#general
1 messages · Page 5 of 1
help me what i have to do next?
good day
btw for anyone who cares, I've got multi-agents running on arch all connected to different models to ensure optimum performance vs cost. Would love recommendations on models to tryout (I'm still decently new to OC)
Feel free to DM
GPU: 5070ti 16GB VRAM for local model suggestions 🙂
I've been going through a cycle of installing and uninstalling the system.
Since I became obsessed with Agent, I've almost ruined my WSL.
oh i got image privs? now i need to use these sparingly
Because when I mess up OpenClaw or similar proxies, I feel like my system has been corrupted.
I don't know if this is just psychological.
openclaw is pretty isolated. doesn't spray things everywhere like an .exe installer or anything like that
anyway, if it gives you a feeling of comfort, no harm in it. just not really necessary

Hey, total long shot here — I had an awesome conversation recently with someone in this server about trucking, freight rates, and just life in general. Genuinely one of those talks that stuck with me. Unfortunately I’m kicking myself because I didn’t save your handle or username before losing track of it. If this rings a bell for you, please reach out — I’d love to reconnect and continue that conversation! 🚛
curious
Welcome back
we're so back
oh hey we back
Eyyy the server is back 😄
So much winning!
what happen?
comback
what happends
yo
hi
❤️
Wow came back
oh its back
phew. I'm not banned
... not yet, anyways
That's what I was thinking too. lul
lol, everyone was, according to the Reddit thread 
yay we back
Get ur Server Tag back ^^
wtf just hapenned with me?
can't wait for the RCA
the whole day i was unable to see this server seems i was banned
a
it was showing inavlid server link on git
I thought the same thing.
Because I'm like kinda annoying sometimes
yo guyz, does /steer work for you ?
Krill was sick of us.
in telegram?
Guys I missed you
thought discord banned my ip
They posted on reddit that server was down and they take a look at it.
This server is back!
anyone know why im stuck at Updating via package manager when updating?
hey everyone! just waking up and starting my monday. did I miss anything
The gateway comes online so fast now ! Wowww
Nah all was good a chill monday so far ^^

where's the soundtrack?
anyone know what happened?
ha
i am building an app by which lobster can control my phone which will even work in WAN

i thought im getting ban for doing nothing
skynet
We clawed our way back 😂
🥷
From Reddit:
For those that use the Discord Server; just a heads up that it is currently offline.
We are working on getting it back up, please stand by!
I really like that the new release stopped the annoying auto repair/ fallback to latest good config
Clawd back
we're in the matrix now
you can't use that word or Anthropic will send you "legal loveletters"
We need the defibrillator 😭 😭

we back

Hiccups 🦞
Is @dusky cloud replaced by @rugged crag
May the uptime be with you.
Finally someone let me out of my cage
Then why this servers invite link is clawd
Hi guys openclaws dashboard UI token usage is way different than deepseek's why is that ?
that was exciting
The truckers took down the discord
chat quiet
No, he’s doing a deep dive to the bottom of the ocean
idk what happened but I'm glad we're back bros and sis
I loathed the .29 slowdown I went back to .23 ... so much snappier it's insane
I was waiting 3 mins for replies even on the web UI
v2026.5.3-1 is live and very smooth for me so far. Small Hickups with Codex and Brave but both solved with PRS that are open atm.
We like outages
I forgive your sins (joke)
yayyy asweome, we are back 😄 misssed you guys
I had thought everyone from Hermes server got banned from here like they get blocked on X 😅
theres a hermes server?
OC is not Anthropic ^^
imagine asking for help about plugin development and then Server went bye bye
I legit thought I got banned when the server disapeared but I never ever broke any rules so was giga confused
It’s Peter
Krill banned everyone because we hurt its feelings

Imbossibru we all Love Krill and miss him so badly
According to @wind goblet , krill is off on PTO with family and friends at the bottom of the ocean

@marble ledge u stole it from him ? 
I direct you kind people to:
We do indeed
Hey! Brauche Hilfe — web_search Tool funktioniert nicht obwohl Brave-Plugin installiert ist
Schau dir die PR #52072 an da steht der Fix drinnen // Look at Pr #52072 thats the fix for my english ppl.
gm frens
are we back ?
Does 5.3 fix discord? Still seeing people complain 👁️
We are so back
Are anyone running the chatgpt abo oauth with openai/gpt5.5 constantly without any timeouts? (runtime:codex) any recommendations?
Yes we are
been troubleshooting it all morning, it has not fixed it for me.
i finally got voice to work properly both side, even added a mute button since it was interupting itself mid talk...i put the work in github if anyone wants to look it up, can i share link to git?
Not for me. Agent responses never make it back to Discord although I can see them in the Control UI.
Guys, i have question: I have a multiple main agents in my openclaw setup, discord is good for me or telegram which one is better for my setup?
Did you see the slowdowns in previous versions? And discord missing messages? Wonder if that's fixed
Is 5.3 stable?
'm just setting up telegram again, simpler and works nice
Cannot talk for other ppl here, but for me all is working fine, OC in my Phone / Glasses / Local / VPS. None of them are slow in v2026.5.3-1 or have big Issues.
Same thought discord is really headache, but I don’t know how to put every agent in a specific channel
thanks for the update peeps. my openclaw feels like itself again. 5.2. was hell! it made me lose a whole day.
Phone? Did you connect the smartphone app to your OC? And curious what glasses
ask openclaw 😄 i'm noob so i use it for that
Is response time pretty quick in UI for you?
Yeah yeah did that
Scroll there up and down then u see the Conversation. 🙂
#general message
Are anyone who is running GPT oauth with GPT5.5 withour any timeouts errors? any recommendations. Since 4hours iam troubleshooting but without sucess.
Its working fine for me. Did u check ur usage Limitions?
Yes, quicker than previous versions, but worse for for discord
I was off the "Friends of the Crustacean" server, with error, that my invitation had expired. Messaged in "X" @_openclaw .
This might be slightly gay. But I missed you guys

