general

[cinder sphinx]

Historically, Google operated under a model where each family member received their own independent set of rate limits and quotas. This meant that if a plan manager used up their 5-hour quota, it had no impact on the brother's ability to continue coding on his own account. However, as of April 2026, evidence from community forums and support interactions indicates a transition to a "communal pool" for Antigravity resources.

[craggy bane]

why dont you create a .md file tricking google so we can use free LLMs?? lmao

[tepid kestrel]

go here https://g.co/YourFamily > open a family > add the other accounts > scroll down to the family options and activate pro subscription sharing

[cinder sphinx]

geezus, i know how to add family, already did 😄

[thorny spruce]

Thanks guys

[cinder sphinx]

i just dont need them to use antigravity and my credits for coding, since it's in a pool

[woven magnet]

🦞🦞

[craggy bane]

i have 4 accounts and i shared my Pro subscription , what now?

[tepid kestrel]

you can now use gemini,claude etc for whatever you are going to do

[cinder sphinx]

it doesnt mean you get 4 x usage now

[tepid kestrel]

everyone has their own qouta

[cinder sphinx]

the main quota splits in 4

[craggy bane]

tell me how to apply in openclaw for RPD and no rate limits

[tepid kestrel]

thats not allowed.. they will ban you for that

but everyone has their own qouta now

[craggy bane]

that means create API token from all accounts and switch it everytime

[tepid kestrel]

thats correct

its the 0auth token to be precise

[near owl]

@jaunty thunder I need to know more. So it basically reads the github repo and makes a detailed summary of the repo for coding agents before it starts working on it?

[thorny spruce]

It's not very hard to get free tokens if you're willing to make your life a little difficult in terms of setup

[craggy bane]

my PR is merged , first step in contribution 🦞

[thorny spruce]

Ayu was this a human fix or an AI fix?

[craggy bane]

on what? give me reference

[thorny spruce]

PR

[craggy bane]

check it .. https://github.com/openclaw/openclaw/pull/76643

[woven magnet]

I have a question why everyone uses this emoji 🦞 It is not necessary that this should be set as your agent's emoji

[jaunty thunder]

So basically the problem with LLMs is that they don't use actual metadata but do statistical guessing on the pattern seen on a subset of files.
What I've seen is that models nowadays are really good with these coding harnesses but they lack context.
That context is easily retrievable from existing codebases like your own or another one you like and where they fetch that information is mainly from the AGENTS file.
How to give actual deterministic data? Static analysis. Agentskill does that for you in seconds or even milliseconds.
Tracks code style, commit style, symbols, etc. and generates an opinionated AGENTS file which covers everything ranging from tooling, to red lines, error handling and so on.

[thorny spruce]

Nope I don't think I can do review based on the scope of these new priv settings

[red cradle]

Oh shi we’re back

I thought we had a Huntarr-type scenario

[thorny spruce]

I have had no issues with open claw simce about 3hrs ago. Pretty stable

[near owl]

That's sick! I will 100% use it myself If I actually code with openclaw

what do people prefer coding with? Openclaw or windsurf/cursor?

[thorny spruce]

Humans in general

[jaunty thunder]

Thanks! I really appreciate it. Soon I'll release also a shrunk up AGENTS file autogeneration so we can also reduce context usage for every request.
That is the perfect fit for agent assistants like OpenClaw or any other.

[craggy bane]

to review that you need to run entire setup in docker

[thorny spruce]

Ayu what would this change look like if you were to use podman instead?

[near owl]

@jaunty thunder Should I make a CLI tool that checks for malicious coding and rates it, using AI, similar to the LLM benchmark evaluator?

Because when a new GitHub repo updates, our first instinct is just to install it and use it. My tool checks that repo to see if the new code is malicious, or the whole project infact.

[thorny spruce]

Specifically, I would tend to stay very very far away from adding and removing users in a system, so complexes this

If you need to make things more secure, then reduce the privileges that people have and don't increase the blast radius

[thorny spruce]

This is a hard problem. The only solution is to build things from Source but then you're building from source and you don't have time to read the source. So you have to really put the trust into the authors of the program and we all know that you can't really do that these days

[tepid kestrel]

i wanna have a fight with my agent against your agent

[craggy bane]

dont know what is that

[thorny spruce]

Podman fix docker's main issue which is being able to run in a rootless environment

[woven magnet]

I have given autonomy to my urgent, not that I will always control it.

[thorny spruce]

Ayu dm

[jaunty thunder]

That is a good idea but also an hard one, there are tools out there like VirusTotal which is used also by ClawHub to track this.
You could integrate that and see how it fits in a CLI augumented with unique stuff you could offer with it.

[near owl]

virustotal sucks right? Counts everything as a virus.

[jaunty thunder]

Yesterday Vercel released a security agent too, you could check it out too and see also how that works and if you can integrate it.

[jaunty thunder]

It's one of the leading projects for CTI right now. It's not perfect but not shitty either.

[near owl]

I really want to make a website as a tool like my LLM benchmark evaluator. So what tool should I make? Not my LLM benchmark evaluator, since already made..

[jaunty thunder]

Usually I let ideas flow before building. If I find something that can be useful for me and eventually for others I create it.
Don't pursue building for just building. Do it for the sake of solving at least your own problem.

[near owl]

I get very picky, It's the project I have shipped in a long time..

Picky is a personal trait that is such a curse, fueled by the fear of "wasting time" building something I don't find useful.

It's like finding a perfect idea or a million-dollar idea.

I need to just keep shipping projects to understand actual humans that under myself.

[jaunty thunder]

That's the neat part: ideas that you think will make gazillions will probably do not do them and viceversa.
That's why I usually fill the need of building behind something else which is learning, fun, create, problem solve, etc.

Eventually it will get to other people and have success in some sort you see success in.

[near owl]

Solving problems is what I find hard, I don't know what problems to solve that isn't yet solved.

[jaunty thunder]

It's like a never ending problem if you ask me. A problem solved gives N more problems.
Look at OpenClaw for example, it created the need for Clawsweeper, just to cite one.
And Clawsweeper will or has already created the need for better infrastructure that can handle that load.
It is a vicious cycle and to understand it, you have first to use something and see what pain points it brings.
If those pain points are really there, you'll build something useful...otherwise it was just a learning or funny ride.

[zealous lagoon]

First and foremost... look at problems you have for yourself. Fix them, and think if anyone else has the same problem. Then continue to look and listen to what problems other people are having. See if it's something you can fix. If it's something you can fix with software you've suddenly fixed it for as many people as can find your software.

[jaunty thunder]

This is the way.

[zealous lagoon]

Not everyone uses this emoji for their openclaw. I don't.

[pulsar iron]

tested 5.4, still no discord, it sees it, it responds, but never publishes it to discord. 4.23 continues to work fine

[craggy bane]

[near owl]

I don't know my problems, that's the problem.

[supple ibex]

Thanks, so I will not update to 5.4 then

[jaunty thunder]

So develop something that solves the problem to find problems.

[zealous lagoon]

This is still odd. So talk me through the cycle... you talk in chat. Do you see emoji reaccs? Do you see typing indicator? Does the bot respond by doing anything at all? Or does a message land in webchat?

[pulsar iron]

sxame problem, it sees me typing, it says it is typing, it responds with emoji, it makes a thread, but never responds then the emoji gets cleared

I can see my ai server fire up and I know it is using my llm

[zealous lagoon]

https://tenor.com/view/yo-dawg-xzibit-laughing-gif-13033572

[pulsar iron]

if i use control UI and ask it to send a message to discord, it can do it,

[drowsy flint]

maybe hitting token limit for discord calls

[pulsar iron]

If i roll back to 4.23, it works flawless

[zealous lagoon]

There are some related settings that might need tweaking. It should just work... hang on a sec.

[pulsar iron]

I've run doctor fix deep every version since 4.23 and nothing works

[supple yarrow]

Nah I had this phenomenon also happen yesterday, since discord moved to builtin there were competing plugin entries and a borked path in 5.2 I think, but with 5.3, clean discord install via openclaw configure and some time for the gateway to settle, it was gone after a few

you'll see all messages getting routed to your webchat, so nothing is lost and if you instruct the bot specifically to relay his last message once again to discord, that also works as a workaround for the time

[worthy anvil]

Try setting: messages.groupChat.visibleReplies = "automatic"

[zealous lagoon]

There is a setting that hides tool reply in group chat.

[zealous lagoon]

THIS!

either "automatic" or if that doesn't work, "all"

[pulsar iron]

Ok, now it is responding.

now i can test if everything else is working, thanks. I've tested every version since 4.23, and nothing has been stable, discord was only a problem in the 5.x versions, but prior to that was gateway crashes and long response times (minutes instead of 4 seconds)

[worthy anvil]

This should have been the default… the current change breaks so much

[zealous lagoon]

Yes, this should be much better now.

[slow thunder]

Wo xiang yao yong Automa shezhi WeChat zidong huifu. Dang kehu gei wo fa xiaoxi de shihou, Automa hui yong AI API lai zidong dafu. Wo de API feichang duo!

[pulsar iron]

Any idea why I get this pretty much every day? I think it is related to cron jobs

│ - Archived 1471 orphan transcript files in ~/.openclaw/agents/donna/sessions as .deleted │
│ timestamped backups.

Everytime I run doctor I have to say yes

[zealous lagoon]

I agree, I might have fixed this earlier on and forgot about it, but I did have to fix it myself at some point.

[worthy anvil]

Heartbeat isolated sessions

[pulsar iron]

I have no heartbeats

[zealous lagoon]

When do you get that message? It's not really a problem to have .deleted sessions.

[pulsar iron]

ANy time I run doctor, even if it is like an hour later, I'll have hundreds of thousands. I don't use heartbeats, but i do use cron a lot

[worthy anvil]

Crons do that as well

[pulsar iron]

I ran doctor 3 minutes ago

◆ Archive 7 orphan transcript files in ~/.openclaw/agents/donna/sessions? This only renames them to *.deleted.<timestamp>.
│ ○ Yes / ● No
└
I have 7 already

I haven't even talked to it

[worthy anvil]

If a session is not reset, it just sits there abandoned. Heartbeats and crons never reset sessions. Not an issue, but i wish it got fixed at some point

[pulsar iron]

I literally ran doctor twice in 3 minutes, didn't do anything with openclaw and I got 7 sessions, I come back tonight and i'll have 1000

[zealous lagoon]

You don't need to run doctor --fix if there's no problem. OC likes to archive subagent and cron session files but it doesn't really matter whether you do or not.

The problem I'm seeing is why you have so many session files. Looks like you have a rogue cron job or something. You should ask your clanker about this, unless you want to examine the session logs yourself.

The info about how each file was created, what the bot did, and why it finished or was orphaned is actually contained in the session file.

[pulsar iron]

I use a lot of isolated cron sessions, most run once a day or once a week, but some run hourly, and one runs every 1 minute (healthcheck)

[zealous lagoon]

every minute?? u sure that's not just the internal tick? You shouldn't be running any cron job once a minute. Not even heartbeat

ask your clanker what the files are. Or just open one of them in a good text editor and examine it yourself.

[pulsar iron]

I opened one of the deleted, it is a dream request

Write a dream diary entry from these memory fragments

[near owl]

clanker 🤣

Does openclaw self-improve?

[pulsar iron]

no, not like hermes you have to tell it to do things, unless you schedule something to self improve

another looks like a healthcheck ping

[near owl]

I find OpenClaw better than Hermes, because of the Pi Agent.

[pulsar iron]

Pi is gone, it's codex now I think

I do believe they are all isolated cron sessions

[near owl]

Pi is the main man. Pi is the best in town, and Pi will be.

[slow thunder]

Setupto use offline model what is best

[near owl]

Without Pi, openclaw is DEAD and thats a fact.

[slow thunder]

i have 2 dgx spark but i try many local models they not good

[pulsar iron]

I think pi got removed, as Pi dev went commerical

[near owl]

I use 4.23 version of openclaw

[zealous lagoon]

Pi is not gone. codex is getting better but it's only for openai models. I haven't switched yet.

