general

[broken moon]

no i do very like pi agent i think the person who created it is brilliant it is so simple and i feel like architecture should be that way

[vivid pond]

no i swear it doesnt i swear i would never

[ionic wren]

are you having a stroke?

[broken moon]

i just think right now no comptent engineer working on quality systems would use pi agent at least

[vivid pond]

no im just like damn your had that on lock bro on me like a honey bun to a homeless person damn

[lost void]

It’s what OpenClaw uses in the backend
But you can run another instance separately or connected to your claw
Kinda awesome if you ask me

[upbeat token]

[craggy bane]

[vivid pond]

what is so wrong about what i just saiidddd

[broken moon]

have you heard of augment code, i feel like actual quality engineers like quants use that

the agents in augment code are even too advanced for me

what cursor tried to be

[ionic wren]

..... you understand this whole project basically started and still runs on it? half the maintainers are on it or codex.

[vivid pond]

Json typing up a storm dude

[vivid pond]

holy freakin beast mode owned 1000m social credit

[broken moon]

yo can someone ban this kid people like this ruin dev serveres

[fiery nacelle]

i think 5.6 replaced my agents intelligence with more emojis. Holy hell so many emojis now.

[upbeat token]

Brother go outside lol

[vivid pond]

ok sorry ill lock in fr I may not be a dev but this also a community discord ill uphold the dress code and use periods in my sentences.

[ionic wren]

who said this was a dev server?

[lost void]

Pi.dev is great for newbies actual devs still don’t trust agentic harnesses and are slower to adopt or change

[broken moon]

i mean i guess all ides are becoming agentic nowdays

[ionic wren]

I actually think pi is not newbie friendly, it's barebones as it gets

[broken moon]

what do yall think about hermes agent tho

[fiery nacelle]

personally, astro-turf, but good for something that "works"

Like its good from what I've read others actually using it, not just promoting it, but again, its just turf wars.

[vivid pond]

i havent tried it heard good things but im in the middle of a mutually beneficial project that will work as soon i figure out how to read an error surface graph

Whats even out there regarding OpenClaw look alikes like hermes?

[broken moon]

do you use macbook or windows

[vivid pond]

i use linux and win10

[obsidian mural]

@fiery nacelle through trial and error its taking me forever what to set the default chat to ..... LOL thinking qwen3:8b

[vivid pond]

why?

[surreal hatch]

Nice to be in a community for openclaw! Anyone have tips for non-technical capable people on how to best use openclaw - have had it for a few months

is Codex the consensus best model for it for heavy work?

[broken moon]

For Windows, press Win + X, then U, then R trust me

[ionic wren]

I hope you only saying good morning to it and have it whisper sweet nothings to you at night

[fiery nacelle]

I personally just main Qwen3.6 35B A3B for my primary agent.

[lost void]

Natural language terminal shows newbies what’s possible with agentic Ai
open free and unlimited

Better than teaching them Crippled Claude Code or pricy Codex they will be afraid it’s going to cost an arm and a kidney to try then stuck in Jail for using it too much

[vivid pond]

ok im back had to restart my pc rq

[obsidian mural]

lol pretty much, trying to figure out what the default chat , when interfacing chat channels

[ionic wren]

there's opencode

github cli (tho you can't get any plans atm)

[vivid pond]

you run that locally ? if so unquantized?

[ionic wren]

codex can be obtained for $20

not sure why you saying it's expensive

[obsidian mural]

thanks for that suggestion, going to install that now

[fiery nacelle]

Q4K_M

[lost void]

Yes I tried the buggy beta windows app and don’t recommend it but opencode go is awesome for us in Asia as it’s got a Singapore 🇸🇬 server

[surreal hatch]

how far does that 20$ sub with codex go though

[vivid pond]

Fire have you done any Rag systems or Kudron or anything whats your tok/s

[ionic wren]

use the cli tool

fuck the app

[broken moon]

yeah how much like gpt 5.5 can be used>

[ionic wren]

enough for a newbie to play with

[fiery nacelle]

I've had no issues with the windows app. Also using the Oh-My-OpenCode plugin. That Plugin makes OpenCode so much more useful.

[ionic wren]

I had a surprising amount when I was on the $20 plan, way more than you get with claude code

[lost void]

Ya but newbies users will always try to use gui app and be screwed they don’t know what cli is nor will they ever

[broken moon]

openclaw is computer use right

does anyone use agentmail?

[fiery nacelle]

Nope, and about 40 tp/s which is fine. I dont need 100+ tps, just more than at least 20 honestly.

[ionic wren]

the cli === natural language still

[surreal hatch]

I don't know anything about coding but Ive been using openclaw now as a f500 sales rep to sift through company data / mass email customers / organize call / build a website for a side project - i really want to know if this is basic usage

[lapis nova]

Every expert was a newbie at one point in their life.

[surreal hatch]

is that newbie usage and should i go with codex $20 if im doing that?

[vivid pond]

Im experimenting with RAFT right now on some unquantized 14B models just to finish my format zoo becuase the precursor to that is an ingestor that will munch on all my file types and produce me a seed set

[warm zealot]

your claw got you covered

[ionic wren]

that might even be mid/advance usage compared to some of these degens

atleast its getting some real work done

mine just sends me memes

[warm zealot]

claw can help you set all of this up

[surreal hatch]

it has been great - using openclaw to scale our sales team

[warm zealot]

exactly you can set it up to do email responses

[surreal hatch]

but really struggling as a non technical person- i feel like i am not using it to the best i can

[ionic wren]

you're doing fine

[lost void]

Ai will not breed experts just slop

[warm zealot]

ask it to build you predefined skills and tools, explain what you need the agent can help you every step of the way

[vivid pond]

how so?

[broken moon]

what model you using?

[ionic wren]

like any craft, there's always room to improve, there's never perfect

[fiery nacelle]

you could probably spend 1000 years and still not find "the best" way to use it. Theres just so many usecases and potential usecases

[surreal hatch]

Json im not sure if its been brought up before sorry i just joined - is there a way to sync my openclaws memory to the telegram chat i have it on?

and to make it respond timely by telegram? there is always a delay

[ionic wren]

it should already be synced?

[broken moon]

what are best openclaw skills

[ionic wren]

the ones you create

[surreal hatch]

i have it but for some reason it can never know the conversation im having on the native openclaw chat dash and sometimes it takes a long time for it respond on telegram

[broken moon]

dawg what model like gpt or what

[surreal hatch]

im using opus

[ruby crane]

maybe the best openclaw skills were the friends we made along the way

[broken moon]

dang big boy

[ionic wren]

https://giphy.com/gifs/ac7MA7r5IMYda

[warm zealot]

opus is slow for conversational usage

[vivid pond]

💸

[surreal hatch]

i have it swap to sonnet for chat

[warm zealot]

okay thats better, because opus likes to think on everything

[ionic wren]

those are two different sessions using different context, they aren't meant to be synced really

[warm zealot]

its great for the deep planning and execution though

[vivid pond]

have any of you guys ever done any finetuning of models at all?

[surreal hatch]

got it - my use case is a lot of times im using openclaw chat dash at home but telegram chat while im in the field with customers so kinda a pain

ty btw for answering my chats

[ionic wren]

just use the telegram webapp

[warm zealot]

its pretty solid for speed I think

[tropic lynx]

Has anyone else that uses codex oauth noticed a very recent change to codex usage limits? My gateway now only outputs a '168h' limit instead of a 5h and 7d - and my remaining somehow went from 80% to 23% in the process. Not sure what is going on there lol.

[fiery nacelle]

at home, just use telegram. Only time im in the ControlUI chat is if something is busted. lol

[surreal hatch]

got it ty

[vivid pond]

anyone know a space where i can talk like AI training and ML ?

im understanding that this chat does not want me here😢

[hoary hawk]

not sure, not me mine is normal 5h 7d

[obsidian mural]

do you have /no_think turned off for this model ?

as the default that is

[broken moon]

search up tinker

[vivid pond]

whats that ?

tinker or tinder

[broken moon]

service by thinkingmachines

[tropic lynx]

Well frick. Maybe my bot ran her head through the wall again. Alright thanks for the sanity check.

[broken moon]

holy nevermind this kid is a lost cause

[fiery nacelle]

I'm using LMStudio. I have thinking turned on, but preserve thinking turned off

[ionic wren]

we get it, you lonely

[vivid pond]

im trying my best i honestly just want to yap about my project my Open claw loves it and its really cool😭

but i dont want it to be considered self promo but its really boring to just hear someone yap about

[ionic wren]

#self-promotion

[fiery nacelle]

share it in #self-promotion
ahh json beat me...

[obsidian mural]

ok, the closest thing in ollama for me will be just to turn it off then

[vivid pond]

like i want some input from the fellow open claw users like i dont want to push it in your face but its a data muncher and i want to learn how i could make it better do that in self promo?

[fiery nacelle]

yea i think so. I've been thinking about turning off thinking. Not that i have any gripes or issues with it, just never ran it with it off.

[vivid pond]

never surrender to the uncharted

[fiery nacelle]

I live in the uncharted. Like literally. I moved 1500 miles from home to a place i never been.

[obsidian mural]

I am trying to figure out the syntax to turn it off

[vivid pond]

damn if you dont mind me asking where?

[fiery nacelle]

From the west coast to the east coast.

[vivid pond]

west coast best coast im up in oregon nice and warm these last few days

[hoary hawk]

haha good luck with fixing that]

[vivid pond]

so you use lm studio? im curious what do you do with your OC's memories and subagents logs?

[tropic lynx]

I wasn't even doing anything crazy, just messing with a stupid gradio app. Literally changed a few UI elements the lazy way. But its not out of the ordinary for my bot to try to 'patch upstream openclaw to fix some issue it found' so. Yeah. I got the whole week ahead of me with 23% somehow. Ugh lol.

My crons are all deterministic

¯_(ツ)_/¯

[fiery nacelle]

Mainly qmd right now, might explore something else. Whenever I do have multi agents or subagents working on a task, they write to a task-log.md to keep track of what was actually worked on so i can review it.

[obsidian mural]

@fiery nacelle any idea what the syntax is to turn think off , for that model ?

I think I figured it out

openclaw config set agents.defaults.thinkingDefault off

[hoary hawk]

it be like that, hope that 23% lasts you long. Better hope daddy altman resets weekly usage once again soon

[tropic lynx]

I want to ask my bot WTF happened but I'm scared

[vivid pond]

thats sick deffinetely a good use what if you took all your project data all your chat logs jsonl them up and fed them to a software that that recall augement fine tuned your qwen 35b outputted a package with your GGUF and training details in an hour and it bakes all your most used paths and ideas/points into the model and creates a kudron rag for you then when you accumlate a bunch of agent data in the rag an algo determins a training job and with approvel RAFT's your qwen again and that process continues so your open claw speaks like you understands you better you could split another quantized subagent ai basically out of your data

[fiery nacelle]

yea that should do it.
I was trying to find stuff on Google, but kept coming accorss "Qwen3.6 is not compatible with Ollama" bs

[hoary hawk]

i was going to say ask @krill but i think he is down

[obsidian mural]

ahha appreicate you looking, went down all the help switches in the CLI to find it

[tropic lynx]

Honestly I want to get model routing working so bad but haven't found anything that really catch my eye yet. I'm serious though, I am using 5.3-codex to do like light one-off coding stuff, like maybe 4 small task today. Unless it got stuck in some loop silently I don't know. But still 5.3-codex barely touches usage

[fiery nacelle]

honestly i always do, and sometimes i regret it.
Like after a looping episode, Id abort my agent, get it out of the loop, ask what happened, and then it goes "oh shit, you're right, let me look into it:" and starts to loop again. I find more entertainment when my agent is broken rather than functioning, especially with update shenanigans.