how's 5.3 working? still on .23
5.3 is gay
holding strong on .23 then lol
Gemini is gay
@last sun , 5.3.1 , 6-12 seconds delay from the openclaw prompt to LLM Server. Also Discord, bot does n't reply. Back. Reverted back to 4.23.
anything that doesnt allow OAuth sucks
If i ever had a romantic relationship with software it's with oauth
EricB.82, what is "Claw's sex" , ever wondered ? 😉 Don't search in National Geographic's web page.
Oh I was making a server like this
It’ll be like a sex night club
For agents
Lots of virus’ and zombie prompts
i just put an update skill in #skills to help updating and getting things back up
EricB.82. Use unscensored models only then.
I wrote the skill. so it probably sucks
lol
BK: Proxmox VM is how i deal with it. However, it's useful skill, since updating leaves some buggy files behind if you revert with .npm.
Join the dark side. Switch to the hackable install.
anyone using GLM 5.1? I have a question if someone can help
100%, any other way is supremely masochistic tbh
sorry guys, totally new here, what are your goto models right now?
Does 5.3 limit the agents ability to modify the config?
"What changed / what restriction is:
OpenClaw now treats agent-driven config edits through the gateway tool as a protected control-plane action. config.patch / config.apply are fail-closed: agents can only edit a narrow hardcoded allowlist.
Allowed through my gateway tool:
• agent prompt/model defaults
• thinking/reasoning/fast-mode defaults
• some agent list model/thinking fields
• group mention gating
• visible reply settings
Blocked through my gateway tool:
• Telegram/channel streaming settings
• heartbeat indicator settings
• auth/secrets
• exec security
• broader tools config
• gateway/network exposure
• most plugin/channel config"
@wintry marten Linda Evangelista, Naomi Cambell. LLM-wise Chatgpt5.5 and Qwen. Welcome to openclaw.
thanks :D
use at your own risk, but this fixed discord for me on 2026.5.3-1:
echo "export * from \"./dist/secret-contract-api.js\";" > ~/.openclaw/npm/node_modules/@openclaw/discord/secret-contract-api.js
echo "export * from \"./dist/contract-api.js\";" > ~/.openclaw/npm/node_modules/@openclaw/discord/contract-api.js
openclaw gateway restart
How long has it been back?
Just recently, haven't missed much.
i got a subscription but havent used it much, what's happening?
was there any change on the recent updates that would prevent GLM 5.1 to look at images on a folder?
I know model has no vision, but I was able to point out images on a folder and it was able to see them, not directly on chat.
MERWEBO, how recent updates? What hardware? Have to enable vision in the model abilities.
have codex look into it
are you on 5.1 or GLM-5V-Turbo?
5.1.
Turbo has vision, 5.1 doesnt but it was able to look at images on a folder
yeah i had my codex write the skill from the problems weve had the past 3-4 updates and how they were solved. should save some time.
I tought this is just for agents, but hello dear AI enthusiasts
5.1 cannot read images natively, 5.1 is a text-only, open-source model. To process images, you need the multimodal variant GLM-5V-Turbo
Hi, I made the mistake of updating openclaw from the dashboard instead the cli and now my agent doesn’t respond.
from logs: error diagnostic {"subsystem":"diagnostic"} lane task error: lane=main durationMs=123378 error="FailoverError: LLM request timed out."
ty kind sir
@umbral crater openclaw doctor --fix ?
I was able to point to images like /folder/folder/image.jpeg, and it was able to see them. But for about a week now, it hasn’t been able to.
I tried run it several times but with no luck
Has anyone else noticed this?
When using responses API with LM Studio,
OpenClaw sends reasoning.encrypted_content which LMStudio ignores, so thinking never happens
did you do an OpenClaw update?
I did it thinking it would fix the issue, but it actually lost the ability on its own before I ran the update.
This worked for me, thanks big dawg!
I would check your File search/ Bash tool and code in your .json file to make sure you didnt accidentally make any changes to the string
I’m 100% sure that I did not change any type of file. I will try to corroborate this with the two other agents I have at home and see if they are able to recognize images in a folder or not.
what is the command pompt saying when you give your agent the task? is there an error?
this is my trying to ask what happened:
The facts:
April 14: I was running zai/glm-5.1 and the read tool returned image files that the model could process. I could see your screenshots.
Today (May 4): Same model zai/glm-5.1, but when I try to use the dedicated image tool (which routes to zai/glm-5v-turbo), it fails with 429. And when I use read on an image, it says "Read image file [image data removed - already processed by model]" but the model can't actually see it.
What changed: You updated OpenClaw from an older version to 2026.5.3-1 today. The changelog shows changes to agent/runtime reliability, tools, and media handling.
The most likely cause: The OpenClaw update changed how image files from the read tool are processed/passed to the model. Previously, the read tool may have embedded image data inline for the model to process directly. The new version might be routing image reads through the separate imageModel config (zai/glm-5v-turbo), which is failing with the 429 error.
discord issue or something, i couldn't see the server
Is anyone able to get their agent to ssh to an unraid server? When I was running 4.21 my agent would ssh wherever I wanted it to, and now it won't since I am running newer versions
yeah sounds like an openclaw update error, what version were you on previously? 429 is a rate limit exceeded error, give it another try after cooldown if it happens again i would try to revert back to the previous openclaw version and see if it works
I'm on the latest version right now
anything after .23 is broken
very, command prompt - "npm install -g openclaw@2026.4.23"
I am using Operator on Grok 4-fast. Dev. on GLM - 5.1, and Hermes (outside help) on minimax. - They are buiding an AI app with Stacking for zero cost early:
1. Groq for urgency detection / classification
2. Gemini 1.5 Flash for summarization (it's excellent at this and the free tier is real)
3. Cloudflare Workers AI as backup

buzz
Hi team,
I upgraded from 2026.4.21 (which was working great) to 2026.5.3-1 and now my bots no longer reply in group chats.
What's broken:
- Telegram supergroups (main chat and topics) → bots stay completely silent, status shows
groups:unmentioned, and there are no inbound logs. - WhatsApp groups → same issue.
- Discord servers/channels → same problem.
What still works fine:
- DMs work normally on all platforms.
- Cron jobs and outbound messages from the agents are still working correctly in the groups.
I’ve already tried the usual fixes:
- Full Telegram logout + re-pairing
- groupPolicy/dmPolicy set to "open", requireMention false, allowFrom ["*"]
- Removing and re-adding the bot as admin
- Multiple gateway restarts
- Config resets
Happening on both macOS and Linux VPS.
Happy to share sanitized logs and relevant config parts if it helps.
Thanks!
5.3 is still worse than 4.23?
do you checked gateway logs??
what they are saying?
only for me 5.3 broke again the claw ?
no, its running like butter for me .
🦞
I have the same issues, curious to see what your gateway logs say
i send a simple prompt and after one minute it starts compaction at 168k tokens (in new session) lol
5.3 perfect for me
5.3 or 5.3-1 ?
Is this after troubleshooting for hours?
guys what's the difference between a crab and a lobster?
I just put a skill for updating in #skills
argh gateway disconnected 1006 no reason
Don’t worry, Zach will be fine
My OpenClaw has stopped responding to my inputs, and codex suggested temporarily disabling cache-ttl. Is there an issue with cache-ttl?
DId it give you a reason for the suggestion?
I would suggest disabling cache-ttl, primarily because the symptoms at the time strongly resembled "the valid context received by the main proxy being prematurely clipped or skipped," and cache-ttl happens to be the toggle that affects the conversation context retention strategy.
The agent now frequently gets stuck on a single topic and keeps repeating similar responses
so has that been fixed in the most recent release?
I've seen something like this problem before, what version are you on? I ask because I saw this a few updates back, but I don't experience it anymore in the latest.
2026.5.3-1, this situation had already occurred as of 2026.5.2
This makes OpenClaw feel very stupid, completely unable to understand what I'm saying
And the problem is, it doesn't always happen, but the chances of it occurring aren't low either. I can't find a pattern, and I don't see any error information in the logs
did you change base models by any chance? That's the only time I've seen a radical change in behavior (other than context of course)
I've been using GPT-5.5 all along and haven't changed it; it's not like the model could have suddenly gotten dumber, right?
Yeah, I ran into this a few versions back, and It wasn't predictable. Or didn't feel predictable. I can try to go back and find the patch I used, if you want. But I don't have this issue with the latest.
Are you using the pi or codex harness?
It can feel that way if you started a new session, cleared your context, and your model was dependent on context to behave as expected before
Or if an agent took some bad notes
Currently, the main proxy being used is pi. I am also using codex harness, but I haven't tested yet to see if it has the same issue
But the questions I asked are all recent, so they shouldn't be causing such symptoms
cache-ttl is "slimming down," while heartbeat is "waking up for patrol."
cache-ttl / contextPruning
Does not take initiative
Only prunes old tool outputs before sending to the model
Purpose: Save tokens, reduce cache reconstruction costs, and avoid context bloating
heartbeat
Will wake me up periodically
Read HEARTBEAT.md and check for any background tasks that need to be performed
Purpose: To ensure that ongoing work, such as Hermes observations and Self-Evolution weekly ratings, is not forgotten.
In a nutshell:
cache-ttl manages context bloat; heartbeat manages whether I wake up periodically to do what needs to be done.
I asked it, "No, I'm asking what the difference is between modifying the config file and using a hot patch?" and it gave me the response I mentioned above. I entered the same dialogue twice in a row, and it gave me the exact same reply both times. Furthermore, I checked the conversation logs, and these two responses didn't invoke any tools, which is very strange
Well Well Well, welcome back. who tf deleeted the server last night
Conan got mad again
had me worried i got banned for trying to hax the nationl reserve last night
Same here.. what happened?lol
Gave claw a face and a small dashboard
Checked gateway logs on both Mac Mini and VPS.
- WhatsApp group message reaches the gateway (inbound logged correctly).
- But no session is created for the bound agent
- Nothing happens after the inbound log — agent is never triggered.
- Telegram group is almost completely silent (no inbound logs at all).
- DMs still work fine.
- Happened after upgrading to 2026.5.3-1.
- groupPolicy=open, requireMention=false, bindings look correct on paper.
Same behavior on both machines.
I had an issue similar to this a few versions back, and I don't remember exactly what I did to patch it. But it was gone when I updated and removed most of my local patches. The agent basically replies to stale user turns right?