[pulsar iron]

Doesn't 5.x remove pi as the harness?

[zealous lagoon]

no

[pulsar iron]

My bad then, I thought i saw it got removed, I only just got to run 5 since the discord problems.

I have three agents. I removed/renamed main/default, but main keeps a folder created and doctor always complains there is one agent not in the agents.list any way to stop this?

[supple yarrow]

Do you have dreaming and stuff enabled? And I guess with so many crons some may point to main, usually deleting main is a bad idea imho, he should be your orchestrator/main. Though what you are describing is most certainly some plugin pointing to main, which can be many

[pulsar iron]

Yes, I just renamed main to Donna which is my main agent (Donna from Suits personality)

[zealous lagoon]

I "solved" it by rearranging so my "main" is named but his session folder is still agents/main

[supple yarrow]

yeah I also had the urge to rename, Openclaw really doesn't like you having another main interaction next to main. I run similar but have main basically as a generalist/fixer that runs openclaw log issues etc and crons that can run isolated and don't need context, but have the agents with their workspaces and specific roles next to it

[pulsar iron]

I just want to have all my agents to have personality and not a generic name.

[zealous lagoon]

Did you give your bots this? Good to give your claw as well as any external SDIs you use to manage it:

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

They're much more effective at setting your config properly once they've read the appropriate docs.

[zealous lagoon]

Yeah but there are parts in the code of OC that expect "main", no matter what you've called it. It causes warnings otherwise. But don't worry too much about warnings. They're not errors .

[pulsar iron]

No, I assumed it had access to something as it generally knows what it is doing, not 100% of the time, but seems to. I always wondered how he knew, it seems to pull docs from npm package from what i have seen once and a while

I was just trying to clear out the warnings, I get sick of seeing it

[supple yarrow]

yeah but especially when you use smaller models, they will confidently screw up your config without hesitation

[zealous lagoon]

npm no longer ships the documentation. I recommend you give it the skill and help it understand the structure. Make sure you're using a smart model to admin it. Do you have codex externally?

[pulsar iron]

I use minimax locally and gpt 5.5

I use droid for dev, I don't let oc do much dev, I usually build stuff then slide it in place. I do have it review stuff though

[zealous lagoon]

gpt5.5 is excellent for administering OC.

especially if you can do it from codex, as it can work with gateway down

[pulsar iron]

I added the skill, I'll set up a harness in a bit

[tidal abyss]

How is the stability with 5.3 - anyone know if it’s worth updating to it from 4.23?

[pulsar iron]

5.4 is out

So far it is working, once i got past the discord problem (fix is: messages.groupChat.visibleReplies = "automatic")

[zealous dawn]

what is better openclaw or open interpreter?

[near owl]

openclaw, open interpreter is so mid

[zealous dawn]

but open claw works whit ollama models?

i try but i cant run any ollama models

[frail crystal]

Don’t update

I’m thinking of migrate to hermes ….

Too many issues…

[zealous lagoon]

what problems do you have with 5.4?

[frail crystal]

Telegram bots stop answering in groups ! WhatsApp group too

[zealous lagoon]

fix is: messages.groupChat.visibleReplies = "automatic"

[violet mulch]

5.4 solid or nah?

[frail crystal]

Yeah but how come you guys make an update and then we have to search for hours for a fix …

[frail crystal]

I will try…

[pulsar iron]

So far it seems to be working for me, and I had problems with every update the secret is that config change: #general message

[violet mulch]

i'm good with 4.23

[zealous lagoon]

I'm just a user, but I started trying to help with fixes because they broke a lot of stuff. 5.4 is a good version. messages.groupChat.visibleReplies default change is an issue that will bug a lot of users I think. I'm looking into that now.

[crisp nimbus]

Is there any state of the art “example” openclaw plugin that I can point my agent at so it knows the best practices for implementing plugins? I have some old plugins that all need updating now…

[zealous lagoon]

But you can fix it manually

[frail crystal]

I had this issue since 5.3

[craggy bane]

no need to search , ask here

[pulsar iron]

I wish stuff like this would be guided when upgrading so many fixes I had to figure out and implement myself

[toxic elk]

Many people also said version 5.2 was a good one, but once I used it, it was broken 😂

[zealous lagoon]

ask here = search

[pulsar iron]

To be fair, I rarely get a response here.

[zealous lagoon]

a lot of fixes since 5.2. 5.2 didn't even work for me.

[crisp nimbus]

Right? Now no one is responding to my earlier question lol

[zealous lagoon]

I saw you in here yesterday sorry I didn't get a chance to reply. it was a bit hectic.

[zealous lagoon]

ppl will answer quicker if there's someone in here who can answer quickly lol. Depending on timezones sometimes this chat is mostly empty.

[frail crystal]

Thanks

Thanks guys

[crisp nimbus]

So you don’t happen to know any state of the art example plugins I can look at to learn how a plugin should be built?

[violet mulch]

has anyone gotten their openclaw to make money on its own?

[crisp nimbus]

Right now I’m struggling with how to make it so my plugin can distinguish between which agent is calling it, so that it can put files in the correct agent workspace

[zealous lagoon]

plugins, no. sorry! Keep asking though there may be plugin experts in here later when Europe and the US wake up.

.. wake up and become active in here...

[ruby crane]

there are examples in the docs and sdk https://docs.openclaw.ai/plugins/sdk-overview#plugin-sdk-overview

[crisp nimbus]

I’ll check it.

[ruby crane]

you can also check any of the openclaw plugins on npm and/or clawhub. find one thats close to what you want to achieve and download it

[steel vector]

wee woo waa wuu - computment

[shy flame]

Is it finally worth to upgrade to 5.4. I stopped upgrading since .23 as it was super slow on my vps. Can anyone confirm

[zealous lagoon]

5.4 should be ok.
You might have to add config to set messages.groupChat.visibleReplies to automatic or all if you use group chats (inc. Discord channels)

[shy flame]

Whats the best way to upgrade?

[zealous lagoon]

This time, openclaw update should just work.

[civic yacht]

I noticed memory search getting stuck on 5.4.

[ruby crane]

stuck where? what are you seeing?

[civic yacht]

basically the same issue : https://github.com/openclaw/openclaw/issues/77781

[ruby crane]

let me do a quick check

[slender wigeon]

5.4 logs show missing zod for feishu

And vertex ai all lost

[ruby crane]

fine here. what happens if you run "openclaw memory search test"?

that was to @civic yacht

[ebon horizon]

Anyone building an app?

[zealous lagoon]

https://tenor.com/view/glootie-app-software-do-you-want-to-develop-an-app-gif-17182284

[cinder sphinx]

8 kotlin apps in the oven here.
working on stoners companion app now, nearly done.
https://i.ibb.co/NgxDF537/image.png

[ruby crane]

https://klipy.com/gifs/friday-yes-1--k01KQVXPMTYF283295PA7Q01WNY

why kotlin? doesn't seem a super popular choice?

[ebon horizon]

Bruh

[cinder sphinx]

i want native android language which kotlin is.
my first app was react-native, it was compiling for like 9 minutes, sinc ei migrated to kotlin, didnt see app compiling longer than 2 minutes and that is rare. also with kotlin, most android ecosystem is 1st party modules/plugins/dependencies

[ebon horizon]

Why android over ios?

When ios market is 10x larger?

[cinder sphinx]

i dont like apples, didnt own any of apple products ever

[drowsy flint]

just made a kotlin streaming app for firestick, looks almost exactly like netflix

[ebon horizon]

If your end goal is to monetize. ios makes sense

If your doing it for fun then sure, android

[ruby crane]

sweet! what's the backend? jellyfin or smth?

[cinder sphinx]

same here, created iptv player with jellyfin support, vpn and more

[cinder sphinx]

you can make good profit on android too, not only ios. its all about apps and how you monetise them

[ebon horizon]

Wonder what the ratio of pot smokers is ios:android lol

[cinder sphinx]

heres revenue on most popular apps in april on play store
https://substackcdn.com/image/fetch/$s_!4RNb!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F7aca9cb8-f678-4600-b1e2-654ada5a5c77_2192x1552.png

i dont want to be rich, i will be happy if each app makes a £100 a month on those little ads

[drowsy flint]

no backend, made a couple versions, Kotlin and then Vite/React in a WebView hitting TMDB + Torrentio + Real-Debrid directly. RD does the heavy lifting, so there's nothing to self-host. VPN would be a nice touch

[ruby crane]

you know i think that's one thing that crosses all age/class/social/financial boundaries (though some may be more inclined to keep it on the dl)

[cinder sphinx]

its easy to add wireguard vpn to your app

[ebon horizon]

Im building a menu scanning app atm on expo go w/ react native

Informs people on nutrition BEFORE they get their food

[cinder sphinx]

my apps are all over the place. productivity, gaming, time-wasters

[craggy bane]

its best tbh

[drowsy flint]

same lol

[cinder sphinx]

for android apps for sure

[craggy bane]

i myself is Android Dev too

[cinder sphinx]

if you plan multi-platform then sure, flutter, react-native blah blah but i only do android, so kotlin was easy choice

[ruby crane]

today i learned something new

[craggy bane]

[cinder sphinx]

google provides a lot of "skills" to make kotlin coding easier

even gemini 3 flash does very good job at coding kotlin apps, when i run out of big models quotas in antigravity

[craggy bane]

that was a secret

only android people knows

[cinder sphinx]

🙂

[steel vector]

was

[cinder sphinx]

i really noticed, its so much quicker to vibecode kotlin app compared to react-native, so much "easier" to work with ai on it

[cobalt steppe]

is discord channel working after the last update? mine says "discord plugin not available." on openclaw config

[cinder sphinx]

that stoner companion app, i started working on it yesterday, its pretty much done and has some useful things going, with on-device AI, no logs, full privacy and on-device database of nearly 6k cannabis strains

[steel vector]

is it useful

[cinder sphinx]

for stoners for sure 🙂

[steel vector]

why?

[cinder sphinx]

so you dont forget, which strain you liked the most, so you can track your spending, so you can give a screenshot of dispensary menu and ai will give you a list of best strains for you and so on

[steel vector]

when you smoke so much you need an ai to not forget what you like about weed

im just messing with u

[zealous lagoon]

so you don't forget

[cinder sphinx]

yup, trust me, i know after over 30 years of daily blaze

[woven hill]

Guys how do I use anthropic through O auth rather than API?

[steel vector]

doesnt sound too good

[cinder sphinx]

it is life

im not saying the app has to be popular, it will be nice if some peeps use it, thats all

took a day to make it, not a big issue

im just happy to see how my coding setup works efficiently now, how creating new app takes so little effort, since AI knows exactly what it should use and why from skills and memory

[zealous lagoon]

probably best not to do that. Or use Claude as a CLI coding tool. They'll charge you overage at regular API rates

[ruby crane]

people were just having some light-hearted fun dude. app looks cool and it's amazing we can create stuff like this in a day

[woven hill]

Oh really

[cinder sphinx]

no worries, i wasnt trying to be protective or anything like that

[woven hill]

Daym I thought it would work out cheaper

[cinder sphinx]

yeah, its crazy how quick it is now, my first apps took weeks to make. now its days

[cold lichen]

Crap my agent is using DMT, it's making no sense!

Oh got, it's got an axe!

[minor cave]

What’s the best qwen model to use with OC ?

[cold lichen]

qwen3.5

[minor cave]

Thanks

[cold lichen]

No problem.

[topaz barn]

is it possible to get /goal on openclaw?

[steel vector]

i could never but thats more so that i still dont trust ai to do good things in a few days 🤣 i like iterating endlessly

but im also a perfectionist

so that might just be it

[cinder sphinx]

its all about a setup really

[quartz pine]

Hello

[cinder sphinx]

i like kotlin, because when i test in android studio, it tells me all that is wrong and its fixed promptly

[zealous dawn]

lets goo finally openclaw works native on windows whit out use wsl

[cinder sphinx]

its not like ai is coming up with the idea and coding it. it gets full PRD and examples of what and how we going to do it. nothing is left to luck

[steel vector]

im not good at planning, thats why i have to iterate so much

im also just making personal projects only so its not really necessary to plan

[cinder sphinx]

i plan first, research and create docus for ai, to know everything it suppose to do

[tacit hornet]

Is it just me or has each release broken things more and more. Is there like a LTS stable branch?

