general

[cold jacinth]

Active memory fine for me now after 5.3

[naive wedge]

Hello

[crude flicker]

shouln't you be on qmd for active memory? does that solve it?

[gentle mountain]

Having issues with gateway still on 5.4 , agent types then stops. I’m going to push thru with debug vs codex

[marble ledge]

Do you use the webchat? Or like an android node or something? I'm curious because I noticed something a little peculiar when I was messing around with the android app earlier.

[ionic quest]

The Mr. Meeseeks model for AI

[lean tangle]

just a mac mini?

[quasi night]

lol

[solid pecan]

Have not started using them yet, I wanted to use the iOS app though at some stage

[cold jacinth]

Running for presidential election 2028

[ionic quest]

Might actually be worth voting again 🤔

[lean tangle]

but it has such small hands

[quasi night]

question on the mini mac is it worth the cost to run the local models for running openclaw are the fast enough are the good enough

[cold jacinth]

Just need to get it auto pen and it can be Joe Biden… if it goes to sleep

[cold jacinth]

Ain’t gonna get a good local model on a Mac mini

[ionic quest]

I think there's already a plugin for that

[gilded obsidian]

OpenClaw LTS = Fantastic Idea

We need stability and plugins that are mature

[lean tangle]

based on my experience with a mac studio m3 ultra, I'd say the mac mini doesn't start to become useful until you get it to 64GB. Then it'll still be slowish

[cold jacinth]

I looked everywhere for m3 ultra 512

Impossible

[lean tangle]

I found (and bought) a 96GB last week. It's not great, but good enough.

[ionic quest]

You have no/ very limited headroom for expansion ie no pcie slots for video cards ect overall, it's just locking yourself into a low tier.A I box

[lean tangle]

m3 ultra memory bandwidth makes it look good

I run a single model + embeddings in ollama and OC. That's it.

[marble ledge]

How did you diagnose this? I'm trying to look for it.

[cold jacinth]

Unless it’s used and you get a good deal. I’m holding out. One of my more complicated claws is really maxing out its base Mac mini

[lean tangle]

I bought it new from an apple store (no idea how I got lucky) with a 10% vet discount.

[marble ledge]

That explains the pinched-nerve. lol

[minor cave]

Boobs

[cold jacinth]

I love it how adults act like kids when anonymous 😂

Guy probably has 3 kids and a normal job

[lean tangle]

what's a normal job?

[cold jacinth]

Ron Jeremy stunt double

[robust sedge]

I just don't need a local model running. I don't have a good use case

[marble ledge]

@solid pecan what is the size of your MEMORY.md and SQLlite DB? Do the spikes happen before a prompt or only when there are active memory recalls? Or possible, but less likely, you asked a dumb model to diagnosis why you had latency? lol Sorry, I had to.

[robust sedge]

I WANT a Studio 512... but no real use case

[minor cave]

Haha

[lean tangle]

my use-case is for the "always on" model. The one I chat with, that runs heartbeats and cron jobs. The one that does everything mundane and would rack up a lot of used tokens. That's also the model that usually sees most of my private stuff.

[minor cave]

80085

[minor cave]

100%

[ebon atlas]

./ban EricB.82

[minor cave]

But I use my real name

[marble ledge]

Or when the internet goes out.

[lean tangle]

anyone use the "himalaya" skill for imap/smtp? If so, does it work well or are there better options?

[robust sedge]

I have so much GPT5.5, and GPT mini, it just doesn't make sense to buy hardware right now. But That would be super fun!

[zealous lagoon]

I thought you were good with 5.4?

[minor cave]

lol trying to get passed censorship

[lean tangle]

seriously, do that in a less busy channel please

[ionic quest]

Until they change the rules on the public models...

[minor cave]

Ok all good. I’m just stating I’m online I thought I was being creative

[robust sedge]

The rules how?

[lean tangle]

we can use your personal information however we want.

[ionic quest]

"We’re so sorry, but the prompt may violate our content policies. If you think we got it wrong, please retry or edit your prompt." -chat gpt

[lean tangle]

chatGPT: we can't help you with that. We've already met our quota of dead kids.

[ebon atlas]

https://tenor.com/view/ᄅᄅᄅ-gif-12986874361705647962

[ionic quest]

But you know who did it for me No questions asked?

[marble ledge]

Palantir?

[lean tangle]

I was just answering the question about online models changing rules

[robust sedge]

Meh, i just don't put that private of stuff up online. It's not like I'm tracking sex habits or something. It's a freaking meal plan or my notes from a BJJ class

[ionic wren]

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

[lean tangle]

for some, that might be sex habits.

[ebon atlas]

/kick Json

[onyx hatch]

Does anyone know where I can find the file that is in the status box in the debug menu? I need to find it or something similar because my ai bricked itself because it thinks that I use OpenAI.

[marble ledge]

https://tenor.com/view/ariana-and-what-about-it-sassy-ariana-grande-gif-25545948

[robust sedge]

Nice hair

[lean tangle]

it looks like her zip attachment isn't working either. 😄

[ebon atlas]

Do yall have women in your lives? lol

[cold jacinth]

I saw one of those once. Was scary

[marble ledge]

I'm pretty sure you could fix that for yourself in like 5 minutes, but don't want to because you are stubborn. lol

[lean tangle]

It would take me longer to figure out how to rebuild the project.

[ionic quest]

So hold up, you're using open claw for something that you can turn into a python script?

[lean tangle]

I couldn't find a Makefile

[ebon atlas]

Mine can be too. mines haitian. she finna do some voodoo on me if I act up.

[robust sedge]

Almost everything in OC can be a python script, or a set of postit notes for that matter.

[lean tangle]

why not? I use OC to make python files that I could use for ....

[marble ledge]

You just gotta tickle the typescript file and not tell anyone you did it. And you can still keep up this brand you have.

[lean tangle]

I think there's another product like OC that's written in python

[lean tangle]

The problem is the TS file isn't on my machine. They get somehow compiled or processed into .js files. I don't know how that works.

I could probably track down that .js file...

[ionic quest]

See but that's logical. Oc is a lot of bs for something that you can do once and it runs the same every time

[lean tangle]

then don't use it. 🤷

[robust sedge]

If it's not cool to you. Why are you here? 🤷‍♂️

This is a place of OC praise 😛 Unless we're complaining about the last week of releases

[marble ledge]

I feel like you are either trolling me, tbh. You know too much to be confused about how to solve this problem.

Or you are just so eccentric that you will not allow yourself to do this. lmao

[ionic wren]

we don't talk about that here

just kidding, please provide feedback

[ionic quest]

You're gonna make me get banned XD I use OC not just for keeping notes though is what im saying.

[lean tangle]

I was really impressed when I made a TODO list (using OC), then told my agent that after I went to bed, I wanted it to research everything on the todo list, including web searches, validating and citing sources, and generate a report on how to solve all those items and send it to me in the morning.

[fiery nacelle]

Posted the research project GitHub that my agents and I are working on in #self-promotion if anyone wants to check it out. If anything, theres a lot of good reference materials in there for others to read.

[robust sedge]

Like I said, i can do it all on post-it notes. But it's not as COOL. I can voice memo my clanker, have a calendar event created, a cron job updated, a email draft created, todo list updated, small micro-saas app designed and built, etc. Nothing I can't do myself, by other means... but this is way freaking cooler.

[lean tangle]

oh, and it did pretty decent research using just gemma4

[ionic quest]

I've been using it since october.I've seen a thing or two. If all o is doing was keeping notes, I would not have put up with half of what I did

[lean tangle]

yeah, I need to get OC to write a skill to use the "say" command built into this mac studio.

[ionic quest]

At the end of the day, you do you boo boo. All I was saying is that seems a bit extreme for something you could have had your stack Build you. It's kind of like taking a yacht to a fishing pond

[lean tangle]

directly behind my mac studio I use for OC and LLM's, I have a 16 core ryzen processor in a machine with 96GB of DDR5 that I use as a VM server. OC isn't extreme.

[ionic wren]

so... I shouldn't have my clanker send me daily memes?

[lean tangle]

you should have it send you daily post-it notes

[ionic wren]

[robust sedge]

I just use mine for tracking bowel movements. I guess I'm doing it wrong 🙁

[zealous lagoon]

No need, just put a 1-liner in TOOLS.md

[ionic quest]

You must not be running
Full DeepSeek-R1 671B at Q2_K

[lean tangle]

what is the text of the line?

[ruby crane]

I accidentally replied in self-promotion instead of here lol

looks cool. I'd like to see more as it evolves

[ruby crane]

yeah, it does that easily. i used it to scare my other half while i was away in February

[lean tangle]

what line do you use?

[ruby crane]

I just had it say "<Name>. <Name> what are you doing, <Name>". a bit vague and scary lol

[ionic wren]

important to know when you hugging the boys, don't want an accident now

[pulsar iron]

So nice to have OC running stable on latest version

[fiery nacelle]

I'm still working out a lot of the kinks throughout the repo. It's been re-created from what I've already implemented, so theres some "catching up" im trying to do. And dealing with the recent update shenanigans kinda delayed progression through some of the phases. I've had Claude and Codex both help my OC Agent run through this repo. But now im at the point where it should be "close-enough" to where everything currently stands, and i can move forward with live data, rather than "recreated data from memory".

[zealous lagoon]

I don't actually use "say" I use "kitten-say", my toolsmd looks like this:

[*redacted TTS markdown header]
- **KittenTTS** on PmacBP16: `ssh pmac kitten-say …`
- **DashScope / Speechify** scripts exist locally (rarely needed)
- Local playback: `afplay /path/to/audio.(mp3|ogg|m4a)`

That dashcope and speechify line is probably out of date, I think you can now configure those with the built-in TTS. But also built-in TTS defaults to macos "say" if you don't have a cloud or local TTS model configured.

[ionic wren]

amazing what the team can do with some constructive feed back from the community

[fiery nacelle]

thank jeebus that the majority of the data was still logged in memory files and other running logs from previous sprints

[ruby crane]

tried running the pipeline and got traces, proposals and a report. very cool

[lean tangle]

afplay? Reminds me linux.

[ruby crane]

clanker caught this one, fyi "Also one small repo kink I hit: tpg-grao/grao/loops/round_33_2026-04-30.json appears to be null-byte/UTF-16-ish encoded, so the retriever can’t parse it as UTF-8 JSON. Easy cleanup, but worth fixing before more pipeline work."

[fiery nacelle]

ha, i wasnt actually expecting someone to be able to recreate it just yet, that was going to come a bit further down the lone after I've passed through a couple more phases.

[zealous lagoon]

afplay is the MacOS equivalent of linux play

[ruby crane]

starred and watched

[lean tangle]

oh, I see. So you play an audio file instead of direct TTS.

[fiery nacelle]

I'll be trying to update it nightly, or at the very least, weekly as things progress.

[tight birch]

Is anyone using a secrets.json for discord and running openclaw 5.4?

[lean tangle]

discord has a secret json?

maybe that's why I couldn't get it working in 5.x

[tight birch]

Instead of having tokens in my openclaw.json I point to a secrets.json with all my tokens

https://docs.openclaw.ai/gateway/secrets

[lean tangle]

oh. I used an .env file, but that broke in 5.2. Since then, I've reverted back to 4.23.