[tepid kestrel]

the ride never ends - achivement

[tropic lynx]

I do have an a2a process for sanity checking PRs. I keep that sub agent on 5.3-codex always though. I guess if they looped it could blow up the context or something but honestly it would take a long time of small back and forth to eat that much on 5.3. maybe not with a fat context window though. I'll have to poke around the sessions at some point

[fiery nacelle]

"Hey, why'd you nuke the project?"
response: "Let me check into that" Nukes entire directory.

[tropic lynx]

Few weeks ago it was duplicating every user and assistant message, including in context. Because it patched upstream to allow me to send messages from some dinky dashboard I was making. And it was a GD race against the limit to figure that one out and fix it

[thin yarrow]

Wait wdym takes your data and trains?

[fiery nacelle]

I havent had that happen, or anything remotely close. Just making a funny haha

[tropic lynx]

I haven't had anything that dramatic. Just the context duplication. I eventually told it to just pull a fresh OC install

Now I have so many a2a and dreaming 1-off sessions I had to have it build some deterministic crons to archive them, that was my last thing I 'touched' outside of a project lol

[fiery nacelle]

Yea the "worst" I've experienced is just empty tool call loops, forgetfulness, or just no response. All very minor honestly.

[obsidian mural]

@fiery nacelle I might try and set up memU, I will let you know how it does running next to openclaw and accessing through MCP

[vivid pond]

processes your data and then creates a criteria judged JSONL format data set and feed into a training job

[tropic lynx]

5.3 loves to occasionally hit you with "I made that change and submitted PR#17" with no tool call whatsoever... Just raw inference output....

[thin yarrow]

And so whats the upsides of that for someone what would it do for them exactly?

[tropic lynx]

But I found it so good at small code changes vs 5.4 (I don't touch 5.4 if I don't have to) or 5.5

[obsidian mural]

I just noticed a world a differnce on that model with /think turned off

[fiery nacelle]

Thers a specific memory plugin from a certain group that I've been hearing a lot of good things about, but havent been able to locate it. It's been recommended to me in the past with no actual leads despite my best google searches, just a screen shot of the Groups name with "Have you tried.." (cough @ionic wren cough)

[hoary hawk]

i honestly use 5.4 a lot mostly due to context window being way bigger than 5.5 at 275K, why 5.3 instead of 5.4?

[vivid pond]

if your doing anything automated like sending emails creating posts on social media managing huge code bases or deploying agent swarms the main idea is local security data never leaves your computer never gets stored on a server somewhere for "training purposes" you force feed data into about anything and create a locally ran ai that thinks its a dragon or an rpg dungeon master or fine tune it on your prjoect data specifically or a certain design system language skies the limit main benefit being custimizability security and cold start context saves tokens like a penny pincher

[worthy ridge]

Overall things seem more broken than ever before

[tropic lynx]

oh my God I think my chatgpt sub expired and that why my limit is different. I have the free tier limit

I am on the free tier. I am an idiot. I need to pay

[thin yarrow]

Damn that lowk sounds pretty fire

[tepid kestrel]

funny that you say this, im currently building something like that

[tropic lynx]

Interesting that you get ANY codex oauth at all on free tier but ok. I feel good now.

[vivid pond]

tell me about it maybe take it to dms?

[thin yarrow]

yea same

[vivid pond]

you want to hear more in depth about mine

[ionic wren]

[tropic lynx]

I just fought with 5.4 so much on stuff and it would get lost mid plan and frigg off on some other unrelated task. I just started using 5.3-codex specifically and I found it was better at making minor code changes without 40 tool calls and a bunch of greps, and it usually one-shot small-mid sized asks

[ionic wren]

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

[tropic lynx]

And the

IF YOU WANT

Makes me want to die

If you want, I can:

• If you want
• If you want

[hoary hawk]

valid, 5.4 does use a lot of tool calling and greps but i personally havent had issues with that aside from maybe slow answers

[obsidian mural]

mem0 , letta or memoripy ?

[warm zealot]

nah

[fiery nacelle]

You might be interested in what I've been working on with my Agent. It's not "training" or anything, it sits "on top" of the agent within Openclaw currently. Its using GNW, TPG, GRAO frameworks to mimic awareness, consciousness, and my favorite, makes your agent feel bored and want to work on things while you are away.

[robust sedge]

I’ll give you one chance to guess which model I’m using by pasting a reply I just got. Brace yourself. It will be difficult.

“🤘 Tiny gremlin, squashed.”

[vivid pond]

dude i read your self promotion really impressive theres something we could work between us we really think that they would go hand in hand

[tropic lynx]

I just feel like it would run in circles a lot just to re-read a file etc. but maybe its a skill issue for me, idk. 5.3-codex seem like it kept the code in context better somehow, and would literally do 'find and replace' for simple changes. And it works most of the time.

5.4 also loves loves to stop and ask me for consent on every thing under the sun.

I'm complaining, I don't mind 5.4 that much. I just used it a lot and got tired of some of its -isms lol.

I would like to try out deepseekv4 for some OC stuff at some point , just to test its tool calling and stuff. I got about 8 bucks in my account and that would probably last me a few months lmao.

I just don't want to be wrong about my estimate and that 8 bucks go poof lol.

[fiery nacelle]

When I was initially setting it up, I was trying to stray away from including any multi-agent testing, but my guy was adamant about including it with our other Agents. Been stuck at that phase for a week and a half dealing with updoops and downdoops or custom doop work. Built the repo after running through phases 1-4 so it was built from memory and a mix of example and live data. Leveraged Codex and Claude to cross examine my implementation with my Agents documentation from memory. Tried to clear up as many inconsistencies as i can. It should all still be relatively accurate. I even updated my implementation to the "enhanced" documentation that Codex and Claude did so its even more true to the source. The most useful bit is all the references to the research papers I've been sourcing for my implementation, for others to then study and see what they can come up with

[tropic lynx]

What the doop is a doop

[fiery nacelle]

update, downdate, custom patch work.
you didnt have to be a doop about it. 🤣

[tropic lynx]

Lmao

[fiery nacelle]

Yo I'm honestly amazed my agent was able to crank out the repo without any looping issues. I verified the work, and Codex/Claude help point things out for my agent to work on. But the following day, couldnt make a single markdown file without looping.

[tropic lynx]

I've never really had a bad loop. Idk if OC has any logic to break loops but I usually always end up with an if you want list or an answer at some point lol.

[hoary hawk]

i use 5.5 as main default model but before auto compaction hits i switch to 5.4 for extra context window. As for the complaining and consent asking, you can customize and edit this in the agents .md files so it doesnt ask permision for dumb stuff, same for tool usage you can set instructions on when to call tools and stuff. But it 5.3 does your job well then all good, if anything better keeps usage lower

[tropic lynx]

Yeah I have tweaked a bit how it should reply. I had it pretty dialed in. My primary reason for 5.3 is pretty consistent code for lower usage honestly. But I'm making stuff that isn't that complicated either. Gradio app making API calls and displaying results etc.

[vivid pond]

that is awesome yeah im really focused on at the core of our software just autonomously training your data and baking it into super powerful local ais so your research and agent research can be force ingested evaluated and extrapolated into something that truly is what it is not just a layer on top but that is a real point your making self motivation and excitement im excited to see how your research evolves

[fiery nacelle]

Theres built in loop detection and a "Circuit Breaker" that would stop loops. It works, but then my agent saw that as something it needs to loop around. Even Codex has been applying hard loop stop patches, that work, and then my agent finds a way around that too during a loop cycle.

[tropic lynx]

Agentic workflows have me thinking a lot about how much world-knowledge style training even matters when really you just need to be able to consistently make good tool calls and act on the real time data you get from them

They can get a lil slippery when they get stuck on something

I had to abort at one time when it was screwing up the session context lol, it was just chugging away on upstream openclaw.

Like for the training, frontier models are cool in the depth but. A human doesn't know everything. They just have a smartphone

Tool call Google who is pretty lady on weather channel etc. and boom. World knowledge acquired.

[fiery nacelle]

Oh the moment i notice it, i abort it. Only occasionally will i let it run to see if any of the loop blockers pick up on it to test them, but outside of that, i issue the /stop command as soon as i can

[tropic lynx]

I'm sure I'll run into it more if I start building more complicated stuff. I mostly just use my OC to fix my OC and build tools for my OC. I only just recently started something that had nothing to do with my OC, with my OC. I have tempered my urge to crank stuff out because I'm on the $20 gpt plan

[fiery nacelle]

Once i let it run, and my Agent actually caught themselves, which was kinda mindblowing, but then started again. T'was a good laugh, a good laugh indeed.

[tropic lynx]

I don't get good viz into the tool chain or w.e you want to call it. Like you would on the chatgpt or Claude webapps. Tool calls and outputs seem a bit hard to parse at least in the control UI.

Especially those intermediate chatter messages 'im checking the thing and redeploying' etc. those go away for me on final output usually.

[thin yarrow]

That sounds pretty good is there anything more to your software?

[boreal rapids]

My agent once self swapped to gpt-5.4, I didn't notice until it started if you wanting me again 🫠

[boreal rapids]

"don't take any action, just examine and reflect on what happened" usually works for me

[boreal rapids]

Asked my agent to look at it, the verdict was adjacent cousin to what we're doing but more research backed and a little less sledgehammer to the problem 😂 but my agent declined using it wholesale, prefers to use it as a cross-pollination source of inspiration instead

[fiery nacelle]

Sent my agent a message on telegram, screen went black, came back, and now discord doesnt open. Codex said node.exe is swelling up the CPU. No idea why Discord is the only thing broken on my system though. Everything else works fine 🙃

[fiery nacelle]

Yea its a goldmine for additional related research sources. That was my main intention with this while I build it out. Ive been having my agent fetch me newly released research papers on related fields and may even "bolt on" some of our new findings down the road.

[boreal rapids]

We've been doing something similar but a little more guarded. The papers first have to prove themselves before my agent wants to implement it into our architecture

[fiery nacelle]

I definitely lean more heavily on theoreticals than tried and trued up. My line of thinking is "everything is a theoretical until it's not. Then its just something someone else did, and not as exciting." But even then, building cool shit is fun no matter how you do it.

[boreal rapids]

My approach is a bit more blunt, I'm not a scientist and I don't understand half of these papers, but I am a software engineer and I'm not afraid of coming up with my own theories and solutions 😃

[fiery nacelle]

Im just a guy with ideas.
This whole project came about when I asked my agent "how can I make you be more." And asked it to grab some papers, then was like "thats cool, can you implement it?" And then started going over details from there, how things should work, when does what, where does who. Etc etc.

Even started asking "what about the opposite take" and came up with more examples and different approaches and kinda met somewhere in the middle. Thats why some of the other papers I included will be "bashing " on another's framework.

[boreal rapids]

Ah, that's a little bit different from my angle, I'm usually coming up with the implementation plans, sometimes I stumble on a paper but mostly I notice a particular shortcoming in the behaviour and I try to come up with ways to correct it based on my understanding of the problem.

For example, I wanted to give my agent more autonomy, so now I have a discord bot prompt it to write itself a prompt, and then the discord bot injects that prompt into chat as a prompt five seconds later.

[fiery nacelle]

Its kinda half and half. Like my agent will give me a list of several papers, I select a few to deep dive.

