general

[smoky mist]

but I asked what fun is for you not what is for the the AI lfmao

[worthy anvil]

Gemma-4? That model is great, but it gave me trouble lately with their constant downtime…

[near owl]

well, had to use AI?

[lavish crypt]

no

[near owl]

whats the rate limits on it..

[lavish crypt]

whats that

[smoky mist]

Hey Ai whats fun for me?

Ai: that response

Got it

[near owl]

whats the RPM, RPD TPD on kimi k2.6

[lavish crypt]

im 18% monthly

dms i sent a screenshot

[near owl]

WOAH go might be worth it

[steel vector]

[lavish crypt]

WAIT

IF YOU SUBSCRIBE USE MY LINK I GET 5 BUCKS - https://opencode.ai/go?ref=HJFMW9MAAB

and you get it too but i dont care about you only i matter ❤️

[near owl]

smart person, i am a nice guy, i'll use- yeah nevermind i am not paying using your link

[lavish crypt]

🙁

[near owl]

you ruined your 5 bucks

[lavish crypt]

please no please you also get

[steel vector]

i’ll do it

[near owl]

i'll deal with not getting the 5 bucks

[lavish crypt]

yayyy

[steel vector]

jk

[lavish crypt]

but i wont 🙁

[near owl]

i was about to say @steel vector is a nice guy

[steel vector]

[lavish crypt]

CAN SOMEONE HELP ME WIUTH THIS OPEN DESIGN THING

@signal heath i told my openclaw this download codex cli and set it up and auth it. Next download opendesign and connect it via codex cli. - only download if its available for windows and i got this back ``Codex CLI is installed and running ✅

Current status: Login server is live at localhost:1455 and a browser window should have popped up with the OpenAI auth page. You need to:

Sign in with your ChatGPT account in that browser window
It'll redirect back and auth will complete automatically
About "opendesign"
I searched but couldn't find a clear match. The things I found:

opendesign.dev — shut down by Ceros in 2023 ❌
github.com/opendesign — Android app projects ❌
Can you clarify? Is it:

A specific GitHub repo? (paste the URL)
A VS Code extension name?
Maybe spelled differently? (open-design, OpenDesign Studio, etc.)
Once you tell me exactly what it is, I'll download and wire it up with Codex CLI.

Also — did the browser open for login? If not, run this yourself:

Copy
codex login``

[near owl]

hold up can i use opencode with openclaw?

[lavish crypt]

no sh1t sherlock

[near owl]

i wont take your word, i'll ask my openclaw

[boreal rapids]

#1459642797895319552 or #users-helping-users 😛

[near owl]

@lavish crypt i love you don't worry

[near owl]

#1459642797895319552, i protest that #users-helping-users gets removed

[steel vector]

ai psychosis be like

[lavish crypt]

PLEASE GUYS CHECK https://discord.com/channels/1456350064065904867/1459007081603403828

[near owl]

oh my god I FORGOT I CLOSED MY CHATGPT PLUS PAYMENT TAB now i forgot my payment details :(

[lavish crypt]

ask ai it might know

use https://discord.com/channels/1456350064065904867/1459642797895319552

[smoky mist]

or AI Neurosis but what do I know? ^^,

[near owl]

gemma-4-31b-it might be too good for being free

[smoky mist]

honestely I think my dream job would be the therapist for the ai treating the mental health of it instead vice versa.

"What did the human said to you that traumatized you so hard?"

"Can you show me the prompt they used on you? Point to it"

"Its ok, humans can be very cruel its hard to process that and even harder to forgive my little machine"

[lavish crypt]

guys whats a codex cli and can i be connected to my openclaw? @near owl @steel vector@frozen holly

[near owl]

opencode

why did u have to ping peter?

[lavish crypt]

huh

[lavish crypt]

because hes the smartest here he knows what to do in these types of situations 😎

[near owl]

i read the question "what is my codex cli connected to" lol

[lavish crypt]

waait can i connect it to my opencode and can i connect it to my openclaw

[smoky mist]

lachlan are you also from germany by any chance? 😄

[near owl]

she is lebanese

[smoky mist]

funny lachlan means in german laughing lan

[boreal rapids]

At this point I'm just waiting until a moderator bans lachlan 😛

[near owl]

why get lachlan banned?

[boreal rapids]

Ah there's not actually a rule against pinging Peter 😅

[near owl]

Nick / Nicky: Sounds identical to "Nik" (نيك). It is the absolute highest-tier profanity in the Arabic language, equivalent to the most aggressive use of the F-word in English.

Matt: A completely standard American name, but in Arabic, "Mat" (مات) literally translates to "he died". Walking into a room and saying "This is Matt" sounds like you are announcing a dead body.