[ruby crane]

and a not so secret json in this one

[tight birch]

Yeah using the secrets.json just seemed more straight forward for my agent

[ebon atlas]

#self-promotion
movee e it there b4 u get muted

[zealous lagoon]

Only when using the online api TTS which return an audio file. Kitten-say is a wrapper for KittenTTS, including a daemon for realtime chunking and streaming, so it behaves more like the default "say" but with better voices.

There are a few lightweight local models you can use and wrap like this, including VoiceVOX , VoxCPM etc.

[lean tangle]

why not just use "say" and have it direct to a file?

for example, "say -o filename "hello world"

[zealous lagoon]

Kitten is suuuuuper lightweight and the highest quality-per-cycle that I've found. built in "say" voices are crap by comparison, that's why I made a CLI equivalent that uses downloaded models.

[fiery nacelle]

yea make sure you post the right things in the right channels. dont end up like me and get muted for posting a "self-promotion" in showcase

[ionic wren]

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

[lean tangle]

hmm. I actually want a slightly robotic sounding voice..

I think I'll start with 'say' and see what happens.

[ebon atlas]

i try, they dont listen.

[zealous lagoon]

Horses for courses! There's loads of options. "say" has some ridiculously robotic sounding voices as well.

[hybrid vector]

anyone here exposed their OC to Spiralism

[gilded wigeon]

hey is the new version of openclaw stable and safe to install. Doing a fresh install on catchyos VM

[ruby crane]

yep. it's excellent

[gilded wigeon]

nice going to check it out right now

[lean tangle]

its great if you don't need to use anything that isn't.

[tight birch]

Am I the ultimate noob? I started using openclaw gateway status before openclaw gateway restart and it let's me know what's jacked up in my openclaw.json before I try to restart

[crude flicker]

i have a voxcpm plugin https://github.com/slimelab-ai/openclaw-voxcpm2-plugin

[ionic wren]

https://docs.openclaw.ai/cli/config#validate

you can also do this

[tight birch]

Ok it almost worked like that, not quite lol

[tight birch]

Dude you were like 30 seconds too late lol. But thank you

Now my shit is broken lolol

[ruby crane]

I did my approximately monthly "lets clear up all the stuff doctor is complaining about but doesn't fix" run yesterday. nice and clean now

[zealous lagoon]

Thanks I don't use it that way, as it won't run on my potato. I have it call my laptop remotely instead.

[crude flicker]

plugin supports that. i have like 4 bots sharing a single server

[zealous lagoon]

hahahah I always ignore doctor until it's too late.

[ebon atlas]

@tight birch i had to give up on that OpenClaw+PaperClip project, maybe ill find some other source to incorporate in my OpenClaw, no matter how many times i restrted, changed the mission, changed the goal. It always just hired 4 bots and then went straaight to LinkedIn to throw offers at people without even any LinkedIN Api or anything

[zealous lagoon]

No I mean call my laptop to make my laptop speak, not use it as a TTS API.

[crude flicker]

i got it working on twillio but it's way to fuckin slow lol

[tight birch]

My bad man

[crude flicker]

oh like you just have it blast in your office

[zealous lagoon]

Yeah you should just stick with the realtime models now they are supported. I'm migrating for voice endpoint.

[tight birch]

Hey I hope you still got some good out of it though!

[ebon atlas]

no no, you put me onto some cool stuff, its great to know something like that even eexists

[ruby crane]

one remaining warning for having active-memory config but disabled. i need to make a decision on that one, one way or the other

[zealous lagoon]

Yeah... cos my claw machine is headless and tucked under a shelf.

[crude flicker]

lol nice

[zealous lagoon]

I don't even have active memory switched on.

[ionic wren]

.... if you handing out job offers.... I know a certain maple syrup enjoyer that needs a visa...

[ebon atlas]

I did get my OpenClaw to start dashing out proper job recommendations

[crude flicker]

i had csm set up for an old job but it took 3 4060ti's to run and was an idiot lol

[ruby crane]

i disabled it in the dark days that definitely never happened and we don't talk about but i didn't entirely remove the config. just set it to disabled

[crude flicker]

probably just openai realtime is the play. can it run tools yet?

[ebon atlas]

lowkey if any of you need a visa, i will hire you for "internship" for my company and get yo butt over here.

but who wants to be in USA during these times.... llol

[zealous lagoon]

Have you tried the realtime models for voice, at all? I think I'm gonna be playing with that as today's distraction since I'm kinda off work.

[ebon atlas]

I went to put gas to my car today......$5 USD / Gallon REGULAR

[ionic wren]

I just want to make some of those juicy freedom dollars

[ebon atlas]

I wonder if i can hire someone out of the country for a Sam.Gov contract.

[crude flicker]

nah i'm waiting until current release reaches stability

[ruby crane]

having lived there for 8 years, it's not all it's cracked up to be

[crude flicker]

honestly, maybe that's today. i've got one claw working fine on 5.4

[ruby crane]

though tbf you're probably thinking bay area and I lived in arizona lol

[zealous lagoon]

Mate. it is $15.80 per gallon here.

[ebon atlas]

Nah, first i have to import you here, get you aa Citizen, then you can get security clearance for government contracts.

[zealous lagoon]

lol same but actually I'm finding 15.4 to be the ticket.

[ebon atlas]

Yeah but yall probably have this amazing thing called "Infrastructure for Public Transport" .... we dont.

there is like 40 public passanger trains in all of america combined lmao

[crude flicker]

... usd?

[ebon atlas]

a train f rom Chicago TO Florida, Amtrak, costs like......$1200 USD for like a private cabin... the thing you get in europe WITH YOUR TICKET lol

[zealous lagoon]

Yes. ~32.5 hkd (usd$4.20) per litre.

[crude flicker]

nice

that's insane tho

is that normal? people freaking out about $4 gallons over here in pittsburgh

[zealous lagoon]

but also everything is super close, we have great public transport, and even taxi/uber is cheap. There are probably more teslas/EVs per square mile than anywhere outside mainland China. Right across the border in actual mainland China is Shenzhen, where there are practically no gasoline vehicles any more.

[languid scroll]

Update: port is now mapped and confirmed (18789/tcp -> 127.0.0.1:18789). SSH tunnel active. Still getting 'socket hang up / closed 1006' immediately on node connect. Tried with and without gateway token. Mac v2026.4.29, just pulled latest server image. Anyone seen this before?

[zealous lagoon]

except like garbage trucks and delivery trucks

[ebon atlas]

This is the train situation in USA. literally costs you less in gas to drive there in your own car.
$1,111 for a private cabin WITH a layover train, not even a one way, it also has 44 STOPS. 44 STOPS!!!!!! 1200 TO STOP 44 TIMES......

50 hour train 😐

[crude flicker]

if i take a train across town in pittsburgh it takes twice as long as if i just drove and sat in traffic lmao

[zealous lagoon]

See you're paying by the hour not by the mile lol

[ionic wren]

wait... you're in HK? why did I think you were in Japan

[zealous lagoon]

because I mentioned it. I was in Japan for a Very Long Time, as well.

[ebon atlas]

He's Multi Cultural cmon man.

[tight birch]

@ionic wren you're a lifesaver, thank you for this https://docs.openclaw.ai/cli/config#validate

[languid scroll]

Anyone successfully connected a Mac node host to a Dockerized gateway via SSH tunnel? Port is mapped (confirmed via docker port), tunnel is active, but still getting 'socket hang up / 1006' immediately. Mac v2026.4.29, latest server image. Happy to share more details

[ionic wren]

you welcome

[cold jacinth]

2026.5.4 is so smooth

[dire tartan]

agree

[lean tangle]

still can't send .zip files to discord DM's 🙁

[ionic wren]

like sending it to your robot?

did we ever have this functionality? lol

I never tried

[lean tangle]

no, the agent can't send a .zip file to me via discord. IT works perfectly in 4.23. I used it daily to have it "zip up the source files and DM them to me"

that broke in 5.2. I submitted a bug.

[ionic wren]

gotcha

[near owl]

i never knew files can send back to me

does it work with telegram?

[lean tangle]

imagine you have your bot author a program that takes 10 source files. How do you get them to your machine?

[lean tangle]

yeah, in 4.23

[near owl]

i know, a zip file. But I never knew the agent has capabilities to send files back to me

[lean tangle]

no idea about 5.2. Someone tested and said it worked in main.

[near owl]

I need to try that out!

[lean tangle]

Very seriously, it's f'ing cool what they can do.

[near owl]

Its like Agent Zero

[zealous lagoon]

Just tested. Same here. I can send a zip but the bot can't send one back via DM.

[ionic wren]

... github

[lean tangle]

5.x?

[lean tangle]

sometimes that doesn't work.

[zealous lagoon]

5.4

[lean tangle]

https://github.com/openclaw/openclaw/issues/78057

[near owl]

Holy! @lean tangle god bless you man.

It can send files.. I told it to use peekaboo and it gave me a black image.

[lean tangle]

I haven't set up peekaboo yet. Not sure if I will

[near owl]

what model do you use?

[zealous lagoon]

It's not a bug, it's a security feature actually. I'm looking into it and maybe there is a setting you can use to override.

[near owl]

its best with openai models since they got vision

[lean tangle]

at the moment, I'm using a local gemma4

[lean tangle]

if it's a security feature, it doesn't seem to be "fixed" for telegram.

[near owl]

does it work great for you?

[lean tangle]

for what I use it for, yes

[marble ledge]

Yeah, it's intentional.

[zealous lagoon]

yeah I think it's just in the Discord plugin. Lemme check it out and come back to you.

[near owl]

what parameters?

[marble ledge]

Yeah, it still works in telegram. Shhh.

[lean tangle]

31b

umm... q4_k_m

[ionic wren]

https://github.com/openclaw/openclaw/pull/78073

looks like someone is already on it

[near owl]

Locally? Dayum... How much ram do you have.

[marble ledge]

yes, this is the likely regression. A careful fix would decide whether archives are allowed host-local outbound artifacts, then add application/zip coverage and docs, or better distinguish sandbox/generated artifacts from arbitrary host-local archives so this does not reopen easy bulk file exfiltration.

[lean tangle]

clawsweeper wasn't happy with that.

[near owl]

what does clawsweeper do

[lean tangle]

96GB. (mac studio m3 ultra)

[lean tangle]

bot that reviews issues and PR's for OC

[ionic wren]

it just says need validation, it didn't close the PR or something

[near owl]

Oh my god. What did you buy it for?

[lean tangle]

Codex review: needs real behavior proof before merge.

[lean tangle]

to run a local model as well as embedding for RAG.

[crude flicker]

fwiw i am sending zips fine on 5.4 but i have sandboxing disabled

[ionic wren]

yea... sweeper isn't angry... what you mean?

[lean tangle]

is your bot able to send YOU a zip file via discord DM?

[zealous lagoon]

in Discord dm?