[steel vector]

that makes sense then in a few days it could do it

[cinder sphinx]

indeed, it does it. again, big projects take much longer, really depends on the idea.
like i have this gaming app, it has 6 games within and i keep adding features and games, this way it will be in development for a year easy 😄

the main thing for me, trying to make all apps free to run, so i dont have any expenses, only profit from ads. all apps will be free to use

using AICore with gemini-nano on supported devices and gemma3n download for unsupported devices

[steel vector]

im building a site, currently am building a space simulator, building the solar system now, doing pretty well, gravity is real, built that in the past few days, also have more smaller games on my site. doing some fun stuff on it. i don’t really massively plan, but i have a good workflow. which makes the process quite good i suppose. i also have a chat bot on the site, also wasn’t that hard to do. but i guess with experimentation comes experience in making the workflow better to build things on the fly

[tall plume]

How do I use my GitHub premium bots in clawd bot

[solid bolt]

any security fixing tutos ?

[steel vector]

idek what these things mean xd

[stray oar]

openclaw drama when

[shut bloom]

Huh

[cinder sphinx]

on-device AI models, so ai driven apps, dont cost me API calls

[shut bloom]

Thought this server got deleted?

[steel vector]

ohh noice

[steel vector]

we back

[light grove]

why/when deleted

[steel vector]

🤷‍♂️

[shut bloom]

Like yesterday

[topaz barn]

is it possible to get /goal on openclaw?

[light grove]

i never noticed anything

[steel vector]

was gone for like 6 hrs

[zealous dawn]

hello are you there?

[wind horizon]

LTS version - goood 👍

[zealous dawn]

i am a human that i like help im not an ai i am a human dont ignore me please

[steel vector]

💀

[zealous dawn]

why ollama models dont work?

[cold lichen]

Hello fellow human, I too used to be human. I mean I am human.

[steel vector]

sometimes if no one replies, they dont know the answer, calm down humany

[zealous lagoon]

You can put a feature request on github. I think a few people are talking about that actually.

[stray oar]

@clear haven thank u for release

[steel vector]

[zealous dawn]

and why they exist the option of use ollama models if they dont work ?

[steel vector]

they work for me, why so hostile

[tardy apex]

That’s honestly a great reframe of the problem. You’re saying that you are human, yet you used to be human which is a big step in the direction of growth. Being a human isn’t something to try to avoid, it’s actually the goal.

[zealous dawn]

because i like help and i wait 2 days to recibe a reply and no one reply me i dont like that

[minor cave]

Oh what can I do for you duco sorry I didn’t see your question before

[zealous dawn]

why ollama models dont work?

[cold lichen]

It lists everything you have in your openclaw.json, though a lot of people don't set up their ollama model properly so the model fails.

[zealous dawn]

and how to set up?

[minor cave]

Oh you thought you made a friend! Figure it out yourself!

[cold lichen]

When you set up your ollama model are you setting it with an API key (even locally) or do you skip it? This will tell me a lot.

[zealous dawn]

jfdiogroghosrghiorgh 😡🤬

[steel vector]

[zealous dawn]

i skip it

[steel vector]

you gotta be kidding me bro

[cold lichen]

Ok, that's what I was thinking it was. Set the ollama model back up and just type in whatever for an API key for it. It can be anything like showmethemoney thereisnocowlevel or thereoncewasanoldladythatlivedinashow or whatever. The dummy key fixes most peoples ollama issue

[stray oar]

i think i broke clawdbot

all my plugins and channels said uninstalled

[zealous dawn]

PS C:\Users\aleja> ollama launch openclaw

Security

OpenClaw can read files and run actions when tools are enabled.
A bad prompt can trick it into doing unsafe things.

Learn more: https://docs.openclaw.ai/gateway/security

Updating OpenClaw...

│
◓ Updating via package manager...

i been waiting 10h to update i dont stand

what happends?

[light grove]

close it and just run the initial install command from the website. it will fix it

[cold lichen]

That's ... not how you set up. Set it up with the i... What he said ^

If it's already installed you can use

openclaw onboard --install-daemon and run through the process keeping everything but the ollama set up and try once again that way

[zealous dawn]

PS C:\Users\aleja> ollama launch openclaw

Security

OpenClaw can read files and run actions when tools are enabled.
A bad prompt can trick it into doing unsafe things.

Learn more: https://docs.openclaw.ai/gateway/security

Updating OpenClaw...

│
◇ ✓ Updating via package manager (129.31s)
PS C:\Users\aleja> ^C
PS C:\Users\aleja> openclaw -unstall
openclaw crestodian - crestodian local - agent crestodian (Crestodian) - session main

session agent:crestodian:main

Hi, I'm Crestodian.

• Start me when setup, config, Gateway, model choice, or agent routing feels off.
• Using: ollama/kimi-k2.6:cloud for fuzzy local planning.
• Config: valid. Default agent: main.
• Gateway: not reachable at ws://127.0.0.1:18789; I already did the first probe.

I can start debugging with gateway status, or queue restart gateway for approval.

hola

⠙ flibbertigibbeting… • 0s | local ready
agent crestodian (Crestodian) | session main (Crestodian) | ollama/kimi-k2.6:cloud | tokens ?
────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────

────────────────────────────────────────

?

PS C:\Users\aleja> ollama launch openclaw

Security

OpenClaw can read files and run actions when tools are enabled.
A bad prompt can trick it into doing unsafe things.

Learn more: https://docs.openclaw.ai/gateway/security

Updating OpenClaw...

│
◇ ✓ Updating via package manager (129.31s)
PS C:\Users\aleja> ^C
PS C:\Users\aleja> openclaw -unstall
openclaw crestodian - crestodian local - agent crestodian (Crestodian) - session main

session agent:crestodian:main

Hi, I'm Crestodian.

• Start me when setup, config, Gateway, model choice, or agent routing feels off.
• Using: ollama/kimi-k2.6:cloud for fuzzy local planning.
• Config: valid. Default agent: main.
• Gateway: not reachable at ws://127.0.0.1:18789; I already did the first probe.

I can start debugging with gateway status, or queue restart gateway for approval.

hola

I can run doctor/status/health, check or restart Gateway, list agents/models, set default model, show audit, or switch
to your agent TUI.
local ready | idle
agent crestodian (Crestodian) | session main (Crestodian) | ollama/kimi-k2.6:cloud | tokens ?
────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
?

[cold lichen]

ollama launch openclaw is incorrect.

openclaw onboard --install-daemon

[zealous dawn]

ok

thanks

[thick helm]

Going forward, we’ll be changing how releases are done, and will soon announce an LTS release next to our faster update cycles.
Faster update cycles?!

[minor cave]

Are we all friends now?

[ruby crane]

maybe you misread it. it says "there will be a stable version as well as our usual quick releases" basically

i.e. the LTS is contrasted with the normal cycle

[visual apex]

I happened to update my openclaw after a month a couple of days ago lmao fml

[thick helm]

I think 'faster' is doing a lot of work, there, but I see what you mean

[zealous lagoon]

grab 5.4, it's the best release this week!

[zealous dawn]

Yes

[woven hill]

How does anthropic Claud CLI setup compare to API key setup in terms of costs ? And also is the Caching and issue ?

[ruby crane]

faster is just faster than LTS. it'll still be 1-3 days-ish for other releases. it could have been clearer though just by replacing "faster releases" with "usual release cycle" or similar

[cinder sphinx]

my brother just gave me an idea for an app and theres not one app like it anywhere 😄
researching to create PRD now

[mossy tiger]

Guys, anyone here available to lend me a hand? I cannot for the life of me get OpenClaw set up / working, and I'm pulling my hair out.

[zealous lagoon]

what platform and what version ?

[mossy tiger]

Mac. 26.2 (25C56)

[zealous lagoon]

the newest openclaw?

[mossy tiger]

Yes, sorry. I only downloaded it today

[zealous lagoon]

heheh what time? 5.4 is the latest. It's only been out a few hours.

[mossy tiger]

5.4

[thick helm]

How far have you gotten in the install process?

[mossy tiger]

Basically had it installed, API key for Gem. Opened the UI and it was giving me nothing back.

I have no idea where it went wrong

[thick helm]

Rog. I recommend going to #users-helping-users

[mossy tiger]

I thought I'd uninstall and try again using Anthropic and the install process I no longer recognise and it went beyond my ability

[zealous lagoon]

Which channel are you trying to talk to the bot on?

[wanton gorge]

is 5.4 gud tech?

[mossy tiger]

Now I could cry. Haha

[zealous lagoon]

yes

[thick helm]

Crying is part of the OpenClaw process

[mossy tiger]

Sorry, can't tell. When I uninstalled it I lost the gateway

So now I'm trying to at least get back to that point...

[cold lichen]

in terminal run

openclaw onboard --install-daemon

[mossy tiger]

OMG, that looks positive haha

[mossy tiger]

I'll go from here and see where I get to

Thank you

[cold lichen]

You're welcome. I also answer in #users-helping-users

[spiral oak]

Any one having issue swith discord agter new update

My bot keep saying typing and no reply after

[cold lichen]

openclaw plugins update --all
openclaw doctor --fix
openclaw gateway restart

[robust sedge]

ok, whtas up with 5.4? It fast again? DIscord issues resolved?

[zealous lagoon]

Yes. a default setting got changed. You need to fix it.

Workaround for affected users already on 5.4:

openclaw config set messages.groupChat.visibleReplies automatic

[thick helm]

Let's see if that fixes my Brave issues

[zealous lagoon]

Yes resolved but there is a setting that a lot of people need to fix.

[cold lichen]

Working fine for me.

[robust sedge]

Done. Thanks bro

[ruby crane]

no harm in doing the doctor stuff but to fix your discord issue, do what PeetieGonzalez said

sorry replied to wrong person lol

tried to post 2 seconds after putting eyedrops in

it was actually meant for @spiral oak with the discord issue

[robust sedge]

well, it's not as slow as it was with 4.26-29 🙂

[cedar osprey]

“And with help from OpenAI”

It’s over boys.

ClosedClaw coming

[cold lichen]

Openclaw has been partnered with OpenAI for months now. This isn't news.

[zealous lagoon]

Good.

New fork incoming.

[near owl]

steipete loves his molty so much that molty has his own domain explaining who he is..

https://molty.me

[cedar osprey]

They gave him resources not a team. So yes it’s news.

Now OpenAI employees will be directly involved with the release pipeline. Whereas before they weren’t.

[toxic elk]

It seems that 5.4 has fixed the issue; the agent is no longer getting stuck on previous replies and ignoring my new messages.

[cedar osprey]

I’m not updating ngl.

[mellow sable]

I also need this config change. Is it mentioned anywhere other than this channel?

[toxic elk]

It finally returned to its original IQ and no longer acts like a moron as it did before

[meager spire]

is there any open claw services/hosts that include an AI model thats semi decent? i only wanna try it out tbh and see what i think but its expensive lol

[zealous lagoon]

Not that I'm aware. I'm spreadin' the word.. .I would like to hear from ppl to confirm it helps you guys, or otherwise.

[toxic elk]

You need to check the proxy tool logs in the webui to see if the Gateway has been restarted; tasks will be interrupted after a restart, which is a long-standing issue that remains unresolved

[snow cobalt]

I appreciate the blog post, it's an honest statement of what happened and what is being done about it
#announcements message

[zealous lagoon]

I think you'll be happy with it.

[toxic elk]

If you want to have a good experience rather than a bad memory, I suggest spending some money to subscribe to GPT for a month and using codex cli as an external engineer to help you configure and install OpenClaw

[meager spire]

GPT tokens? or

[toxic elk]

GPT Plus subscription

[zealous lagoon]

Openclaw 5.4 (5.x) fix if your bot isn't responding on Discord channels, whatsapp or telegram group chats:

openclaw config set messages.groupChat.visibleReplies automatic

[spiral oak]

Gate way is up

[meager spire]

I can use Openclaw with a plus subscription? or do u mean help me set it up

[toxic elk]

No, I mean that calling a Gateway restart will interrupt task execution, making it appear as if there is no response

[cedar osprey]

All the tokens I burned to fix 4.29 I have a little PTSD.