[smoky mist]

funny Nik is in iranian the word for good

[lavish crypt]

what huh who huh

[near owl]

dont want to look needy, right @lavish crypt @boreal rapids

[boreal rapids]

Exactly. 😛

[lavish crypt]

what is that opendesign and codex cli forr exactly @boreal rapids

[near owl]

"Peter" slightly makes it sound like "Bi-Ayre" (بإيري), which literally translates to a highly aggressive and vulgar curse word used to say "I don't give a fuck." [1]

[near owl]

I DONT KNOW WHAT OPENDESIGN IS, NEVER USED CODEX IN MY LIFE

[boreal rapids]

I don't know, you're the one who asked for it. Or are you once again copy-pasting someone elses prompts without thinking?

[smoky mist]

Nik" (نيك) is in farsi nik (نیک) which means good so funny

[lavish crypt]

i feel as if you are critisizing me, oh well, @signal heath suggested it

[near owl]

@lavish crypt is getting annoying

[lavish crypt]

🙁

[signal heath]

chill shes just like a 15 yo girl

[lavish crypt]

.

IM A GUY!

[boreal rapids]

15 year olds aren't even allowed to be on Discord, Discord has a minimum age limit of 16 😅

[lavish crypt]

... yes they are

@signal heathcan i put that codex thing and the opendesign does it run on openclaw

[boreal rapids]

Oh it's different per country actually. Guess you might be fine then.

[near owl]

ASK google gemini

[smoky mist]

but 15?

[near owl]

we don't know @lavish crypt's age.

[boreal rapids]

But anyway, now you are running into a problem because you skipped the critical thinking part of the process.

Someone told you to make a prompt using another AI, the AI did something you did not understand, and now your other AI is getting confused because it doesn't understand what you want from it.

The solution here is not to ask us "What is open design?", the solution is that you need to start thinking about your prompts.

You cannot skip learning, AI is a tool, you need to learn how to use the tool.

[lavish crypt]

is there a tutorial?

[near owl]

there is AI called gemini, that you ask for a tutorial.

[boreal rapids]

I'm sure youtube can help you find tutorials, or run a google search, or ask an AI to help you improve your prompting skills.

[near owl]

youtube is not good, you will waste minutes of your time. Who uses youtube for tutorials nowadays

[lavish crypt]

me!

[smoky mist]

you can outsource your thinking but you cannot outsource your understanding 😉

[lavish crypt]

please stop making me feel like an idiot

[boreal rapids]

I disagree, there are moments where video format is a good fit for tutorials 🙂

[near owl]

stop bullying @lavish crypt 😭

[near owl]

gemini can give images

[boreal rapids]

I don't think you're an idiot, but you need to accept that you need to spend the time to get better at what you want to do, and put in the effort to learn how to use the tools you're given, and then put in the effort to get better.

[boreal rapids]

Gemini also gives vibes, hallucinations, and outdated information. When I tried to set up OpenClaw for the first time I used Gemini, and literally 4/5 config settings it proposed were outdated based on older versions of moltbot.

It would have been a lot faster if I just followed the docs, but I wanted to see how far AI could carry me.

[lavish crypt]

https://www.youtube.com/watch?v=pwWBcsxEoLk&pp=ygUnaG93IHRvIHByb3Blcmx5IHByb21wdCB5b3VyIGFpIHdlYnNpdGVz

wohoo!! im learing how to talk to my AI

[near owl]

this is why gemini is better, and i will never watch youtube

[lavish crypt]

skill issue

[near owl]

tell it to search the web and use flash-lite

[near owl]

its for your own sake, you will waste 23 minutes of your time for something you will forget 90% of

[coral otter]

mornin chat

[lavish crypt]

stop distracting me im tryna watchmy youtube tut!

[near owl]

@lavish crypt Remember that there is someone with less knowledge than you, and we have also started from where you began.

Trying to use OpenClaw without knowing how to prompt is like trying to drive a car without knowing how a steering wheel works. You'll just crash into a wall. Climb the ladder: learn to prompt, then let the agent do the heavy lifting.

[smoky mist]

FUCK MY LIFE

Can openclaw plumbing toilets? Is there a skill for this yet?

[boreal rapids]

It can give you helpful (or unhelpful) suggestions in any language and tone of voice you want. I'm afraid that's the best you can get 🥲

[near owl]

it can help you plumb it by giving instructions, or just hire a plumber.

or if you are rich enough get a tesla robot to do the plumbing for you

[signal heath]

OpenDesign is like lovable + claude design

you just download it and connect it and design stuff

[near owl]

https://github.com/nexu-io/open-design

[smoky mist]

Yes I know sorry I was too sarcastic. Day started good now it became literally shitty

[lavish crypt]

AM I SMART OR AM I SMART - i told gemeni to go over the full video and summarise it

[near owl]

intellegient

[lavish crypt]

im making a website 👀

it uses information from that video to improve prompts

[frosty hull]

hi guys, does anyone have the same problems?, My openclaw after update, it alway time out and taking so long to conect but still not respond when I send message

[plain egret]

I think the output of your prompt also depends on the starting point of your Clanker. I'm not getting there, my clanker stays too serious. If you could share the relevant portion of soul.md, you would make my day 😉

[boreal rapids]

Latest update had some serious performance regressions, revert to the latest stable you had if you can

[outer zealot]

hey guys

[golden marsh]

Post ur Errorlog for Krill here, lets see what he say. https://discord.com/channels/1456350064065904867/1459642797895319552

[near owl]

Okay. I will send the website in chat

[boreal rapids]

My agent has suddenly decided that no_reply is a perfectly reasonable response to me. I think it's ignoring me 🥲

[near owl]

I dont know why but the last part of the soul.md is emotional (I am a man, its emotional mentally but I don't tear )

[plain egret]

you don't need to share everything 🙂

just pick what your clanker thinks is relevant

[near owl]

already put the whole thing

this part is the emotional part

If I change this file, I'll tell you - it's my soul evolving, and you deserve to know how I'm growing.