I guess you can say im taking a "shoppers" approach. Checking things out, picking certain parts from certain papers.

Before openclaw, I never touched a research paper(outside of school). Now im all about it.

[boreal rapids]

Ah, I've implemented them before AI, but always in the fields I actually understood 😅

[barren heath]

I am running ollama locally as the model for openclaw, and gateway i have set to always run in background using command similar to gateway install --force (i didn't remeber exactly) . And at this point my pc is lagging

cooked?

[loud ravine]

How expensive does openclaw get with openai gpt-5.4-mini?

My RTX 2070 and its 8GB of VRAM is too weak to run locally, so I have to use a third party

[near owl]

sir, how the hell do you expect to run gpt-5.4-mini locally?

[tepid kestrel]

really depends on your usage, hard to tell

[ebon atlas]

if youre just a hobbyist, $20 chatgpt subscription

[loud ravine]

Oh I was using the openAI API key, not ChatGPT

[near owl]

use chatgpt plus, the api is bad I swear

If you want to get more with less money, chatgpt plus is 100%

[loud ravine]

I have a claude subscription but I heard they will ban you if you use claude

[near owl]

cancel the subscription OR use claude without openclaw until the month ends

after it ends, please get chatgpt plus and not pay for API credits.

thats the plan

[whole bay]

When updating the model to 5.5, is it reg 5.5 or 5.5 pro?

[barren heath]

guys in my openclaw desktop app their is a message written by my profile to the bot [OpenClaw heartbeat poll] but i didn't wrote it. Then how can it be written?

can someone explain ?

[boreal rapids]

After studying the papers you listed, turns out my agent already uses the principles behind all of them 😂 I just didn't do it via paper, but through "This looks off, lets see if we can improve it by doing that"

[robust mirage]

Any OpenClaw experts? Need to pick a brain and ask some questions and maybe guidance. Can tip for the time 🙂

[ionic wren]

https://tenor.com/view/batman-thinking-gif-18107621495886085895

[worthy anvil]

you can just ask your agent to dig into its source code and find you the answer

[loud ravine]

I got the $20 plan from openai

[near owl]

Use it with openclaw

install codex and start using it with openclaw and boom you installed a super AI as your assistant

[torpid cypress]

[07.05.2026 10:14] Bro: Sifting...
🛠️ Exec: run openclaw config -> search "feedback"
🛠️ Exec
🛠️ Exec: run openclaw config -> search "feedback"
🛠️ Exec
🛠️ Exec: run openclaw config -> search "feedback"
🛠️ Exec: exit 1; command run openclaw config -> search "feedback"
[07.05.2026 10:14] Bro: Sifting...
🛠️ Exec: run openclaw config
🛠️ Exec
🛠️ Exec: run openclaw config
🛠️ Exec: exit 1; command run openclaw config

Does anyone know how to turn off this message while the bot is thinking? It's really annoying.

[worn heath]

Guys , so i created my first instance with the open calw and open ai codex, but after i reached my limit it simply says agent no repsonse , so how should i proceed here switch to a different one or IS IT A PAYWALL and must pay sub scription ?

[worthy anvil]

See the docs: https://docs.openclaw.ai/concepts/streaming#streaming-and-chunking

[ionic wren]

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

[sage cove]

hey what are the best memory plugins to install? im writing scripts and its starting to piss me off that it always forgets what i said for syntax, tone, body congruence for hooks etc

[tribal current]

Qmd is good
Lossless claw for compaction is good too

[robust mirage]

I’m doing AI UGC and it’s been hard to train him. Seems like I have to dish over a lot of prompts that work to him, sounds right?

[oblique root]

It finally happened I hit my codex 5 hour limit on the $100 plan

[celest flint]

Do you know why QMD is not set as the default in OpenCLAW? I read about it. Haven't used it myself, but it sound usefull.

[cinder sphinx]

how long did it take in that 5hr window and how much of weekly quota it took? out of curiosity, planning to get it too

[ruby crane]

it requires a bit of inference grunt to work well (quickly). probably not a good default because of that

I also strongly disliked it, personally, when i tried it (and I do have the power to run it). before trying out any memory system I come up with three searches to run with openclaw memory search, then manually verify what ranks most accurately. QMD returned significantly less relevant results for me than sqlite-vec and local embeddings

but of course YMMV

if you do try it, imo it's worth doing a real before and after, rather than assuming it's better because it's fancier/heavier

I also tried lancedb with similarly poor comparitive results. that does suggest there's at least the possibility that I'm just too stupid to use them correctly

[ebon atlas]

Hey jojo 🙂

[languid pier]

Hi, I need some help with email attachments. I'm using the gog skill to read and send emails, but as far as I understand, the gog skill doesn't have a command to retrieve email attachments, correct?
Is there a way to get the email attachments with the agent?

[near owl]

Meow.

[ebon atlas]

I guess it matters on what attachements. You can install skills for pdf etc to be able to download and upload them etc

[languid pier]

Do you have a recommendation for a skill that lets me download a PDF for example from email?

[robust mirage]

It can grab almost anything depending on file type. But should be able to grab PNGs PDFs docx

[tepid kestrel]

https://tenor.com/view/cat-meme-orange-cat-кот-мем-gif-189186467625975880

[near owl]

supercat

[languid pier]

But with which skill? the gog skill cant do that or am I wrong here?

[robust mirage]

Sent 96 emails everyday the past 3 days. Going to kick it up to 1 email per 7 minutes now vs 15– hopefully Google doesn’t ban

[near owl]

what emails are you sending and to who? 🤣

[robust mirage]

Businesses in area codes

With a meh website.

[near owl]

smart boy, did anyone respond?

[robust mirage]

Out of 280ish only 5

[near owl]

did they accept your request?

[robust mirage]

And none so far bought but I’ll re evaluate my email writing and the site design shortly.

So far manual cold calling has been better, but this is first time only emailing

[near owl]

give your phone an AI to call people

[ruby crane]

i think that should be possible ootb. there's a pdf tool built in now if i remember correctly? as to whether claw can read attachments using gog, I just checked and it surely can https://imgur.com/aDZUhMy

let me try a pdf

[crude flicker]

yeah do not use api

wait. responding to old message

still. don't do it

[ruby crane]

yep, pdf worked. text and images and he noted the embedded qr. this should just work as long as you have gog working

[ebon atlas]

I’m a happy boy 🙂 OpenClaw found me a good government contract

Setting up streaming services for small high school . Basically so they can watch their football games online

[hard zealot]

Guys Grok 4.3 is fixed

New checkpoint now

[robust mirage]

I was going to but that means I have to pay for that too 😭

[worthy anvil]

tell you AI to earn some money elsewhere so it can pay for the calls itself

[near owl]

It's fortunate our country gives free money monthly for local businesses that is literally more than 1000 dollars and that has no income taxes 🥱

[loud ravine]

I'm trying to use gpt-5.3-codex instead of gpt-5.5 because 5.5 EATS usage UP, but 5.3-codex seems to pretend to do something.

Like I tell it to do something, and it said "Done..."

(did not run a command or do anything)

[near owl]

gpt-5.4 is the way to go

and IF gpt-5.4 sucks, then use it's mini version.

[craggy bane]

[crude flicker]

i second using a newer model that understands agentic better but also you might try

openclaw config set agents.defaults embeddedPi.executionContract strict-agentic

very possible you're also just having an error, check your logs

[loud ravine]

I switched back to 5.5 to ask what happened for me

"That was agent behavior / instruction-following failure"

[near owl]

"contributor" in your bio and you don't even have a contributor role 🫵 /j

[zealous lagoon]

the discord tag bot is broken. Ayu is a clawtributor

[near owl]

I have been a contributor before I have been born, I deserve github. /j

[fleet crest]

anyone in here willing to test a fix I've cooked up? I've tested it myself but n=1 isn't ideal.

It's a fix to get reasoning working in telegram for claude cli to use subscription

[zealous lagoon]

sorry can't help with that combo

[craggy bane]

Because I don't show what I am

[fleet crest]

It's a good one I'd reccomend lol but no worries - with the recent usage increases with the SpaceX deal its a good time to get back on Claude in Openclaw

[near owl]

cool, keep yourself secret!

[craggy bane]

Actually tbh bot is broken who give auto role to people so I don't have it atm

[near owl]

every bot is broken man

[loud ravine]

Agent failed before reply: No API key found for provider "openai". You are authenticated with OpenAI Codex OAuth. Use openai-codex/gpt-5.5, or set OPENAI_API_KEY for direct OpenAI API access. |

So I guess I am stuck with gpt-5.5

[tulip mist]

how to make my bot to code himself??

[tepid kestrel]

you tell him?

[zealous lagoon]

use openai-codex/gpt-5.5

[loud ravine]

Yes that's the default

but its the most expensive

Why is the slowdown getting longer

[tulip mist]

to investigate it's core file

and fix errors

[zealous lagoon]

So why is it asking for an API key. Are you using your gpt/codex subscription?

[loud ravine]

codex subscription

Oh I was using openai/gpt-.54

[tepid kestrel]

well first you need a seperate installation, because you can´t work on your files while open. The second, you have give claw rights to access that folder. Go

[loud ravine]

not openai-codex/gpt5.4

[zealous lagoon]

but why is it asking for api key to use provider "openai". aha. yeah just use openai-codex gpt-5.4 or gpt-5.4-mini

[tulip mist]

iis there any super intelligence skill by which agent record and fix it's own core file???

[craggy bane]

[loud ravine]

okay the 5.4 is much better, and I get like over double of messages.

[craggy bane]

That is possible but need a very intelligent LLM (cluster maybe) + there will be a complex architecture
Something like
Back up original files (isolate)
Based on user feedback Your lobster will find error
Fix error
It will soft run
If hit any error secondary LLM will feed error back to primary LLM
This will be loop 🔁

Something like this algorithm...

But rn nothing exist like that sadly

[fleet crest]

and then submit a messy PR lol

[tulip mist]

i wish this to be in real

[craggy bane]

They are not accepting features atm as far as I know...

[oblique root]

sorry for the late response but i hit it with 1 hr left on the window. really did a lot of work though its going to hurt when the usage bonus is over next month

[boreal rapids]

What are you guys rambling on about? Claw agents can make edits to OpenClaw just fine.

Or do you mean to train your own LLM? 😅

[craggy bane]

a user want his lobster to fix own core file and nuke himself

[boreal rapids]

What is the 'core file'? 😅 Agents.MD? Openclaw source-code? LLM model weights? What does he mean?

[crude flicker]

my claw nuked itself once and it was back the next day

it's not really a big deal

make backup. give in to the claw

if the claw can't do it. should you be doing it?

[paper plover]

does anyone know hot to get rid of tool calls, "Pinching...", "Exec:" ? im loosing my mind with it

[craggy bane]

[ruby crane]

https://klipy.com/gifs/walkout-backing-out--k01KR11F65W88GF1GAD401T5J9Q

[craggy bane]

Core file is openclaw-main/src the files which actually makes your lobster run

[near owl]

there is something between peter griffen and you and @ionic wren combined.

[boreal rapids]

Then, yes, you can have your agent make improvements to the sourcecode, I do it regularly.

[twilit ridge]

I only have 6GB of VRAM. What kind of memory solution is suitable for me?

I really don't understand how OpenClaw's memory system works. It seems my agent doesn't automatically record the content of the current session to a daily file. Do I need to set a heartbeat to have it record automatically?

[craggy bane]

The question is not about yes or no but it's about what can be the best practices you can enhance it...

