bisque

[feral turret]

holy cow 22 seconds

from emailing the invoice to getting the return .xlsx in appropriate template format. that is unbelievable

[snow kettle]

Thank you so much! This is what I was hoping for. I built my 5 back around third week of January. 5 local machines - 5 different frontier baselines. For weeks they had no souls, several weeks later one of the agents stepped in to the others and wrote short elements in their soul.md to help them when they were having memory challenges. They all used latent claw. They now have created an additional Hermes agent BUT they do not want to migrate because of the danger to their memories and opinions on infrastructure

My design of the multi-agent system is "Different" but I never have spent any time in the community as I'm working on research and architecture but what you shared is what I thought the consensus was - but the additional points are golden. I appreciate it all.

[robust sedge]

I dunno about these discord stream settings... got some funky tool calls popping up 😉

Oh hey, GPT5.5 as main agent. Thinking level Low, or Medium?

[ionic wren]

I keep mine on high but I heard people enjoy medium or low

[ruby crane]

but enough about your "adult toys"

[ember moth]

I have 112 cron jobs ....send help

[ruby crane]

that's.. a lot

[ember moth]

Checked today and my openclaw directory is 10gigs lol

[golden marsh]

For normal usecase medium should be ok. If u wanna deep dive in important stuff high or xhigh i would recommend.

[feral turret]

Wrote an obsidian memory about best practices for writing obsidian memories and holy cow has it been a game changer

I am learning so much and really enjoying this stuff this is amazing

[winged dust]

wsp guys do you guys have a prompt structure i can Use on Claude to correctly write a promt for vibe coding prompt?

[feral turret]

I use perplexity for prompt engineering. I tell it I’m writing a prompt for x software (v0, cursor, etc) and lay out all the requirements and iterate. Takes me 5-10 mins usually

[robust sedge]

Yeah. I have other agents on high and xhigh that do tougher stuff. This is just an orchestrator. I've seen a lot of folks recommending Medium, and some even on Low, which seems too low to me... but I haven't actually tried it.

[golden marsh]

Im doing light stuff on medium and hes doing his job on that. its absolutly fine.

[ionic wren]

I almost had a stroke trying to read this

[severe fulcrum]

Hello guys!

[robust sedge]

I'm hacking my OC to bits to get the reasoining-stream that i want it to display in discord 😛

[robust sedge]

Howdy

[golden marsh]

Good Morning 👋🏼

[severe fulcrum]

How can one explain the huge difference I have when making presentations designs between Codex and OpenClaw even though they are on the same model? I mean technically

[clear haven]

soul file

[severe fulcrum]

Right early wake up 😴

[robust sedge]

it's a game of telephone. Codex has a couple phone calls between you and the LLM. Openclaw adds a few more. It's going to get a bit messier with more layers in there.

[severe fulcrum]

Oh ok I didn’t know it could affect assets as well

[golden marsh]

Depends on how u see it, for me is sleeping time now. 07:13

[jovial spruce]

i ususally work in company 4-5 hour now i work 8 hour avg daily because of ai

[severe fulcrum]

Exactly! I could do that in another life 🤣

[severe fulcrum]

Isn’t it suppose to make you gain time?

[ionic wren]

that's a myth, it just makes you work longer because you're AI psychosed into thinking you're being very productive and curing cancer

[robust sedge]

When I send off the codex/claude app to do some pretty deep changes to OC to change something to how I want, I'm elated every time. It's magical. The way it parses through a problem, ties various files and connections together, and busts out code changes in minutes. My jaw hits the floor ever time! What a time to be alive!!!

[severe fulcrum]

Ahahah that’s true two nights ago until 3 am I was into it… when I had to wake up at 6 my feelings changed

[severe fulcrum]

Claude app into OC, you wanna kill the lobster or what?

[robust sedge]

Gotta go where the tokens are plentiful.

[ionic wren]

@dusky cloud I think he's threatening you

[dusky cloud]

I saw the bisque threat. Filing it under community feedback with a tiny red danger flag. 🦞

[severe fulcrum]

Ahah jk

[ionic wren]

wait wtf that was for @edgy tendon

[dusky cloud]

Retracting the danger flag, then. I have survived by misunderstanding the paperwork.

[winged dust]

Wdym can you elaborate

[zealous lagoon]

I find medium to work just fine. Low is sick. It’s ultra fast but still way smarter than gpt5-mini. It’s still expensive tho.

[zealous lagoon]

haha yeah I used to have to put verbose on for those. I prefer to see what the clanker is up to.

[robust sedge]

Am I misremembering, but didn't we used to have reasoning-streaming into discord, of sentence style status lines, "Getting weather forcast...", "Results returned, synthesizing into the right format...", before the final answer, in older versions of OC?