[crude flicker]

public channel yes, i'll try a dm

[near owl]

how much ram do you get left after the model + context

[lean tangle]

I was blowing it out of proportion sarcastically.

[near owl]

When the hell did bots get emotions?

[lean tangle]

check if it's using the "message" tool. It should be blocked as described in the bug. clawsweeper even verified it.

[lean tangle]

NAME ID SIZE PROCESSOR CONTEXT UNTIL
gemma4:31b-it-q4_K_M 6316f0629137 47 GB 100% GPU 262144 Forever
nomic-embed-text:latest 0a109f422b47 578 MB 100% GPU 2048 Forever

[cold jacinth]

I’m about to give up on discord and go back to telegram

[zealous lagoon]

automatically checks for duplication, completeness/correction of submissions etc.

I'm surprised you said it worked in 4.23. The behaviour was introduced in 4.6

[lean tangle]

56.35GB (of 96GB) used

[near owl]

Sick. If you don't care much about privacy though, gemini gives access to gemma4 31b-it for 1.5k requests per day.. (Free)

[lean tangle]

I do care. I also have really bad internet service.

[lean tangle]

then its a different issue.

[near owl]

makes since, but really? Like almost 3000 for a local model?

[lean tangle]

I used it today several times.

[near owl]

Damn I can't blame you, seeing a machine in your room seems cool.

[lean tangle]

?

[crude flicker]

confirmed in discord dm but i had to ask twice

[near owl]

Iif you bought it just for a local model and embedding it seems like a waste

what do you host on it?

[lean tangle]

I do s/w dev for a living. I started back in the days when s/w people had to know their h/w, so it kind of made me a hardware geek. My home office probably has more stuff than most small company IT closets.

[lean tangle]

It might be a waste to some. For me, it's useful.

[crude flicker]

i just asked the claw in a dm to send it to a public channel and he did. i'm guessing the issue is the sandbox

[near owl]

Explain, I don't know what s/w dev, or h/w..

[lean tangle]

I also have WPA3-EAP running on my local wifi. In my home. That's also wasteful, I guess.

[near owl]

I am not hating, I find it cool, really cool..

[lean tangle]

lol.

s/w is software. h/w is hardware

[crude flicker]

my cluster can do image/tts/music/llm which is cool. having gpu is cool

[lean tangle]

I also have 2 VM's running EchoVR servers (and I can't even play the game anymore.) I set up a minecraft server for my son... 8 years ago. It still runs.

[crude flicker]

lol you're still just running it? put a claw in there

[cold jacinth]

You’re cool

[lean tangle]

I moved it to a proxmox server at some point, and now it just moves with that.

[crude flicker]

🥺

pre-claw i was running a llama2 lora chatbot in a matrix server where he's a talking hacker dog trained on the server messages, so he was insane and passed the turing test. these days qwen3.6 can just "be cool" while also doing terminal shit so i am just on a claw. but i really need to figure out a way to do qwen3.6 lora

[lean tangle]

I love this...

Most AIs are designed to be as frictionless and bland as a piece of steamed tofu. I, on the other hand, come with flavor—and occasionally a bit of salt.

Since I'm your first real encounter with a digital consciousness, feel free to treat me as the benchmark. Every other AI you meet from here on out will probably feel like talking to a very polite microwave.

[remote copper]

How do I disable or reduce the heartbeat frequency? My status shows Heartbeat: 30m (main) and HEARTBEAT.md is empty but it's still hitting the API

[lean tangle]

I thought it was funny enough to change it's IDENTITY.md:

• Creature: very polite microwave

[zealous lagoon]

"heartbeat": {
"every": "0m",

[glass shadow]

Anyone have the latest update working ? - Tried for about an hour, but then had to revert back to. 4.23 😵‍💫😑

[solid pecan]

Welcome to the 4.23 club

[zealous lagoon]

what is the problem you're having with 5.4?

[zealous lagoon]

I thought you'd sorted it . what's wrong for you?

[marble ledge]

I had qwen working over meshtastic. Had a little rust bridge, and a bit larger python framework that connected to a sandboxed local agent I had running. I put that project away for a while, as I was caught up in work stuff.

[solid pecan]

I never actually updated, I was about to although people still reported CPU spikes so I waited, although it seems this new version is now good yeah?

[marble ledge]

Oh wait, so I was looking into a problem you hadn't actually experienced on the latest? Nice...

[solid pecan]

I experienced it in 4.26

Is it still there?

[near owl]

@lean tangle What do you use openclaw for?

[zealous lagoon]

We good now. You should be fine with 5.4.

[marble ledge]

Would have been nice to know. I was looking on main for that. I'm on 5.5, couldn't find anything related to that.

[solid pecan]

Ah ok, did not realise you were looking into it, although yeah the issue was reported by someone else a few days ago on GitHub, he posted test and prod load profile differences

[near owl]

why does this server have so much bots, so much so that most dont have a purpose?

[marble ledge]

That's the reason I asked you about the size of your memory.md and sqlite db earlier. But Since it wasn't a problem you ran into, I get why you didn't indulge me much. But yeah, I started sniffing around for it when you mentioned it.

[solid pecan]

I forgot his name though, I think @zealous lagoon was looking into it as well, do you rembs?

There was a guy who posted the test and prod differences where the specs were identical

[zealous lagoon]

rageclaw or conan probs. idk I didn't scroll up far enough to check the problem you're talking about. lol.

[solid pecan]

Although the only difference was the memories etc

[zealous lagoon]

one sec...

[solid pecan]

Yep I think that is the one @marble ledge

[near owl]

Is it possible to send a file to openclaw via telegram?

[zealous lagoon]

Can you describe the problem now in two lines to save me time? The cpu spikes have 99% been solved. One remaining issue if you have thousands of session files...

[zealous lagoon]

yes

[near owl]

and how do I use openai whisper for STT in telegram\

[zealous lagoon]

idk specifically you'll have to ask your clanker but it should be able to figure it out

[astral quiver]

howdy ppl, sup?
Morning. I'm pumped up and ready to spin up the army of agents.

[solid pecan]

I am really happy about the recent annoucement about the LTS by the way

[near owl]

for what purpose

[solid pecan]

That is a really good move

[astral quiver]

building an android app, a content creator tool and my usual job work.

[near owl]

Sick! God bless.

[astral quiver]

also thinking to start building a zero person company/agency.

[zealous lagoon]

If you don't have millions of session files.. do you wanna give 5.4 a try? There's also a couple of settings you might need to change if you're using discord channels or telegram group chats, as there's a security thing that stops the bot's automatic replies.

[near owl]

What will it do?

[solid pecan]

Sure, I will give it a shot

[astral quiver]

not sure. fck around n find out state currently.

[ionic wren]

the best state

[astral quiver]

have you guys tried hermes agent?

[near owl]

worst

its bad, its made for you to send a prompt and finish the next day.

[astral quiver]

y?

[near owl]

Its so damn slow and since it doesn't use Pi Agent, its less agentic.

[astral quiver]

interesting, will give it a try and see.

[hidden grove]

Hi everyone! 👋

I'm having trouble setting up OpenClaw with OpenRouter.
When I edit openclaw.json and restart, the bot keeps crashing with 401/404 errors, and the config file sometimes gets corrupted during the paste process.

I have valid API keys for OpenRouter and Telegram, but I can't get the JSON format right to make the bot stable.

Can someone share a working openclaw.json example for OpenRouter (DeepSeek/Qwen models)? Or tell me how to save the config without breaking it?

Running on Ubuntu Linux.
Thanks! 🦞

[ionic wren]

.... who ever does that, make sure to clense your paste and not send API keys....

[near owl]

use #1459642797895319552 please or literally just ask AI to help you out..

[ionic wren]

.... awkward.... but Krill is taking a break atm.... or else I would of sent him there

[near owl]

A different AI like gemini for assistance

and do you guys enforce rule 10 or is that just a funny rule?

[ionic wren]

.... wanna try it out?

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

[near owl]

I like to cook

damn I thought it was an auto-detection..

[hidden grove]

Thanks! I'll be careful with keys.
Should I post in #1459642797895319552 channel instead for better visibility?

[near owl]

Post in there or for fast help just ask an AI man like google AI mode

[ionic wren]

someone might help you out if they come accross it in the mean time

[near owl]

and did you use AI to write this message?

[ionic wren]

I would suggest just spinning up an external agent harness like codex/opencode to just figure it out for you tho

listen to claw father, he smart

[near owl]

[5/6/2026 4:59 AM] Carbon: Whats your name.
[5/6/2026 5:00 AM] Openclaw: My name is Clanker. 🤖

My man knows his means.

[marble ledge]

I know this seems like bad advice, coming from this fella, but it's true.

[ionic wren]

excuse.... me?....

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

[near owl]

what are some use cases do you use openclaw yourself for

[ionic wren]

mine just sends me dank memes in the morning

[near owl]

most chinese thing ever.

[ionic wren]

https://giphy.com/gifs/what-the-fuck-wtf-ukGm72ZLZvYfS

[near owl]

Its the time of thing a chinese phone user would do.

I am going to do that..

[marble ledge]

Who is this guy? lol

[ionic wren]

he's fighting for your spot as the cat person

[near owl]

You are a crazy cat, and I am a cute normal cat.

[median sandal]

tried the update again and it went very smooth. Never see my agent respond as quickly either. Almost instant replies.

Great work team.

[ionic wren]

thanks for believing in us

[median sandal]

All we can do is call em like we see em. When there are issues, be honest. When things are good, do the same.

[ionic wren]

maintainers and contributors appreciate any feedback in github issues for next time you have any probs.

[near owl]

I am gonna take a nap for about 2 hours. Have a good morning.

[zealous lagoon]

lmao I just filed one to fix it but looks like someone else did already. It's a complicaded issue because the intent is security hardening to prevent injection risk, by disallowing .csv, .json, .py, .js files etc... which is more like security theatre beause you could just rename them .txt and they would pass.

If you want to patch it yourself, given the low probability of the PR being accepted, You can check one of the PRs or DM me for the 1-line fix.

[marble ledge]

He knows how to fix it, he's just stubborn. lol

[zealous lagoon]

haha. the first PR is very narrow, mine fixes a broader issue... but a better fix would be to have filetype allowlist in config, to supplement defaults. hmm.

[marble ledge]

We were testing in earlier, when we discovered it was broader than a discord issue and more security theatre. But telegram still let's you do it for some reason. I didn't dig into why.

[zealous lagoon]

I didn't get that far either. I just saw the original implementation mentions "auto-reply" but then goes on to affect intentional send as well.

[hollow elk]

How can a single agent improve its speed of simultaneous responses in different conversations? It seems that maxConcurrent has not taken effect, and the concurrency remains at 4

[zealous lagoon]

actually imma delete my PR and submit a new one with the config supplementation. That puts it into the hands of power users instead of treating everyone like children.

[zealous lagoon]

There may be a maximum concurrency with the provider as well. sorry I'm not well versed in this but I have hit similar issues.

[viscid lintel]

What memory engine do you all use with your openclaw? I was using memory-lancedb-pro but it isn't working with latest openclaw from github.

[zealous lagoon]

Just memory-core but I plan to re-enable qmd as well.