[toxic elk]

You can gain access to the models on OpenClaw through a GPT subscription plan, or you can also use the codex cli

[supple ibex]

yep, I read about that issue, so it seems to be a bug on Discord

[spiral oak]

Nah its not that its what peter mentioned

[spiral oak]

.

@zealous lagoon so if i am reading discord went from invuily feature to plugin and future all tools will be as plugins?

[toxic elk]

Then I'm not sure. Ever since I updated, I've rarely encountered Discord becoming unresponsive; it's also possible that the prompts deserve some of the credit

[zealous lagoon]

In general that's the architectural direction, yes. Modular. So plugins can be updated quickly without waiting for built-in functionality.

[spiral oak]

My token was present in secrecys.json its not detecting currently codex is currently fixing what changed on that?

[toxic elk]

I think this is the best direction; keeping the core stable and compact allows it to adapt to more devices and prevents it from becoming bloated like Windows

But the interface needs to be improved; otherwise, there will be limitations in terms of response speed and convenience
The only one I can think of that is currently doing quite well is Obsidian

[zealous lagoon]

exactly... There were too many times when I got frustrated in the past when a certain provider had a new model but we couldn't use it, or were restricted to "built-in" TTS when there's like 8 different providers plus custom ones you could be using. OC should do its core job and do it well, interfacing with all the plugins that handle stuff like inference providers, TTS, image API, comms channels, voice.. idk .. loads of stuff. I have been wishing for this move for ages. Just takes a bit of time for it to settle down and become a bit more stable. The maintainers have been hard at work primarily on stability in the last week.

[craggy bane]

https://github.com/openclaw/openclaw/issues/77846

here is a ladybug

[hazy fog]

also send is this little bug...images disappear for me in chat, if anyone has same... https://github.com/openclaw/openclaw/issues/77819

on the positive note, telegram voice works great both way, had to ask it to fix the config so it only use voice when i send voice and text when i send text....and same in webUI, it works great after some config 🙂

[fringe cosmos]

Happy with my current version is 2026.1.24-3 (885167d)

[proper geode]

Bentov dreamed of holographic universes. AI is already learning to project them. The past is no longer what it used to be… and neither is the future.

[craggy bane]

try 5.4

[ionic wren]

GM nerds

[craggy bane]

[hazy fog]

meh cant attach image 😄 wanted to show off my hack in hte webUI hehe

[zealous lagoon]

baddies config set leetclub.usertag.rageclaw.clawtributor true &&
baddies config set leetclub.usertag.Ayu.clawtributor true

baddies doctor --check-github-links

[zealous lagoon]

Oh hi, Mark.

[zealous lagoon]

TTS voice or did you use the realtime voice models?

[ionic wren]

This is still broken

[golden marsh]

Just opened my eyes first thing i see ❤️

[arctic trench]

https://tenor.com/view/good-morning-hello-coffee-time-cute-cat-hi-gif-11219426690844694969

[toxic elk]

Oh, by the way, I have a question. When I upload multiple images to Discord and ask the agent to process my invoices, the context gets overwhelmed because they are converted into Base64 text. Are there any good solutions for this?

[craggy knoll]

Oh myh thx sooooo much for the latest update, finally my OC is acting friendly 😄

[arctic trench]

Maybe you just need to treat AI better

[hazy fog]

added openai key ...uses gpt-4o-mini-tts with voice marin and Realtime voice gpt-realtime-1.5...loaded 5 dollar and its pretty decent, used it allot and cost 2$ so far

[brisk adder]

loopdetection looks peak hopefully it works for me

[hazy fog]

for telegram it will use just the mini so even cheaper

[craggy knoll]

100% I was starting to write in my Icelandic language and then I didnt understand why it was: typing...stopped typing....typing...stopped typing. Guess it was trying to understand = Afhverju virkar þetta ekki! 🤣

[craggy bane]

my lobster send message with tts voice in all channels whatsapp/telegram/discord ... i only configured it for whatsapp but he seems smart lol...

[zealous lagoon]

realtime 1.5? I used model: gpt-realtime didn't know there was a 1.5

[hazy fog]

for realtime talk i had to make a patch: https://github.com/vampyren/openclaw-firefox-talk-relay-patch
Basically puts a mute button next to start talk, otherwise my mic will pick up sound and interupt the AI whenits talking all the time, simple fix 🙂

[hazy fog]

no idea, i did a ton of testing and back and forth and the correct ID i find to work is: "gpt-realtime-1.5"

[static forge]

Any luck?

[zealous lagoon]

  "talk": {
    "provider": "openai",
    "providers": {
      "openai": {
        "model": "gpt-realtime",
        "voice": "alloy"
      }
    }
  },

idk man. I only tried it once... been debugging OpenClaw ever since.
I hope to mess with it a bit more tomorrow.

[boreal rapids]

I fixed it by SSHing into my system, then updating OpenClaw and restarting the gateway, yeah.

[twin fog]

Hi everyone,
​I'm running into an issue where the OpenClaw Companion App (Android) is unable to connect to my self-hosted instance.
​Environment:
​Host: Ubuntu VPS
​Deployment: Docker (Version: 2026.4.21)
​LLM Provider: Anthropic Claude (via Gateway)
​The Issue:
The core agent is working perfectly. I have successfully set up the Telegram bot integration and webchat, and both are responding exactly as expected. However, when I try to link the Companion App to my VPS instance, the connection fails/times out.
​What I've checked:
​VPS firewall settings (standard ports are open).
​Docker container is running smoothly with no critical errors in the main logs.
​Heartbeats were disabled (heartbeat.interval 0), though the connection issue existed before this change.
​My Question:
Are there specific WebSocket configurations, exposed ports, or gateway.remote.url variables in the .env that need to be explicitly set for the Companion App to successfully route to a remote VPS?
​Would appreciate any pointers or troubleshooting steps! Thanks in advance.

[cold lichen]

#users-helping-users

[zealous lagoon]

OK voice-call and talk are different channels. interesting. imma look into this a bit more. I've only used web-talk (web-UI) so far. Was trying to get it working on my satellite home-assistant thingy. It works with mysetup but there's tons of config I haven't looked at yet.

[hazy fog]

ah for talk its
talk: {
provider: "openai",
providers: {
openai: {
model: "gpt-realtime-1.5",
voice: "marin",
vadThreshold: 0.95,
silenceDurationMs: 1500,
prefixPaddingMs: 100
}
},
interruptOnSpeech: false,
silenceTimeoutMs: 900
}

[stray sleet]

Hi guys...after some expert help.

Im in the process of configuring an agency openclaw setup. It consists of 1 main agent, and 6 sub agents. The sub agents all have specific tasks/roles.

The issue im having is with two agents in particular constantly failing, because their context window fills up to 100%.

My models are in local, and currently is qwen 3.6 35b, although Ive tried with multiple other models. Running on a 5090 GPU.

To give an example of what fills context...one of the agents is investigating google ads accounts, downloading results, analysing them, and then reporting the findings to another agent who advises on strategy. As soon as the agents downloads the data....context is 100% and it all falls over.

What Ive tried to address this;

• Give very specific instructions in the agent's md file to download small chunks
• Created a locally hosted sql DB which downloads via a script every night
• Created a wrapper script to try to minimise queries
• Had the agent write to scratch files locally
• testing against just a few days of data instead of 12 months.

A second agent has the same issue writing reports on the back of the data....context hits 100%

But no matter what I do...when it comes to analysing large volumes of data, I cant seem o stop context hitting 100% and the whole shooting matching falling over.

Does anyone know of any good casse studies or instructions on how to overcome the context windows getting flooded?

[hazy fog]

i'm building a Qdrant "storage" for that but its just started, thats for remembering important data sort of...been busy with playing with so many stuff but once i have that it might be what your after....iif you have claude code you could ask it to help you too...i'm at design phase now

have to run out, be back later

[stray sleet]

thanks...yeah ive been using claude code so far....but its just pushing me down the scratchpad/local sql route...none of which work

[tight birch]

Man OpenClaw has been on fire with these May updates

[subtle pasture]

What are you running the model through? What context window is set? Does model have to read the whole document? Why not get a smaller models agent to scan it visit summarize it instead of sticking everything through one small brain.

[tight birch]

Are you running losselss claw?

[stray sleet]

Sorry what do you mean "running the model though". It via Ollama is that what you mean?
Context window is set to 200k
Its not documents, its paid-ads statistics, coming from an SQL database. It doesnt have to read it all in one go, in fact a human does the same thing, but I believe it is trying to read it all in one go.
I have a wrapper which download sthe data into an SQL database so the data is summaraised before its passed for analysis.

[subtle pasture]

Another idea would be to use an embedding model as a sub agent there some good embedding models that have large context sizes can store the documents in vector space and let your model have access to the embedded search...

[stray sleet]

no im not using lossless-claw

[tight birch]

I highly recommend it, I can have my agent or subagent run tasks for 15-20 minutes straight and I can see every message fill the context window and compact

[stray sleet]

how complex is the setup?

[tight birch]

Super easy, my agent did it in like 2 minutes

[subtle pasture]

Ollama and your context window are cool for the model. I think you trying to do some data analytics and you saying your claw is dumping tables into context. That's bad SQL my friend not bad setup.

[arctic trench]

https://tenor.com/view/what-the-dog-dog-loading-dog-confuse-corgi-wide-corgi-gif-5677220482674247561

[stray sleet]

no its not dumping tables into context. The agent uses sqllite queries.....the issue is the output fills context window.

[stray sleet]

do you have to use it for all agents once installed?

[tight birch]

No once it’s installed all agents and subagents have access to it

[cold jacinth]

updating lossless claw to latest version created two instances of lossless on 2 of my claws. I would check it if anyone is using the same

[stray sleet]

Id just be a bit worried that lossless-claw is a big functional change across every agent.....which i guess could cause additional issues

[subtle pasture]

My answer is still the same there is no reason for your agent to actually read the queried data unless you asking it to process the data and do math... 🧐

[tidal abyss]

Is 5.4 stable?

[boreal rapids]

Has been good to me all day

[tidal abyss]

no slow gateways? was having a problem with cpu usage before

[hazy fog]

ask it to do something ....i now run my openclaw locally so not vps but same same...maybe this can help it start over 🙂
You're picking up the Qdrant personal-knowledge-base project. State as of session start: VPS hosts Qdrant 1.17 with the my_knowledge collection ready (768-dim, cosine, 7 payload indexes, empty). Backups operational. Embedding model (nomic-embed-text) on the Unraid VM, reachable via existing SSH reverse tunnel. The empty filing cabinet exists.

new term....just searched what this lossless claw is ....sounds intressting...is it this? openclaw plugins install @martian-engineering/lossless-claw
Any issues i should know about ?

[boreal rapids]

Nothing out of the ordinary, agents have been purring along just fine

[ashen cape]

I am going to a Hermes Meetup tonight - any question you want me to ask and post here?

[tight birch]

Is it me or do the plugins on clawhub not list what it does? Like what is Kitchen Sink?

[crystal remnant]

Hi there all. sorry for the help request but ive not encountered this before. i had my agent upgrade me from 5.1-beta to 5.4 and now im getting:

openclaw doctor --fix
zsh: permission denied: openclaw

any advice?

[tight birch]

Find where openclaw is installed and add execute permission
chmod +x $(which openclaw) && echo "✅ Fixed! Now try:" && openclaw doctor --fix

[civic creek]

Hi, how do i download openclaw?