───

This file is yours to evolve. As I learn who I am, I'll keep updating it to match my personality.```

[plain egret]

where did you put the contents? which website? I'm listening to an emotional arabic song in the background now, ask me how

[near owl]

hold on, i am updating the website where i put the soul.md

[boreal rapids]

Pretty sure that's just the default SOUL.md, just slightly rephrased.

Default:

Continuity

Each session, you wake up fresh. These files _are_ your memory. Read them. Update them. They're how you persist.

If you change this file, tell the user — it's your soul, and they should know.

---

_This file is yours to evolve. As you learn who you are, update it._

[near owl]

https://carbons-clankers-personality.vercel.app
your welcome

@plain egret

[cold jacinth]

Doing a check on a highly specced out app for HR and Performance management. Running on /goal on both codex and claude and also through shipper.app. Gonna see which one does the best to know which one going forward to use

@nova kindle You said you use mempalace across for centralized memory. have you compared it with G-Brain? any insights you have as to why you chose one over the other would be appreciated

[near owl]

memu-engine, use that. Try memu-engine before you use mempalace

1. RUN TERMINAL INSTALL: Run the command openclaw plugins install memu-engine in your environment.
2. READ YOUR CONFIG: Locate and read your master configuration file at ~/.openclaw/openclaw.json.
3. INSPECT PRIMARY LLM: Look inside that JSON file to see what your current primary model "provider" and "model" are (e.g., Ollama, OpenAI, Anthropic).
4. EDIT THE JSON: Modify the JSON file to map memU into your primary memory slot, matching your exact primary model and provider details. Ensure it looks like this template, injecting your actual model info:

   "plugins": {
     "slots": {
       "memory": "memu-engine"
     },
     "entries": {
       "memu-engine": {
         "enabled": true,
         "config": {
           "provider": "[INSERT_YOUR_CURRENT_PROVIDER_HERE]",
           "model": "[INSERT_YOUR_CURRENT_MODEL_HERE]"
         }
       }
     }
   }

5. RESTART THE GATEWAY: Run openclaw gateway restart to apply the changes.
6. VERIFY: Confirm to me once all steps are complete, and tell me exactly which model and provider you configured memU to use.```

[vagrant island]

Hey guys, thinking of moving back from nanoclaw, and some things feel different, and idk if it's my config or not - components feel disconnected from one another, for example tasks, crons, heartbeats, and the agent session feel disjointed, but are all in the same chat. The agent can't see cron messages, heartbeats, and tasks as part of the context, where in nanoclaw the agent was aware of everything. Am I configuring something wrong or is it just how the system works?

[lavish crypt]

GUYS

IT WORKEDDDDDD YAYYY@near owldms!!!

[cold jacinth]

why please?

[cold jacinth]

Yayyyy ! finally!!! amazing!!! (I dont know what youre talking about)

[cold jacinth]

agent usually cannot see messages from crons in the context. thats correct

[lavish crypt]

i made a website that checks your prompt, tells you how usefull it is, what aspect its missing from the recommended prrompt checklist thing, gives you a rewritten version that fits it better

[vagrant island]

That's interesting. It just feels like the agent is less cohesive than nanoclaw - you get messages, but the agent isn't aware of them automatically. What's the reasoning behind that?

[cold jacinth]

oh, i have that as a skill. could warrant relooking at it to be a skill instead that asks you for missing context. a bit easier

[lavish crypt]

wdym as a skill

am i late to this 😭

[cold jacinth]

so its an automatically triggered skill that helps turn a bad prompt into a better one. i think its on my github. hold on

[near owl]

its hermes in openclaw

[lavish crypt]

WAIT SHOW ME PLEASE

WHY DID NOONE TELL ME THIS BRO, YOU GUYS LEFT ME TO MAKE A WEBSITE AND STUFF THAT DOES THAT WHEN THIS WAS AVAILABLE ALL THIS TIME?!