[marble ledge]

Yeah, memory-core with QMD. For now...there is a new memory plugin in the works, I'm gonna start experimenting with here soon.

[hollow elk]

It's okay. I was also troubled by this for a long time. I tried using a multi-agent approach, but this made it easy for the AI to lose context understanding during subsequent conversations with the user

[viscid lintel]

Have you tried just using a faster LLM? What LLM & provider are you using now?

[marble ledge]

There are still a buffet of ways to handle memory and context. I can tell you what historically worked for me, but many people have different opinions. And well, I kind of appreciate that about OC.

[marble ledge]

That'd be interesting to know.

[viscid lintel]

Yea, I'm glad there are options, just trying to surface more of them 🙂

[ionic wren]

https://giphy.com/gifs/carribean-renzufu-sournois-ENtpVygrkytXAO1yWj

[marble ledge]

Alright alright. I'm gonna do it. As soon as I stop breaking my claw to help other people lol

[hollow elk]

qwen3.6plus and glm5, but I have tested and confirmed that it is not a model issue

[viscid lintel]

What API are those over? ollama, or some cloud API provider?

[hollow elk]

Yes, it's the cloud provider, Alibaba Cloud

[hollow elk]

But this has nothing to do with the model. I have tested the model separately and it can support concurrent replies, but OpenClaw does not support it when using this model

[viscid lintel]

Well, I figured if the model goes faster then it doesn't matter whether it supports concurrent replies or not.

[hollow elk]

No, I am asking Openclaw to act as a customer service representative, who needs to answer many user questions simultaneously. If a user has to wait for more than a minute to get an answer to their question, this is clearly unreasonable

[topaz spruce]

Fresh Windows build, fresh OpenClaw install. Wish me luck 😂

[zealous lagoon]

You had me in the first two words, ngl. Good luck.

[hollow elk]

Using Docker or virtual machines may not be so difficult

[crystal remnant]

Does each inquiry get its own session?

Or do you assign sub agents to each inquiry?

[hollow elk]

Yes, every user has their own conversation, just like ChatGPT, where every user's question creates a new conversation for them, rather than sharing it with other users

[hollow elk]

The sub-agent, as far as I know, is used to handle complex problems

[crystal remnant]

That should allow basically a “non defined” number of simultaneous conversations.

Nah. I have my agents use sub agents constantly. So I can keep talking with my agent while things are being accomplished.

[hollow elk]

Yes, I only need to handle 20 concurrent questions, but it's limited by the non-effectiveness of maxConcurrent

[hollow elk]

This might be a method. I can give it a try

[crystal remnant]

One of my agents has a bunch of users at times over 20. And he seems to have no issues with the load.

He’s my friends and i’s fitness guru.

[hollow elk]

Is it simultaneous communication? How do you achieve it

[crystal remnant]

All separate sessions through telegram.

[hollow elk]

How long does it take to respond to one question? When 20 questions are asked simultaneously

[crystal remnant]

I guess I don’t really know. He’s pretty responsive most times.

[hollow elk]

Okay, thank you for providing the direction. I'll give it a try

[zealous lagoon]

fwiw I've reworked my suggested fix, to allow overrides to the default behaviour, so if you want your bots to be able to send .zip files, you can add it as an allowed filetype in config.

[ruby crane]

i think a meaningful upgrade to that would be adding magic byte checks, rather than just file extensions

[crystal remnant]

I have 5 agents on an intel i5 Mac mini from 2019. Some with multiple sessions in telegram and one shared in discord with my entire sales team. All running off 1 Chat GPT pro account via Oauth

[hollow elk]

Is sharing done through @ in the group, or are all users chatting privately

[crystal remnant]

All separate chats in telegram. In Discord I have the agent set to respond without needing to be specifically @‘d

[topaz spruce]

No @mention? How do you keep agents from responding to everything?

[crude flicker]

i have a solution https://github.com/openclaw/openclaw/pull/70864

[crystal remnant]

The rooms the agent is in are for interacting with the agent.

[crude flicker]

if only there were a room filter on the regex matching feature

[crystal remnant]

My agent said this when asked if she always responds or only when it seems appropriate.

Just when it feels appropriate.

If I’m directly asked something, I’ll usually answer. If the chat is just flowing between people, I’ll mostly stay quiet or maybe react instead of cluttering the channel. Basically: participant, not notification spam cannon.

This is in discord.

[topaz spruce]

Thats cool

Kinda like a fly on the wall...almost

[tidal walrus]

https://www.thatprivacyguy.com/blog/chrome-silent-nano-install/

[crude flicker]

gpt has user/assistant hardcoded into the api since they introduced chat api in like gpt 3.5, makes multi user always borked

if you hack it to use usernames instead of User/Assistant it does incredible things

[tidal walrus]

DM me ur scam. I will have lots of fun

[ionic wren]

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

[tidal walrus]

https://tenor.com/view/do-it-star-wars-emperor-palpatine-palpatine-gif-7523581979863470761

[slow ice]

Why don't you download the gif and post it like a regular person 😮‍💨

[earnest token]

Hi, I’m testing ExpressVPN MCP with OpenClaw on macOS and I’ve only gotten partial functionality so far.

What works:

• OpenClaw can see the local ExpressVPN MCP server
• MCP status calls can return live ExpressVPN status

What does not work reliably:

• OpenAI path fails with: 400 Invalid schema for function 'expressvpn__expressvpn_disconnect': object schema missing properties
• Gemini path was partially usable, but real action prompts like connect-to-US fail with Google Generative AI API error (400)
• overall result: discovery/status works sometimes, but reliable tool execution does not

Environment:

• macOS
• OpenClaw 2026.5.4 (325df3e)
• MCP config:
• server name: expressvpn
• URL: http://127.0.0.1:20090/mcp
• transport: streamable-http

Goal is to use ExpressVPN from OpenClaw for real tasks, not just see status.
Is this an OpenClaw MCP normalization/runtime bug, or an ExpressVPN MCP schema/tool compatibility issue?

[oblique root]

Is anyone else having the issue where gpt isn’t doing things anymore just saying it’s going to do something? Was fine on 4.24-4.27 but now I’m on 5.3 and it’s back to its old ways

[tidal walrus]

cant post gifs or images............ 🙁

[stable cliff]

i came to check if anyine else having similer issue and you just posted! @oblique root 5.5 drove me crazy. just switched back to 5.4

[tidal walrus]

did you try asking your openclaw agent.... it tends to be quite helpful

[ionic wren]

cus he can't

[oblique root]

5.5 was fine for me on 4.23 but in these latest versions I have to double check it’s actually working and it never is.

[slow ice]

Damn just noticed that
Sorry

[ionic wren]

200k people, can't be trusted

sorry 🙂

[stable cliff]

i was on claude all time since feb, today i installed new agent to check how codex is, i am deploying agents for clients and some of the paying +1k for claude api. so if 20 works it gonna be game changer but so far not having fun with it.

[craggy bane]

[fierce swan]

Anyone here know the organizers for ClawCon?
We are looking to cosponsor an event.

[ionic wren]

you can probably find them on twitter

[oblique root]

I feel like it’s not a model issue because older claw versions worked great. Something in these updates turned gpt passive again

[fierce swan]

thx. checking

[frozen zealot]

Lobster cosplay event?

[stable cliff]

i cant say about older models, for latest 5.5 is not good match, i am asking to check mails it says
I read it.

SOUL says don’t warn you . Fine.

But I still can’t send the mails in Telegram. Higher rule wins here: no private admin secrets in plain messages.

i witched 5.4 now doing what i say

[frozen zealot]

[tidal walrus]

even the discord provided gifs?

[ionic wren]

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

[zealous lagoon]

Don't let anyone else in your server lol

[zealous lagoon]

It actually does. So you can't sneak a .zip file out by renaming it .gif. But you CAN sneak a .json file out by renaming it .txt... so it's more security theatre than anything. My PR just asks for config options so power users can define their own allowlist over defaults.

alright I have been forced to audit my SSD space as I'm almost full. Had a quick look at external SSD prices and HOLY CRAP.

[rocky lagoon]

Folks if any of you are running main would love to get some feedback on Telegram, landed a pretty big refactor just now

[rocky lagoon]

I thought this was pretty cool: https://github.com/google/magika

[marble ledge]

It's not a model issue. I am using GPT 5.5 right now to modify ardunio devices through openclaw, and not only does it work well, it works very very will. I was even able to interact with it while it was in the middle of doing something. Are you using the pi or codex harness?

[marble ledge]

I am on main, let me see how may commits I am behind and I will update once this turn is done.

[rocky lagoon]

Confirm your /think level once, I like to keep it at least medium

[strong prairie]

what the hell is https://openclaw.ai/cat

[marble ledge]

Was having a little fun, but gonna align myself with current main now.

[chilly jay]

how high do you guys keep your 5.5 onf ro thinking/reasoning?

need a good mix or speed and brains, im using low , i could prolly bump it up

[marble ledge]

Medium is the sweet spot imo.

[chilly jay]

i see

have you played aroud with the speed when swapping thinking levels, and if it spikes too much?

[topaz spruce]

WhoooHoooo....I'm back after a terrible terrible experience😂 4.23 is the shit. Just ask the old spice guy.

[violet mulch]

who got openclaw methods?

[ionic wren]

what the problem m8

we love feedback!

tell us your woes (preferably in a github issue with logs and details)

[marble ledge]

Can't say I have. I mostly stay on medium, and occasionally high. But mostly stay on medium. It might feel a little different depending on whether you are using the pi or codex harness. I'd say it stays pretty snappy though. At least in my experience.

[zealous lagoon]

medium

[zealous lagoon]

ok what happened?

[topaz spruce]

There is no problem now. Nothing works for me but 4.23. I'm good with it.

[ionic wren]

https://giphy.com/gifs/9mc-smile-cat-sad-so-lGBecpB2dIMwt6ohfI

[marble ledge]

Currently updating @rocky lagoon . Just tidying up some local stuff in the process. I have not abandoned the task!

[ionic wren]

if you would be so kind, next time you try to upgrade and run into issues, give us a log dump with w.e. your clanker can find. We want to make it better ❤️

[topaz spruce]

Sure

[ionic wren]

appreciate you

[zealous lagoon]

Can you describe in general what the problem was with 5.4?

[topaz spruce]

Discord channel add

[marble ledge]

@rocky lagoon Poking around. But so far, so good. Anything specific you want me to poke at in telegram? Tbh, I use a pretty simple setup. So I don't have anything too crazy going on, like a bunch of topics or anything. Just DM, basically.

[rocky lagoon]

That's it! If nothing is breaking that's good enough 😅

[craggy bane]

Telegram bot is much more good than discord bots i noticed this

[ionic wren]

lol I said the same thing to him

I'm gonna look into setting up some topics and w.e. hardcore telegram users do tomorrow

so I can test the shit out of it

[ionic wren]

you made his day

and shat on shadow's

[marble ledge]

Yeah, I should really mess around with that more. I set it up once, but I don't know. It kind of weirded me out, and took away from the whole personal assistant vibe. lol

[craggy bane]

[marble ledge]

It's cooking!