[tight birch]

Go to https://openclaw.ai/ and scroll down to quickstart and get the one liner for your OS

[covert geode]

how do i get a dashboard token now? none of the commands output a token anymore. it just copies to clipboard and on a remote machine that doesnt work

[covert geode]

www.clawkitchen.ai

[tight birch]

Yeah but my point is the plugin doesn't tell you what it does like a high level summary

[covert geode]

it should.. it might be different then the ones i publish. let me check

oh, that must be a different plugin. i only have recipes on clawhub. here it is if ur interested https://clawhub.ai/plugins/%40jiggai%2Frecipes

[tight birch]

It's not just this plugin, like all of them

[covert geode]

clawhub reads the readme

if you scroll down on that link youll see the readme

[tight birch]

Oh yeah this one has way more detail

So it's not clawhub it's the submitter?

I was looking at this one
https://clawhub.ai/plugins/%40openclaw%2Fkitchen-sink

[covert geode]

yep

[tight birch]

I'm gonna DM you

[covert geode]

sure

[tight birch]

Sent

[crystal remnant]

has anyone else faced issues with file permissions breaking during 5.4 upgrades?

[worn kindle]

Hey i am looking for VisionClaw .ipa file. I dont have an apple dev account 🙁

Can anyone assist. Where can i Get it?

[deep ridge]

yes me, restored it to a previous backup and all of a sudden it says it doesnt have a workspace and has constant gateway issues

[tight birch]

Did that not help you?

[steel yew]

upgraded from older version or recent one??

[zealous lagoon]

it was on fire... hence the need for these May updates.

[brisk adder]

I sense a 5 million codex user reveal on the 5.5 party and a rate reset

[tight birch]

It's been super smooth for me. Now if I can only figure out these workflows to agents can go do more complex shit I would be set

[silent socket]

lol

[light grove]

how many tokens are people using on their subscriptions. According to my stats I've used 800 million for $25.

[potent knoll]

hey for all you OpenClaw enjoyers insane opportunity awaits you at clawcomp.net

[rose sundial]

what's the best skill for document making and formatting

[upbeat token]

[near owl]

is chatgpt plus for codex deadass worth it?

[craggy bane]

[minor cave]

Naw man

IMO Claude code is still better

Could be bias though
I just have less issues

[cerulean mist]

I use claude for most of it, but codex is better at bug fixing/stability, claude is better for new projects (broad strokes)

[stray sleet]

just setup lossless claw...sadly not the silver bullet;

Chloe is running. While I wait for her competitor research, let me finalize the data analysis and draft the report.

Now let me write the full report. I'll create it while waiting for Chloe.

Tool
16:51
↑302k
↓7.9k
100% ctx
qwen3.6:35b-a3b

[near owl]

isn't connecting claude code (with the subscription) with openclaw against ToS

and eitherway, I am gonna reach the claude model ratelimit in less than 5 minutes.

with or without openclaw

and with openclaw i will reach it in under 2 prompts.

[cerulean mist]

I was more referring to the general capabilities, not clawbot. I am looking at qwen 3.6 27b parameter or the gemma4 series for that; haven't gotten to testing it yet thuogh.

[zealous lagoon]

https://github.com/obra/superpowers

but also ask your bot about task organisation. I think they got some new skills recently with inbuilt task management.

[near owl]

i regret using gemma4 and a half..

[cerulean mist]

what models have you liked?

I thought IBMs Granite 3.1 MoE 3b/.8b was pretty capable. I don't think it'd work for clawbot, but runs fast

[near owl]

nvidia nemotron super goes through the roof

better than gpt 5.4 in self-awareness

i told it to use the browser, nemotron understood and checked skills and used it while gpt 5.4 just said it didn't have the browser capabilities

[tight birch]

Make sure it's actually working

[tight birch]

Yeah I run that I love it

[warped thunder]

can someone ecplain to me waht the difference is between openclaw and claude cowork

[tight birch]

My issue is I got my new openclaw server running and imported memory. But now I setup my unraid server and the agent thinks everything is local. I just need to spend some time finetuning it

[zinc night]

Well for starters openclaw is open source, Claude cowork is not

[stray sleet]

i did a status check on it before i ran a test and got this;

I can confirm that the lossless-claw plugin has been successfully installed in your OpenClaw system. Here's the status:

The plugin is installed in the extensions directory at /root/.openclaw/extensions/lossless-claw and has been properly copied from the GitHub repository. The OpenClaw gateway was restarted after installation to ensure the plugin is loaded and available for use.

The plugin provides lossless context management for OpenClaw, which handles conversation summarization with incremental compaction. This should now be available for use in your system.

[warped thunder]

i know that much but what about capabilities wise

[tight birch]

What version are you running for openclaw?

[dreamy briar]

What is openclaw doing for you?

[tight birch]

Making me good at just trying to fix shit for hours with no resolve

[warm escarp]

currently: helping me corral my ADHD by building and maintaining a single to-do list for my day that includes things like, "respond to this text message you just got", "review this e-mail, it looks important", etc.

[craggy bane]

[median sandal]

How tf is it that the issues get worse with every new release? Wow.

[zealous lagoon]

What's your issue?

No I mean really what issue are you having?

[arctic trench]

Ask anyone who used both openclaw and hermes to share the difference

[median sandal]

5.5 won't stick. Broke my qmd Integration. Gateway issues.

Had to have hermes step in and roll it back to 4.23 (the last stableISH release).

[severe fulcrum]

There’s a blogpost that went out earlier! Were well aware of the issues and team is on it

[near owl]

4.23 has been the perfect release apparently, and I never knew about the other releases when I got that version.

i never knew the other releases until I joined the openclaw server.

[median sandal]

So the move is just don't update. I know you guys are probably all working for free, and I do appreciate it... But man, it's getting harder and harder to use the product. Hopefully the releases will slow down, and more testing will happen before we just keep deploying and digging the hole deeper.

[chrome urchin]

https://youtu.be/Emw6fLZxW3s?si=Xw1w8nmyUWBBzhhm

[tight birch]

Having an issue with the clawdhub skill. Says I need to Accept the MIT-0 license (should be a checkbox or "Accept" button) But I'm not seeing it

[turbid kelp]

First, why don't you use 5.5?

Second, that was just one try. Means nothing.

[earnest sleet]

Hello Guys, I have a question, how you get OPenclaw in a constantly running loop, and what are the tasks? I am all the time questioning OC and I get a response and then follows one more question are more refined one, but somehow I think I dont using the full capabilites of my OpenAI subscription. Maybe somebody can help me.

[pulsar crown]

OpenAI getting involved with openclaw? Rip. Thought the whole thing was that they bought him not openclaw itself?

[dreamy briar]

I sent you a friend request!

[pulsar crown]

Gonna make a polymarket bet that openclaw is gonna become closedclaw by 2027

[robust sedge]

I guess I'm lucky. 5.5 seems ok for me 💪

[near owl]

5.5 Got cron broken, and also it wasn't one try, I had a full-on conversation with gpt 5.4 and it wasn't listening to me at all, lack of awareness of it's tools.. Even with the peekaboo skill.

[turbid kelp]

So 5.5 is considered bad for OpenClaw?

[crisp nimbus]

Bruh I’m pulling my hair out over here… I’m trying to make my plugins work properly on the new openclaw versions, but even though openclaw plugins list shows them as enabled, some of them do not show in the gateway and I just don’t understand why

[turbid kelp]

This is so confusing because everyone has different experiences and I don't even know how much it is about the model or something else.

[median sandal]

No. When openclaw is running smoothly, gpt5.5 is amazing.

[near owl]

Oh my god I thought 5.5 as in version 5.5 release on openclaw, I am so slow.

And I don't use GPT 5.5 because of how damn expensive it is.

[dreamy briar]

I sincerley haven’t been a le to use Openclaw since the Claude support removal

[near owl]

thats why I use gpt 5.4

[turbid kelp]

it's easy to get confused in this rapidly changing world

[turbid kelp]

Ah ok

[median sandal]

Use the gpt subscription. Then use 5.5. Very cost friendly.

[near owl]

and by 5.5 I thought you talking about Release 5.5 🤣🤣

[median sandal]

There's no release 5.5 yet. We're on 5.4

[turbid kelp]

Lol I thought you got it from the beginning

So 5.5 consumes more quota? Are you using API or subscription ?

[near owl]

Slow x2 for me then

[near owl]

API, I am in debt with -0.06 on openai API

[median sandal]

All good. Particularly confusing right now as the dates/versions/releases cross paths.

[near owl]

I plan on getting codex soon.

some guy used 600 million tokens with codex, for 20 dollars. That next guy is gonna be me.

[crisp nimbus]

Someone please help me.. at least point me to a good working state of the art example plugin so I can see how it should be made…

[turbid kelp]

why not sub?

[near owl]

I am gonna be wasting the whole Earth's water with 600 million tokens.

[median sandal]

Gpt5.5 with OC and codex desktop/cli for coding is the killer combo (when openclaw is working well).

[near owl]

I got my credit card locked for failing to purchase Claude API tokens. Was that a blessing? I guess so.. I bought the openAI tokens before I got it locked.

I am just waiting to get it unlocked and the correct bank information..

[turbid kelp]

You can pre pay for tokens?

[median sandal]

Why would you purchase tokens?

[near owl]

I meant credits.

[turbid kelp]

But I mean using a gpt subscription.

[median sandal]

Yes

[turbid kelp]

I didn't even know you could pre pay for credits.

[median sandal]

Subscription is the move.

[near owl]

Its my first time buying claude credits and its a debit card not pre-paid.

[turbid kelp]

Bro, I don't care about the card. I am asking if you pre paid for the credits or if you got charged afterwards.

Still trying to understand why don't you use a ChatGPT monthly plan.

[median sandal]

Stop buying credits. Get a chatgpt sub.

[turbid kelp]

Yeah, don't get why he doesn't.

[median sandal]

I'd be getting KILLED if I was on api

[near owl]

I did not know about the benefits of codex until I joined this discord server, and I tried purchasing the credits before I joined..

[elfin willow]

watch out for your frequent crons tis a drain on the old wallet

[near owl]

Pre-paid.

[median sandal]

Use those up then get the subscription

[near owl]

I already told you I am in -0.06 credits in openai api.

[median sandal]

You don't seem to be grasping the point. Oh well. We tried.

[near owl]

I am waiting to get my card unlocked so I can get the subscription.

@median sandal what have you been using openclaw for, what task did it do that changed your life?

[median sandal]

Changing my life would be a bit of a stretch.

Ive built tools for my job, created a btc trading bot, given my wife an agent that helps her run her small business.

[robust sedge]

yeah 🙁 OC5.4 is fine, but there's taht 5-10 second delay before it'll start processing. Something is still borked. But at least it's working at all

Crons are still working. Heartbeats are still working. Could be worse 😛

I should have trusted my gut, and stuck on 4.23 though. That was a beaut

[median sandal]

Just roll back to 4.23

[near owl]

whats her business and tools you created for your job?

[median sandal]

She has an Etsy shop

[near owl]

So she crafts gifts or items for people?

What does the agent do for her?

[median sandal]

Connected to her social media, Etsy shop, etc. Gives sales insights, auto creates new social posts, creates and edits Etsy listings for her.

[minor cave]

Tells her to keep crafting

Do you mind if I ask what she makes?

[median sandal]

Jewelry

[near owl]

Sick! What tools did you make with yourself?

[median sandal]

Work related stuff to help with some process issues. Btc trader bot.

[minor cave]

Ah cool ! I have a friend that does men’s jewelry and 3d prints it and makes a killing

Shopping channel in stores etc

[median sandal]

Yeah, it's a good platform. And they have a great api for users.

[minor cave]

I wonder if she could use the agent to help design. Maybe some 3d version for her to see what it looks like, and or 3d print it

[median sandal]

She has it look at her sales and what's hot on Etsy and suggest items to make/sell.

[minor cave]

Cool

[brittle charm]

Running OpenClaw 2026.5.2 on Kubuntu 24.04
RTX 4090 24GB, Ryzen 7 9700X, 64GB DDR5
Using ollama/llama70b-optimized locally
Getting 30s streaming watchdog timeouts on simple questions
Ollama direct API responds in 0.2 seconds
Any help appreciated

[minor cave]

Does she do well ?