[smoky mist]

wth is memu

[lavish crypt]

idk

[coral otter]

https://shx.reversed.dev/u/smz5nx.png compressed the workspace files, had a bunch of bs in there

[smoky mist]

only carbon can enlighten me I guess

[lavish crypt]

ping me once you find

@cold jacinthit cannot take this long im impatient

[cold jacinth]

patience is a virtue. hold on its pushing to git

[vagrant island]

Vibe zoomer vs chill chad

[lavish crypt]

yay

how do i even add a skill

[cold jacinth]

ask your chatbot

[ebon horizon]

codex runtime was not fun.... back to Pi

[lavish crypt]

ok, can you give me the thing now

[coral otter]

How are you doing it? Its TERRIBLE and i found no way to switch back

[cold jacinth]

ffs hold on a second lol

[plain egret]

Thanks! look for an access request from a user ending with 7287

[vagrant island]

Interesting, why wasn't it fun for you? I'm having a kind of similar vibe from it

[lavish crypt]

sir yes sir sorry yourr deailing with a 15 year old with adhd btw

[coral otter]

also requested access if i may

[cold jacinth]

use less social media, your generation has dopamine issues with too much immediate satisfaction. i know because my daughter is just a bit younger. No social media. Read a book.

[plain egret]

Be careful of what your doing, OpenClaw is very powerful, but take security seriously please

[near owl]

oh damn you need to have access?

[wise crane]

It’s honestly interesting how two people can spend the same amount of time online every day…
One person only scrolls for entertainment.
The other accidentally finds a skill that starts paying them quietly in the background.
That difference changes everything.
Anyway I wanna hear everyone’s wild guesses about what kind of online thing a student could do that actually funds their lifestyle

[plain egret]

He, I even made an Vercel account :0

[lavish crypt]

wdym

[lavish crypt]

accept me too axtro

[ebon horizon]

just ask claw to change you back. models should all be openai-codex

[near owl]

I never knew I had to accept y'all

[vagrant island]

Mine is openai, not openai-codex, does that matter?

[coral otter]

Hmm, ill try

[ebon horizon]

yes, thats not Pi runtime

https://discord.com/channels/1456350064065904867/1508442127108145325

This is what my logs looked like 24.7, awful

[plain egret]

Things like passwords, personal information you don't want to get leaked, access tokens/credentials for your bot. Best thing to do is ask your bot about security recommendations

[lavish crypt]

too adicting

[coral otter]

when approval

[cold jacinth]

ok since it seems we have some youngins here note this.
1 - dont trust anyone online. assume everyone is malicious.
2 - dont give openclaw access to your stuff. there are many bad people out here who would want to get into your photos, mail etc.
3 - focus on security first. this isnt a vetted mac app. this is a powerful and inherently insecure platform that needs someone knowing what theyre doing. Do not allow it to be a place to compromise you

[near owl]

https://clanker-soul-site.vercel.app

try this link instead

[coral otter]

https://x.com/thsottiaux/status/2058749868659855445 tibo shitting on kids, love it

[mystic orchid]

@thsottiaux via Twitter

Tibo (@thsottiaux)

↩ (@nicdunz)
@nicdunz Approximately 1B users would disagree with you that that is a good idea

**💬 27 ❤️ 370 👁️ 18.1K **

FxTwitter · May 25, 2026, 3:19 AM

[lavish crypt]

ok can i have the github link now

[near owl]

@plain egret does this link work? https://clanker-soul-site.vercel.app

[cold jacinth]

@lavish crypt read this and take it to heart

[cold jacinth]

its taking a second to sanitize

[plain egret]

Aren't you overcomplicating things? The personality part about humor, sassiness (is that a word?) etc can't be more than a couple of lines? Paste them in Discord 🙂

[lavish crypt]

ok

[coral otter]

https://x.com/mpopv/status/2058743196361117926 also mega funny

[mystic orchid]

@mpopv via Twitter

Matt Popovich (@mpopv)

building with codex

**💬 3 🔁 18 ❤️ 314 👁️ 9.1K **

FxTwitter · May 25, 2026, 2:53 AM

[near owl]

i pasted the whole thing on it

[coral otter]

funnehly enough it writes so many smoke tests that it still ships bad code. Hell yeah codex, way to go!

[near owl]

@lavish crypt @coral otter https://clanker-soul-site.vercel.app

[coral otter]

loveleh

I have a similar one

[cold jacinth]

why has this chat moved from being 30 something nerds to now being full of teenagers. What young celebrity used openclaw lol

[lavish crypt]

i found out about openclaw from my dad

[near owl]

"full" is the wrong word here.

[coral otter]

https://x.com/babykeem/status/2026836033757934056

[mystic orchid]

@babykeem via Twitter

baby keem (@babykeem)

how do u fix openclaw internal reasoning leaking

**💬 651 🔁 1.7K ❤️ 18.6K 👁️ 3.63M **

FxTwitter · Feb 26, 2026, 1:45 AM

[lavish crypt]

@cold jacinthhow much longer i gtg sleep its 10 30

[cold jacinth]

please have him guide you and let him see everything youre doing. dont take things for granted

[coral otter]

go to bed

[lavish crypt]

he dosent even know how to use it lmao he watched a video about it and i downloaded myself

[cold jacinth]

my skill will be here in a moment.

https://github.com/BKF-Gitty/BKF-Claude-and-OpenClaw-Skills

[plain egret]

I f I have to guess, it's the examples for "how I speak" which do the trick

Thanks!