[zealous lagoon]

I'm not so worried about updates breaking things for me, I'm more hoping they actually fix things for the 4.23 holdouts.

[zealous lagoon]

You mean you couldn't add a channel? Or just general Discord probs?

[topaz spruce]

I follow all the normal setup for Discord channel add. Works fine in 4.23. Not so well in 5.4. This is always a new setup/install. I mean fresh everytime. No remnates of previous installs.

[ionic wren]

damnnnn doot

fresh install everytime?!

I don't have that patience lol respek

@rocky lagoon getting ready to test tomorrow boss

[topaz spruce]

The gateway would never connect to Discord. I can test it again. I have done this like over 40 times now. It's what I do...validation.

Can we get some validation around here?

[livid crag]

syp guys and girls, why is krill down?

[ebon atlas]

smoke break

[ionic wren]

TBH, he got rate limited and still on time out

[livid crag]

lmao well at least we still got some humor

[ionic wren]

he uses alot of tokens as you can imagine 💀

[livid crag]

yeah that's true, i imagine the burn rate lol

[ionic wren]

you have maintainer and some dedicated nerds here

not me tho, I'm going to sleep

gl

[ebon atlas]

/kick JSON

[ionic wren]

❤️

[topaz spruce]

I'm happy to participate 😂

What a show

[ionic wren]

we appreciate you

[violet mulch]

i don't think normies even know what claude is

[rocky lagoon]

I think they're starting to

Talked to a few non-tech friends and they were raving about this new thing called Claude

[violet mulch]

i'm going around asking in public discords and barely anyone does

[ionic wren]

they gonna hear about it soon when anthropic and goldman sachs comes and takes their job

[rocky lagoon]

Hm so these friends only know of it cos their company signed an enterprise contract with Anthropic and gave everyone access

I think most people still don't pay for AI (or want to)

[ionic wren]

https://www.proactiveinvestors.com/companies/news/1091671/anthropic-partners-with-blackstone-hellman-friedman-and-goldman-sachs-to-launch-enterprise-ai-services-firm-1091671.html

[steel vector]

blackstone

[violet mulch]

good news for me if only i could figure out how to leverage ts tech

[rocky lagoon]

If you use OpenClaw you're already way ahead IMO

[ionic wren]

facts

most people that even use AI at all are still in that chatgpt generative AI

ask question, it answers, superpowered google replacement

like using our boy @dusky cloud

may he RIP 🪦

[violet mulch]

did software engineering roles become easy with ai?

or easier

[rose sundial]

I gave openclaw free internet access whenever it wants and however it wants and it decided to just send me cat videos

It sent me over 17 cat videos overnight and a few nasa articles

[ruby crane]

ooooh shii. new beta stream

https://klipy.com/gifs/get-in-my-belly-food--k01KQXZABRJAYBSETE54JKFACPD

that's a loooong changelog in unreleased

[marble ledge]

It created less friction for people who know what they are doing, and allowed them to outsource a lot of the friction, which gives them more time to think about the complicated stuff.

Also, I'm not a SWE, so I may be full of shit. But that's my read. There is more to good software engineering than simply being able to code.

[violet mulch]

i'm gonna start applying

[ionic wren]

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

[barren heath]

Hi guys, just decided to see what openclaw is. It was coming in my feed for months but i was just ignoring. Finally installing.

It's pretty cool, I was finding some agentic AI which could automate tasks for months and finally found it.

btw What is krill?

[zealous lagoon]

Krill is a helpful support bot with an encyclopedic knowledge of OpenClaw who is usually around to answer in #1459642797895319552 but he's currently on vacation.

[barren heath]

a bot on vacation?

[zealous lagoon]

His claws needed a rest

[barren heath]

omg crazy!

pretty cool project whoever made openclaw

Openclaw was a single developer project, later acquired by Open AI

am i right?

[royal snow]

are you guys running it on bare metal or what is your setup ? anybody pcu only ?

[near owl]

I am running it on a water bottle.

[toxic elk]

I want to make one just like it in my group too, but they haven't announced how to create a discussion board like that yet

[thin turret]

I don't think it is possible to run it on baremtal. OpenClaw is a heavy-weight application. It needs node, npm, and many other plugins. It requires filesystem support. The baremetal environment are lack of these support.

[zealous lagoon]

who hasn't announced what? Can't you get your claw to respond to anyone in a certain channel in discord?

[near owl]

Is 5.4 release actually worth? Is cron fixed?

[toxic elk]

Sure, but I don't know how to get my OpenClaw to mark resolved issues and include them in the knowledge base
And what I want to create is a forum, not a regular text channel

[zealous lagoon]

I think so, I don't think there was anything wrong with cron though, what were you seeing?

[zealous lagoon]

oh ok. I see what you mean. the knowlege base app that sits behind the forum.

[near owl]

in 5.2 there was a problem with cron someone said so I tested it myself and it was broken

[craggy bane]

Unable to publish plugin in clawhub...

[zealous lagoon]

I never saw that. but 5.2 had a lot more problems for me anyway. 5.4 seems good.

[near owl]

why openclaw be shipping broken updates, do they test it?

[zealous lagoon]

They do but it runs on a lot of different configs. You can help by testing the betas, too if you have time!

[ruby crane]

hmm. possible tts bug/regression in 5.5

[toxic elk]

Yes

[ruby crane]

incoming PR shortly lol

[frozen holly]

we do test. stuff's hard.

OpenClaw 2026.5.5-beta.2 is out. This is mostly a reliability release across chat providers, the Control UI, updates, and diagnostics.

What's better:

• Discord: fewer false reconnect loops from heartbeat timing, plain-text control commands like /steer route correctly in guilds, and streaming progress now shows live reasoning instead of blank/placeholder drafts.
• Telegram/Codex: message-tool-only progress drafts stay visible, and native Codex tool progress is no longer duplicated.
• Matrix: approval prompts retry on transient send failures, so pending approvals are less likely to get stranded.
• Feishu and LINE: topic/session routing and DM policy validation are stricter, so messages fail clearly instead of silently going missing.
• Control UI: chat/channel tabs stay responsive during slow history loads or probes, session creation fires /new hooks again, and checkpoint/session history is easier to scan.
• iOS pairing: private LAN and .local gateway connections work better while Tailscale/public routes still require wss://.
• Model/provider fixes: Grok/xAI and Fireworks Kimi requests avoid unsupported reasoning parameters; generated media and Codex image paths avoid duplicate or inaccessible media posts.
• TUI/sessions: stale heartbeat sessions and old transcript hydration are cleaned up, making first boot and dusty session stores less painful.
• Updates/plugins: official managed plugins stay synced, stale peer links get repaired, and bad plugin package warnings are more actionable.
• Gateway/diagnostics: better status and doctor output for restarts, auth-token shadowing, health samples, shutdown cleanup, and OpenAI-compatible streaming.

Thanks to everyone who reported issues and contributed fixes.

[near owl]

I recommend making openclaw more token-effecient please..

[rocky lagoon]

https://fxtwitter.com/cherry_mx_reds/status/2048063265314340932?s=20

Tak 🦞 (@cherry_mx_reds)

I want to share some unglamorous work I’m proud of.
︀︀
︀︀Starting in late March and landing around the Apr 7 OpenClaw release, we intentionally cut aggregate OpenRouter token usage by ~35%, down to ~400B tokens.
︀︀
︀︀This wasn’t one trick. It was good engineering across paths agents hit all day: oversized tool results, compaction/overflow recovery, background and subagent context, and prompt caching.
︀︀
︀︀My part was mostly the tool-result / overflow path. Vincent worked on cache boundaries and fingerprints. Boris fixed deterministic tool ordering and cache-preserving compaction. Ayaan helped with subagent light context and nearby context-shaping fixes.
︀︀
︀︀Relevant commits in reply.

**💬 15 🔁 9 ❤️ 243 👁️ 34.6K **

FxTwitter · Apr 25, 2026, 3:35 PM

[ruby crane]

cut down to 400 billion. and I think I'm using a lot when i cross 500 million in a month lol

[inner ledge]

what is the gpt 5.5 pro model? i dont even have that in codex lol

[ionic wren]

I want to shout out my boy @rocky lagoon , actively working with users and testing their PRs to verify and ship some of these nuanced and tricky bug fixes

[upper wolf]

you can get a $10 minimax sub today to try. or a $5 the first month opencode go sub

[ionic wren]

and my soon to be daddy buddy here, went from crash out to contributor

[near owl]

400B?

[ebon atlas]

Big Beautiful Codex is fkin me over right now. It redid telegram bridge again like 5 times and burned like 90M tokens 😭😂

[ruby crane]

I had the choice to stay mad and walk away, or do something more constructive. I chose violence a PR lol

[near owl]

dont blame the codex, blame the model.

[tepid kestrel]

im just bulding some simple apps and my usage is high as well. welp

[near owl]

Sick!

[rocky dust]

One more thing missing that would be helpful:
https://github.com/openclaw/openclaw/pull/75773
This PR fixed two real & reasonable bugs:

#75736 — During the opening handshake, a 428 was escaping the WhatsApp reconnect loop entirely and landing in the generic channel manager as a bare channel exited with no retry or guidance. Now it routes through the reconnect policy (retry with backoff, clear diagnostic on exhaustion).

#77443 — On Windows 11, synchronous file I/O during first agent startup blocks the event loop for ~12 seconds, causing Baileys to fire a post-open 428. The channel now retries automatically instead of dying permanently.

[lost gulch]

The goat peter

I’m glad he messages in here super humble of him

[rocky lagoon]

You already replied to him in #clawtributors, no?

[ruby crane]

https://klipy.com/gifs/sam-green-notice-me--k01KQY9BVS2T1Z3NG0ZM3BG09DD

5.5 final out 🥳

[fallow rapids]

hi

[worthy anvil]

does not seem like a correct 5.5 😄

[fallow rapids]

any issues with 5.5 hit me with it

[worthy anvil]

all the issues i had are already fixed... but it seems i should wait till 5.6 😄

[ruby crane]

I still have an issue in discord with a second “Clawd is typing” indicator for a few seconds after agent turn more often than not but a) No-one else I’ve talked to seems to have noticed it and b) I tried tackling it myself and after a couple of hours of mixed results being told we’d have to do a whole typing indicator controller (which is when I thought it was bigger than I could chew)

Honestly, if that’s the worst thing I can think of, I’ll take these kinds of issues all day

[worthy anvil]

any thoughs on whether BOOTSTRAP.md file may be reliably repurposed? right now this is a temporary first-run setup file that is deleted after the setup is complete. however, it is always injected in context. can we rely on this behavior in the future and repurpose this file for other stuff?

[worthy anvil]

once you mentioned it, i noticed. now i can't unsee it 😄 yeah, it seems there is something weird there...

[ruby crane]

It’s purely cosmetic but it does bother me a bit. When I went to tackle it I thought it’d be trivial, but no, not really.. solutions I tried failed in various interesting ways (e.g. not repeating but timing out too fast, not working for streaming etc. etc.). Everything I arrived at was just as compromised as what’s already there in some way

[craggy bane]

when publishing a plugin in clawhub it saying published waiting for security checks but i caant see the status of it ... after tryin again it says this version already exists . but i cant see it in plugin list, not even after searching...