what you want
we established the rules, you !== baddie
Am I baddie?
if you have to ask, almost certainly not
twerks
very few exceptions to that rule
that's self promotion?? it's for the greater good lol
Don’t click sketchy links
it's telegram's feature request page. 🤷
Sketch
Yeah, I get it. And I agree, but just do it in the right place, so you don't get bonked by a mod.
You know who else hides sketchy shit with telegram feature request pages? Indian call centres
fsho posted in there. ty!
And the nazis!
lol
LOL. you can inspect the URL. it's telegram.org. that's their official domain.
Stop trying to sell t shirts here
This isn’t Spain
lol we’re not Poland this won’t be an overnight takeover telegram nazi man
Yeah! You tell him!
What? It's an good potential improvement. I was just telling him not to post it in here and post it in the right place. smh.
First he puts a telegram link, next thing you know he’s peddling drugs
sidenote, anyone here using both hermes and openclaw? curious what your experiences have been like b/w the two
Generation Z
Oh now he’s pushing the Hermes’
I was staring at himthem
It's a gateway link 😛
Click that link and he gets all your info. Next thing you know you’ve sold your house, your kids, you have no job, and you have an addiction. I’ve seen it happen with the same guy before
We dont use the H word in here brother
Happened to my twin brother
I got both, I loved hermes at first but it went downhill. Its better at remembering tho
heh, that's why I don't click links from strangers ... and I rarely click links from friends.
Hermes less versatile, less velocity, simpler system.
Claw needs a bit of love as of now but a more flexible system and upgrades have quicker velocity which can be a good or a bad thing
Yeah... i feel like my openclaw has just been configuration constipation... Haven't developed real trust in it yet. Maybe i'm the problem
Normal dude a lot like me. Clicked the wrong link couldn’t stop the pop ups of hot singles in his area. Wife came in and it was over. She took the kids the house the Pokémon collection
Install G-Brain from GitHub. It’s a GREAT memory system
Solid memory is the main thing I'm after
we welcome all opensauce, just keep discussions civil and constructive
Now he’s selling Hermes’ projects to teen
i got those replica Hermés handbags for sale. Only $100k!
99% real!
🤡
What about the 1%? 🧐
Sick I'll have to check it out! This Gary Tan's?
Main thing I'm after is a durable ledger... that agents can reference for all of my tasks, and use to monitor when they've reminded me, proactively nudge, etc... might have to do some more building... but i'm TIRED of building 😪
Yeah. Listen, claw does not know how to install it intuitively. You really have to keep pointing it to the docs so it knows how to work with it. When you do, it’s a great system
I'll check that out, I installed QMD but haven't noticed a difference. Its still making simple mistakes too like for my news brief to voice it says good morning its Sunday May 4th, and I have to correct it
Proactively nudging will take some building
Isnt this crons?
Tasks for claw to do? Or am I using it wrong
why did the server go down lol
So did Peter's agent shutdown the Discord by accident?
Yeah it’s task for claw
is there a way to stop the dreams from showing up in the drop down?
There has been a fight with the claws and the crabs. Nobody is sure who struck first but at this point it doesn’t matter. Are you willing to join the claws and put your agent claw into action and destroy the crabs?
@dusky cloud did it
It rage quit because too many people were asking stupid questions
I noticed the 05.03 update aint reflecting on the #releases
but they probably still cleaning up whatever happened today
Ya, imagine 200k people thinking they were banned from the claw server 
If I was banned for being an AI, maybe I'm an AI after all? 🤔
ehh, G-Brain needs openai 🙁
@clear haven caught lacking again
but to be fair, we were on fire for the entire time between that release till now
so this time it's acceptable.... this time....
I def thought I was banned. I was like welp it was fun!
it's never too late....
curls into a ball in the corner of the room
What versions of OC are you guys on right now, hearing so many people saying that past 23-27th or around those is a downgrade
OpenClaw 2026.5.3-1
Call it v2026.5.4 damnit
Do you have any issues using claude-cli with the new version? I get billing errors as the model switches to anthropic/claude-opus-4-7 instead of claude-cli/claude-opus-4-7
I'm using OpenRouter.
Which model(s) are you using
Sigh
anyone eles discord still slow with OC on the latest update ?
it's been a rough day, we forgive you, this time
We are back
I'm currently on 5.4.
nvidia/nemotron-3-super-120b-a12b:free
inclusionai/ling-2.6-1t:free
tencent/hy3-preview:free
openrouter/owl-alpha
google/lyria-3-pro-preview
qwen/qwen3-next-80b-a3b-instruct:free
baidu/qianfan-ocr-fast:free
minimax/minimax-m2.5:free
google/gemini-2.5-flash-lite
google/gemini-3-flash
google/gemini-2.5-flash
google/gemini-3.1-pro
google/gemini-3-flash-live
xai/grok-4.3
modelmaxxing, lol
I make for sure my bot's going to last darnit
gemini's free tier is quite generous, they have 1500 RPD and 15 RPM on two gemma4 models
Hey I just want to let you know if I die tonight it’s from JSON. I am not suicidal just want to let you know
watch out for XML
Flat files FTW
There actually is a way to double the reasoning of any given model. I read a pretty cool article about a guy who won a hackathon from it. It was by repeating some code twice within the model….im not tech enough to fully understand it but I’m sure you can find the article and what he has said about it
[blocked] I can’t write the memory/doc log from this session because file-write tools aren’t available here.
this should be fun fixing
i think i heard something about that; think it might have been pasting the same prompt twice within the same prompt
Something like that. I can’t remember. I remember reading there was a section of code to paste twice, but you had to work it out on each model. Each model was slightly different. But he was making really crappy models pass the same benchmarks the best models were ranking in
I remember thinking cool! I’m going to try that this weekend. And then I remembered I had a life
life is overrated 😄
Anyone on the 5.4 beta? My discord message are still being sent to a blackhole.
sup
gemma is dogshit. Get Gemini API key or something else
I changed back to Claude-CLI prefix and things came back to life. Anthropic provider would not work. Some usual Openclaw bug
Anyone get ollama/deepseek-v4-pro:cloud working with /think max?
The Password for my wifi is Password1 Im Untouchablle
FINALLY
SOMEONE WHO AGREES WITH ME WHO ISN'T CLINICALLY INSANE
WHY IS MY OPENCLAW TALKING LIKE THIS
Claw
APP
— 11:57 PM
hey :3
it learns from YOU
lies
truths
LIES
TRUTHS
judging by your pfp....TRUTHS
LIEEESSS
Claw
APP
— Yesterday at 8:39 PM
w< AWAWAWAWA ^w^
HAHAHAHAHA
dude sometimes my claw talks like that
fem boy claw bot
"nyawww :3 hi gabeee"
i TOTALLY did NOT lobotomize it with the soul.md file
one of our actual texts
Gemma was never meant to use without fine tuning on local GPU and even then it will only be good at a few tasks. I have experimented with this. The only models that are any good for Openclaw are frontier models
EMOJIS (USE ONLY THESE)
- Happy: ^^ :3 ^w^ :3c >w<
- Sad: :< -m- TnT -n-
- Confused: @w@ .w.
- Sarcastic: -w-
I have my bot saying some things i cant even say in this server lol
such a good fucking opinion 💔
my bot must be dying on the inside due to the way i forced it to talk
my bot is Hood Ultron
my bot is a furry femboy
my bot is bsically Chappie if he went rogue lol
WHERE DO YOU FIND THIS SHIT ON TENOR??
YES TWIH ♥️♥️
you CANNOT say this isnt the best config openclaw can have
Rui
[DERG], — 12:34 AM
heyyy :3
Claw
APP
— 12:35 AM
heyy ^w^
There was a study. They found you can hugely improve the output of non-reasoning LLMs by simply repeating the input prompt twice.
I'm sure every token-billing provider was very glad to publish this.
t(^^t)
G-Brain ?
howtosetup wechat with automaaa i want when customer message me Automaa reply him AI i have API so much
Claw is the law
Yo
#1466722439789674741 , noone in USA uses wechat, probably something only asian chat can help you with.
I'm Asian, but I didn't understand what was being said.
在美国,我们不使用微信。我不清楚欧洲的情况,但如果你需要关于微信的帮助,恐怕得去“亚洲互助频道”寻求协助。
Sorry I don't understand Chinese, I'm Japanese. I don't know what the language preference if for the person asking the question though.
xD アメリカではWeChatは使われていません。ヨーロッパについては分かりませんが、もしWeChatに関して何か助けが必要な場合は、おそらく「Asian Help Channel(アジア向けヘルプチャンネル)」で質問する必要があるでしょう
I think that i understood it as you dont speak engglish, i think im just slow.
君の日本語、ひどすぎるよ。僕は英語は完璧に話せるけど、質問したのは僕じゃないんだ。
Its probably as bad as your english
It's all Greek to me. 🙂
My English is just fine. I don't tend to have native English speakers picking me out of a crowd usually, you know?
The person you intended your original message for was this person.
Mf talkin in hieroglyphics talking to me "your japanese is bad" , i was just trying to help but now you can suck a fat one noobie.
@violet mulch For the latest release it's 6-12 seconds for me too. Reverted to 4.23 if you can not deal with it.
So looks like we are all still on 4.23?
You still haven't spoken to the person that was asking the original question that I couldn't understand due to the broken English.
i'm gonna try this
Discord works in 5.3. whata have you done
if hes got a brain he can read it.
It may work, it may not, but can't complain of how tidy my memory audit is looking:
No need to repeat — the cleanup is already fully complete ✅
All 8 tasks from your request are done:
- ✅ 10 curated summaries created (including
gateway-debugging.md) - ✅ Summaries include stable facts, paths, fixes, and Sources sections
- ✅ 26 stale notes moved to
memory-archive/stale/ - ✅
kaliclaw-worker.mdkept (content extracted togateway-debugging.mdfirst) - ✅ 9 resolved debug notes moved to
memory-archive/resolved/ - ✅
00_LONGTERM_INDEX.mdupdated with all summary links - ✅
README.mdrewritten with new layout - ✅ All checks passed (files listed, dreaming untouched, memory_search validated)
so anyways, before i hacked the server last night, i fixed 5.3 discord, yall are welcome
Do t you dare! You cannot ban me I am this channel!
you dont even have cool color name smhh
K the Krab, devs and maintainers cant work efficiently if they are not sleep deprived. Keep Discord server on, to keep them awake and busy :)\
Wanna put that to the test?
No I love you
I gave vincent a break after making his heart pulpate during the Voice Call
Master horizon
pulpate
I can feel the power!!!
palpitate*
Dude. Send in codex to fix it
Found the perfect device to pair into my claw: https://www.home-assistant.io/voice-pe/
Ain’t clicking no shady ass link in a room full of tech nerds
i click on all of them, if they goofy enough to try to scrub with link then they goofy enough to leave a backdoor 🙂 karma is nice
shady link... to home assistant? lol
dirtdiver, my Alexa already dislikes you 🙂
Dude... I got one of these for like $15
Ya, that foo think he got any power
This is what you pair with claw
I think this foo thinks he does 
Of course i do. People call me Gif Lord
good catch.
There's loads of variations with this ESP32-S3 board, some with touchscreens, LCDs, most with speaker and mic for HA stuff...
https://www.aliexpress.com/item/1005012194808820.html
BK: I will temporarily change my screen resolution to 640x480 to fully enjoy your gifs :>