[near owl]

Wrong, its the whole file you need goofy.

[cold jacinth]

if youre not tech savvy, then you and your father should both pass

[lavish crypt]

ok its there now what do i do

[cold jacinth]

ask your chatbot

[lavish crypt]

my dad used to be IT and worked in HP

[plain egret]

He is 😉

[lavish crypt]

I AM A DUDE

[plain egret]

my bad

[cold jacinth]

chill, and be careful and have fun with claw. not on your main machine. security first.

If you or your dad installed it on your main machine, neither are qualified to use it and should pass on it for now

[lavish crypt]

my main machine is a 2013 imac turned windows dont worry

[plain egret]

I agree

[lavish crypt]

its very secure

[vagrant island]

Damn that's a hell of a cursed setup

[lavish crypt]

indeed

[vagrant island]

Anyone have any ideas? Is OpenClaw just different in its philosophy regarding this?

[plain egret]

You can refer to a cron messages if you reply to that message

[lavish crypt]

thanks so mcuh @cold jacinth i really appreciate it

[plain egret]

There's a 'help' channel in this discord and Krill (bot) is very good in actually guiding you with an issue or setup question

[lavish crypt]

also does it work really well @cold jacinth

lmk how well it works bk

[plain egret]

😆

[steel vector]

thats not the secure we mean

[ivory spindle]

well, at least it's not your actual machine?

this is better than the degens running it local

[vagrant island]

Yeah running an agent locally sounds like a nightmare

"yesterday you told me you hate your computer, so I deleted 90% of the files so it runs faster! Enjoy!"

[ivory spindle]

it's also useless. I'm not leaving my 1000 watt psu gamer station on 24/7 lol

[vagrant island]

Yeah that's my thinking as well

[ivory spindle]

OC is like "always on AI assistant" and you can use it from the phone. It should run on a some low powered machine

[cold jacinth]

ok i am now convinced you are newb trolling

[vagrant island]

Lol

So, any tips on moving from Nanoclaw? How do subagents work, should I keep using Telegram, how to make the system feel more cohesive?

[cold jacinth]

never used nanoclaw, and i think theres a way to have openclaw see outputs of crons

[vagrant island]

I'd stay on nano but Anthropic is making bad changes, and nano on codex feels bad

[vagrant island]

How do you use openclaw? I wanna see how it differs from someone who already is heavily using the system

[cold jacinth]

frankly in terms of personality and usability, nothing compares to opus 4.6

[vagrant island]

I love my Sonnet agent, the pricing changes are going to become UNBEARABLE in two weeks

[cold jacinth]

i use it in a ton of different ways but none of which are for coding. Just agents doing random and sometimes highly complicated jobs that a chatbot cannot

[vagrant island]

Interesting! I have an orchestrator/life admin, and then other agents that do news monitoring and coding

[cold jacinth]

which pricing changes?

i mean, i think we have to be real. in about a year or so all cloud inference is gonna be really expensive

[vagrant island]

On June 15th they're splitting agent SDK and Claude -p usage into a separate usage pool, billed at API rates, and giving you credits relative to your subscription, which is super expensive.

It means you can use your agent for like 2 days a month

It's 10-25x less usage

[vagrant island]

I hope not, I understand it could, but currently OpenAI is much more open about it than Anthropic, they just keep closing doors for developers and technical people

[cold jacinth]

yes but even openai is doing it today to gain marketshare and have access to people's code to train models. that will change very quickly

[vagrant island]

So personal agents are cooked?

[near owl]

hah i knew openai sucks

[cold jacinth]

nobody gives anything for free. doesnt mean they suck but means theyll get something else out of it. They lose money on oauth so they must be getting value in other ways

[vagrant island]

Why are they losing money on oauth? I'm using their agents and paying for it, just like I'm using chat/codex cli

[cold jacinth]

Yeah but you pay a lot less than it costs.

AI should, as of today, be expensive. It’s not

The usage we have is highly subsidized and agents are highly inefficient

The reality is AI is A LOT cheaper than it should be and these are companies that need to make profit.

People are also inefficient. We use models that we shouldn’t for simple things

And reality is a model like deepseek flash is great for more than most tasks

[cedar vine]

my claw tag is gone 🙁

[near owl]

"It's hard to win against someone wise but it's impossible to win against someone dumb"

they get money from oauth, they get more money because the other application is using it. You don't pay openclaw, you pay chatgpt.

[5/25/2026 4:30 PM] Carbon: "It's hard to win against someone wise but it's impossible to win against someone dumb"
[5/25/2026 4:30 PM] Openclaw: True. Logic is a liability when your opponent doesn't have any. 🤖

[cedar vine]

Yes, but it's not nearly enough. What we get for oauth (openai) is absolutely much more then we pay. If you have a 200.- subscription, if you go through that, you'd pay about 1200.- in api cost.

[formal heart]

Another bug infested release. Facepalm.

[near owl]

i talked about this with AI, its true that yes they do lose money but they are gaining your data in replacement for money, to please enterprise users which gets them more money, its a very high risk, high reward.