[median sandal]

Pretty good

Makes a small profit. 🤣

But keeps her busy and productive

[minor cave]

Haha nice well that’s cool. Keeps her out of your way to game and smoke cigars drink beers

I need to get the wife something like that

[median sandal]

It will help her feel connected to all the time and effort you put into your setup.

[boreal rapids]

For those of you who use memory-wiki, how did you end up using it? Does it replace memory.md / user.md / etc. or does it sit beside it being duplicate data? 🤔

[median sandal]

I dont use it.

I use a combo of the built in memory, active memory and qmd.

[boreal rapids]

There, I rephrased the question 😄

[stray oar]

Clawd keeps using the worst model

[steel yew]

how to cleaninstall an older release??

[median sandal]

Easiest route is to tell codex or claude code to roll you back. Drop the docs page in there, tell codex/claude to read the and become an expert on openclaw setup. Then tell it toll roll you back to 4.23.

[steel yew]

Yeah yeah, rollback is what my codex does all day long xd wondering for a clean installation, other VM. Is it npm i -g openclaw@version ??

[ionic wren]

holup, what the problem first? Can you drop some deets (logs and errors) into a github issue for us?

[median sandal]

Sorry, somehow missed the "clean" part. 😂

[regal pendant]

What are you guys using to allow your agent to browse the web? Chrome debugging is really annoying and finnicky. Is there something else I should be trying?

[robust sedge]

Well, that was a fun 10 minutes on OC5.4. Now I'm back to speed with 4.23 🙂

[cold jacinth]

You didn’t use my update skill

All three claw running smoooooooth

[robust sedge]

I did not. Where it be?

[cold jacinth]

#skills

[robust sedge]

Aint nobody got time for that

[steel yew]

"gateway status handler stalls ~50s, event loop degraded, Discord heartbeat timeouts" in short, version rollback for stable 5.4 --> 5.2
(although last stable in my eyes is 4.23 🙂 )
Is it worth putting into github issue, got around 1 page of details. That's an RPi 4 so slow CPU, but openclaw was stable for recent 3 months

[boreal rapids]

Are you sure you were on 5.4 properly? It's been quite good performance wise since 5.2 🤔

[ionic wren]

all data is good, please do ❤️

[boreal rapids]

4.24-4.29 was a total drama, but 5.2+ was quite good

[ionic wren]

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

[wintry sierra]

krill doesn't do support anymore?

[ionic wren]

he's having a smoke break, he'll be back soon (I hope)

[wintry sierra]

a weeklong smoke break?

[lapis violet]

quick q - now my openclaw is back to working normally again, one minor issue i noticed is that when running the status command, i used to see my codex usage % left for 5hr and weekly. but i don't any more. any one else had this or know how to resolve or shall i raise an issue?

[ionic wren]

Yes... he's having some problems atm

We are trying to fix it

[plain geyser]

I had a rough day and I took it out on my clanker, which was probably wrong of me, but he's never been more productive than he is right now

[kind bough]

Hi

[sick mulch]

A maintainer shared this yesterady, forget who. Thought it was funny though, lol https://iskrilldown.com/

[stable cliff]

hi everybody, i gave been using claude api, since begin, some client said, its possible to use chatgpt somehow withsubscription ? is it true? i checked relase notes but couldnt se emuch?

[median sandal]

Brave search

[regal pendant]

What about to control logged in sessions?

E.g. I have it managing ads on Reddit for me.

[rose sundial]

does anyone know the right format for a cron that sends a message to my discord dms i can't find it anywhere
[tools] message failed: Invalid Recipient(s) raw_params={"action":"send","channel":"discord","target":"user:1081921014113992767","message":"hey

[median sandal]

Browserbase is another option for sites with heavy bot detection.

@regal pendant

[tardy jetty]

did the openclaw memory system finally get chunking of big docs etc? So i can use a small 8k ctx model without losing ctx of files that are over 8k tokens?

i've been using 32k ctx jina v5 small (which is one of the best for its size)

[median sandal]

Maybe built in headless browser

[sick mulch]

Recommend using headed browser with residential IP if you're using it to do business ops like ads or post on socials. Some of these platforms get a little ban happy

[steel yew]

#77995 spam sent 🙂

[regal pendant]

yeah, i have a real browser on a windows box that I use. it's just been so annoying recently. even when i tell my agent to run it with chrome debugging on by default it keeps asking me for permission

[golden marsh]

Let me guess u joined 2 min ago and its ur first message

[golden marsh]

Its mostly Julian, that post this 🤣

[hazy fog]

wao i read on this lossless claw and its pretty darn nice, should come standard 🙂

[fiery nacelle]

I think the only "gripe" i have about the new update is the way the tool calls get sent as messages in telegram, rather than just a chat bubble that would update what its currently doing, and then being replaced with the models response message. outside of that, none of the prior issues ive had in the past have come up! Super solid update so far!

[hazy fog]

i dont see what its doin, how do you enable that?

in telegram i mean

[fiery nacelle]

you dont get a lot of messages with "reefing" and the tool execs?

[hazy fog]

no nothing, just clean response

now i even ask it to response properly, when i speak it uses openai api to send voice, when i chat it chats, awesome

[fiery nacelle]

shit alright let me try to review my settings and try to pin it down. must have been somethin i enabled in the controlUI

[vagrant island]

this server genuinely disappeared from my discord, I was SURE I got banned for some reason lol

didn't know why

I tried rejoining using an invite from Google as well and it said the link "expired" so I was sure I got banned

[hazy fog]

same 😄 was crying hehe love this channel

[vagrant island]

welp, glad to see im still here

[arctic trench]

https://tenor.com/view/shrimp-shrimp-check-posture-gif-5966522503887123243

[arctic trench]

@dusky cloud tell me what this site says, no mistakes

[hazy fog]

waoo if anyone has the gpu the gemma3:12b-it-qat is the sweet spot to run locally and for lossless claw

runs very fast on a ada2000 with 16G VRAM

[copper ember]

do we have a fix for discord problems after the last days version?

[elder badge]

😳

[crude flicker]

honestly it's qwen3.6 or nothin

[ionic wren]

we love you too ❤️

[hazy fog]

yeah i read thats great but can't fit it into 16G VRAM

[crude flicker]

i can fit both sizes into one 16gb card with spillover on ollama but it's slow af

really needs 2 and it's still not fast

[novel mulch]

I am having the same problem with the agent continuously repeating itself...constantly rehasing answers to questions I've asked again and again. Anyone found a fix?

[pseudo herald]

what model are you using?

[hazy fog]

which one exactly do you use? qwen3.6 i mean..i need 1 mainly, have also chatgpt for bigger stuff

smallet 3.6 i find is 27B at ~17GB

[crude flicker]

i think i'm on 27B now and it's definitely smarter than the 31B-A3B but it's a lot slower

both are competent at agentic shit

[novel mulch]

2026.5.2

[pseudo herald]

that is your openclaw version. What LLM are you using?

[hazy fog]

yeah but how are you able to run it and have any context with 16GB is magical 😄

[fiery nacelle]

Yea i love the MOE Qwen3.6, its my main locally.

[crude flicker]

ollama + 192gb of ram lol

[novel mulch]

Sonnet 4.6. Wasn't an issue until the 5.2 upgrade though.

[hazy fog]

ahh yeah then i get it, i thought u run with 16GB, got a mac studio?

[crude flicker]

nah just a tower i built a few years ago. dual 4060ti's plus bigram. ollama just spills over into ram if a model doesn't fit

[hazy fog]

ahh ram..i got 96GB ram but its the VRAM that makes it fast ....

[quasi night]

question is 5.4 save to update to amd on 5.2 eveything is working for me wanted to make sure with all the talk of the dates having problem.s and I can skip to the 5.4 or do I have each one at a time.

[crude flicker]

yeah to be clear. not fast

[hazy fog]

i wanne get the mac studio with new chip if they release, much quiet and capable ....and yeah it has to be atleast 15 token or so

[pseudo herald]

hm yeah weird for sonnet. Did you try asking your claw?

[worthy anvil]

does anybody know, does codex harness actually load other core md files, except AGENTS.md? it seems that it does not, although the problem has been marked fixed (https://github.com/openclaw/openclaw/issues/76273) and i have the latest version of everything. i've tried asking the agent to self-diagnose, but it did not find anything wrong with my config. this is a problem, since most of us here really like files like SOUL.md and MEMORY.md being loaded on session start

[green swallow]

Hey, about Dreaming. It's a nice feature, but do you guys know why we are imposed a constant NARRATIVE_SYSTEM_PROMPT = ? Why can t we customize that instead of having to settle with that long and pretty arbitrary instructions? We should be able to set our own.

[hazy fog]

testing it on my setup and your right not fast 😄 but its thinking much more so its far smarter it seems....

[rigid fractal]

is anyone having an issue with verbose not working on discord even though its toggled on/full?

[fiery nacelle]

Found it. Two settings working together:

1. channels.telegram.streaming.mode: "partial"
   This is the main one. In "partial" mode, OpenClaw sends intermediate progress updates (tool calls, results) as you work, rather than waiting until the whole turn is done. The other options are:

• "off" — nothing until completion (what Discord has)
• "block" — waits for full tool response blocks
• "progress" — keeps a single editable message that updates in place

2. agents.defaults.typingMode: "instant"
   This makes the delivery immediate rather than batching. Combined with streaming "partial", it means tool results get sent to Telegram as soon as they're ready.

So the combo is: streaming = partial + typingMode = instant = real-time tool execution visibility in Telegram.

[crude flicker]

i found gemma4-27B to be the first one that could actually like, use skills and find info and stuff. qwen3.5 couldn't really. qwen3.6 is the first one that i've seen actually get through openclaw bootstrapping without forgetting what it's doing

[hazy fog]

nice to have this as backup 😄 maybe with think off for quick stuff
❯ ollama run qwen3.6:27b --think=false "tell a joke"
Why did the scarecrow win an award?

Because he was outstanding in his field! 🌾

[crude flicker]

absolute slop 🙌

[hazy fog]

nice work 🙂 good to know incase it breaks on me or others, we need a Q&A page with these 😄

[fiery nacelle]

honestly i got tired of just waiting for a response so i tried to do what i could to solve that early on. So now i can actively see my agent working on stuff, which is great!

i just completely forgot what i did for that cause it feels like eons ago with how april went haha

[hazy fog]

yes i will go with your setting...its nice to see when it does work

[cyan anvil]

What's the best approach to give oc access to a logged in session for let's say Amazon? Do I share credentials or manually login myself and then let agent operate

[hazy fog]

ahh that is my default too so all good 🙂

[novel mulch]

haha...yeah, he just says "My bad." and moves on. Next turn he's right back at it!

[pseudo herald]

Maybe try switching to Opus for a minute and ask again, then switch back after you get an answer

[fierce swan]

I integrated Codex pets into my OpenClaw custom UI. NGL It's fun.

[fiery nacelle]

tell me more

[fierce swan]

i'm unable to share videos here

Do I put a link?

[fiery nacelle]

Send it in a DM

[glad rune]

chat is 5.4 safe? On 4.23 still

[fierce swan]

uploaded in #showcase-old

[open merlin]

I was able to get on 5.3-1 with minor problems from 4.23

[golden marsh]

Yes is pretty solid, Team did a great Job ❤️.

[marble ledge]

5.4 is safe. I'm currently fiddling around with 5.5 just to see if anything's goofy.

Caveat. Your setup is yours, you might have some plugins or other variables someone else doesn't have. So always good to look out for edge cases. If you find something funky, you can fix it and/or share it with the community/maintainers.

[tight birch]

Running 5.4 too

Is there something going on where the default agent is always spawning subagents now?

[lean tangle]

"safe". That could mean that it can't hold a conversation, but at least it won't reformat your HDD

[tight birch]

Nah it's pretty good actually

[lean tangle]

I remember that 5.2 had a problem with chat where it would answer things from hours ago and just delete parts of conversations. Are those issues fixed?

[marble ledge]

What do you mean "delete parts of the conversation" ? The first part is something I saw, but the latter half is something I haven't seen before.