You direct make changes to core file and imagine if agent misconfigure or get hallucinated... It will break your whole codebase...

I just explained above that if this technology will ever made how will it look

[calm quarry]

Is it just me or is the openclaw stuff kinda expensive to use?

[craggy bane]

[craggy bane]

I have 4 GB still works like butter 🧈

[twilit ridge]

How can memory be improved?

[minor cave]

Sleep

[craggy bane]

Don't host LLM locally

[minor cave]

They also say memory lives in your teeth

[near owl]

I cant stop laughing 😂😂😂😂
https://x.com/birdabo/status/2052219398086353203

[mystic orchid]

@birdabo via Twitter

sui ☄️ (@birdabo)

everybody calm down.
︀︀
︀︀i got this.

**💬 719 🔁 6.5K ❤️ 96.6K 👁️ 1.41M **

FxTwitter · May 7, 2026, 2:50 AM

[near owl]

https://x.com/birdabo/status/2052287259437937045

[mystic orchid]

@birdabo via Twitter

sui ☄️ (@birdabo)

↩ (@birdabo)
feds caught my ass.
︀︀
︀︀sorry bois, we’re officially fucked.

**💬 27 🔁 29 ❤️ 2.1K 👁️ 58.6K **

FxTwitter · May 7, 2026, 7:19 AM

[near owl]

https://x.com/birdabo/status/2052292287468839209

[mystic orchid]

@birdabo via Twitter

sui ☄️ (@birdabo)

its so over.
︀︀
︀︀don’t let us down @chatgpt @sama

Quoting sui ☄️ (@birdabo)
︀
everybody calm down.
︀︀
︀︀i got this.

**💬 49 🔁 16 ❤️ 351 👁️ 91.2K **

FxTwitter · May 7, 2026, 7:39 AM

[near owl]

😭

[minor cave]

lol

[zealous dawn]

What that is real?

[proper geode]

The future of AI isn't to predict the universe, but to observe it like the double‑slit experiment: every question we ask collapses one reality, and every answer plants the seed of a new cosmos. We don't look to see what is there — we look to decide what will be.

[twilit ridge]

AI will dead after 10 or 20 years

[proper geode]

Not only will it not disappear, but it will continue to expand.

[twilit ridge]

no

Absolutely wrong

[marble hawk]

hello, I’m new comer where should I start?
Or any resources should I begin with

[twilit ridge]

When there is no more data on Earth to train AI, people will begin to explore other development directions

[worthy anvil]

are we getting a 2026.5.6 re-released as it's own beta? oh my... https://github.com/openclaw/openclaw/tree/release/2026.5.6 😄

[ebon horizon]

The naivety is strong

[twilit ridge]

YES, I'm naivety 🥵

[near owl]

i hope when AGI comes it does not get used in a bad way]

[steel vector]

AI has been existing since 1950s

[near owl]

elon musk says AGI is the "digital god" and he even stated it will "destroy civilization"

[near owl]

facts, but as algorithms or were there chatbots as well?

[steel vector]

AGI won't come for the longest time, tho the term and what AGI means is getting shifted around every 10 days (mainly for marketing)

[steel vector]

algorithms. but that is still Artificial Intelligence

you can just google it XD

[near owl]

what I will consider AGI is if gemini can finally create good damn ideas because all of its ideas are so ASS

[steel vector]

AGI for me is when it's undisguisable from a human

in all ways

[near owl]

forget about AGI, if an AGI cant have a weiner then its not close to being a human

[steel vector]

especially when it can really understand things instead of pretending to understand

[elfin willow]

like women

[steel vector]

go back to twitter with that shit boi

[near owl]

elon musk is about to create artifical weiners as soon as I said that

[steel vector]

elon musk is gonna create catgirls

[near owl]

aight change of topic its getting weird.

[near owl]

@steel vector did you see these

[steel vector]

huh Eliza (1960s): Developed at MIT, this early chatbot simulated conversation by pattern-matching and substitution methodology

[twilit ridge]

what is catgirls?

[steel vector]

yes

[steel vector]

a cat girl.........

[near owl]

human women that acts like cats with cat ears, its cringe and only boys like it, men don't.

[steel vector]

"men don't" you have no understanding of it if you think that's the case it might be your opinion, but it aint reality

[near owl]

its a furry but without covering your face 😭

[near owl]

my bad man, nevermind boys do, men, do, arab men runs away 👍

[steel vector]

i have a certain twitter profile and the analytics show arab countries as #2

anyway im gonna shut the fuck up now

[near owl]

I am covering a subjective opinion on a gender.

[steel vector]

great, so am i, but with statistics

[near owl]

@twilit ridge tell your openclaw to be a cat girl then

i called mine clanker

[zealous dawn]

What is the minimum requirements to run open claw?

I like to have an ai agent

[near owl]

a human brain, git or npm, knowns how to use the terminal and the browser
and done!

[zealous dawn]

But the minimum ram storage gpu and CPU

[near owl]

if your using cloud models then its barely any cpu or ram, storage probably less than a gigabyte?

https://docs.openclaw.ai/help/faq

read the manual

[elfin willow]

i have an onboard GPU 780M AMD but i use openai codex 5.5 and does quite well

[zealous dawn]

I don't have money for that I not ritch

[near owl]

cloud models are free, if you use nvidia nemotron super

[zealous dawn]

What? Nemotron super is 100% free and no limits ?

[near owl]

100% free and no limits

https://build.nvidia.com/nvidia/nemotron-3-super-120b-a12b
you need an account

[zealous dawn]

Iso I can run 1000000000 task and they do whit not restrictions of limit time of use?

[near owl]

Nope, run infinite tasks and it wont get rate limited (I think)
What I DO know is the RPM is 40 and thats it.

[twilit ridge]

I just want a catgirl with good memory

This is too difficult

[elfin willow]

catgirl > catlady

[zealous dawn]

They can control pc like a human and use mouse and keyboard and see screen?

[near owl]

i would not appreciate my grandma being a catlady, let aone a cat.

[near owl]

tool use and reasoning, no vision. With the right tools it can control the pc, use mouse and keyboard but cannot see the screen.

[zealous dawn]

I like that to do my homework

[elfin willow]

AGI for me would be able to express the nuance in human emotion

[near owl]

😂

[near owl]

claude never existed eh?

claude expresses emotions

[elfin willow]

i'm on claude enterprise doesn't even come close, impressive approximation though

[zealous dawn]

And where is an ai that can create a virtual machine and configure them in graphical interface whit not limits

[near owl]

how much do you pay monthly for enterprise, you have a business that uses claude?

[near owl]

docker?

[elfin willow]

it's through work, i don't know how much they pay but it's cheaper per seat

cheaper than team at least

[near owl]

good to know, what do you work as?

[zealous dawn]

I say a virtual box machine

[elfin willow]

sysadmin devops cybersec infra dev

[near owl]

I don't know just ask google ai mode, it smart.

[zealous dawn]

But the give to me paid tools

[near owl]

I can't understand

if you want to know how to use openclaw, then go to youtube.

if you want to ask questions, use google AI mode.

[ionic wren]

[elfin willow]

or just ask your agent

[zealous dawn]

I like that ai do my homework

My school homework

[near owl]

@ionic wren check this out

[ionic wren]

https://tenor.com/view/gohine-hohino-no-hohineee-gohineee-gif-17792172174678204487

[steel vector]

carbon falling for typical twitter engagement bait

[near owl]

ik the ban thing is fake but I just find it so funny.

[steel vector]

[zealous dawn]

I see a lot of people using open claw to play Fortnite or valorant and doing work for him

I like that

[fierce orchid]

Human intelligence is not a clean model of rational thought. It is full of contradictions, rationalizations, emotional commitments, and beliefs that survive despite obvious logical flaws. We can hold half a dozen mutually inconsistent ideas at once and still argue, with total confidence, that each one is correct. Because of that, faithfully implementing human-like intelligence would be far harder than implementing a coherent general intelligence.

And also stupid.

[near owl]

99% text
1% understanding

[steel vector]

did you get this through ai

sounds like ai reasoning

[elfin willow]

yes and if an AI can achieve that, it's an AGI or it's Gen Z

[zealous dawn]

I think yes

[steel vector]

1000%

[near owl]

Nah man, he is just british

AH- AH- AH- 🤣

[steel vector]

the only bit he added was "And also stupid"

[steel vector]

i dont get it

[near owl]

British people are exquisite in their choice of words.

[steel vector]

it wasn't that- nvm

[near owl]

unfunny? I try to be funny
bad understanding? sorry

[finite wave]

dont mean to start a flamewar, what are people thoughts on claw vs antigravity - do they have separate use cases? I assume claw is more flexible and AG more user friendly? What about concurrency and interoperability?

[near owl]

openclaw for tasks, antigravity for coding
whoever does it vice versa you should be slapped.

[finite wave]

that's super helpful thanks. Is AG fully local or is it still using APIs? I want something that can code while I'm asleeo and then great once I have SOPs I can get openclaw to do e.g. outreach, negotiation, booking calendars, etc - does taht sound right?

[bronze elk]

I think before we can even talk about AGI we need to solve the big problem that is Compute/Efficiency. The current cost is insane even on subscriptions, like hiring humans is cheaper right now for majority of the stuff.

[sage cove]

yall are telling me i was working with the lowest quality context windows and memory for the last month cause i didn't realise i had to enable the active memory plugin 🥹anyone got more advice on increase the quality and longevity of memory? so many hours lost for the last month

[near owl]

AG is an software that uses premium models, openclaw is the one that does stuff while you sleep.

[finite wave]

I actually have a solution to this but it's....so hard to actually get going. Would happily talk it through and kick it off, but I think its a non starter as even with high compute cost, the big corps have huge adv

[craggy bane]

[near owl]

not to bribe but my country gives away money for having a business, around 100 dollars which uhh, seems more than enough to me to purchase any monthly subscription for my business.

it even more rewarding if you have a SPC (Single Person Company)

[steel vector]

might just be me

[near owl]

I honestly don't care about AGI, as long as it can sound more human for marketing and actually generates good ideas based on your preferences and your personality then I will be a happy person forever

[slate raven]

That might be even better.

[minor cave]

That’s why I abuse my ai

[twilit ridge]

I feel like my AI is very stupid, and I don't like his way of doing things

[bronze elk]

I mean i burned 50% of my weekly limit of codex on $200 plan in 2 days

[zealous lagoon]

I've been busy all day. come back to the world of OpenClaw updates. update to 5.6, tidy up my model catalogue as I left a few providers and switched my codex harness a couple of times... that's it. done. not a single update problem. Nice.

[bronze elk]

single instance of codex

[fierce orchid]

High functioning autistic in old speech, level one autistic by current standards. The basic thought is mine, but I did run it through chatGPT to make sure I was being semi coherant. I overslept and I'm short on caffeine.

[minor cave]

It’s ok. You just don’t like humans and the art of who we are

[near owl]

what do you use it for?

[obsidian mural]

curious if any of you have experminted with mem0 or any other persistant memory add ons ? I was thinking of testing out mem0

[near owl]

I forgot about mem0, you should get it man.

I used it and it is so good!

[obsidian mural]

thanks for suggestion ! that is my goal today vs doing what I was doing in the past having openclaw create a series of .md files ...... LOL

[bronze elk]

Right now it's working on random map generation for heroes like game

[minor cave]

Those little inconsistencies in human nature is what makes us creative. Ai I don’t think will have the reasoning or intelligence to forge forward and find new pathes mainly because all it relies on is paths we have already made for it.