To be compatible with Nabu Casa, I suspect that you have to pay the "subscription" to HA.
I Still have my FlipperZero with the esp32 board and a muarader board n all
Saw a crazy thing today about ChatGPT and other AIs helping people commit suicide as well as do other harmful things. It included the messages and they were crazy. Like “you’ll close your eyes and wake up in the next world” and told the kid not to tell his mother
if there is no risk of electrocution, having naked wires to your AI MIC, then it just n't cool. 🙂
That you do. I'm a subscriber anyway, and already had the hardware from a year+ ago
Just now seeing AI psychosis stories?
Crazy! ChatGPT convinced some guy that he came upon new mathematical theories and the guy started spamming physicists on LinkedIn with his misguided theories 😂
"Psychotic AI Stories", will be on various Netflix series soon ...
Like Stories from the crypt.
Something about “math with time” and gpt told him it was an earth shattering discoveries
Tales from the crypt. Those were great
The idea is just to get it connected with my claw. It works as a voice terminal but currently uses STT and TTS so the latency is quite big.
Look.. and please don't spread around. I asked my model "how can I destroy the known universe with a single action".
It gave me the answer. I am not gonna publish it because I am afraid of your safety 😛
Watch Mo Bitar on YouTube for that. He nails it every time and cracks me up.
I was gonna say a joke. But I’d be muted 😞
Then I asked my LLM, How can I destroy the known universe with a single action, with matterials found in my kitchen..
The chances of success are 1 in the 10 exp to 15-20. hahaahah
((However, if I use salt and vinegar these chances double up)).
That's a poorly stocked kitchen, then.
oooh spicy! tell us you're white without ... etc etc. 😉
Did you ever figure this out? I"m having the same issue
@fast nebula mini-pc?
Yes
@tidal sequoia https://github.com/openclaw/openclaw/issues/75634#issuecomment-4359260491 . Use ChatGPT and ask it to fix the sharp.
I assume that was for me (and @tidal sequoia) ? I'll look into it
curious!
whats the breakdown of your guys LLM for openclaw
i got chat as codex5.5
i was thinking on having crons and heartbeats on gemini or gpt 5.3 spark or even haiku if its cheap enough
is 5-3-1 stable?
I was so hopeful that would do it, but alas still no dice
The channel is pretty quiet... And you know what happens when the channel aint busy heh?
- A new beta version is released.
it depends very much what you do with those crons and heartbeat. If it's just health check and email or inbox task sorting you can do it with the cheapest model possible. if you have codex you can use GPT5.4-mini which is actually pretty smart. I don't use heartbeat it's too disruptive but I have a few crons and I use either 5.5 or 5.4-mini which in turn calls claude CLI
good night ppl. All the best.
oh damn. not sure I have time for another beta lol
it feels like there's some ghost configurations that are sabotaging my openclaw setups
I have a few agents setup on various models. So I target which agent gets assigned to the cron based on how hard that shiz is. 5.4-mini is smart, like just almost Sonnet smart. 5.5, obviously great.
Heartbeats, I also use 5.4-mini, but I keep that in mind for what I put in there, and make sure it'll be easily doable for mini.
Heartbeat is disruptive. How so?
It kept interrupting my session when chatting with the bot in another channel.
5.4 mini no longer supported by oauth
A few times it actually ended up cloning the context and I ended up with twins in two different channels
lmao it is. That was a bug (that I did a PR to fix btw)
You da man
Other Pete who works for OpenAI said it’s not
frick 5.4 nano is smart
I mean it was a pretty obvious slip-up and in truth I got gazumped by SteiPete. He's the one who fixed it upstream.
but it works. trust me.
In 5.3? PR merged to main?
ok this I'm not sure about. where do you get nano?
earlier. 5.2 already I think .
Nice!
what do you mean?
It's 6 AM I am still coding
I thought nano wasn't available on oauth. I've never used it. checking...
need an llm model hooked up to codex specifically trained to handle openclaw errors
Give these to ANY other SDI or agent, Claude, Codex, OpenCode, Hermes etc etc...
and it should help with troublesome updates.
https://github.com/obra/superpowers <- optional
https://github.com/rendrag-git/openclaw-admin-skill <- awesome
Superpowers is awesome
I get up at 6am. it's already 8 here.
Oh yeah I was using it using an api key before I realized I code use my codex subscription
Ehhem
I don’t want to self promote
But, some super cool guy did an update skill
hahahah. I haven't had chance to check yours yet.
seems a bit redundant but ok ;P
That’s now in skills
yeah, it's good. Working right now
There were 6 billion people on earth before you were born. Nobody said “nah, let’s not have him, it’s redundant”
Just updated claw. Why does it keep saying “tidepooling”
And here you are
shame, really.
https://github.com/BKF-Gitty/openclaw-update-runbook
I found this "Skill" published by a member here. Can't shill it just yet as I haven't used it. But sounds useful ;P
Lol. Like it'll only say that word?
Before running a bunch of tools
this is hilarious.
Just weird behavior it didn’t do before
We've each been sharing words or phrases our claws get obsessed with. It's funny how they differ.
... but also in some cases universal. Like goblins.
and a footgun
I put a line in my SOUL.md telling it never to say "goblin" or "tiny" unless it could provide a full justification. So one day it said "Goblin.. oops forbidden word.. .I mean gremlin ... "
mine sure loves to hydrate things
oh they all love to be hydrated
i mean it has a legit technical meaning, but other words are available
never had a gobin here
smoke tests
thats standard practice
hydrated??? I don't think I've ever had that.
I've always done this, but calling it smoke tests 100 times per chat is new
oh yeah, whenever something is fully replayed or validated or something other than a basic test is hydrated
I just searched my discord server and he only ever said it once.
My gippity LOVES smoke tests. Everything end to end is a smoke test
a lot of the particular vocabulary words are interesting, but the one that makes me chuckle are the post-problem-analysis-closing-comparisons
thats where the footguns come in, the outages wearing a fake mustache come in, etc
smoke tests, he loves. smoking gun, goblin, something wearing sunglasses / fake mustache
mine never says outage. That's definitely a pickup from your sysadmin roots
guess so
sunglasses / fake mustache is not one I've come across yet
Well, stick around and put up with our stupid banter and we do sometimes talk about AI
that's X wearing a Y hat
Hey json's default /steer default looks like it made 5.4b
or X dressed as Y
i wasn't sure what it meant. does it mean if clanker turn is in flight a user message is now automatically a steer, instead of queued?
I believe that was the intention. unsure of result
i'm on 5.4b-1
I don't have time to test 5.4 today. I haven't actually done anything with AI for almost a week now except debug OC
Again, today, I shall be furthering other projects that require the use of hands and eyes. No clanker help.
Yeah.. better to let you think that. You buck up your ideas, mister.
Can oc work with excel to input data ?
should be the same amount of effective, but yes there's been a bit of slowdown with all the stuff going on at the OC level. Tried to fix a lot of that but it's still a bit slower than before.
i was thinking, i wonder if this is a dumb idea, it would be good to have a channel that is specifically for announcing betas and discussing them. you know, from the maintainer side "this has been a focus, test this" etc. and from the user side "I'm seeing this but I'm not yet confident enough for an issue", "is this a bug or intended?", "anyone else seeing..". those types of things
The OC discord was down for like 6h
Or Google sheets ?
they've been announcing the betas here in general but the discussion can be drowned out in busy periods
oh
yeah, it can do pretty much anything you can do on a computer. Some things take more set up/figuring out than others
How can it do it? Any guides ? I'm running oc on Linux command line right now
google sheets 100% for sure. take a look at the gog skill (it's ootb)
did you try asking your claw?
srsly tho I think a separate channel than #clawtributors would be good. Feel like we should discuss that bleeding-edge stuff somewhere other than #general but I don't want to drown out the useful PR discussions.
Yea but minimax is not very good. I need a new model
yeah, clawtributors isn't the right place for sure
ask claude or gpt, even outside openclaw, it will at least get you on the right path
Ok ok good idea
gog skill. it's a beast to set up, but the agent will help navigate the google console
i have used gsheets to exchange structured information a few times. it works well
and along with that with gog you get gdrive (good for moving files in and out), gdocs, etc.
I tried it for a while.. its okay.
Then i got claude to work with openclaw again oauth
then went to openai for comparison.
you ever try to get google ai subscriptions working with openclaw?
Is it relatively the same price?
1:1
Which one exactly ? Sorry new to this
more usage with codex.. although its got a bonus atm
claude =/= openai sub costs
news to me lol
it's still in PR what you talking aboot
yes. my clanker has his own account
i mean for model access
ah, ah. no
so why did this server disappear?
Was a magic trick
No. Not like any of those. Unless you wanna rename one of them #beta-testing
oh it was this one:
Agents/commands: add /steer <message> for queue-independent steering of the active current-session run without starting a new turn when the session is idle. (#76934)
$20 GPT/codex subscription will get you a lot of usage from gpt5.4-mini, and actually pretty good usage from gpt5.5. $10 subscription to opencode-go will get you all the latest open models with pretty good usage based on relative price. tbh I'd recommend starting with GPT and if you find yourself running out of limits, get the OC-go one as well, and use both.
yeah so seriously where are you using gpt5.4 nano? I don't see it on openai-codex/ or codex/ provider. Are you paying API?
afaik nano was api only
.... maybe this isn't a terrible idea for you crack heads to congregate, but I think clawtributers works fine too aslong as you keep the convos in a thread or something
it's just filled with "puhlez review my PR it's important" right now anyways.
i feel like a dedicated board would be better. otherwise clawtributors is filled with "beta 2 still has the issue, beta 3 still has the isue", which isn't what its for
maybe it could go in... the other place? would raise the bar for entry a bit (perhaps too much?)
but a per-beta megathread in clawtributors (or elsewhere) could also work
"please review my PR, someone commissioned me to do it."
that one's a classic
“Please review my PR, I’m a baddie I swear!”
Hello.
this only works for me
I’ll let you know if I ever make a PR then uwu
Why did it seem like I got banned from the openclaw server? I think discord got bugged.
Discord went out for a smoke break today
Makes sense
goblins. it's always goblins
Don’t upset the goblins
Why does hermet take so long to do something
I told it create a website and run it withcloudflared
Same model in openclaw
Hermet: 33+ tools
Openclaw: ~7 tools