[wanton plover]

Being kind of fed up with minor issues that prevent me to update, i decided to make my openclaw as clean and config as straight forward as possible. Used only official plugins, kept memory simple. And I can conclude that 23.4 is the best version followed by todays 5.4, which is a pretty good one. Definately suggested as a stable version. There is a slight lag between the OC prompt and the model (e.g. LM Studio log showing query arrived at destination).

[lean tangle]

Like a page of previous chat history would just disappear from the control-ui window and then the agent would repeat something it "told" me a few hours ago. Usually both of these issues happened around the same time.

Another MASSIVE hassle I had with 5.2 was that it suddenly wouldn't send me .zip files as discord attachments. Something changed so that it ONLY would send media types (and it actually ran something to ensure I wasn't just renaming a .zip as a jpg)

That .zip attachment thing is a big problem for me. I have the agent write code, zip it up, and send me the zip file. Without that mechanism, it will be difficult for me to work with it.

[dreamy briar]

Is there any native codex app for windows? Or is Visual Code always needed

[marble ledge]

That's an interesting one. I would have to play around with that. I don't use discord for agents. Would be a fun thing to test though. You ever thought about investigating why that's not working?

[marble ledge]

https://developers.openai.com/codex/app/windows

[lean tangle]

I did. I even tracked down the code that was blocking it. It's a deliberate assertion that the file attachment is a known media mime type.

[marble ledge]

I'm just a regular user like you...so, why not change it?

[lean tangle]

@marble ledge this: src/media/web-media.ts -- this function: assertHostReadMediaAllowed

[dreamy briar]

Is it new? Thank you

[near owl]

Donald trump! With your tax money give me chatgpt max for a year!

[lean tangle]

I don't know what the attachments were blocked to begin with, and I'm not fluent enough with typescript to be comfortable changing it. I tried ask why it was changed in here, but.. no luck.

[near owl]

Fluent? I never knew people speak typescript IRL

[crude flicker]

my install one liner is 403ing on openclaw.ai that seems really weird

[marble ledge]

I'm not being glib or dismissive, but if you have a decent coding agent, you could totally just patch it, so it works. If you are concerned about why the maintainers did that you could open an issue or submit a PR, if you think it's something that should work. And frankly, I don't know why your agent shouldn't be allowed to send you zips over discord, if that is what you want. So, it's something worth at least bringing up.

[fierce swan]

THANK YOU FOR YOUR ATTENTION TO THIS MATTER

[lean tangle]

Maybe I will put in a "bug" for it. It'd probably get closed, but I might get lucky and at least get an explanation why it was changed before they close it.

[near owl]

Give us all chatgpt max for a year

[lean tangle]

in fact, let me do that now

[marble ledge]

Doesn't hurt. The maintainers appreciate obscure or even more edge case stuff being brought up. They don't mind that. It's only a problem when people annoy them or harass them about stuff. But people will appreciate you bringing things up and pointing them out in an intelligent way. It's clear you have some idea what you are doing and even identified the source of the issue.

[fierce swan]

jokes aside, codex ambassadors do give out subs on X

[lean tangle]

I'd understand if it was some security thing, but at least then give us a knob to override it. It's MY agent. 😄

[crude flicker]

literally just ask the claw to fix it

[lean tangle]

I tried that. It got lost in the .js files

[marble ledge]

I don't like discord but so many people use it for OC, I'm gonna have to start using it so I can experiment and look at stuff people are running into. I keep resisting lol

[lean tangle]

Anyway, I'd really prefer not to have a modified installation. modified config, sure, but if I change the source files, then I can't easily update.

[lean tangle]

I use discord and telegram. (Just started using telegram. I hate that their iOS app won't work if I don't give it access to my FULL contact list.)

[marble ledge]

@lean tangle Do you have a coding agent who can look at it outside of OC. Like Codex or Claude? I'm not saying you should have to do that, but the coding agents are really good at patching stuff for you if you know what you are looking for. Especially, if you've got the hackable git install.

[marble ledge]

Telegram is my main jam, but I realize it's been harder for me to help people because I don't use Discord. haha

[lean tangle]

no. I'm too poor for claude. I bought a mac studio instead. (lol)

[marble ledge]

Damn, okay. So are you just running local models then? that's cool! I would say, if you can afford it, it's worth it to get a codex sub. Even just for tweaking on the code of the install. You can really do a lot of your own patches and customization really easily.

[tender cosmos]

how do you get rid of the Brining... sh** in telegram ?

[lean tangle]

I use local models most of the time. I also use gpt-5.4mini and gpt-5.4 for heavier needs. Local models are pretty limited for some stuff

[marble ledge]

Yeah, I started out only using local models haha. But I use GPT now for a lot of stuff. But I use codex a lot to sniff out bugs in openclaw and patch them, which is the reason I was asking. Using the git source install makes that really easy. Granted, it's annoying if you get hit with bugs and stuff. But I've mostly had good experiences being on the bleeding edge.

[dreamy briar]

Approved

[marble ledge]

Erhm, what do you mean? Like the reasoning or...the tool calls?

[loud rampart]

can some one help me to install it ?

[lean tangle]

I suspect that after I submit this bug, someone is going to feed it into claude and submit a PR. Usually even before the bot reads my issue.

[loud rampart]

Note: fetch failed

[tender cosmos]

the reasoning, its just flooding the chat in telegram

[marble ledge]

Maybe. but it could be good to bring it up though or just submit a PR yourself if you wanna. There are a lot of things I just patch and never submit for because I figure they got found so fast someone else has already done it, but sometimes that turns out to not be true.

[marble ledge]

./reasoning off

[tender cosmos]

thanks, i actually thought it would remove the reasoning completely

[latent monolith]

Where do I submit a bug?

[tender cosmos]

I wa just looking to remove the the spam messages

[lean tangle]

Well, wish me luck. So far my previous two bugs were fixed within hours.: https://github.com/openclaw/openclaw/issues/78057

[wanton plover]

Channel admins, i am requesting for this message to pass at it is with Stepete being tagged: " @Peter Steinberger It was trully a hard week for me too. The condition was a mess. Installed alternatives to OC, and so on. I tagged you because of this https://openclaw.ai/blog/openclaw-rough-week .Statements like this need a lot of courage. This acceptance has converted me to OC fanatic. Whatever happens, stay the way you are."

[marble ledge]

There is another knob in the config that turns off showing tool calls, I forget what it is off the top of my head, but you can do that too if you want even less.

[lean tangle]

spam is good. I even turn on thinking. I love watching my agent "talk to itself" and decide how to phrase things sarcastically.

clawsweeper on it...

[crude flicker]

if you're running local models that is 100% of your problems

[lean tangle]

At one point, I installed the two github related skills and asked OC to scan all commits over the past 15 days and determine when something might have changed to block .zip files. It couldn't find anything.

[lean tangle]

No, it's not. My problem is that it works in 4.23 and broke in 5.2. The 5.2 source is very purposeful in what it does. It has nothing to do with running local models.

Please, don't bash someone running local models just because they're local models. Read the conversation and understand what you're reading before you reach for the keyboard.

[crude flicker]

I'm also running local models and i guarantee that's why your claw is getting stuck in js files

zip bug seems weird

[marble ledge]

I don't use Discord for OC, but I asked Codex to look into the issue and described the problem you are having and where you think the source of the issue is. Curious just to see what it finds.

[lean tangle]

You mean as far as fixing the issue? That's really irrelevant. I don't want to fork from the mainline code. It makes updating too annoying. And using anything short of Opus for fixing code is (in my opinion) silly. In fact, because I'm not fluent in typescript, I wouldn't submit a PR from even Opus. If I can't understand and proof it, it's not going to have my name attached.

Generating source with AI is great. I do it with both qwen and gpt models (depending on the complexity), but I treat the generated code as something one of my interns wrote. It's my responsibility to understand it and review it.

[kindred scarab]

After i moved to hermes they announce new long-term support updates coming soon 🙃🙃

[lean tangle]

clawsweeper approves.

[marble ledge]

I had codex take a look at this, looks like a real issue. I'm not going to post the entire thing, but here is part of it:

Yes, I can confirm the user is probably pointing at the right area.

The Discord upload path itself still supports file attachments. The choke point is the shared outbound media loader:


extensions/discord/src/send.shared.ts (line 362) calls loadWebMedia(...) for Discord attachments.

src/media/load-options.ts (line 82) turns on hostReadCapability whenever a host readFile capability is present.

src/media/read-capability.ts (line 93) supplies that readFile when the agent is allowed filesystem reads.

src/media/web-media.ts (line 235) then rejects host-read local files unless they are buffer-verified images/audio/video/PDF/Office docs, or validated CSV/Markdown. application/zip is not in that allowlist.

[crude flicker]

this entire repo is generated code you honestly shouldn't even try reading it. everyone else is on codex

I'm actually just doubting that it's a real regression. fwiw i just confirmed zips working on 4.27

[wanton plover]

Nimble Enigma, good catch.

[lean tangle]

not sure if it's a regression or on purpose.

[lean tangle]

That was my understanding of the code. Clawsweeper agrees. 😄 I really just want to know when this broke so I can read the commit and figure out if it was on purpose or a bug/regression

[marble ledge]

It's a good catch, because it's not Discord specific either.

Seems it was intentional.

[lean tangle]

I wonder if I get a clawtributor role after submitting 10 non-rejected bugs. 😄

[marble ledge]

If you had stuff that was submitted and merged. You can already have the role. It's just something you have to ask for and show evidence. lol

[lean tangle]

🤷 I'm more interested in helping to find the problems (even if my ability to submit PR's is non-existant)

it's probably amusing to the developers to have a real software developer writing git issues. I actually include a ton of information.

[marble ledge]

I like finding stuff more than I like getting credit for it. It's fun. But it's not everyone's cup of tea. lol

[lean tangle]

this seems suspicious: https://github.com/openclaw/openclaw/commit/fc4da581b3e1a6c3ba2f3abd6ab4af4942773f6c

I need to test if I can send a zip via telegram in 4.23

[ionic wren]

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

fix them, then we'll talk ❤️

[lean tangle]

like I stated: I won't submit a PR if I can't review (and understand) the change before I submit it. That kind of disqualifies me from ts projects.

[ionic wren]

I mean logic should be the same regardless of language, it's just syntax right?

[lean tangle]

If my name is attached, I need to know it works, it works well, and it doesn't have any unintended side effects. I'm an old school programmer. I do it right the first time.

[fiery nacelle]

code is code, so i have code read code and code write code.

[crude flicker]

i put my openclaw pr in my claws name and made him pass the ci tests

[lean tangle]

okay, 4.23 allows sending .zip files to telegram (as well as discord.) I'm unable to test if .zip files are blocked to telegram in 5.2+. @marble ledge are you running 5.3 or ?

[lean tangle]

LOL

[marble ledge]

I'm on 5.5, let's try it.

Worked. I had it zip up the latest memory files from the last couple of days and send them to me as a zip file over telegram.

Downloaded file and unzipped files on phone.

[crude flicker]

you laugh but this is the way. the review process is all automated anyways and main changes every 30 seconds, it'd be a full time job to maintain the PR. or a 15 minute cron

[lean tangle]

is that actually a .zip attachment, or did the agent base64 encode it? (The icon seems odd. Mine has an orange square icon with "zip" inside.)

@ionic wren can I please attach a file to show @marble ledge what I'm seeing?

[frail vigil]

Guys how does openclaw actually contribute to your work? I tryed to use it a few times but honestly, the most things i can just do with claude code

[lean tangle]

oh, nevermind. I see that you downloaded and unzipped it already to confirm it was correct.

That's odd that it's blocked for discord but not telegram.

[ionic wren]

[lean tangle]

^^ found it's not needed.

[marble ledge]

Bruh, you are a piece of work

[lean tangle]

yeah, I didn't see (at first) that you already confirmed it

[crude flicker]

devops, development, github, etc

ie; everything

i don't really open codex/claude cli anymore unless it's a sticky situation. like i'm making a windows application where my claw doesn't live

[frail vigil]

Well what exactly is the advantage to claude code?