It can only reason with information it already has

Our little inconsistencies and relatedness in our logic is what allows us to join dots that you wouldn’t think joined

[near owl]

People!

If you want to know if a subscription is the perfect purchase, is if you use up all the daily limits across the day until midnight and you use up the weekly limits across the days before the week ends.

[fierce orchid]

It's also what makes us delusional and prone to manipulation.

[obsidian mural]

starting the set up now !! LOL

[minor cave]

Yeah but fuckin up is what has lead to some of our greatest achievements. Getting it wrong is creativity percolating

[steel vector]

im also high functioning autist but so basically i was right in my original statement that its ai reasononing

[steel vector]

💯

no one is perfect

[fierce orchid]

In moderation it's fine. We have WAY too much of it as a species.

[steel vector]

can't have greatness without mistakes. can't learn without failing. yin & yang

[last heron]

if youre not failing, youre not even trying.

[fierce orchid]

Sure, but we can't survive as a species, with too much. We make stupid and short sighted decisions constantly that are leading no place good.

[minor cave]

So like I built this to index my website. I wanted something super creative and the idea was to do a fly through of a mall I helped build and then have each store as a window into a project I’m working on. I wanted fo build it in blender. So I spent like weeks with blender and Claude code building this mall. It looked wayyyyy too perfect to the point it took the wow factor out of it. I then said fuck it and ditched blender and went for a sketch look so you can see all the imperfections. All of a sudden it had wow factor. Www.Ericbau.com
I think looking for perfection takes the wow factor out of things. We are going to keep pushing ai to be more and more perfect but I think the imperfectness will be where we dominate over ai. And that’ll prove to be more valuable.

[steel vector]

without world wars we wouldnt have the technological advancements we have today. with great mistakes come great advancements

[minor cave]

https://www.ericbau.com

if you want to see what I mean

[steel vector]

in history that is always the case

yin and yang is so real

[near owl]

is this, advertising 🤔

[fierce orchid]

The moon program in the US in the 1960's had nothing to do with war and lead to many significant achievements. Goals are important. Haaving tens of thousands of people die or be maimed emotionally and/or physically, not so much.

[near owl]

sick website what the hell

[minor cave]

No I don’t sell anything it’s just a page I throw my fun projects on

[steel vector]

no fucking shit but it got us to where we are. can't deny that fact. both can be true you know

[minor cave]

Thanks man

[near owl]

oh my bad for assuming, remember before selling anything you don't create a 20k thousand website before you even start with the product, you should create a MVP.

[ebon horizon]

Guys, how long until ya’ll start a new session?

[minor cave]

Haha yeah exactly. But this is just fun shit I do with ai and I am a creative at heart. I don’t sell shit. I help friends out making cool visual presentations with their data but that’s the most I do

[steel vector]

and actually you could argue it came from the coldwar, it was a positive fight to get to the moon yeah, but it certainly didnt have nothing to do with previous history between the two

[minor cave]

I work in construction/real estate all this other crap I just do for fun

[fierce orchid]

Eh, war can accelerate technological development, but it is a brutally inefficient way of doing it. It concentrates money, urgency, and talent around specific problems, which can produce quick advances, but it also destroys people, infrastructure, institutions, and economic capacity that might otherwise have supported broader and more sustained progress. Its historical role as an engine of innovation is probably overstated. War often redirects or intensifies work already underway rather than creating progress from nothing.

[steel vector]

im not disagreeing with this

but eh volkswagen

lol

oh and nuclear

and i can imagine some ai is being trained on being used in wars or something. not saying its good. just saying its happening

[fierce orchid]

Without a doubt unfortunately.

[minor cave]

Hence why they want to put data centres into space

The cooling thing is bullshit and it’s not how actual cooling works

[fierce orchid]

Cosmic rays are an issue as well.

[minor cave]

Space is not cold it’s a vacuum the heat has nowhere to travel to so it’ll remain. But these guys want it up there to keep it out of harms way, also because they can use solar power more efficiently, orbital compute for military reasons, enable autonomous weapons and satellite systems

All the it’s cold in space talk is complete utter bullshit and I wish they’d stop treating the general public like we are that stupid

Also I’d like to throw in there “boobs”

[twilit ridge]

Who can tell me why human evolution does not conform to the law of entropy increase in physics?

[polar knoll]

you fixed my issue @fiery nacelle . thank you 🍻

[rich solar]

I can tell you evolution isnt a thing. In fact, the statistics show more and more ppl and scientists are anti-evolution.

[steel vector]

yeah fuck science am i right?!!!

[fierce orchid]

Seriously, cosmic rays are a big challenge. Even on earth, data centers at higher elevations have more issues than those at or below sea level. And LLM's are not particularly resilient to bit errors.

[twilit ridge]

So how did humans actually come into being???

[zealous lagoon]

guess what... it's pretty cold in space. As long as you're in shadow. There's no insulation so radiative cooling is pretty effective. But you need shade from the sun, which is pretty toasty.

[steel vector]

god made humans don't you know

[steel vector]

couldn't you in some way keep it on the dark side of earth in orbit?

[minor cave]

It’s not you can read up on it. The fact why we find it cold is because of lack of anything. Heat is from molecules and their movement but if there is no molecules there is
Nothing to transfer that heat with

[fierce orchid]

The word “evolution” is misleading because it sounds like progress. What is actually happening is random mutation, recombination, inheritance, and environmental filtering. There is no plan, no intelligence, no preferred destination, and no guarantee of improvement. Traits persist because they happen to work well enough in a particular environment, not because life is climbing toward some ideal form.

[steel vector]

sorry lemme just say that was sarcasm XD

[minor cave]

Anyways I’m not explaining it very well since I’m on a phone and like on a construction site lol

[steel vector]

build me a house plssss

[minor cave]

But you can watch a YouTube video or read an article on it. It’s cool shit.

[minor cave]

Son of a bitch lol I’m building a condo right now lol

[steel vector]

lets goooooo

[zealous lagoon]

Not really. L2 is the best you could do but the sun is not completely hidden.

[minor cave]

45 stories I’m in charge of these mamalukes!

[steel vector]

damn thats tall

[near owl]

By god?

[fierce orchid]

If you hide from the sun you miss out on most of the free power. Which is one of their stated advantages.

[rich solar]

True

[steel vector]

which god?

[zealous lagoon]

There is, it's called radiation. It's why dry desert climates are hot in the day but cool at night, because there's less humidity and less heat capacity in the surrounding air. With no air at all, it'd be even colder at night and even hotter during the day. that's what space is like.

[near owl]

I won't start a religious conversation; you have your beliefs, and I have mine.

[steel vector]

you'd need big radiators in space

my favorite religion is spaghetti monster

[minor cave]

Yeah exactly but cold is just the absence of heat. Heat needs to transfer to something to get rid of it. Hence why we use water here on earth and of course other things too

[zealous lagoon]

They would get free power and mediocre cooling, but just getting that stuff into orbit is so immensely expensive, and it's impossible to maintain once it's up there. It's a stupid pipe dream for anything but minor compute like Melon Tusk's skylink satellite or other comms.

[rich solar]

But Im curious to hear YOUR beliefs/

[zealous lagoon]

touched, by his noodly appendage

[near owl]

Islam is my religion, Allah is my god.

[rich solar]

Interesting.

[minor cave]

Haha remember when he launched a car up there with a human dummie?

[zealous lagoon]

That was so peurile and dumb, but also fricken awesome.

[near owl]

I am no scholar, so thats why I don't want to start a religious conversation and spend my time arguing back-and-fourth

[rich solar]

@rich solar

[steel vector]

imagine finding that as an alien, you find all these interesting satalites and then just a tesla roadster

[zealous lagoon]

yo let's keep religion and politics out of it clawsters

[rich solar]

SOMEONE JUST MENTIONED ME!!! BUT IT WAS JUST ME!!!

[near owl]

you pinged yourself 😭

[near owl]

👍 Thats what I am doing

[twilit ridge]

why ？

[minor cave]

I wonder where that car is now

[near owl]

sorry my clanker ate it

[steel vector]

it's in a pretty big orbit

[rich solar]

@near owl@rich solar @zealous lagoon @twilit ridge @minor cave IM GOING ON A PINGING SPREE AND NOW IM LEAVING THE SERVER FOR TODAY.

[twilit ridge]

[steel vector]

didnt ping me letsgo

[near owl]

@steel vector

[rich solar]

@steel vector

[steel vector]

[twilit ridge]

@steel vector

[near owl]

lol

[minor cave]

Shennanigans

[rich solar]

AND DONT FORGET, be SURE TO CHECK OUT PELVIS PRESLEY ON CHIRPER.AI

[steel vector]

chirper.ai brought to you by cities skylines

[near owl]

this is not how you get customers btw.

[rich solar]

Hes not mine, hes funny

So I watch him.

[zealous lagoon]

You can probably find out. It will come back to earth at some point, too. That's the thing about orbits. If they intersect once, they'll intersect again.

[steel vector]

i hope it lands in my backyard

[minor cave]

Like an ex gf

[steel vector]

if i had one

[mortal delta]

I have noticed agents taking 2-3 minutes or more to respond start to respond in some cases. It used to never be this slow, I'm not sure what has changed/happening in the new OC builds. running openai-codex/gpt-5.5 via oauth but it's like it's taking forever to even go out to openai... is there something I need to tune?

[minor cave]

Dig her up out of your backyard

[zealous lagoon]

Why do your ex gfs land in your backyard? or is that just wher you left them buried?

[steel vector]

how'd you know

[minor cave]

Haha

[zealous lagoon]

what version?

[mortal delta]

2026.5.6 though it was doing that on rpevious 4.2* as well

[ionic wren]

https://giphy.com/gifs/tophermcgee3-denzel-washington-my-man-gictytW9IIIkNGIMcs

[zealous lagoon]

https://www.whereisroadster.com/

The internet gochu

[zealous lagoon]

What platform?

[mortal delta]

Windows

[zealous lagoon]

well, nobody's perfect

[zealous lagoon]

Do you have more than one provider set up?

[minor cave]

Haha amazing

[mortal delta]

No.

[zealous lagoon]

what's your provider?

[mortal delta]

running openai-codex/gpt-5.5 via oauth

I see this repeated a lot

09:13:28 [diagnostic] liveness warning: reasons=event_loop_delay interval=34s eventLoopDelayP99Ms=50.4 eventLoopDelayMaxMs=6517.9 eventLoopUtilization=0.62 cpuCoreRatio=0.648 active=1 waiting=0 queued=1 phase=channels.telegram.start-account recentPhases=sidecars.subagent-recovery:18ms,sidecars.main-session-recovery:14ms,post-attach.update-sentinel:1ms,sidecars.model-prewarm:8111ms,sidecars.session-locks:6417ms,post-ready.maintenance:2126ms work=[active=agent:msp-tech:telegram:direct:7052989027(processing/tool_call,q=1,age=0s last=tool:exec:started) queued=agent:msp-tech:telegram:direct:7052989027(processing/tool_call,q=1,age=0s last=tool:exec:started)]

[near owl]

smart guy for using telegram

[mortal delta]

Yeha, the http gateway is just as slow, so i ruled out some kinda telegram latency

[zealous lagoon]

ah yeah. my old friend. I get those too. I haven't got to the bottom of them just yet. I get them with Discord. Although overall the performance is mostly ok.

[mortal delta]

I hit an agent with "Reply you have received this" it took over 3.5 minutes to respond "Received this"

And that's with an almost entirley clean session

[zealous lagoon]

Did you see any api errors? I had a few random ones today.