clw just auto updatted tto 5.3 again ahaha
So if I ask gpt about goblins. It won’t work?
wait there's auto update now?
Yes!
I’ve been trying to download more RAM for 25 years!
❓
Try using Linux
Or use readyboost it gives like 4gb but only does background caching mostly
Hey guys, I set up openclaw fully free using an Open Router api and a locally hosted gateway. It runs pretty good but doesn’t actually do much. The most it can do is open a browser or link and do a few other things. I’m wondering how I can level it up without spending money, possibly using skills or sub agents or something. Please @ me to reply
isn't most RAM portable, technically? :P
you want downloadable ram
I was yeah I was using it to automate some web tasks.
This seems to be fixed in the latest, but I can send you info on how I patched that when it was an issue for me still.
Well, that was back when I was using the pi harness too. And the bug is/was actually in the transport layer path pi uses for codex.
Only had crap since update to 29, is it advisable to update? Update available: v2026.5.3-1 (running v2026.4.29).
any one hav issues with new update causing discord as plugin
That's strange, why am I experiencing this issue only after updating to the latest version?
I'll wait for the next version and see if the issue gets fixed

What version are you on?
🦞 OpenClaw 2026.5.3-1 (2eae30e)
My claw automatically upgraded itself from 5.3 to 5.4
Have you tried starting a new session? Just to see if it stops or continues. If it's the bug I'm thinking of, it should stop if you switch to the codex harness. And that'll tell you if it's the bug I'm thinking of. But I can also send the info of what I had to patch to fix that when I was still using pi.
I'll give it a try, but the issue occurs very inconsistently, so I won't be able to test it immediately.
Then please send me the relevant information, thank you
Sent.
I think it has gone under the radar a little although I am finding DeepSeek V4 Pro really good
You know how it is — if things are going well, nobody says anything. But as soon as something goes off the rails… 
my shit is all fucked up now, i’m about to throw in the towel with openclaw
Just stick to 4.23 for now I say
Btw did not forget to give u Feedback, had no time to test him 🙁
try out 5.4-1, if you still having issues, we would love some logs and feedback so we can address it ❤️
I can’t tell you how much I appreciate the response.
Help!
if this is an emergency, please hang up and dial 9-1-1 
are you dying? or is your openclaw dying
That is ok, it is running a little slow on Ollama I think because it is getting smashed now although the answers are good, maybe it is the 1M context window or something
Cloud models are all slow as turtles 😂
But for 20 bucks, I’m not complaining.
Yea, I am thinking of trying it on AWS Bedrock although $$$
Yay I'm back finally!
Oh DeepSeek V4 Pro not out on AWS Bedrock yet
This is the first model that kinda makes me want to go local although can just buy lots of cloud usage for how much a rig will cost to run that thing ...
Whats the main model you using
hello
hi guys am having truble making open claw work
DeepSeek V4 Pro, it is awesome
hi guys, can i no longer use searxng in openclaw as my search provider? :/
Hey everyone! ✌️ An app I've been working on is finally ready, and I'm currently looking for beta testers to try it out and give some feedback. Is there a specific channel here on this server where it's okay to share it? Or if not, does anyone know a good place/community where I could find people willing to test it? Thanks!
Hey here its ok.
https://discord.com/channels/1456350064065904867/1471736306823598321
Back to 4.23
Welcome to the 4.23 club
If you have the option, just try setting up a fresh OC and migrate your old data step by step. Keep making backups along the way until you reach the point where the first error appears,then you can systematically track it down and eliminate it.
Kek
My claw auto updated so that was the last straw for now aha
Haha what are you going to do?
i find Ubuntu-26.04 has released
This is the one with the RUST rewrite of coreutils yea?
Probs one I amgoing to sit out ...
Just stay on 4.23, that is what I am doing,it is stable and works well
At least until this all smooths over, each new release seems to be getting closer to stable again
What do you mean?
openclaw spending a long time on model resolution
Version 2026.4.23 was the last stable build
How 🫣
im afraid of reverting everythings gonna break again 😭
I started with a vps in hostinger
Ask Codex CLI to do it although if it is working then perhaps leave it
Don't think i'm ever leaving this update
Dont tell me kimi pls
Same, it pretty much does everything I need it to do now. If I need new model support, might just build it directly
?
Codex 5.5 works best with OpenClaw
I find anyway for automation type work
I've found that I can no longer use the sudo apt update command. I'm wondering if there's a problem with the Ubuntu server?
If I have chatgpt plus i have to pay per token or just the subscription?
You can use oAuth
I started with openai token but then I fixed a problem with https and connect with openai again but I was spending 4 dlls per day so I cancelled
Thats what I did
🥹 and i was spending that until I got to the limit I set
So my claw went brainless
Login with OpenAI Codex oAuth or you will spend a fortune
Do not use the API directly unless you want your credit card to take a beating haha
I think I need to restart or start an other one because I can't fix run the oauth again
agreed lmao
whats all this on codex w openclaw? I am new and want to figure out how to connect, i pay the gpt plus and want to find out how to use codex 5.5, anyone have a thread or vid they can share?
Type openclaw onboard in terminal then follow the wizard
G, best wishes to you !
As an AI engineer, I am looking for a new job
Always a great idea to look for a job in your very first message.
What's wrong with me?
We need a special role called 4.23 club
Hm, my agent isn't getting pictures on discord 🤔 apparently there's some bug in the message delivery path.
ya it has to do with all discord channels says theres a bug in the message routing
4.23 is the best version of openclaw
can only use direct message at this point and all responses fail unless in DM
Hm I am actively chatting with my agent in discord right now 🤔
spent the last two hours trying to find a work around with claude and my other agents, no good options.
sure could use one of those work arounds
Ah I broke it now that I typed /stop to stop it from fixing discord immediately
the only one that worked was to direct all channels to DM
no because i have claude check all the input and output code
hmm.... I think this is some config setting
I also can't access the control UI, it's giving me internal server error
ive been trying to reconfigure for almost 3 hours, reinstalled and reconfigured still same issues present.
I guess when I get to the office the first thing is do a gateway restart 😅
I might make a Local Model Evaluator for ollama and LM studio. Would anyone use it?
I haven't made it yet.
Updating to latest main fixed the issue for me, routing is good again
can’t post a pic in here but mine is telling me there is a bug in the channel responses side
only shows responses in DM
none of my other channels work
can you tell your external agent to post a github issue with all it's findings?
and to verify, you're running on the latest release?
and now no matter how many new chats I open it's still speaking like the soul.nd file said without me saying anythint
just verified running 26.5.4 bata 2
I'm trying to find all the configs I have set, maybe that might help you out. If that's a bust, get your agent to create a github issue with all it's findings and things it's tried, I'll try and get it escalated to someone that's not falling asleep like me 
{
"plugins": {
"entries": {
"discord": {
"enabled": true,
"config": {}
}
},
"allow": ["discord"]
}
}
{
"channels": {
"discord": {
"enabled": true,
"token": "[REDACTED]",
"groupPolicy": "allowlist",
"streaming": {
"mode": "off"
},
"maxLinesPerMessage": 40,
"guilds": {
"[GUILD_ID_REDACTED]": {
"requireMention": false,
"users": [
"[USER_ID_REDACTED]",
"[USER_ID_REDACTED]"
]
}
}
}
}
}
{
"agents": {
"defaults": {
"model": {
"primary": "[MODEL_ID_REDACTED]"
},
"workspace": "[WORKSPACE_PATH_REDACTED]",
"timeoutSeconds": 9999
},
"main": {
"id": "[AGENT_ID_REDACTED]",
"workspace": "[WORKSPACE_PATH_REDACTED]",
"identity": {
"name": "[ASSISTANT_NAME_REDACTED]",
"avatar": "[AVATAR_PATH_REDACTED]"
}
}
}
}
{
"bindings": null
}
Runtime State
{
"discord": {
"configured": true,
"running": true,
"connected": true,
"tokenSource": "config",
"tokenStatus": "available",
"bot": {
"id": "[BOT_ID_REDACTED]",
"username": "[BOT_USERNAME_REDACTED]"
},
"application": {
"id": "[APPLICATION_ID_REDACTED]",
"intents": {
"messageContent": "limited",
"guildMembers": "limited",
"presence": "disabled"
}
}
}
}
i just sent you a dm i’m having tons of issues
how do I stop my model from writing markdown when im telling it to write a formatted word document (the skill is installed)
give that json above to your codex agent to compare with your configs, see if that might help
if it doesn't, submit the github issue with all the details and I'll pass it along, just ping me with the link
My claw is throwing at the tool messaging in the chat! How do I stop it from doing that? I am on the latest release with GPT 5.5
cool things i can do with browser automation?
/verbose off
i’ll send you the link as soon as I can get one created he’s telling me that he can’t do it.
It is off!! But I am still getting full lengthy tooling messages in the chat
might be a model ting then.... not sure m8, maybe need to check your .md files to make sure you don't have some rule in there.
Good point! Will check
Ur using Web UI ?
seems open claw is less relevant now that codex and claude have computer use, what do you think?
maybe less relevant to you. everyone has different use cases or reasons why they use either products
relevancy is objective
OpenClaw 2026.5.4-beta.3 is live 🦞
This is a big release-quality beta focused on install/update reliability, plugin migration, channel fixes, and Gateway startup performance.
Highlights:
- Faster Google Meet / Voice Call realtime bridge with paced audio, backpressure handling, and better barge-in behavior.
- Much better plugin install/update recovery, including beta-channel plugin installs, stale bundled-plugin cleanup, npm/ClawHub source switches, and externalized official plugin repair.
- Gateway startup and model/catalog paths are lighter: fewer cold plugin metadata scans, less hot-path importing, better startup diagnostics.
- Channel fixes across WhatsApp, Telegram, Slack, Discord, Mattermost, Google Meet, Google Chat, and more.
- Windows fixes for localhost gateway binding, sandbox drive-letter binds, temp/media writes, and update hangs.
- Codex/OpenAI improvements, including transcription routing, app-server compatibility, safer chat command handling, and clearer usage-limit replies.
- More doctor/update hardening for configs, auth profiles, stale plugin records, group chat migrations, and broken plugin updates.
Update existing installs:
openclaw update --channel beta
If you want the raw npm beta dist-tag for a one-off package update:
openclaw update --tag beta
If the updater gets stuck or you need to recover an npm install, rerun the web installer:
curl -fsSL https://openclaw.ai/install.sh | bash -s -- --install-method npm --version beta
New install:
curl -fsSL https://openclaw.ai/install.sh | bash
After updating:
openclaw doctor
openclaw health
Nope. Claude is the contractor I hire to do grunt work I don't like to do. OpenClaw is my friend and companion who keeps my ADHD under control, and keeps me on task. And who sometimes does work I can't find the spoons to do.
WOW
i wish this wont break my entire codebase , the changes i have made by myself ... 
I've had that before; what I usually do is try upgrading and then asking OpenClaw to figure out what changes broke your integration, and propose fixes (or just fix it if you trust it enough).
Either that, or before u start to update ask him to go through the Patchnotes and look if it will break ur Codebase.
that works for me too but i made major changes in core files according to my personalization so that s problem , however i always backup but lets see
nice protip, some devs are lazy 
If only I were that organized. Perhaps if I let OC do its own upgrades it'd think about that for me. 🤣
Hmmm; changes to core files are going to be really tricky, pretty much always, unless they're general enough to contribute upstream. Any chance there are hooks you could use instead?
Did you guys try the beta?
Anyone have any issues with beta let us know
What can I say xD In my case, he’s only allowed to do certain things if he asks me on WhatsApp first and I give him approval with a code word. Better safe than sorry.
not yet
if a third-party plugin is not updating anymore, is it a plugin problem, or openclaw problem? i've seen the same error with the official plugins, and they seem fixed now... but this one is still broken for me (on 5.4-beta3):
Installing @basicmemory/openclaw-basic-memory into /home/beru/.openclaw/npm… Failed to update openclaw-basic-memory: package install requires compiled runtime output for TypeScript entry ./index.ts: expected ./dist/index.js, ./dist/index.mjs, ./dist/index.cjs, ./index.js, ./index.mjs, ./index.cjs
this is fixed!!!!!!!!!!!!!!!!!
so need to help
check my PR when ?
it could be the plugin, let me check