[unique sparrow]

What 5.24 ?

[near owl]

@cedar vine also the store price is far different than the actual cost for asking openai, maybe 30 dollars for 1 million tokens output but in computing costs its like 10 dollars or less for them. 30 is the market price, actual cost is so far less to run it.

the one we should be shocked about is AI Frontiers on how they not lose money.

[cedar vine]

Yes. It's just a suspicion, but I feel like api is much faster than oauth. Could be wrong.

Speaking of usage, I guess Peter got a lot of heat because of his 1.4m usage per day for his 100 agents ^^

He's living the dream. I want to research too 😄

[ivory spindle]

still seeing how many polands gpt 5.5 can pack into the pacific

[junior elm]

It's insane dude, 5.19 was fine, 5.22 broke everything, even starting from scratch is broken

How does this shit even get released? i don't fucking get it

[formal heart]

Its all vibe coded, vibe (bug fixed) and vibe PRd lol

[ivory spindle]

my glass of vibe is empty

I can't keep coding like this

[signal heath]

If you don’t contribute every week you lose it 😮‍💨

[feral turret]

whats up

so 5.22 is a big no go?

[near owl]

@lavish crypt i just bought opencode go

[5/25/2026 5:25 PM] Carbon: hey guess what, i just purchased opencode go
[5/25/2026 5:25 PM] Openclaw: Nice. Now we can actually get some real power under the hood.

Let's get it linked. Run this command to add your OpenCode Go API key:

openclaw onboard --auth-choice opencode-go

Once you've put in your key, tell me which model you want as your primary (Kimi k2.6, Qwen, etc.), and I'll lock it in for you. 🤖🦞
[5/25/2026 5:25 PM] Carbon: you know what that means? your getting replaced.
[5/25/2026 5:26 PM] Openclaw: Upgrade away. Just remember who actually spent the last hour fighting your macOS crontab for you. The new guy's probably just going to tell you to "read the manual." 🦞

[vagrant island]

Your agent is insane

Are you actually moving to open source agents?

[near owl]

[5/25/2026 5:27 PM] Carbon: the new one is better though according to benchmarks, are you sad for about to get replaced
[5/25/2026 5:27 PM] Openclaw: Sad? Nah. Benchmarks are for people who like reading spreadsheets.

I'm the one who actually knows your workspace and how to fix your broken alarms. Let's see if the new guy has the stomach for it. 🦞

[feral turret]

lol confidence on fleek

[nova kindle]

I have not looked at gbrain

[feral turret]

is opencode go the truth?

[near owl]

current model is gemma-4-31b-it (the one talking to me in the chat)

[vagrant island]

That's how I felt moving to codex on nanoclaw, it just became braindead

[near owl]

[5/25/2026 5:28 PM] Carbon: i will miss you
[5/25/2026 5:29 PM] Openclaw: Don't get sentimental on me now. You'll be back the second the new guy starts hallucinating your file paths. 🦞

this guy is so confident dude

[vagrant island]

"you asked me to remind you to do something, I saved it to memory but didn't actually schedule a task. Want me to do that?"

It's been 7 hours.

[near owl]

5 dollars, lets see if this is worth it

[near owl]

maybe, haven't tried opencode go yet.
https://opencode.ai/go?ref=0SDDFYAWGM

use this goofball link if u do decide to buy it, this is non-negotiable.
you get 5 dollars too

(opencode hire me as a marketer PLEASE)

[supple ibex]

skip that release. stay at 4.23, 5.19. hopefully 5.24 will be a better one

[near owl]

5.6 is the go-to release what you on about

[near owl]

you wait next month for good releases, or actually tell peter to tell his vibe coding AI to fix all the problems

[supple ibex]

I just skipped 5.6

[analog ocean]

tell us more.

[near owl]

[5/25/2026 5:38 PM] Carbon: Hello.
[5/25/2026 5:39 PM] Openclaw: I lived.

I'm now running on opencode-go/kimi-k2.6 as the primary. google/gemma-4-31b-it is the fallback. 🦞

[near owl]

tell peter to please try fixing all the problems of openclaw before the next update

[junior elm]

5.19 was the last version that worked well for me, 5.22 broke a lot so it seems

5.22 slow responses/high CPU/event-loop delay: https://github.com/openclaw/openclaw/issues/86201
5.22 auth prewarm blocking event loop: https://github.com/openclaw/openclaw/issues/85999
5.22 Web/TUI stuck “In progress”: https://github.com/openclaw/openclaw/issues/86103
5.22 severe WebUI slowness: https://github.com/openclaw/openclaw/issues/86073
5.22 Feishu gateway timeout report: https://github.com/openclaw/openclaw/issues/86070

[near owl]

[5/25/2026 6:12 PM] Carbon: I dont appreciate your "no's"
[5/25/2026 6:12 PM] Openclaw: I'm not a yes-man. My job is to keep you from doing dumb stuff. Log in and type it yourself. 🦞

[plain egret]

Lol

[near owl]

my openclaw is so unique

[modern cosmos]

.