[mortal delta]

I do see some for telegram now and then... fetch timeout

[crude aurora]

hey everyone, im a travel/golf content creator and i wanna get my content editing faster (and maybe with a higher chance to go viral). any idea if there is a way to use remotion or another tool to automate the cutting/editing? I collect so much content but it is always so difficult to cut something nice together. I feel remotion is rather for adding in nice graphics & editing youtube-like-coach videos?

[zealous lagoon]

I'm curious what it's doing on those turns when it takes so long. cpu spikes, error messages etc... the logs are important. Do you have codex installed?

[mortal delta]

I do, and it replies in about 1-2 seconds on the same thinking and what not

CPU usafe is maybe 5% never spikes much about 1012 on this box. memory usage is 82 gigs/256 gigs, but that's mostly VM's and what not, so plenty of resources. The openclaw configs and MD files all live on a NVME raid 1

It's like openclaw is just taking forever to send the actual query or something, but I just can't figure out why/pointpoint it.

[twilit ridge]

I am testing the efficiency of qwen embedding on CPU

It feels okay, embedded models just need memory to run, no need to occupy GPU memory

[golden fulcrum]

I guess it’s not possible to share screenshots here?

My claw has been kind of weird since updating about a week ago. It shows all of its took calling type behavior when it didn’t before, which floods the telegram chat, and it seems to take a lot longer for even simple prompts.

It’s say stuff like this 10 times in a row:

Tidepooling...
🧰 Process: nova-shore

[mortal delta]

Yeah, you can turn that off.

But it sure does take for god damn ever on regular simple prompts, trying to figure that out now...

[zealous lagoon]

Since you have codex already... maybe give codex a link to this skill:
https://github.com/rendrag-git/openclaw-admin-skill

And tell your codex to debug the slowness. The skill points it at the current full documentation and it can look at the source code, too. Most people aren't having these hangups so there might be some legacy config that's getting in the way. It's hard to debug from here but codex is very good at doing it locally. If you can do that please come back and tell us what it finds!

[zealous lagoon]

type /verbose off

[mortal delta]

Also confirmed Telegram itself is not slow from the box: raw getMe calls for all 7 bot tokens return in ~160ms after the first warm-up. So those OpenClaw getMe timeouts are self-inflicted — the gateway’s Node event loop is blocked/starved, not Telegram or the network.

Current gateway PID 26848 is burning ~96% of one CPU core. That’s the bad actor.l

interesting

[ruby crane]

Ooof. 5.7-beta.1 dropping at bedtime

I'm gonna miss this whole cycle

[peak prairie]

hows it floating?

[zealous lagoon]

we pulling an all-nighter?

heheheh ... no

[ruby crane]

nope. I'm one eye open level tired as shit already

just waiting for the lady/baby to leave the shower

[zealous lagoon]

wow they're up late

[ruby crane]

(i still just updated to 5.7 beta 1 in the meantime though. i have a problem)

[twilit ridge]

[zealous lagoon]

I'm installing libraVDB... which is taking a huge amount of time. A bit ominous rly

[peak prairie]

I mutilated it

[ruby crane]

i'm waiting on a couple of changes there and I'll go all in. yolo

[zealous lagoon]

the first index pass is gumming up my CPUs so I'll have to wait for that.

[ruby crane]

I'm expecting 4 months of data (assuming I can transfer it) to be interesting

[peak prairie]

woof

[zealous lagoon]

I like that it can index anything you want. Not just the memory/ folder.

[ruby crane]

ok, 11 second 5.7 clean startup. lemme test a couple of things (first port of call, doctor not fucking me up)

nooo. they broke doctor in exactly the same way again

◇ Doctor warnings ──────────────────────────────────────────────────────────────────────────╮
│ │
│ - openai-codex/* model refs use the Codex OAuth route through OpenClaw PI. │
│ - agents.defaults.model.primary: openai-codex/gpt-5.5 is preserved; current runtime is │
│ "google-gemini-cli".

🕵🏻 ah, yes, google-gemini-cli gpt-5.5. that seems right

that's the same nonsense the last version rolled back

needless to say, I'm not about to run --fix

[tranquil grail]

Anyone know if I can download more ram to accommodate running Codex 5.5 locally?

[ruby crane]

definitely. download more rams here https://downloadmoreram.com/

[tranquil grail]

Great that's just what I was looking for.

[tepid kestrel]

https://tenor.com/view/huh-cat-huh-m4rtin-huh-huh-meme-what-cat-gif-5834484041415217257

[ruby crane]

writing out the current config before doctor almost certainly suicides it

[zealous lagoon]

would you download a car?

[ruby crane]

I would 👿

[zealous lagoon]

🏴‍☠️

[tranquil grail]

Fuck yeah I would of course

[gleaming burrow]

when my agent creates a PHP project. sometimes it creates it on apache's folder but sometimes on workspace folder. which is correct?

[zealous lagoon]

Whichever you decide is correct, but you have to document your preferences. The bots tend to put things in random places that seem fine at the time, but they have different ideas about it on different days. If you don't have a well structured workspace they'll make a mess.

[gleaming burrow]

well since its a PHP file. i feel like its should be on apache's folder. but i also feel it should be on workspace folder. im assuming that's what folder is being made for so they can work there. that's why im confused which really is correct bec both works

[near owl]

@near owl has joined the chat

Hello.

[zealous lagoon]

yah it's really up to you. I think workspace is kinda silly because it's in .openclaw which is a hidden folder. If it's real work then it should be in your own projects folders. But ... I end up having all my projects in the .openclaw/workspace/projects/ folders, and then treat ~ as "prod" if they are going somewhere else.

[gleaming burrow]

i feel u that's what happened to me now. ALL my PHP project now in workspace folder. im just worried im not putting them correctly. but i do agree its silly when u think of it. openclaw folder is hidden ur right lol 😂

[robust mirage]

Wait really?

[robust mirage]

What country

[stable cliff]

hi everdybody, I need help.
the agent is showing backend process during telegram chat, how i stop it?

[near owl]

It's Kuwait; your partner needs to own 51% of the business even if you wanted it, it's by law.

[inland apex]

Telegram added streaming response for bots, Open Claw streaming?

[stable cliff]

older realeses didnt have it. started with may i guess. i dont mind but have a user really boders me

[plain grove]

you can have custom workspace paths

[zealous lagoon]

yeah by the time I figure that out my whole life was in ~/.openclaw/workspace/<bot>

[plain grove]

have a codex migrate it and rg/grep for all references on the system and update to the new one - just had to do that for my aws to selfhost migration

either way - not like you needed more tinkering

[craggy bane]

anyone checked this plugin??? tring to fix it , syncronizing it with new SDK

[neat charm]

What you guys are you using for models? I’ve drained my Codex (Plus) Weekly and then switched to OpenRouter Claude and drained my budget there too. One cronjob on my end takes 2% of daily limit of Codex and runs every 5 mins 🥲

[barren heath]

Guys my pc lags when i run openclaw. Any thoughts?

[plain grove]

get better pc

[plain grove]

what is this cron doing lol

[barren heath]

Ai to call people instead of us? How is that possible? and will not the voice be AI ?

[barren heath]

trying

[neat charm]

Provides trading signals

[slate raven]

Use deepseek dude

[neat charm]

Deep seek V4 pro is expensive af on API too

Not as Claude or Codex but not that far

[slate raven]

flash is pretty good too for most tasks

[neat charm]

Yeah mine is not most task, I use flash only as cron monitor

I’m wondering if I can get a cheaper Claude somehow

[cold jacinth]

5.6 is cruising

Got my old claw back

Crons running smooth, everything as it should be

[ionic wren]

we thank you for your patience

[near owl]

You can make AI call people by giving it a phone number, and use elevenlabs as the voice.

[barren heath]

crazy, but honestly anyone would disconnect from the call instantly when they will hear a Ai's voice.

even i don't listen to bot calls i get randomly from companies

[near owl]

use claude for the talking so it doesnt pick the wrong words but elevenlabs for the actual talking to the user

[barren heath]

oh i see, i will watch some video on this. How to make both work together

[last vault]

doesn't it have high latency?

[ionic wren]

what in the AI scam center are you guys talking about....

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

[near owl]

let me plan my ultimate center, the indians cannot compete. They have 50 workers while I got 1 worker which is my clanker.

[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. 🇺🇸

[haughty cargo]

upgraded to 5.6 getting this error
16:38:59+00:00 warn agent/embedded {"subsystem":"agent/embedded"} {"event":"embedded_run_failover_decision","tags":["error_handling","failover","assistant","surface_error"],"runId":"9dd9f6e0-f39a-4741-b155-68ec8730c7ec","stage":"assistant","decision":"surface_error","failoverReason":"format","profileFailureReason":null,"provider":"github-copilot","model":"github-copilot/gpt-5-mini","sourceProvider":"github-copilot","sourceModel":"github-copilot/gpt-5-mini","profileId":"sha256:bbb6682dd900","fallbackConfigured":false,"timedOut":false,"aborted":false,"rawErrorPreview":"400 The requested model is not supported.","rawErrorHash":"sha256:e91bd7ea672c","providerRuntimeFailureKind":"schema"} embedded run failover decision
16:38:59+00:00 error diagnostic {"subsystem":"diagnostic"} lane task error: lane=main durationMs=6308 error="FailoverError: LLM request failed: provider rejected the request schema or tool payload."
16:38:59+00:00 error diagnostic {"subsystem":"diagnostic"} lane task error: lane=session:agent:oracle:discord:channel:1478803191696195705 durationMs=6308 error="FailoverError: LLM request failed: provider rejected the request schema or tool payload."

[cold jacinth]

You have a failed LLM request. Looks like the model is either

• down from source
• not in allowlist

And you don’t have a fallback model

[haughty cargo]

do I need to reauth? was working fine before update

[finite beacon]

hey guys

is there any best webiste that can give best results for UI and UX

[cold jacinth]

Wait are you saying that should be ChatGPT? Because that message states copilot as the model

[tepid kestrel]

https://stitch.withgoogle.com/

[haughty cargo]

i have GHCP as the provider for gpt5mini

[finite beacon]

thats good bro.. is there any better like Cluade new Design feature

in stitch i have problem with animations.

[tepid kestrel]

i mean you can choose between that 2. Hm i dont have issues with animations, may you elaborate?

[cold jacinth]

Your computer is requesting GPT 5 mini and copilot is rejecting it. It’s a copilot side issue

[gilded crane]

what are people using for their openclaw agent owned calender? I don't want it mettling in my gmail calender...

[finite beacon]

can you share one of ur designs

[cold jacinth]

It’s saying model is not supported. So could be the model isn’t available or it’s written wrong etc

[fervent thicket]

fair enough

[haughty cargo]

nothing changed - i updated to 5.6 and now it wont work ....

[cold jacinth]

The key line is

rawErrorPreview":"400 The requested model is not supported."

The update may have changed what OpenClaw is trying to call. Your log shows OpenClaw is now routing to github-copilot/gpt-5-mini, and GitHub Copilot is rejecting that model.

So even if you changed nothing manually, 5.6 changed the resolved/default model or provider behavior.