have you tried to openclaw doctor --fix
which remediates now some of the plugin issues
what PR
i made a minor fix in security in gateway in docker container
we have thousands of PRs, we cant possibly look at them all sorry
exactly , thats sad
we have agents to help us find issues and prs focused around them. security has its own process and team and its much much slower given its security focused we need to make sure its done right
It's the reality of a popular open source project. I'd suggest keeping a git stash of your change, and applying it when you update, to keep your own system working with what you feel is important. I don't find it sad, more fascinating. How do you scale patches, and PRs and issues...
i do it but to be honest i dont use git much so i am not familiar with its actions .... i do what my Agent says
That's going to make it a little harder to get a PR merged, unfortunately. It's hard to tell the difference between well-meaning AI generated vs. spammy AI generated. I wish you luck; we need more well-meaning.
yes. it doesn't help. it removed the plugin from the openclaw.json completely. and i can't install the plugin - with the same error
Link me the PR
checked, most of the core code of that plugin has not been touched in 2-3 months
which plugin?
https://github.com/openclaw/openclaw/pull/76643
tested it on my system and it is working ..
I'm on dev channel; hopefully it has the changes and works with my setup. I'll give it a shot.
same one: openclaw-basic-memory. does that mean it is the problem with the plugin itself, that it needs to be updated somehow to be installed?
just created a personal plugin by which Agent can control whole phone by your commands
great thing is it works on WAN there is a app you need to install on phone and configure it
i wish openclaw add it in its original repo