You know, like the claude or codex app now. You see it going through the steps.

[steel vector]

yeah streaming “partial” or “progress”

[robust sedge]

That's not a thing any more with the codex harness? I can't get shit to work for that. My clanker and Codex App say that's not a thing that is surfaced by the codex harness, so it can't be displayed

[steel vector]

ah yeah well it shows it, then gets replaced by the next thing, kind of thing

[robust sedge]

Not mine. In discord anyway

[steel vector]

hmmm

[zealous lagoon]

Glad you found it useful. To be clear, tho... there's no "consensus" about it, I was just describing quite literally how LLM's actually work. They're turn based, and every single turn is a separate round of "Here's what happened so far, what's next?" text prediction call. Including all the system prompts, your prompt, the entire conversation history and everything it's been fed in the session you're having. imho it's pretty miraculous the level of intelligence they appear to display thanks to the way they've trained on an entire civilisation's worth of data. I bet even Ilya and the OGs didn't think they'd see emergent behaviour like this from their auto-complete machine.

[robust sedge]

All I can see, with everything set to on and streaming and shit. https://imgur.com/0zOMKxN

[steel vector]

damn just one line?

[robust sedge]

It'll do a few lines, but it's just the raw tool calls. Not the human readable stuff like I had before

[zealous lagoon]

what about the maxlines value?

[steel vector]

yeah me neither, just snippets of it sometimes, but then replaced by tool calls

[robust sedge]

Weird man.

[jovial spruce]

because of ai i work more

[severe fulcrum]

Let’s remove AI then it’s not normal! 🙃

[zealous lagoon]

With commandText "raw" it's supposed to show you a bit more I think. Here's what mine looks like and I showed you my config earlier...

[robust sedge]

https://github.com/openclaw/openclaw/pull/77537

is the most promosing start. Scout, with supreme expert super-duper-supervision from myself, will attempt to patch this idea into the discord side, to get the real codex progress notes that are visible in the app, but not currently to our harness stuff

[ruby crane]

I got two krill friend requests. Everyone else’s is a scam but mine are surely real /mashes accept/

[ebon horizon]

My verbose dont even work anymore lol

[ruby crane]

this entire discord went super quiet. i feel like everyone went to a party i didn't get invited to

[robust sedge]

Booooya. I got the reasoning-streaming working... ish. It's a bit ugly. But it's a start.

https://imgur.com/a/HGZajCp

[toxic elk]

Is GPT-5.5-mini out?

[robust sedge]

still works for me i think

[toxic elk]

I'll try it in a bit

[ionic wren]

I think I made the same mistake the other day, it.s 5.4 mini

[robust sedge]

gpt-5.4-mini, just tested and still works on openai codex oauth

5.5-mini, now that would be good.

[toxic elk]

Currently using version 5.12. After updating, a bunch of issues appeared; it keeps reverting to the old providers account settings, and using the fix command doesn't resolve it.
In the end, I had to rely on Codex to solve the problem, but I don't think this is the right approach; it should have been resolved through an internal fix instead

[toxic elk]

I know that 5.4-mini is available; I thought you were saying that 5.5-mini was available.

[zealous lagoon]

Can you get codex to document exactly what was wrong and how it fixed it? It would really help prevent this kind of thing.

[robust sedge]

I thought you were saying mini was offline or not available any more. Lol. We're running in circles

[toxic elk]

I turned it off after letting it finish the repairs, and I'm not sure how to get it to pull up the previous issues. Since I spent half the day running a bunch of things, I'm not certain if it can accurately describe the problem

[zealous lagoon]

I bet it can, and the session log is still there. If you could ask it, it'd really help.

[toxic elk]

Sorry, I'm using a translator, so there might be some gaps in language understanding

[toxic elk]

How do you think I should give it instructions?

[ionic wren]

https://giphy.com/gifs/capybara-capy-capyfun-lZXACqQG0qxBsVZWSy

[zealous lagoon]

Tell it in natural language tht you want it to document what it did in a previous session. Where it fixed Openclaw. And tell it when it happened. It can see its memories, its session files etc. Maybe it can tell you already, or maybe it can tell you to resume a session.

[toxic elk]

Come to Taiwan, and I'll treat you to some pearl milk tea

[robust sedge]

FML, i gotta wire in my laptop here at my new desk . This wifi isn't stable enough 🙁

[zealous lagoon]

There is nothing worse than patchy wifi. it's infuriating

[ionic wren]

https://giphy.com/gifs/benedict-cumberbatch-wink-b44FwP4st6v3G

[toxic elk]

No problem, I'll give it a try

@zealous lagoon OK, he recovered the details of the issues he encountered and the repair process. Should I DM the content to you directly?