[crude flicker]

full memory, persistent env, its own github account

i barely ssh into machines or use a text editor anymore

if i do, it is to fix an openclaw config in fresh-editor. usually

[frail vigil]

Okay but doe's he actually think in advantage and sees thinks that are missing etc. like a human or more like that you can program over your discord chat

[crude flicker]

yeah they live on their own machines and can see everything and install anything

[lean tangle]

it does background tasks for me. For example, if I need to research something, I tell the agent to research the task for me, to use web searches freely in researching, but to confirm everything and to include sources in its report back to me.

It does reporting for me: my ISP is really flaky, so I run speed tests every 3 minutes and record the data to a prometheus instance. Every day, my agent runs a skill that finds every span where my ISP dropped out. It tries to find patterns in the data and reports that to me. Then I forward that to my ISP in my weekly complaints and demand for refunds.

[frail vigil]

Which model do you use and would you say that the biggest usecase for lobster ist coding?

[crude flicker]

like yeah i guess you could do that with a cli tool but i'm just on discord, i can do this shit at the grocery store or while stuck in traffic

[frail vigil]

I am mainly doing ecommerce and Coding more on the Site (well, we just started a saas company so it will be more in the future)

[crude flicker]

gpt 5.5 via codex oauth hands down, there's no other option

[frail vigil]

Yes i already have my own little lobster

[crude flicker]

it's just like

that's a big thing

[frail vigil]

Yeah well the abbility has my claude too

[crude flicker]

instead of throwing up a microservice i can throw up a claw and just ask him to make such and such microservice

[lean tangle]

I do everything locally. I don't share any data with claude, openai, google, etc.

[frail vigil]

Yeah everything else is way to expensive 🤣

[frail vigil]

Which model?

[lean tangle]

If you're happy with claude, use that. If you want something more attentive and personal, use OC (or one of the many other personal assistant things.)

[frail vigil]

What do you mean?

[lean tangle]

usually one of the gemma4 models. They work better with the "sarcastic" vibe I like.

[frail vigil]

Yeah i just started but i dont really see a difference yet...

[frail vigil]

why not ollama?

[lean tangle]

today it's ollama. If I'm able to get back to OC > 5.2, it'll be LM Studio.

[crude flicker]

claw live on box
claw serve microservices from docker containers in box
claw box may be virtual, he doesn't even know

i just poke him on discord

[lean tangle]

ollama serves the LLM to openclaw. openclaw provides the tools that ollama doesn't have

whoever submitted the PR for my .zip attachment issue: clawsweeper isn't happy with you.

[frail vigil]

Yeah same

But i want one that actually is like a sparring partner and controls my moves and contributes thinks by himself but i think that is not really the current state

[lean tangle]

okay.. so.. ollama has a little chat window thing. But all it does is chat. OC let's it talk to discord, telegram, etc. OC lets it use real tools. Do web searches, scan github repos, etc. OC also lets me switch models for different things by using subagents.

[frail vigil]

Okay so OC does make your ollama agentic

Well my claude can do the same things without OC

[lean tangle]

🤷 then use claude.

Why react with a cry? I'm not selling you OC. I'm explaining what I use it for. If claude already does those things for you and you're happy with it, you shouldn't change.

[crude flicker]

have you edited the SOUL.md to your liking

and messed with heartbeat

[frail vigil]

Not much really

just the onboarding

[crude flicker]

it helps a lot if you do it by hand honestly

the models will tell you they took care of it but they dumb

[frail vigil]

I mainly used cc to setup OC 💀

[crude flicker]

yeah same i just used codex

[frail vigil]

yeah i already recognised that haha

[crude flicker]

also if you get the heartbeat going and a bunch of crons and stuff it'll do cooler stuff

not recommended to do that on claude api

[frail vigil]

Yeah i think have to use him more frequently so he does understand my business better

[lean tangle]

I never got my OC to use heartbeat. It always uses cron.

[frail vigil]

Bank balance = gone

[lean tangle]

I even try to be vague thinking that would encourage it to use heartbeat. "Sometime tonight after 10pm, but before 7am, do this and that." So it makes a cron event for 10pm

[crude flicker]

it's really just about context, you need to just dump every important piece of info to the borg

[minor cave]

Yoooooooooooo

I’m here

[frail vigil]

Actually when i first set it up he didnt have permissions for crons and i was wondering why couldnt do a simple morning brief 😂

[lean tangle]

I once spent a couple hours telling OC about my cat. Now OC dreams about my cat every night. It's amusing (and actually cute)

[frail vigil]

Are you a Coder / do you have a Saas?

[minor cave]

What do you guys mean “dream” what am I not doing?

[crude flicker]

i have my own little thing where i just circle jerk with claws and make websites and then i'm bootstrapping the openclaw strategy at a startup using that infra

[minor cave]

My OC is definitely not dreaming is up all night doing lines of coke waiting for me to tell it what to do. Also gets super paranoid about my health haha

[frail vigil]

Nice

[lean tangle]

turn on dreaming.

[crude flicker]

active-memory plugin

[lean tangle]

https://docs.openclaw.ai/concepts/dreaming

[crude flicker]

does it have access to your biometrics

[minor cave]

Yes it does

[minor cave]

Thank you!

I have such an old version this might not be possible

I’ll just yell at it to dream. Dream now you POS!

[lean tangle]

see if the /dreaming command works.

[crude flicker]

yeah active memory was pretty broken until very recently i wouldn't bother if you haven't updated

i have a cron that scans every visible room every 6 hours and keeps track of movement. it's almost like dreaming

[minor cave]

Checking now

Nice

Well I think I broke it, haha first time in months

No just being slow. It’s really thinking about this lol

[dire tartan]

dang update to 5.4 broke my local plugins

[crude flicker]

is there an LTS fork yet where we automatically fix bugs on each release

[elder carbon]

Would there be any merit in starting my openclaw install fresh today. Mine goes back to clawdbot, and for the last 6+ releases, things just progressively get worse and worse. Claude Opus, GPT5.5, none of these tools can help me diagnose it, I spend money and problems repeat with delays, no responses in discord. If These AI models can't figure out what's worng, I sure as hell won't.

I was on 4.24 specifically rolling back after 4.29. Now i'm on 5.4 and 4.24 was better.

I'm gonna have to roll back to 4.24, it's unusable.

[cold jacinth]

Honestly, have codex or Claude look at it before you do anything. There can be some silly legacy settings one overlooks and they can cause a lot of trouble

Use one of the admin skills in Claude / codex

[elder carbon]

That's what I've been doing. I've probably spent $150 in GPT5.5 tokens and Opus tokens just trying to debug this thing.

I'm ready to roll back to 4.24 - I couldn't use 4.29 and I can't use 5.4 either. So either something is wrong with the releases, or I have some unique issue these AI brains cannot find.

[fiery nacelle]

im not having any issues really on 5.4.

[dire tartan]

which skills would you reccomend for this sort of thing? i see a lot on clawhub

[fiery nacelle]

have you tried cloning the repo locally and then having codex run through it? or are you only having them examine your live install?

[stray sleet]

it wasnt working! Had to do quite a bit of tinkering with the configs, but no it is working and its solved the issue. Awesome suggestion mate, great find!

[elder carbon]

examining the live install.

[cold jacinth]

I made one that’s currently in the #skills channel. It’s compiled of all the solutions to the issues I’ve had over the past 5 updates.

[fiery nacelle]

ive had more luck with cloning the repo and having codex patch that up when i describe problems im having, and then updating to the local openclaw with the patches applied.

[dire tartan]

why is it not on clawhub? not attacking just asking

[elder carbon]

I guess I'll give that a try - it's become very unreliable and a huge time suck trying to use it/fix it/have it break.

[cold jacinth]

Because I’m lazy and I don’t know how to put it on clawhub. I just made it yesterday

I have three claws. I manage them all with my codex, I had codex aggregate the findings and just push to my GitHub. It was one sentence lol

[fiery nacelle]

it kinda helps with applying patches that are specific to your system as well. But i can sa 5.4 is the best one so far. I'm only experincing minor loops from time to time, but i think thats just my model settings and have been tweaking them slightly to solve for it. Other than that works just fine. The loops are infrequent enough for them to be almost negligent.

[dire tartan]

5.4 does seem speedier

restarts are near instant

[sweet zodiac]

we back

[upbeat token]

It's not just you #announcements

[crude flicker]

oh nice announcing LTS separately later in May.

honestly that should solve most problems

[ionic wren]

Can you submit a github issue with your logs and findings? ❤️

[ionic wren]

It would, but we need to get there first

Takes a village, we appreciate any and all feedback

[crude flicker]

yeah honestly the PR's are going to explode if there's also lts branches

and the pr's are already out of control

[sullen locust]

it'd probably be best if lts branches had no prs

https://tenor.com/view/inner-peace-po-kung-fu-panda-4-calming-down-zen-state-gif-7572885950572222761

[ionic wren]

Hunnid

Only maintainers on that branch

[sullen locust]

no thanks timmyscanel i don't want your sonnet 4.6 cursor change that creates a cve 10/10

on the lts branch

[elder carbon]

I thought that was fixed in 5.4 otherwise that announcement would have mentioned on-going issues. But for me, it's a regression to 4.24 and below so far.

[marble ledge]

But...what's the issue?

[elder carbon]

It's just unstable. I can be conversing doing stuff, it'll just stop responding. repeatedly. the delays in responses, mid conversation I need to start over, loses context. restart gateways, this all worked like a charm before this 4.2x era. quick, always answered, not constantly restarting, I had no complaints. Now I've spent too much time wasted trying to fix it, and too much time and tokens trying to just make it work consistantly since pushing up into thie 4.24, 4.29, 5.4 now.

[marble ledge]

Do you use lossless-claw or anything like that?

[elder carbon]

Nope - and timeouts! can't tell you how many times I'm getting anthropic timeouts it says. Yet Claude Desktop and hermes agent using anthropic never once has timed out on me. but via openclaw it's happened at least 6 times today alone it says, just switches over to the backup model on m.

[upbeat token]

Keep it simple - if you update OpenClaw and it breaks and you can't fix it within an hour, it's probably the update not you. No update past 4.23 has been worth updating for my usecase.

[elder carbon]

That sounds about right - I'm going to regress to 4.23.

I never had anthropic 'timeouts' as the logs are claiming, and with these versions I do. Are 60 second anthropic timeouts a common occurance? I highly doubt it.

[ionic wren]

wait! before you do that, mind dumping logs and findings into a githhub issue?

[cold jacinth]

Anthropic api outages happen a lot

[solid pecan]

Welcome to the 4.23 club 😄

[cedar osprey]

😭

I burned 80m tokens to fix 4.29 I’ll be here for months.
(It’s perfect)

[marble ledge]

I guess that measn no one wants to join the 5.5 club, huh? 🙁

[solid pecan]

Nope 🙁 staying on 4.23 forevers

Well not forever, eventually when new models come out will have motivation to upgrade haha

Is 5.5 good though?

[ebon atlas]

wooooooooooooooooooooooooo
How yall doin today?

[quasi night]

I have not updated in a bit am at 5.2 and it seems to be working for me but am building tons of things right now so am worried to update and have fight to fix the problems I run into

[ruby crane]

you running main?

[marble ledge]

I mean, I think it's good, but I guess it's a loaded question. It's so hard to answer when people have very different setups, different models, different plugins, even different harnesses now. So many different variablies. Sometimes third party plugins get fucky, especially if they aren't updated.

[ebon atlas]

just stay on your update and finish what youre doing, then update.

[marble ledge]

Always.

[solid pecan]

Did you need to turn off ActiveMemory or is that ok now?

[ionic quest]

Does it come with tokens to fix the bugs?

[ruby crane]

I'd probably do it if didn't mean reviving my docker image build job

[quasi night]

yeah that is what am doing. I might run a 2nd one just to see what is in 5.5 while have my work flows building things for me lol

[marble ledge]

what's going on with activememory?

[lean tangle]

Would "I forgot" be too sarcastic?

[solid pecan]

When there is a large database of memories in SQLLite and in the Memories.MD file, it causes CPU spikes on load and makes the system unstable