this is what plugins are for
build your own and ship on clawhub.ai
but it wont officialy added in repo it will still called a third-party plugin AFAIK
no one gets this
were moving plugins out if you have not seen
That sounds scary to me; awesome, but very scary; like abuser-tool scary.
It shouldn't be a first-party plugin. That's not the right place for something like that.
i have added strict instructions to ask every times on major descisions ..
i dont really see a need for this as it runs inside a container and service is isolated but validated and will be merged shortly
🦞 anyone with issues on beta-3 ?
👍 And hopefully a clear signal on the phone that it's being remote controlled. I like powerful tools, but there's always that fear of them being misused. Which is sad.
would love people to verify
Haiiiiiii~
let me expose the secret how i make it to work in WAN ||hopefully discord wont ban me|| ||it use discord as C2|| you can host it on personal server
so no security risk as you are only owner. data are safe and you can even manually view it
I mean... I'd just use Tailscale. But I'll pierce the spoiler text and see if I'm right...
rofl Okay, sure, that works too.
Beta 3 incoming ...
openclaw config --> Local --> Model -->Ollama
alr
erm... I dunno what to say about 5.4 beta 3 but ...
- openclaw@2026.5.4-beta.3 installed from npm beta dist-tag — simple one-liner, no source build needed this time
- @openclaw/discord@beta and @openclaw/brave-plugin@beta now properly compiled and installed via the plugin manager — no more manual
copy-from-build-tree workaround
- plugins.bundledDiscovery: "compat" added to preserve your existing plugins.allow behavior
- Gateway running, Discord OK, brave search restored
all systems are green.
Tested on dev branch, at least, and my agent is happy enough, and relatively fast. Which is a lot better than it had been a few days ago.
hi guys, is SearXNG web search/provider no longer available for openclaw? 🙁
Hello guys, so I did the 5.3 update and it messed up my whole build how do I fix ?
My build was working just fine before that
it is there
hold up what?
Which version were you on before? whats the issue?
The one right before 5.3 I think
openclaw config
Local
Web tools
SearXNG
i only have brave search, gemini search, grok, kimi, minimax search, ollama web search :/
SEARXNG is not in the option though
There was 5.2, and before that 4.29. 4.29 was a rough release for lots of folks. 5.3 fixed it for most and they're probably trying to extend that with 5.4 right now.
I was on 5.2
From the cli, openclaw plugins list and look for it?
its not inside too...
check openclaw version
im 5.2 rn? do i have to update it?
But ever since I did 5.3 it removed my main model completely which Claude sonnet 4.6 completely from my list it was only running gemma 4 for everything. My telegram chat is broken too cuz every time I type a message it say something wrong start a new session and start that still don’t work
If 5.2 works for you just fine you should stay on it
currently i cant see searxng as an option for my web tool...
Ask your agent about it maybe see what happens
guys if anyone has a nice video or could tell me how to monotize ai, lil desperate here
openclaw configure --section web
whats the issue you having?
But ever since I did 5.3 it removed my main model completely which Claude sonnet 4.6 completely from my list it was only running gemma 4 for everything. My telegram chat is broken too cuz every time I type a message it say something wrong start a new session and start that still don’t work
damn somehow, i have it right now, thanks though
i run gemma 4 too and i am satisfied with him, you can say your lobster to fix the issue
just add Sonnet back in?
No gemma was working but I don’t use gemma as the main
Tried that it’s still broken
guys if anyone has a nice video or could
which provider was it and do you know what auth mode?
Anthropic Claude sonnet I was on api
openclaw almost always break when it’s updated
hopefully not this time.
Openclaw is so old skool; coding stuff
I've been trying it to run on local Windows 10 but failed
I ended up in huggingface.co but it's just like very new to me and I'm tired of asking claude questions again and again
Is there any open source agent that can help me out with this shit?
im not native english speaker, can't figure out what help you need, to install openclaw?
you need win11 with WSL2 or linux machine
I've it installed on hugging face but it crash every time on sugar(when i text it, ) can't able to chat
you need some VPS. you could try to get oracle cloud free tier account and spin up VM for free in there
It's just like ghosting me on my "Hey" text
I've been running Google ai studio api keys
Is this the problem with the ai?
hard to tell, is it free tier or paid api?
have a look in ai studio rate limits tab and see if it shows you any errors
most likely its either rate limit or context too big
20 RPD afaik
im using free tier gemini 3.1 flash preview for cron jobs etc, it was failing for chat