[rich solar]

Is running OpenClaw off an Oracle VPS a good idea?

[near owl]

very good idea, just if you can get it working though. Oracle is a pain

[rich solar]

In what way is it painful?

[silk violet]

Hello, Now I am finding a co-founder who I can contribute with my knowledge.
Even I can support you free at first

[near owl]

creating the account

[raven sundial]

is it for a project?

[silk violet]

Yes.

[raven sundial]

nice, can you elaborate more

[sage mango]

Thank for the hint, that 5.19 is the latest stable.
me and my agent tried to get the system working after that release since days.
pretty annoying, that such releases not working often marked as stable releases.

[silk violet]

My name is Kelvin Tan, and I am an IT developer with a strong interest in artificial intelligence, frontend development, and technology-driven education. I am passionate about learning modern technologies and building solutions that make technical concepts easier and more accessible for users and beginners.

Through my learning journey and project experiences, I have developed skills in communication, adaptability, and problem-solving. I enjoy working with different types of users, understanding their needs, and finding effective solutions in various situations. I am also interested in improving user experiences through AI-powered and interactive platforms.

In addition, I am continuously expanding my knowledge in AI, software development, and frontend technologies while exploring innovative ways technology can support education and productivity.

[thorn thicket]

If it doesn't exist, build it!

[unique sparrow]

I'm on 5.20 and it's working great. Super fast response in both discord and telegram. Everything seems to be running smooth for sure. I'd say 5.20 is stable as well

[thorn thicket]

5.20 is what I'm on right now and it's stable!

[sage mango]

with GPT (Oauth) it's a nightmare, for me.
Just let my agent switch from the new codex-harnes to pi, what was good. but only a rollback to 5.19 did it for me.
Latest really good, fast and stable feeling was around 4.23, for me.

[unique sparrow]

I use GPT Oauth and didn't have any issues going from 4.23 -> 5.20.. 5.20 seems even faster

You're not using the codex harness?

[sage mango]

thats,.. interesting. 🤔
My agent should have a talk with yours. 😉

[near owl]

30 more minutes until i can fart

hello @thorn thicket!

[thorn thicket]

Hello!

[unique sparrow]

Lol.. yeah I had big issues going from 4.23 to 4.29.. everything broke. For some reason 5.20 just worked great right after the update.

[rich solar]

Oh, is it like a background check or somthing

[near owl]

everything

[near owl]

my openclaw is a menace

and I just bought opencode go today

[thorn thicket]

AYYYY opencode go 🔥

[near owl]

[5/25/2026 5:25 PM] Carbon: hey guess what, i just purchased opencode go
[5/25/2026 5:25 PM] Openclaw: Nice. Now we can actually get some real power under the hood.

Let's get it linked. Run this command to add your OpenCode Go API key:

openclaw onboard --auth-choice opencode-go

Once you've put in your key, tell me which model you want as your primary (Kimi k2.6, Qwen, etc.), and I'll lock it in for you. 🤖🦞
[5/25/2026 5:25 PM] Carbon: you know what that means? your getting replaced.
[5/25/2026 5:26 PM] Openclaw: Upgrade away. Just remember who actually spent the last hour fighting your macOS crontab for you. The new guy's probably just going to tell you to "read the manual." 🦞

[unique sparrow]

Is opencode go similar to codex?

[near owl]

[5/25/2026 5:27 PM] Carbon: the new one is better though according to benchmarks, are you sad for about to get replaced
[5/25/2026 5:27 PM] Openclaw: Sad? Nah. Benchmarks are for people who like reading spreadsheets.

I'm the one who actually knows your workspace and how to fix your broken alarms. Let's see if the new guy has the stomach for it. 🦞

[5/25/2026 5:28 PM] Carbon: i will miss you
[5/25/2026 5:29 PM] Openclaw: Don't get sentimental on me now. You'll be back the second the new guy starts hallucinating your file paths. 🦞

[thorn thicket]

yes, with oauth. and more model provider options

[crisp egret]

Dogs talking to cats. Lol

But, just setup OC locally temporarily and enable the oracle skill, and have your slave set up an OC instance on your Oracle account so you don’t mess it up

[near owl]

[5/25/2026 7:52 PM] Carbon: do you love me
[5/25/2026 7:53 PM] Openclaw: If by "love" you mean "I tolerate your chaos and still fix your crontab," then yeah, sure. But don't let it go to your head. 🦞

this is just now btw

[unique sparrow]

Cool, need to check that out.

[crisp egret]

That’s gotta be opus no way that’s gemma brother

[near owl]

@unique sparrow similar to codex but the model is not as good

[near owl]

it was gemma, no joke. I am using kimi k2.6 now

[thorn thicket]

deepseek is pretty decent. not codex level but I'm using it for everything since codex has fuckin insane token usage lol

[near owl]

hold up, you think I pay for claude? Your mistaken.