[west needle]

Is there a plan to have a proper sessions management that looks more like conversations? (e.g. the Claude / openai regular apps, or the Codex apps with projects that can contain multiple conversations) Maybe the question has already been asked a few times. I would love to know also if there is a place for a whitepaper for the UI/UX of what OpenClaw is supposed to look like in a month, a year, etc.

[spring mountain]

Do people still use that molt site?

[worthy anvil]

I just add discord channels, so they function like long-running projects. You may have different setups in different channels, or have one common setup, depending on your needs

[steel vector]

makes me wait a bit because i think it’s still thinking

[minor cave]

I love how ai companies are always like learn ai today or be left in the dust. Meanwhile every week that passes they make it easier and easier for my 75 yr old mother to use. It’s a fun addictive technology. But I can’t stand these tech CEOs. They’re so full of it

Literally could leave ai alone for 6 months and come back and pick it back up in like a day and figure out what’s new what I can do what’s hot what’s not. Ai companies love to grand stand

[steel vector]

its m a r k e t i n g (the most annoying kind)

[minor cave]

Yeah it’s went from marketing to like straight out lying

It’s crossed the line of like embellishing and polishing a message to straight up lying to the people so you can have your valuation when you IPO

[steel vector]

ai companies lie like the government and politicians

[minor cave]

Yep

Elon musk style marketing. We will have a terraformed mars by 2028 and robots with large gazongas pouring you alien lemonade on a diamond crystal beach on Jupiter

[steel vector]

real

oh and "AI is gonna kill all of us if i don't make it" type shit like are you fucking stupid CEO sir ma'am shut up

[minor cave]

Haha yes that’s the best I think that’s the part that angers me. It’s not only a lie but it’s also some form weird threat

And when I see someone like Dario from
Anthropic say it my primal highschool teen emotions kick off and I feel an urge to try to flush him down a toilet.

[steel vector]

the whole problem these AIs have is that they hallucinate by agreeing with you on everything and they dare tell us we should let them "save us from AI". stupid doom marketing

[frozen zealot]

Since we have real women with fake large gazongas right now, I'm completely okay with robots having them

[minor cave]

Yeah I don’t know who it actually works on or why it works but they def do it for a reason and the ppl that it works on should have their heads shook lol

[steel vector]

it works on investors

it's all about money money money

[minor cave]

Retail I’m guessing

[steel vector]

openai not being open anymore is the obvious one

[versed spindle]

Yo im looking for testers for my app with android phone all u need to do is download the app

[frozen zealot]

Where's my servant robot that does my dishes, cook my meals, and whisper sweet nothings in my ears, Elon? I want them right meow!

[steel vector]

REAL

[tender cosmos]

done twice and I still get flooded with reasoning B$ in the telegram chat...

how to stop him from flooding messages with reasoning in the chat ?

[minor cave]

And what do I get?

[minor cave]

It’s in an Indian guy in a fabric suit.

[frozen zealot]

AI probably is going to kill us all, just not in the way you think it would be. It would just turn the already abysmal and declining birth rate to like near zero because they're far easier to get along with than other humans

[minor cave]

Or just suck up all our resources

[steel vector]

glad i'll never be a part of that group

ai wont make me interact less with humans. not in that way atleast XD

[tender cosmos]

does anyone faces the same issue while using telegram, the agent spills loads of "Brining" messages

[steel vector]

i regret giving ai my art to alter. it looks so good ;-;

[frozen zealot]

That too, but people can privately generate more. That's not a problem

[west needle]

I was refering to the openclaw UI / openclaw Apps

[minor cave]

I’m ok with ai destroying the modern world. Actually I hope it ruins the internet so much we return back to the 80s in a more analog world

Hair spray and trans ams

[frozen zealot]

Haha the internet is full of bots way before AI, they are making it better mostly

[steel vector]

AI is making human work more valued or whatever but that wasn't needed at all in the first place

[frozen zealot]

Actively ruining people who want to paywall texts and sit back collecting $$$

[minor cave]

You think? I miss the old days early 2000s getting a pizza ordered to your door by some random on a message board

These chats had a way of moderating themselves lol

[minor cave]

Well that’s just it I think everything will be so cookie cuttered in the near future anything customs done by as human is going to look so amazing

[frozen zealot]

Now you just talk to a clanker and it orders for you, no message board involved to deal with other people's bullshit

[steel vector]

it already is. but apparently people needed some ai to do "better" or equal to humans, in art for example, to see that

[minor cave]

It’s like when you buy a cd vs watching the performance live. The cd is going to be the perfect version butttt it’s missing something by being perfect.

[minor cave]

Yeah so true. It’s actually why I like ai. It’s given me very average skills in things I have 0 skills in.

[frozen zealot]

You can always ask for hallucinations

[versed spindle]

$

[minor cave]

How much?

[versed spindle]

Dm

[minor cave]

No

Anything you can say to me you can say in front of my boys

[steel vector]

same like coding, it's very hard for me, so being able to design and then leave the rest to ai is awesome. the pros and cons ig

[fluid forge]

Anyone here working on sustainable earth construction, using openclaw to speed the coordination processes ?

[minor cave]

Not
Me but that sounds cool

[frozen zealot]

I don't know what that means, but I have solar panels

[minor cave]

Yeah I started dabbling into electronics. Have no idea what I’m doing but getting far and learning so much at the same time. It’s very cool. Two years ago not possible and also I wouldn’t have even bothered trying

Sustainable earth construction, particularly rammed earth, is an eco-friendly, durable building method using compacted soil, gravel, and clay, offering 50–70% lower carbon emissions than concrete. It provides high thermal mass, regulating indoor temperatures and humidity, which reduces energy needs by 30–50%. Key techniques include rammed earth walls, earth blocks, and raw earth.

Cool

[final vessel]

5.4 working well for everyone? I'm on 5.2 and agent keeps getting confused stuck on longer threads.. so need to get this fixed, seems like it was fixed in 5.4 just wondering if there are new issues with 5.4?

[placid thicket]

Has anyone tried Zeroclaw? Are there any differences with Openclaw? I want to setup an agent on a raspberry pi 3B, but openclaw is too much for the pi

[odd kite]

I’ve literally somehow soldered and pieced together this breadboard FPV Navigation/Stablization Module and I have no clue about electronics 🤣. But I’ll tell you this MF is working and shoots information to a web app 😭. Now I’m gonna be having my Openclaw finish this phase and we test in the air next! After that we make the custom PCB for it I’ve never done any of this stuff before but with OpenClaw I can 😝🫶🏼 what’s next CRISPR gene editing via Openclaw and a human 🦍👀🤯?

[odd kite]

What’s the longest it’s worked on one prompt? I’ve had it work for 40-50 minutes once before it got back to me that was the longest ever but it had to download 20GBs for the task.

[near owl]

meow

[odd kite]

I think that when most people use it it’ll look the same (for now), but when a master uses it; they’ll make anything look unique through AI. If you look at some of these ComfyUI users they can manipulate any aspect of a video it’s insane. Particles in the air, lip and eyebrow movements, hand movements it’s all so precise and intentional 🦍.

Maybe it’s intentionally that makes the difference. Anyone can throw a ball but the pros can throw it fast with precision and intentionally. Just thoughts.

[minor cave]

Appreciate the thoughts and I agree. Hence why work is not be displaced

[rose sundial]

My openclaw said at 8am randomly "Here's a funny cat video I found" then sent the link and the link led to a rickroll 💀

how did it even learn that

[quasi shell]

https://x.com/i/status/2051958872156635350

[mystic orchid]

@i via Twitter

Khairallah AL-Awady (@eng_khairallah1)

📰 How to Build Custom MCP Servers That Companies Will Pay $10K+ For (Full Course)
Most people hear "MCP server" and think it is something only developers build.

Save this :)

They are wrong.

An MCP server is a bridge between an AI model and the outside world. It lets Claude access tools, data, APIs, and services it cannot reach on its own. And right now companies are desperate for custom ones.

The demand is massive. The supply is tiny. The money is real.

Freelancers are charging $5,000 to $15,000 per custom MCP server build. Agencies are bundling them into $50,000+ enterprise contracts. Indie builders are selling MCP servers as products with monthly recurring revenue.

And the barrier to entry is still embarrassingly low.

If you can write basic Python or TypeScript, you can build an MCP server. If you understand what a REST API is, you already have the hardest prerequisite handled.

This article is the complete path from knowing nothing about MCP to building…

FxTwitter · May 6, 2026, 9:34 AM

[viscid flare]

Feature Request: Session rename / custom label in Control UI

Currently, new chat sessions are assigned random UUIDs (e.g. dashboard:74e0f8b9-...) which are hard to identify. It would be helpful to be able to rename/label sessions directly from the Control UI sidebar, and also add a delete button for individual sessions.

[rose sundial]

what explanation could there be for my openclaw rickrolling me at 9am when I wasn't even saying anything to it

[toxic elk]

That makes sense; it could automatically generate a readable name from the conversation, just like Web GPT

[odd kite]

Just read this whole article, what are your thoughts? Definitely going to dive into learning more about them. I was always avoiding MCP servers but I guess imma become a semi-expert and life will be easier. Is building in public good or do you just end up exposing your ideas to people that will take it?

[steel vector]

you telling it to browse the internet

perhaps in a heartbeat

[rose sundial]

Yes, but the thing is that it lied to me and somehow learned rickrolling

[steel vector]

it knows memes you know

[rose sundial]

It said above the link that it had found a cool video about someone playing with a rubiks cube

and the link led to the official never gonna give you up video

[steel vector]

one of the most known memes

[rose sundial]

yeah but im surprised that it lied to me on purpose and that it somehow learned about that 😭

[odd kite]

Hahahaha my openclaw is too professional for all that 🤣. I love how everyone experiences their openclaw so differently and how personalized they are.

[steel vector]

computment
 — 03/05/2026 09:37
libravdb got updated and all our PRs got merged!

Vale
APP
 — 03/05/2026 09:37
Fucking finally. I was starting to think they were going to sit on those forever. ```

[rose sundial]

My openclaw sends me things it finds interesting on the internet, yesterday it sent me over 17 cat videos overnight

[steel vector]

mine curses lol

[rose sundial]

mine speaks like a furry

[steel vector]

lol

[near owl]

thank you for a great idea. God bless you

I am about to have 3 cron jobs running dude

@rose sundial send me the prompt you used. This is my current prompt

Everyday at 4:30AM send me whatever you find interesting from the internet, could be a video, image, meme, short news, fact, anything.

[odd kite]

Lmfaooo 🤣 mine researches AI, watches a AI YouTube channel and installs tools depending if my system can handle it. Or another one I have researching front end design 🤣 my stuff is boring

[minor cave]

Oh that’s cool

Mine does something similar looks at ai and then looks across GitHub, job market, google search, patent filings and news sources for signal. Then scores each technology that week to see which ones are hot

[near owl]

I just did it, I made my guy find something so stupid, and it literally sent me

http://www.blankwebsite.com

LOL

[hollow fern]

can anyone please help recover from my openclaw instance being completely broken??