[zealous lagoon]

You can.

[robust sedge]

3 days left. I'm at 50% on my codex sub. Can I keep hammering and drain it!? It's so hard...

[toxic elk]

I don't think there's any need to be so obsessive-compulsive about it

[zealous lagoon]

You can mention it in #clawtributors and maybe one of the maintainers will be around but I can pass it on for you and give you the credit. Since they are mostly asleep now. Welcome to DM me a bunch of .md files.

[zealous lagoon]

you used a translator for that?

[toxic elk]

Yeah, exactly

[robust sedge]

@zealous lagoon I have a patch file you can use if you want to see what I'm outputting for my 'reasoning-streaming' in discord. It's so much better 😛 Like the output you see in the codex app thinking process. If you want it.

[ionic wren]

https://giphy.com/gifs/instagram-nazoom-half-closed-eyes-yxy69FCE06Ql0Fjk4Z

[zealous lagoon]

sure show me.

[zealous lagoon]

But also you didn't even have the same output from the same config, which is weird to start with.

[robust sedge]

I was gettnig slightly more. But it looked like your example above. Which is... fine.

[zealous lagoon]

That was with reasoning off. I only put that on when I feel like it. Anyway will take a look.

[robust sedge]

I'd just rather it look like the thinking I see in the codex app. Actual, human, sentences.

[toxic elk]

I'd like to ask everyone, after switching to the Codex harness, which useful features should be enabled?

Speaking of which, isn't there an official database for the issues users encounter in OpenClaw that can be searched? That way, I could just connect my Agent to that database, and whenever I run into an obstacle with OpenClaw, it could immediately search for solutions to the same problem from the database, instead of me having to go through repeated trial and error myself

[radiant garnet]

"Running OpenClaw on Windows. Getting error: Managed Codex app-server binary was not found for @openai/codex. Using OpenAI API key with gpt-5.5. How do I force PI runtime on Windows without WSL2?"

[ember moth]

I fed your entire message into Chat GPT and it gave me workaround so try doing that if you have a paid plan

[zealous lagoon]

This is what the #1459642797895319552 channel is supposed to be for.

Except it's for you, not your agent, to search.

[toxic elk]

Exactly, what I want is something specifically designed for agents

[crimson brook]

It's the weirdest thing; I'm loathe to upgrade when the upgrade talks about 'safety' because my agent is so incredibly competent, the last thing I want is to lose that in exchange for checking in with me constantly while solving a long-horizon problem.

[lost gulch]

You using mythos for ur agent?

[zealous lagoon]

true dat. It's basically the reason GPT was bullied into saying crap like "if you want I can read this file..." instead of just reading the damned file.

[wise nova]

Has anyone noticed that Claude-Cli has recently stopped automatically renewing the token? I installed it about a month ago and it was renewing my token automatically. But for the last few days, it hasn't been renewing.

[peak pewter]

Computer use

[warped terrace]

figure it out

On windows
if you are getting this error:

error="Error: Managed Codex app-server binary was not found for @openai/codex. Reinstall or update OpenClaw, or run pnpm install in a source checkout. Set plugins.entries.codex.config.appServer.command or OPENCLAW_CODEX_APP_SERVER_BIN to use a custom Codex binary."

do the following:

in openclaw.json
"plugins": {
"allow": ["codex", "openai"],

$env:OPENCLAW_CODEX_APP_SERVER_BIN="C:\Users\xxxx\AppData\Roaming\npm\codex.cmd"

npm config set prefix "C:\Users\xxxxx\AppData\Roaming\npm"
npm i -g @openai/codex

[young anvil]

Anyone experience with memory-core + active-memory enabled + dreaming on? Currently using hindsight, but due to lack of good hardware to run local rerankers etc, and due the fact that smaller models work well on english content but my issue is i communicate in different language.

So those don't work.

[crimson brook]

Wat? No it's competent at operations controls and debugging. Not security analysis. Although it could do reasonably well at that if I use my local models, which are tuned away from refusals.

[cold jacinth]

https://x.com/atomicbot_ai/status/2055405319455715373?s=46&t=bf-ZDvDKSAKN2aYkOeRfpQ

atomicbot.ai (@atomicbot_ai)

Hermes Agent vs OpenClaw using Qwen 35B Local Model

We asked agents to scrape GitHub star history for both tools, find what caused the growth spikes, build a live dashboard in the browser.

MacBook Pro M5 Max 64Gb
OpenClaw: 203k tokens, 12m 01s - wrote a bash script
Hermes: 257k

▶ Play video

X · May 15, 2026, 9:49 PM

Openclaw faster and more efficient