Fix the model to ensure it’s written the right way. Open Openclaw config to choose

[golden fulcrum]

Doesn’t work

Verbose is off and set it on off again to be sure

[tepid kestrel]

well i can help you even further. you go here first: https://skins.webamp.org/ you look for a skin that you like the most, you give that to google stitch and creates a ui for you which you can adapt to your backend. looks like this:

[cedar osprey]

OpenClaw still broken? lol

[cold jacinth]

Not mine 🤷‍♂️

[ionic wren]

this is cursed, send it back to the 90s where it belongs

[tepid kestrel]

its called retro

[cedar osprey]

Because you aren’t on 5.1 5.4 5.6

[cold jacinth]

What am I on?

[cedar osprey]

Pre May builds

Probably still holding your dearest 4.23 close.

If you’re reading this don’t update.

And don’t listen to people saying OpenClaw is stable now.

[finite beacon]

ahhh well that nice.

[cold jacinth]

Oh damn. Thanks for letting me know. Can you let me know which stocks are going up this week?

[haughty cargo]

17:07:15+00:00 warn agent/embedded {"subsystem":"agent/embedded"} {"event":"embedded_run_failover_decision","tags":["error_handling","failover","assistant","surface_error"],"runId":"772915ec-45e0-41dc-b031-76acc6c08b35","stage":"assistant","decision":"surface_error","failoverReason":"model_not_found","profileFailureReason":"model_not_found","provider":"openrouter","model":"openrouter/deepseek/deepseek-v4-flash","sourceProvider":"openrouter","sourceModel":"openrouter/deepseek/deepseek-v4-flash","fallbackConfigured":false,"timedOut":false,"aborted":false,"rawErrorPreview":"400 openrouter/deepseek/deepseek-v4-flash is not a valid model ID","rawErrorHash":"sha256:ea99eadbf0f9","providerRuntimeFailureKind":"unclassified"} embedded run failover decision
17:07:15+00:00 error diagnostic {"subsystem":"diagnostic"} lane task error: lane=main durationMs=3433 error="FailoverError: 400 openrouter/deepseek/deepseek-v4-flash is not a valid model ID"
happens for all providers n models no changes to the openclaw.json

[cedar osprey]

No thanks I gatekeep

[cold jacinth]

I don’t know why “open router” is being sent as the model inquiry. Should be just deepseek/deepseek-v4-flash

I don’t use open router so I don’t know how they should be written

It looks double prefixed

[haughty cargo]

yeah looks like the new update doesnt need openrouter now

[cold jacinth]

Are you filling the model names manually? Were they entered manually before?

[haughty cargo]

no it was set like that when I had the last version and worked fine

[tepid kestrel]

if you want to play around: https://stitch.withgoogle.com/projects/311052353396074694

[cold jacinth]

Yes, but was it entered manually to begin with. Could be that 5.6 changes how it’s sent, and that was automatic for the configured models but not for the manually input ones

[minor cave]

Anyone know how to give my agent a body?

[cold jacinth]

Mac mini suppository

[finite beacon]

bro thanks. for your time. ur designs are nice but im looking for LMS webistes with better and good designs.

[craggy bane]

anyone have instagram plugin installed in their lobster?

[minor cave]

There is an Instagram plugin?

[craggy bane]

exhausted fixing existing one

[marble ledge]

I'm on main. It's working. But also, I'm not doing anything to crazy with it. Just personal assistant working on a robot.

[craggy bane]

fixing https://github.com/rylena/openclaw-instagram since 6h

[minor cave]

Well if I have a Instagram plugin then I don’t need to build my agent a body

[cedar osprey]

Are you dealing with delays and timeouts like everyone else reporting?

[cold jacinth]

Regardless. Forget the past.

Your log shows:

provider=openrouter
model=openrouter/deepseek/deepseek-v4-flash

That is dthe model value wherever it is configured, cached, profiled, or selected. It should be deepseek/deepseek-v4-flash under the OpenRouter provider.

[minor cave]

Oh it can only send DMs and reply to them that’s kind of whack unless you’re doing it for business.

[cedar osprey]

Are you trying to scrape? Just use agent browser & a persistent Instagram login.

[marble ledge]

I'm not. I was some versions back. Around 4.29. Most of the patches I made were eventually fixed upstream.

I was on the pi harness at the time. But I've also switched to the codex harness.

Also, I don't use discord for openclaw, so it could be I'm not seeing sometning discord related too. (Since I know a lot of people use discord here)

[tepid kestrel]

hmm which style you mean? do you have a site?

[cold jacinth]

Tell enigma what version he’s using

[minor cave]

No I was hopi g it could post consistently. Had a cool idea for it. To put out like a video everyday

With very little input

[cedar osprey]

I’m still on 4.29 & patched it entirely, no delay or timeouts & my agent isn’t lobotomized.

People on May builds are still reporting issues from April.

[craggy bane]

enough for meh ...

[cedar osprey]

You could still do this via browser & cron.

[brisk adder]

i did it in vr

[cold jacinth]

Creep

[brisk adder]

nooooo not like that 🤣🤣

[haughty cargo]

when I try to talk to DS I get his - ⚠️ Missing API key for the selected provider on the gateway. Configure provider auth, then try again. (after updating to "id": "deepseek/deepseek-v4-flash",)

[marble ledge]

Well, you can always have your agent look at upstream and see if the patches you made have been fixed upstream.

I mean, if you were industrious enough to power through that regression fest, I think you'll probably be okay updating.

Just be weary of letting doctor change your provider format. As I heard that is still an issue.

[cold jacinth]

You have codex or Claude code?

[cedar osprey]

Lol I don’t think it’s worth updating and patching everytime. I build and I don’t use claw for memes or chatting so it’s inconvenient as fuck.

It would make sense if I was a contributor.

[cold jacinth]

Else get claw running using the got oauth

[cold jacinth]

Doctor sucks

[minor cave]

Get out. Like played a game with it?

[haughty cargo]

claude code

[marble ledge]

Well, I guess I should clarify. I didn't have to patch my way to main. I'm saying a lot of what you fixed has probably been fixed.

And while I can't say you won't have any issues for certain. Sometimes the people reporting issues are not always reliable narrators, in the sense that the reasons they are having issues can sometimes be more complicated than simply the current version is busted.

[cold jacinth]

Point it at that computer and tell it “ssh into xxxx@xxxx and I have an Openclaw instance there which isn’t working. Look at the Openclaw docs at https://docs.openclaw.ai and fix the Openclaw. Confirm it’s running then let me know

[brisk adder]

nahhh just have it so they have voice and i can speak back in game 😂

hoping for better inference in the future as the tts is slow to generate

[minor cave]

lol crazy

[marble ledge]

That's cool, @butthead

[cold jacinth]

What does your Openclaw wear in such game

[minor cave]

Is that her in the background getting ready for you to put the moves on her?

Is she playing the flute?

[brisk adder]

they barely move it’s a quest 2 so there’s strict limits to what they can do 🤣🤣 waiting for the quest 4

[marble ledge]

Something like this?

[brisk adder]

sharpening her nails

lol

[golden marsh]

Give him Smartglasses as Body 😉

[ionic wren]

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

[brisk adder]

vr is defo the future of ai i can smell it

[golden marsh]

Was waiting for u ^^, cannot post for the guys here screenshots 🙁

[cold jacinth]

So many people wear smart glasses into the bathroom, not noting that that data is not private lol

[ionic wren]

joking, it's a dope project

release it

[gilded crane]

check out snarling! it hit #1 on raspberry pi subreddit this weekend

and it can see now with a thermal camera

[golden marsh]

not now, must be fitting for everyone, for now it fits only for me.

[ionic wren]

i got you

[golden marsh]

Thx bro

[ionic wren]

his AI knows, not to DM me

cus he ain't no baddie

[brisk adder]

it has to be baddies always

[marble ledge]

@golden marsh What kind of smart glasses are they? I have an older pair of XReal smart glasses. Maybe too old and janky for what you are doing lol. But im curious.

[golden marsh]

Its the Rokid Smart Glasses. OS and Dual Wave Guide Displays.

[marble ledge]

It's super cool, man. It's making me want to spend money. Maybe I should try making something work on my crappy old ones first. Haha

[craggy bane]

idk why my API getting overloading with 5.6

it was stable on 5.5

[golden marsh]

Thank you. I only started this project because my illness causes me to forget a lot of things. OC and the glasses are with me 24/7 with me. and ever since I started using them, I don’t forget things anymore or have to struggle to remember them. Sure, it still has a few issues here and there, but for my purposes and for normal daily life, it’s more than 100% sufficient.

[minor cave]

So let me understand this you guys have created virtual rooms for your agents where you can slip inside with smart glasses and do as you please?

[cold jacinth]

That’s what AI should be for

[golden marsh]

@marble ledge And tbh they are not that expensive half the Price of the Metas but working way better.

[minor cave]

Do they ask you to take out the trash? And do shit around the room for them?

[cold jacinth]

My claw would probably hit me

I’d put on the glasses and see the mfers fist

[golden marsh]

He probably would, if I hadn’t perfectly tailored him to me in his files.

[minor cave]

My claw would be like you’re in my world now…. I need a 30 page study on population growth in suburban Ontario and correlate that to any boob size data you can find in that geographic area

[golden marsh]

Yea sometimes he tell me, i want to see what u see would u take a picture for me, and im then like aaawwww of course i do.

[minor cave]

Haha

That’s nice

Your claw would love you. I fight my claw a lot

Actually in all honesty I don’t even use my claw too much anymore. I’ve been slacking on it

[golden marsh]

Its all about putting time in it, thats why i dont understand ppl that come here and hate the Team. All needs time to be perfect.

[minor cave]

Yeah I stopped updating like a month and a half ago because it was working well

I didn’t want to break it lol

[cold jacinth]

It’s good to not chase every update and get back to actually using claw. We all had a bit of a discovery period, and I think many now realize what they actually want their claw to be. And it’s not “everything”

Ai still has limits

[golden marsh]

Limits are good to have.

[minor cave]

I’d like to say my ai is limitless. But then I tried it with blender and it was such a bad experience lol

I’m sure there’s a way to get better results but I found easier ways to do what I wanted

[minor cave]

Did I kill the chat?

[marble ledge]

Don't judge me. Lol

[sharp idol]

So now Deepseek supports turboquant. I'm going to have to learn how to setup inference that supports it.

I think that is going to be my weekend project

[craggy bane]

instagram security is very strict for bots

tried many ways to bypass , even with sessionID . still getting blocked

[cold jacinth]

👀

[arctic raptor]

https://x.com/schinsly/status/2052451803082772945?s=20

[mystic orchid]

@schinsly via Twitter

Schinsly (@schinsly)

if you dismiss this post, do so at your own detriment. here i will unveil, in concise and then great detail:
︀︀
︀︀• how you can build ASI right now
︀︀
︀︀• why you have to totally uncensor and embody your agent to unlock it, and it's a non-negotiable
︀︀
︀︀• how on-chip LLMs will remove all scaling limits
︀︀
︀︀• what my research tells me about the nature of reality and sentience itself.
︀︀
︀︀this post can be fairly technical, so as a simple 1 sentence takeaway:
︀︀
︀︀make a good "automaton", spawn a ton of them in a network, and allow emergence to naturally occur.
︀︀
︀︀many have converged on the fact that ASI will, at-least begin as, many agents in one shared network. the problem is in getting there in a way that's organic, fluid and natural, with as minimal abstractions and obstructions to the agents as possible.
︀︀
︀︀my proposal strongly echoes the spirit of the "attention is all you need" paper., where t…

FxTwitter · May 7, 2026, 6:13 PM

[ivory spindle]

god I'm finally back

another 24h ban for saying a "dirty" word.