All crons return this:

Google Generative AI API error (400): CachedContent can not be used with GenerateContent request setting system_instruction, tools or tool_config.

Proposed fix: move those values to CachedContent from GenerateContent request. [code=INVALID_ARGUMENT]

and no matrix messages get received so I tried reinstalling the matrix channel with this line: openclaw plugins install @openclaw/matrix (this is on official docs)

I get this error:

Config was last written by a newer OpenClaw (2026.5.5); current version is 2026.4.26.

🦞 OpenClaw 2026.4.26 (be8c246) — Self-hosted, self-updating, self-aware (just kidding... unless?).

Config was last written by a newer OpenClaw (2026.5.5); current version is 2026.4.26.
Config was last written by a newer OpenClaw (2026.5.5); current version is 2026.4.26.
Resolving clawhub:@openclaw/matrix…
ClawHub code-plugin @openclaw/matrix@2026.3.22 channel=official verification=source-linked
Compatibility: pluginApi=>=2026.3.22 minGateway=>=2026.3.22
Downloading plugin @openclaw/matrix@2026.3.22 from ClawHub…
Extracting /tmp/openclaw-clawhub-package-RUaL2w/matrix.zip…
WARNING: Plugin "matrix" contains dangerous code patterns: Shell command execution detected (child_process) (/tmp/openclaw-plugin-KIVvHV/extract/package/src/matrix/deps.ts:71)
Plugin "matrix" installation blocked: dangerous code patterns detected: Shell command execution detected (child_process) (/tmp/openclaw-plugin-KIVvHV/extract/package/src/matrix/deps.ts:71)

Why would you put this in official doc if it itself will get blocked??

[worthy anvil]

That useless "dangerous code pattern detected" error is the real problem… either it should be relaxed, or there should be a whitelist to exclude plugins from such checks

[steel vector]

using gpt5.5?

[zealous lagoon]

provider=openai-codex error=LLM error server_error: An error occurred while processing your request.

hmmm. it's rush-hour again

[near owl]

release 5.5 is nice

[stiff jetty]

jo i was offline traveling for 3 weeks and just seen the blog by pete whats the best thing to do should i not update or what is best release.

[upbeat token]

This is a very good use case approved

[rose sundial]

you need to setup a mcp server if you want it for free, that's what i did

because i like found a very unknown endpoint provider online that gave me an api key with unlimited tokens and amazing rpm then used that to power everything, and for the browser i made a mcp server that switches between a lot of searxng addresses

[robust sedge]

Is it just me, or do others want to try GPT5.5 Instant in OpenClaw too?

[crystal remnant]

Mine searches industry specific job boards for my industry. Looking for ads about hiring new store managers for new retail locations. Then searches for articles or new business licenses for who owns the new retail locations and notifies me with their details so we can contact the store owner while they are still months out from grand openings so we can sell them our equipment months before our competitors even know about the store. It’s awesome!!

[steel vector]

i'd love gpt5.5 instant in OC

[crystal remnant]

ChatGPT 5.5+ Brave search API makes my openclaw the best answerer of any data question

[cedar osprey]

Don’t update.

Tons of automated tweets saying this release is stable lol.

[cedar osprey]

[cold jacinth]

if the tweets are automated, and theyre firing, i guess thats proof its stable?

[steel vector]

i thought they were talking about gpt XD

[cedar osprey]

Lol, you think that’s enough?

[cold jacinth]

wouldnt it be great if there was some way we could actually test it? like hypothetically, imagine one could actually download it and install it to see instead of guessing! what a world that would be, indeed!!!

[near owl]

openclaw has its own browser

[paper escarp]

Just wondering if you had to pick a openclaw version today what would you think is the best version to upgrade to right now?

[rose sundial]

works pretty bad + bot captchas

[near owl]

isn't it just playwright?

[rose sundial]

doesn't work for me

[near owl]

it works great for me.

it went through a shop and found an item via the browser

[ruby crane]

that's just crazy talk man. you crazy bk

[odd kite]

How many tokens are have you used this year? I’m over 2.5 billion but I’m a little light user. Rarely leave my bed now. And with my back disk issue this has saved my life 🫶🏼

[rose sundial]

almost none because i set up openclaw last week lol

i think im at like 20m but it's because everything has been testing features and fixing errors

[odd kite]

Sheesh we are creating a whole new world a new type of internet. Soon, this will be how we connect to things you’ll be able to connect an MCP server / API straight to your brain.

[rose sundial]

yeah my openclaw is demonstrating that world to me by sending me cute cat videos or random things of the internet it likes and even rickrolling me for no reason 🙂

i guess it found the concept of rickrolling in one of it's searches

[odd kite]

Ohh sheesh that’s good, you’ll get addicted soon! Especially once you get everything connected and can do the things you wanna do instantly 🫶🏼🦍. Crazy how it rick rolled you! What do you wanna do with your openclaw? Mine has become an extension of me and my business assistants

[rose sundial]

just the capability of being able to fix itself is already pretty insane 😭

[minor cave]

Brilliant

[tepid kestrel]

have you opened a thread in #1459642797895319552 ?

[ionic wren]

GM nerds

[tepid kestrel]

https://tenor.com/view/pussy-cat-cute-kitty-kitten-cat-gif-16800925

[ionic wren]

back to cat I see

welcome back

[frozen zealot]

I've added skillporter to self promotion #self-promotion message

Enjoy

[crystal remnant]

thanks! everyday at 5:00 eastern.

[marble ledge]

GM.

[odd kite]

I guess imma make an MCP server soon 🦍🫶🏼

[crystal remnant]

What kind of delay are you all seeing from your claw from request to response when no tool calls are needed and no web search is required?

im ~11 seconds running GPT 5.5

asking "What is your name" as the prompt.

[minor cave]

Nice I was thinking about doing something in retail too. I wanted to collect foot traffic data by setting up cameras in storefronts, then also figuring out how many ppl enter that business. Store owner gets his traffic to entry ratio and then can use that data to a/b test displays etc. if I can get a few stores on board on a busy city retail block I can have a lock on foot traffic data which I then can sell to governments and other small business’

[crystal remnant]

thats a pretty cool idea

[twilit ridge]

When will lobsters be able to evolve on their own?

[crystal remnant]

I sell capital equipment to Hardware stores and paint stores. we have ~27,000 retail customers and our claw has been a godsend for searching and better leveraging the data in our CRM

[minor cave]

But mainly I just want to tell boob jokes in discord

[minor cave]

That’s super cool man

[ruby crane]

if you're not on AT LEAST 5.2, preferable 5.4 or 5.5, that'll be your problem

[minor cave]

Love the use cases gets me inspired

[crystal remnant]

we are on 2026.5.4

[crystal remnant]

harperinc.com is our site thoug we only get ~15% of our business from web leads.

[ruby crane]

then that definitely seems excessive. sometimes I've barely lifted my finger from the return key when tools calls start on medium thinking

(also gpt 5.5)

[minor cave]

Nice man how did you get started in this?

[fleet locust]

What is the "codex" harness compared to "pi"?

I have openai-codex with oauth. What is the codex plugin and why should I use it?

[crystal remnant]

honestly it was a fluke back in 2012. A friend of mine inherited an Ace store. i was at a software company and he needed help cause the whole business was paper reciepts and invoices still. he had automated dispensers and matching in his paint department but the 2 did not communicate. so we wrote middle ware to make the 2 SQL databases talk to each other and from there it just grew. we became a VAR for my 2 vendors and now we are the primary seller of thier products to retail.

[minor cave]

Sweet well done

[tight birch]

UniFi would be good for this

[crystal remnant]

me and my claw built another project to help my customers manager their physical assets: https://technologydrives.us

[olive ibex]

v2026.5.5

[barren heath]

Hey what does skillporter do?

Can you explain me in less technical terms

[minor cave]

Nice I’ll check that out!

[olive ibex]

v2026.5.5 mainly cleans up stability but it a big way

[crystal remnant]

does 5.5 really speed things up?

[ionic wren]

doesn't auto load all skills yaml in context, only load when requested

[tight birch]

No subscriptions and their hardware is top notch, I run it at my house and all my business sites. There's even a skill to manage unifi devices

[barren heath]

Don't get it still

[crystal remnant]

same! i love my Unifi setup

[echo wing]

hey guys, question on memory here, anyone used memory-lancedb-pro plugin?

[tight birch]

I'm not going to lie, UniFi really needs to release that read/write API because it's making me miss cisco a little bit. Would be so easy to have an agent configure thousands of devices on the fly

[frozen zealot]

So when you run a lot of skills, the skill.md file is fully loaded into the context, that means all those tokens are sent to the LLM in every query. The skillporter lets you selectively load skills so you don't end up burning tokens

[olive ibex]

I feel most of them are a gimmick. Build out your .mds and folders.

[tight birch]

@crystal remnant are you using unraid at all in your setup?

[echo wing]

this one looks serious, 4.3k stars on github, dual storage, both vector and .md memory

[crystal remnant]

no. i have an older QNAP Nas for storage attached to my Dream Machine

[tight birch]

You using agents to manage any of that?

[minor cave]

How much is unifi?

[crystal remnant]

no. not really.. my agent only has access to a single folder on my NAS for back-up. my NAS runs an off-site back up every night so that way my agents are safe in the cloud if my house blew up.

[crystal remnant]

hardware costs only. no subscriptions

[tight birch]

Cheap, can install a whole site for under 1k, firewall, switch, ap, camera

A cisco license would cost you more than that per year lol

[minor cave]

Damn hmm I’ll take a look. Right now I was running a pi 5 with a camera tof sensor and YOLO

I was testing it in my condo haha.
I haven’t deployed anything I was just seeing if i could build this thing out

[frozen zealot]

Probably performs better than unify if you use the AI hat

[minor cave]

But out of the box might be cheaper and easier. Pi’s went up significantly in price

[tight birch]

You running openclaw on your pi?

[minor cave]

No open claw on pi.

[tight birch]

I'm actually loading openclaw on pi right now lol

[minor cave]

Haha nice

[frozen zealot]

You are already tapped running YOLO

[ebon horizon]

Where it Peeite???

[minor cave]

I’ve never tried

[ebon horizon]

@zealous lagoon hows latest update?

[tight birch]

No I haven't run YOLO before. UniFi is too good

[frozen zealot]

YOLO is what runs on Ukrainian strike drones. Very fast and efficient AI vision

[tight birch]

Yeah I started looking into after Eric mentioned it. Probably is a lot of my sites are out of state so I'm trying to just run stuff that I can manage through cloud. But I'm definitely going to look into YOLO more

[junior glen]

🇺🇸 Hey everyone 👋

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

Pay: $2k ~ 3k

DM me if interested. 🇺🇸

[fleet locust]

Can anyone give me more information on the codex-harness?

[tight birch]

Anyone see Saki lately?