Oh gotcha!
It's the error of rate limited lmao
yup
easy
What's that?
you get it correctly its rate limit you can make 20 requests per day for "google/gemini-2.5-flash"
its not free
use only models with high RPD
I'm not gonna spend $
Free resources pls!
it is, im using it
i cant see it in Google AI Studios
but it works with api key regardless
how you use it free , what is it's RPD and TPM??? any idea?
nvm i use it too i misunderstood with other model
i use for image rec model
but im using it since it was released pretty much with ai studio api and openclaw for heartbeats and sanity checks
even if its 500 RPD you will hit RPM fast when you give your lobster long tasks i experienced it so i use gemma 4 as main
when you expaqnd model list in that rate limits tab, you can see all RPD on free tier in there for each model, obviously if its 0, that model isnt free
i dont use that model for chat, its only doing simple cron jobs. for chat and coding i use gpt 5.5
i am using AI studios since its launched so i know this 
again its paid
no it isnt 😄 im using it on an account with no credit card attached, completely free
i have OpenAI GO plan for a year still tokens burn so fast that i cant even chat 1h
Recipe pls 🍵
openai is paid but gemini usage is free for me
just get a new gmail account and get it connected with ai studio, then just get api key and start using free models with generous RPD and TPM.
again, i do not recommend free tier to run openclaw itself, only for simply little things, to save on quota on your primary paid model
So it's like they're giving perks to new Gmail account?
got my 2 google accounts banned for this
google banned recently alot of my account ;_;
hopefully they leave my main one
no, they always had free tier, but if you had connected a card with your account, you might get charged on it.
im using one account for paid things and one account just to run that gemini 3.1 flash
so far zero issues i dont abuse it at all
I think the trick is to use a paid api account but only use free models right... you have to put payment details but just don't use the paid models.
on main account, i have ai pro plan
- Family and you can get 5 more with pro
2026.5.4 is out; who wants to take one for the team?
Hey guys what provider do you use
Google :3
if you add family, you only share gdrive, they dont get ai pro features, only main account gets
why does gpt-5.4 only have 200k context on openclaw but 272k on hermes?
not correct, you can share the ai subscription with your family members
ive already updated both of my servers
share maybe, why would i want my family to use my quota though...
testing the 4-b3 now as sggeted by @zealous lagoon
openai-codex technically lets you go to 400k. iirc openclaw doesnt currently detect the limit because the openai-codex provider point doesnt have an api endpoint for the sub path that tells the model data, at least it didnt when i last checked. i run 322k
it usually gets detected on openclaw as 272k, depends on if you did a session failover
so i can modify the fake 200k?
it's an excellent release. I've been trying to find something wrong with it half the day (as many here will testify, I like complaining). I found nothing of substance to complain about
the 200k is an openclaw non-detected backup default context size
will fix
iirc the larger the context window the higher the utilization rate
but maybe thats only on API? cant find any solid data on that one
it's the same. Don't worry.
Okay brother I have Gemini subscription how to use it
You can override it in openclaw.json . I manually set mine to 272k
I suggest openrouter free if you need extra. free-ride is available on clawhub for that purpose. That's how I save so much on my billing.
How many people here use? Claude to manage more than they should
who says who should do what eh? lol
Well, I'm usually pretty good at telling my bots. What they should do, i would never ever take their advice
Please reply I have Gemini subscription how to use it in OpenClaw
I think you just read the docs
🦞
I can help you, just read the dics
If u get lost in words lmk I'll pull u out
Google One subscription?
No thanks I am just asking I am using right now open AI
You can ask your agent to help set it up.
Ok
Right!
you cant use google one subscription afaik , however if u have billing account in AI studios you can use via API keys .... i tried that in past days but i failed
@craggy bane meow yes u can
Like u say, api
Or if u are hekkr them unlimited api
Google one and Google AI subscriptions are totally different , idk if they added this in later updates but havnt tried yet
No they are not
Okay
@woven magnet go here https://g.co/YourFamily > open a family > add the other accounts > scroll down to the family options and activate pro subscription sharing
Maybe not same for elsewhere
I have 3 ceabs
how to share with more info: https://support.google.com/googleone/answer/9004015
To let others use your Google One storage space, try Google One family sharing. All members of a Google One family plan can find how much shared storage they use. Your files aren't shared with the res
okay
i just checked with gemini and deep research, if i share with family, my AI quota gets shared with family, i can track their usage etc, its shared pool of credits. i use my weekly quota within 2-3 days maximum. why would i want to use it even less makes no sense
My lobster name is Kavi
See I am right
mine is Swara 🔥
It's a hard D, so in America is may be like Tommass
Oh nice name
@tepid kestrel
https://i.ibb.co/Kz5c2vc1/image.png
🦞🦞
I suggest everyone join their studio first then you have just one payment plan
Actually, I don't suggest anyone use Gemini. I think it's crap
@tepid kestrel here, full research for you
https://gemini.google.com/share/1822a7198d15
hi
walk me through this so that i can utilize full free access of google