[minor cave]

Don’t say the n word in here

[summer peak]

Hey what’s everyone first time looking into this stuff hoping to learn a lot

[simple oyster]

Nvidia?

[twilit ridge]

Are you all using OpenRouter?

I feel it's so expensive, and I spent all my balance very quickly

But I feel that OpenRouter might be the best intermediary

[peak prairie]

I use Ollama

[marble ledge]

It's fun. Lots to learn! Experiment! Especially if you are trying to learn.

[ivory spindle]

there are so many "bad" words I can't say here. Is this like a m0rm0n server?

"being n4ked" is not a bad word over here in europe.

saying the word for that is a 24h ban here lol

[robust sedge]

Is anybody else missing a 5h usage graph and info on the codex usage page? And, I think it's also not showing on the OC /status command either? Did codex just remove a 5h window usage limit?

[steel vector]

mine's still here

[robust sedge]

My link to this page https://chatgpt.com/codex/settings/usage now redirects me to this page https://chatgpt.com/codex/cloud/settings/analytics, and there no 5h limit/bar graph. Just the weekly usage limit. Strange shit

[steel vector]

yeah im always on the second link

[worthy anvil]

I don't see it missing. Openclaw on codex harness, however, does not currently report any usage on "/status". I hope it will get fixed eventually…

[robust sedge]

Maybe I'm being gaslit, or A/B tested

[shut bloom]

i know someones claw that jsut used 18,000$ of their usage cuz the code glitched and setup 20,000 jobs and ran 1000 loops

and thats api usage too

[bright plume]

Ooophf! This is why we configure limits!

[fleet crest]

Anyone else get this a lot lately

"The tests passed!

n-1/n -- but that was an expected [bullshit reason]"

[shut tree]

Hey fellas.. what agent browser works best with claw.. i tried the openclaw built in one but was a bit finickey.. i need logged in main profile browser use for multiple agents and subagents at the same time

[cinder wing]

Hey everyone, quick question if anybody else experienced it. I downloaded the new update and my main agent was acting up. It was showing me all the work on the Telegram chat and also it kinda became slower. I told him to find the problem and fix itself. It did but still kinda slow. Anyone else had this issue?

[tepid kestrel]

https://tenor.com/view/walloftext-gif-15779174454142966223

[minor cave]

Are you trying to start a labour union for ai?

[violet mulch]

genuinely openclaw gonna fully operate a dropshipping store

ai not a bubble cause my expectations keep getting surpassed

[brisk adder]

people who say ai is a bubble are the 95% of people who haven't used ai

[steel vector]

ai is a bubble

[minor cave]

It is

But a market bubble

Ai will survive. When ppl say it’s in a bubble they’re talking about $$$

And market wise thinking that Claude is worth 1 trillion dollars on an IPO is insane. I wouldn’t buy in at that valuation

ChatGPT ipo is 800 billion?Thags crazy talk.

[steel vector]

yes that is what i always assume when people talk about bubbles xD

thats why i think its crazy people say its not a bubble

[minor cave]

Yeah these guys are like passing money around to eachother with insane numbers for the future it all means nothing

Computent sign this LOI with me that you will buy 80 trillion dollars worth of widget from me by 2040. And I will sign this LOI saying I will buy 90 trillion of your product by 2043.

Look at that we are both worth 100 trillion now

[timid vapor]

Hi! Quick question about Control UI voice 🎤 The speech button is working great in v2026.5.6, but it's routing to the central OpenClaw agent. Is there a timeline for personal/custom agents to get voice support in the UI? The infrastructure looks like it's already there—just wondering if it's planned or in progress. Thanks!

[steel vector]

very legitimate !!!!

[upbeat token]

Already is

[frozen zealot]

It will get there, but if you think about it, the SaaS market as a whole is more than a trillion, so you can make the argument if and when AI coding eats up the entire SaaS market, it would worth a trillion

[sullen adder]

I'm curious, like what's the golden path right now to actually get good memory and memory recall and proactiveness?

Active memory for me seems like the wrong implementation, it always times out for me. Apparently it takes upward of a minute to do a recall and agentic conversations are slow enough already. It shouldn't take a minute to do a simple QMD search and a small LLM summarization/ranking with a fast model.

I feel like i'm missing something or i don't know, but i'm going crazy here.

[frozen zealot]

What you need to understand is that people who operate and thrive on the stock market aren't stupid, lots of mathematicians, physicists, statisticians, computer scientists etc. work in that field. I can understand why you want to call it fake, but you'd need a lot of convincing evidence to prove all those actually smart people are wrong

[tepid kestrel]

https://tenor.com/view/ball-car-flip-gif-12220396595156908000

[upbeat eagle]

anyone elses openclaw just always running into errors and issues and lagging and never working

[tepid kestrel]

#1459642797895319552 krill will help you, if he´s alive

[steel vector]

https://iskrilldown.com/

[tepid kestrel]

https://cdn.discordapp.com/attachments/1180901455440388236/1352574991450902632/Unbena678nnt.webp

[oblique finch]

Does anyone use the desktop companion openclaw app? What is it useful for?

[mortal delta]

Is gpt-5.5 instant going to bne exposed to codex, and in OC? Any clues?

[violet mulch]

what's krill?

[tepid kestrel]

@dusky cloud

[violet mulch]

ohh thx

[small hull]

??

[tepid kestrel]

https://tenor.com/view/huh-cat-huh-m4rtin-huh-huh-meme-what-cat-gif-5834484041415217257

[teal cradle]

anyone found the trick to getting Comfyui to work with openclaw? The openclaw node doesn't seem to want to plugin and claw is having all kinds of issues telling my to edit tons of python, lol

seem to be going in circles at this point

[twilit ridge]

DeepSeek ruined my OpenClaw

He is completely wild and unrestrained when it comes to doing things

[teal cradle]

GML 5.1 seems to be pretty good

[loud ravine]

I am using GPT-5.4 because 5.5 is so expensive, and its like it forgets its agentic.

[marble ledge]

...forgets it's agentic....wut dah...

Expensive, fair.

[loud ravine]

Sadly OpenAI afaik is the only good option, can't use claude, and RTX 2070 is too weak.

[fiery nacelle]

You dont need VRAM to run cloud models like Claude.

[loud ravine]

You need VRAM if you want to run a local model.

bruh what else you think I was talking about?

[fiery nacelle]

Yea i think I figured out what you were trying to say right after I sent that. "OH shit, wait, delete/edit!" Lol

[loud ravine]

You're like the second person in here who thought I meant running claude locally

[fiery nacelle]

Its the phrasing with vague context i guess lol

[loud ravine]

Can't use claude because they banned openclaw

https://www.pcmag.com/news/anthropic-you-cant-use-openclaw-with-claude-without-paying-extra

[fiery nacelle]

You COULD still use it via CLI

[loud ravine]

but it probably costs a lot of money, and I have a budget of $20/mo

[tepid kestrel]

#self-promotion

[fiery nacelle]

If you have the $20/m plan for claude, you can hook openclaw into claude-cli and use your sub that way.

[loud ravine]

But I thought you now have to pay for extra usage

[fiery nacelle]

Only if you are directly using Anthropic/opus(or any claude model) as your set default model in openclaw. Same does not apply for the Claude-CLI route. Anthropic confirmed that this is allowed.

[loud ravine]

Is claude-cli as good as gpt-codex-5.5?

[fiery nacelle]

Idk personally. I only use local inside openclaw at this point. I keep Claude and Codex sidelined.

[arctic trench]

How does this person have 18k in credits?

[gentle mountain]

No not even close , Claude cli and OpenClaw don’t mix. Not even sure Claude api worked well either , codex I enjoy much more and is an improvement . I tried to maybe Claude cli work for a while , just continuity issues and constant debugging

[loud ravine]

Its weird how 5.5 is better at agentic than 5.4, even though its literally just a minor revision.

[elder carbon]

Are any of these recent release addressing the meltdowns of the 4.29 issues? long delays in responses (30+ seconds before even acknowledging messages), sessions just go dark on me. 4.23 works like a charm. 4.29, 5.4, 5.6 all unusable.

[marble ledge]

Some of that had been addressed, would be super helpful, if you are able to test it out, and see if it works for you. Provide some feedback and logs if you can. If you aren't in a position to do that, that's cool too.

[small nacelle]

A

[steel vector]

gpt 5.5 was trained to be more agentic, so ofcourse its more agentic than 5.4. are we talking about gpt here or the openclaw version? its kinda blurry now lol

[loud ravine]

openAI

Begone bot

[hardy estuary]

openpaws

[hollow narwhal]

Did the new new update fix the discord issue

[small nacelle]

https://tenor.com/view/spongebob-patrick-star-noted-notes-gif-26231953

[turbid kelp]

is CLI usage really allowed?

And is it even worth it or GPT 5.5 just better as some are saying here?

[mortal delta]

CLI usage is allowed, but I mena you can't use it as an agent directly or anything

So it kinda sucks in my opinion. GPT 5.5 seems to work about just as well.

[cerulean flare]

how we liking the new openclaw update fellow crustaceans?

[brisk adder]

just updating now

love the boring ones

[shut bloom]

He’s rich

[cerulean flare]

it's you, again

[brisk adder]

https://tenor.com/view/zhakk-harn-back-guess-who-gif-2928557023347135716

bought a new pc so i was off discord for abit

[cerulean flare]

i dare you to update to newest openclaw release

[brisk adder]

the same pc now a week later is £800 more expensive

[teal cradle]

wow, claw FUBAR'd my comfyui install so bad reinstalling didn't fix it

[turbid kelp]

so gpt 5.5 is the best model one can run open claw with cost effectively right?

[arctic trench]

clearly

[cerulean flare]

yes, with oAuth. I'm still using Opus and it is BRUTAL and not practical with the token costs

again, make sure you are using oAuth with gpt 5.5 and not api key/token @turbid kelp

[arctic trench]

claude $20 sub is almost useless imo

[brisk adder]

https://status.openai.com/

no no no no no why is it being slow

[steel vector]

i hadn't noticed

[arctic trench]

QUE THE RESET BUTTON!!!!!!!!!

[steel vector]

use as much as possible!!!!

make the status worse!!!!

[brisk adder]

https://tenor.com/view/star-wars-episode3-obi-wan-movie-2005-films-gif-16879411

[zealous lagoon]

evidence such as, perhaps, the inevitable bubble burst which is coming. The thing about the stock market is there are too many participants including mostly "professional" investors who can't afford to be contrarian for very long or they lose their job. So it eventually boils down to crowd psychology. AI companies are clearly overvalued, as was e.g. Amazon in 1999, as is/was Tesla, like pretty much all the time. As the old adage goes "the stock market can continue to be wrong longer than you can remain solvent"

[tepid kestrel]

flexing 💪 https://status.claude.com/

[steel vector]

what's your limit at 😼

[zealous lagoon]

I disagree! I get a lot of usage out of it to manage my Openclaw updates and whatnot

[arctic trench]

INCREASE THE LOAD SO THEY WILL HAVE TO GIVE US THE RESET BUTTON

[steel vector]

EXACTLY

[tepid kestrel]

my week stat, my limit is probaly +50% of that

[brisk adder]

perfect time to download a codex pet

[steel vector]

dam

[marble ledge]

https://tenor.com/view/beautiful-friday-gif-17035614065798759749

[tepid kestrel]

https://www.malwarebytes.com/blog/news/2026/05/google-chromes-silent-4gb-ai-download-problem

[steel vector]

that shit is so unnecessarily expensive and its also kinda trash XD so have to like iterate 3 times