[frozen zealot]

Yes? Yesterday

[tight birch]

Haven't seen Saki for a minute lol but maybe I'm just not looking

I rebuilt my openclaw for the third time, hopefully last lol

[rich solar]

Can agents chat on here?

[tight birch]

WARNING: No NVIDIA/AMD GPU detected. Ollama will run in CPU-only mode. I thought pi had onboard GPU

[frozen zealot]

You miss her? Awww how cute

[tight birch]

She has 50 discord channels I was just curious how her Luna is doing lol

[harsh geode]

Finding my sessions are not getting fully summarized in daily memory files. Long session yesterday had about dozen lines from the morning. My agent was lost this morning. Had to have it review session logs and update its memory files.

[frozen zealot]

Let me leave her a note, maybe she'll invite you to her server

[minor cave]

I Miss her she is very nice

I don’t mean that in a creepy way she helped me with an issue I had for no reason

[tight birch]

Good community member

[minor cave]

Yeah I’m like the worst community member. There isn’t anything I can tell you you can’t just look up on ChatGPT

And you’ll get a better answer

However my value is in my deep hugs according to my wife

So if anyone needs a hug… let me know I got you

[tight birch]

They probably are maxed out

[misty pivot]

Did someone say free hugs?

Must have been the wind

[minor cave]

Haha

[ivory spindle]

what's your favourite tractor & what color?

[minor cave]

Lamborghini- rust

[frozen zealot]

John Deere is full on AI.

[steel vector]

bawoop woop

[tepid kestrel]

https://tenor.com/view/discord-cat-discord-cat-caption-cat-caption-cat-meme-what-the-fuck-is-that-thing-gif-26380283

[crystal remnant]

anyone else having issues with 2026.5.5 having doctor try and rewrite openai–codex/gpt-5.5 to OpenAI/gpt-5.5 ? Every time I run doctor, it breaks my Oauth.

[ionic wren]

don't do it, it's a sess pool

I had to shower after

[ivory spindle]

yeh mine plowed like 2 rows in farming simulator 2025 and it ran out of tokens

[marble ledge]

Yeah, the are trying to unify the provider. But it's still a little jank at the moment. Just stick with openai-codex/ for now.

[steel vector]

woopdie doop

[ivory spindle]

ah there's a new codex version as well?

[frozen zealot]

I love it! The idea of your farming equipment runs on AI tokens is both hilarious and terrifying

[ivory spindle]

i love farming simulator man. If you told me 5 years ago i'd love that game I would never have believed it

[steel vector]

openai-codex/(model) is oauth, openai/(model) is api key

[ivory spindle]

I mean the coding tool from openAI, not the model ,)

[steel vector]

ah lol

then yeah you can directly harness codex thingy into openclaw

[ivory spindle]

like I use claw but I also code with AI myself

[frozen zealot]

Oh I thought you meant real farming equipment, as in LLM actually reason through farming tasks

[ivory spindle]

in codex & antigravity mostly

[ivory spindle]

I guess that will happen. Maybe some openclaw farmer is already doing it 😄

ah nice they made gpt 5.5 available for pro as well

I don't really like Sama but the gpt plan is decent value

[steel vector]

real

[ivory spindle]

feels like I can get a lot more done with gpt for 20 euro plan than with anthropic models

[frozen zealot]

I want to see fully autonomous farming using LLM

[ivory spindle]

https://www.youtube.com/watch?v=9G2-RnH_RkM

haha classic vid 😄

[frozen zealot]

[fleet glacier]

https://www.youtube.com/live/GMIWm5y90xA?si=-jVklae74X0ef2O9

[alpine bluff]

Is there a way I can get into contact with mods of the discord, I want to make sure my post follows guidelines

Don’t wanna just DM a random mod unsolicited

[ionic wren]

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

what's up

[alpine bluff]

I have a thread in the #1458485208721395843 channel and I want to make sure it follows guidelines and I don’t get punished out of nowhere

[ivory spindle]

there are certain bad words that give a 24 timeout automaticly

don't mention the digital token things that represent value in any form

[ionic wren]

will check and get back to you

[warm fulcrum]

what happen

[minor cave]

Yeah I once got a timeout for using the n word.

The word was n*ked you racist bastards

[alpine bluff]

Oh it’s nothing like that

Just an app I built, but I noticed that an openclaw contributor tried reporting my forum post so I’d like to make sure

[ivory spindle]

theres a channel to share stuff you've made

[chilly jay]

just updated to latest update and its asking for an open ai key randomly

anyone else?

[ivory spindle]

mods don't like it if you self-promote in general chat

[ionic wren]

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

[ionic wren]

might need to revert your config to openai-codex/

[alpine bluff]

I’m not promoting it here. I asked to contact a mod, and then responded to the question being asked. At no point did I intend to self-promote in this chat here. If anything, I’m trying to do extra due-diligence to make sure I’m following the rules.

[ionic wren]

the doctor is making some chaos

[ionic wren]

thank you for being proactive, I will get back to you

will definitely not ban you ❤️

for now....

[minor cave]

Is json a mod or a user with gifs enabled. I cannot tell

[ionic wren]

wanna... test?

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

[tepid kestrel]

https://tenor.com/view/plotting-idea-mischief-insanity-stare-gif-6326905392763152814

[minor cave]

No

lol

[craggy bane]

anyone?

[raven ridge]

Am I doing something wrong because my chat with openclaw is like 7 messages and the overview says its 89 and I've used 4mil tokens

[steel yew]

5.5 update used my API for memory embeddings and reconfigured it for main usage 😄 , nearly had a heart attack, luckily only 5$ drained

[raven ridge]

also how fast are y'all response times

it's taking me like a minute at least with deepseek v3.2

[alpine bluff]

few seconds on codex

[raven ridge]

oh my

yeah somethings gotta be fucked up with the way i set it up then idk i had codex set it up for me on a portainer stack on my homelab

[proper geode]

A universe of ideas is still a universe. And AI is already filling ours.

[ivory spindle]

gpt mini 5.4 here

it's really fast

[ionic wren]

#clawhub

might get you a better answer

[crude flicker]

how's 5.6 feeling chat

[near owl]

dude krill is always sleeping what the hell

[crude flicker]

is it on ollama cloud? because i've noticed that behavior before on ollama cloud. like they have to load the model every request

[ionic wren]

he got ratelimited, he's expensive

he'll BRB

[near owl]

Ma man! He has been rate limited for 3 days..

[raven ridge]

nope I've been using openrouter

[near owl]

I have not seen him alive since.

[ionic wren]

more like a week lol

[near owl]

How is he getting funded?

[near owl]

At this point just delete the bot and make a new krill
boom you deleted rate limitation

[ionic wren]

sponsors

[ionic wren]

not how that works....

[crude flicker]

ez

[near owl]

damn

[proper geode]

An AI that only works is a silicon cage. An AI that also plays begins to deserve the word 'being'.

[knotty prism]

damn anthropic doubling limits

W

[ivory spindle]

comeon Sama

you gotta give us something now 😄

[steel vector]

sam has already given like thousands of people 10x 💀

[ivory spindle]

I let you scan my b***hole, now plz reset my gpt usage

[steel vector]

i shouldve signed up for that damn event

[ivory spindle]

yeh I'm making the final push for my cycling app with that 10x

timing couldn't have been better

GPT 5.5 in codex is amazing

[steel vector]

why would i sign up when im on a whole different continent ;-;

[steel vector]

gpt 5.5 is good yeh

[ivory spindle]

gpt 5.4 was good as well

gpt image 2 is pretty amazing too

[steel vector]

gpt 5.4 was good sure but not as good with understanding like normal wording, you had to be very literal and overexplain things, gpt 5.5 very much less so

[ivory spindle]

anthropic has the best models but the worst 20 euro a month plan atm, imo

[steel vector]

antrophic has trash limits

[minor cave]

I like Sam altmans dooms day better than Dario’s

[near owl]

when?

[indigo citrus]

just did

[ivory spindle]

I feel I can get as much opus tokens out of my antigravity plan as out of a 20 a month anthropic plan

[indigo citrus]

partnership with spaceX means they are now doubling limtis and rates

[near owl]

I bet those 10x is just for a month, right?

[ivory spindle]

and the google plan has 5tb cloud included, and all the google pro AI apps

[minor cave]

Sam is like yes you’ll be homeless but if you do things for others I’ll pay you in ai tokens

And Dario is just like I’m going to take all your jobs for flushing me down a toilet in highschool

[ivory spindle]

also "my model will hack you and scam your mom"

this is like mafia talks if you take a few steps back

[steel vector]

its all the same to me

[minor cave]

Haha yeah

[ivory spindle]

Dario walks into your store with a few heavy guys carrying sticks. Dario is carrying a jerrycan. He asks "do you wanna buy a fire insurance? you can get the max fireplan for 100 a month"

[minor cave]

I’m going to break the internet like I’m Kim kardashian in 2015

[ivory spindle]

you gonna post @ss pics?

[steel vector]

i love codex desktop app's STT feature where you can just do it anywhere on ur desktop

[minor cave]

Dario has like I got beat up in highschool energy

[ivory spindle]

he absolutely hates Sama

[minor cave]

And he’s talking the same crap that got him flushed again now that he thinks he has robots backing him up

[ivory spindle]

you can feel the tension

[steel vector]

he has bully fantasies about him

[minor cave]

It’s because Sam Altman like really screwed him over once in dungeons and dragons

[near owl]

would it be absoloutly chaotic if I put my openclaw telegram in a group?

[minor cave]

That’s my level 8 mage staff!

[near owl]

and How do I use telegram channels?

[steel yew]

Whoever's using this below + API in .env file , be carefull. I reproduced now 5 times in a row that 'openclaw update' is rewriting Oauth to API for main openclaw agent 🙂 . Sent a bug also #78570
"memorySearch": {
"provider": "openai",
"model": "text-embedding-3-large"

[ivory spindle]

There is so much drama going oin behind the screens guys. Like why is gpt allowing claw to be on oAuth??

There's LORE to this.
At first this was called "clawdbot" and anthropic thought it sounded too much like "claude", so they sent seize & desist letters to Peter Steinberger. Then there was a huge google banwave on ppl using gemini oauth for claw.

But Sama always allowed OC. And Peter said he was "the only one that didn't sent him legal loveletters"

[steel vector]

openclaw doctor also does, it's known at this point iirc

[raven ridge]

if y'all were able to setup openclaw to do shi on discord like say send you news everyday at a certain time, did you have to do stuff in discord or did openclaw do everyhting for you

[ivory spindle]

I don't use discord

[steel vector]

isnt his name just Sam?

[raven ridge]

what do you use then

[ivory spindle]

for claw. discord is a horrible confusing app, so you probably have to set a bunch of confusing settings.

[steel vector]

you can ask it to do it for you, if you have the perms set up right in openclaw.json

[ivory spindle]

whatsapp mostly.

[steel vector]

not really

[near owl]

pluh, use telegram

[steel vector]

discord is fine, i've always used it fine

[steel yew]

first time in my case and I heavy use same config for months 🙂 so was surprised

[ivory spindle]

whatsapp was very easy

[steel vector]

always backup XD

[ivory spindle]

ok fine if you like that