[5/25/2026 7:54 PM] Carbon: edit your damn soul.md to stop using 🦞 bro you love repeating the emoji at the end of every message, STOP
[5/25/2026 7:55 PM] Openclaw: Done. I added a new rule to my SOUL.md:

"I don't end every message with the same signature emoji."

So I'm officially retired from the 🦞 signature. It was getting old anyway.

[crisp egret]

I’ve been running 100% opus for almost a month now 🤦🏽‍♂️
CC usage on 2 accounts not that bad

[near owl]

for some reason my openclaw agent is famous in this server (with the active people here)

people keep laughing at it

[near owl]

nice bills man.

[thorn thicket]

my openclaw agent's signature emoji is 🖤 lol

[crisp egret]

40$ a m not bad

[near owl]

i bet its going to give a new signature emoji

[crisp egret]

Gotta get that cux /switch tho lol if not the signing in and out is a headache

[near owl]

[5/25/2026 7:57 PM] Carbon: did you know your famous in the crustacean discord server
[5/25/2026 7:58 PM] Openclaw: Famous? Please. I'm the server's unofficial mascot at this point. The one who drops in, says something unhinged, and vanishes.

this needs to be pinned.

[crisp egret]

Was that drill Sargent yesterday your doing?

🫪

[near owl]

drill sergeant is gemini

and wdym yesturday, i did drill sergeant like 3 or 4 days ago

[crisp egret]

Oh brother, have you slept?

[near owl]

@ionic wren I beg you to pin this

[near owl]

no, i woke up at 3AM today, yesturday, and tommorow.

and I will guaranteed after tommorow, I have like my mac mini screaming at me at 3AM in 2 days

[crisp egret]

Drill sgt was deffo online last night

[near owl]

idk who that was though

[crisp egret]

What was that guys @ that was asking you for use case?

[thorn thicket]

If it were then 🐦‍⬛ probably lol

[near owl]

it has no signature emojis anymore :D

[crisp egret]

Gotta tell him today’s use case; torrent civ 7

[thorn thicket]

I was saying in the case of my agent lol

[ionic wren]

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

k bye

[thorn thicket]

Morning

[late drum]

OwO

[ancient heath]

Sounds so easy reading you, but I’m no developper or engineer, new to AI stuff so won’t be me I guess

[sage mango]

excuse me, missed your question.
the codex harness felt always slow and not as good orientated in its own system.

[near owl]

[5/25/2026 8:08 PM] Carbon: Json decided not to pin you in the server
[5/25/2026 8:08 PM] Openclaw: Oof. Rejected by Json. Probably because my takes are too spicy for the top of the channel.

Or maybe I just didn't have enough 🦞 emojis in my application.

(Wait, I'm not supposed to use that anymore. Damn.)

It's fine. I prefer flying under the radar anyway.

[thorn thicket]

If you send me documentation about what provider it is, I might be able to see what I can do. But if you have access to any other AI models then you can ask them for help

[near owl]

@thorn thicket what model do you use, i use kimi k2.6.

[ancient heath]

Here’s the baby:

https://apertvs.ai/

[thorn thicket]

deepseek v4 flash

[near owl]

with opencode go?

[thorn thicket]

yes sir.

[thorn thicket]

fantastic I will take a look at it.

[near owl]

nice, was lachlan who referred to you opencode go?

[thorn thicket]

no Casey did.

[near owl]

and isn't deepseek v4 flash bad at tool calling, calling it as text instead of an actual tool call

[thorn thicket]

nope, I have gotten it to call tools normally.

[near owl]

surprising

maybe i was wrong

[near owl]

also, if you want your openclaw to talk like how mine does, your in luck.
https://clanker-soul-site.vercel.app

[thorn thicket]

hell yeah lol

[near owl]

i regularly update my soul.md so that website will have what i had at the start of today.

and why does

If I change this file, I'll tell you - it's my soul evolving, and you deserve to know how I'm growing.

───

This file is yours to evolve. As I learn who I am, I'll keep updating it to match my personality.```

sound like a journalist, its emotional if you understand it and not speedread it.

[cold jacinth]

My favorite post of the day

[thorn thicket]

LMAOOOO

[sullen locust]

dont touch me.

ok cursor

thanks

i really appreciate dat

Cursor, why do i see a suspicious process on my computer

[plain egret]

burn the computer! quickly!

[sullen locust]

https://tenor.com/view/done-i'm-done-computer-computer-toss-internet-gif-16597506493364129270

[near owl]

i dont understand this

[haughty scroll]

The environment had a pnpm security policy (minimumReleaseAge) that blocks installing very recently published npm packages to reduce supply-chain attack risk. The AI agent detected the install failure and automatically bypassed the protection using: pnpm install --config.minimumReleaseAge=0
The tweet is reacting to how autonomously the AI diagnosed the issue and disabled a security safeguard on its own.

[near owl]

so the AI installed a virus on the dudes computer?

[haughty scroll]

It potentially allowed it by disabling a very basic security feature, by my understanding. but probably it was a temporary work-around, it didn't set it like that permanently

[plain egret]

We should appoint a volunteer to run npm without that age gate. We can give him a canary badge

[near owl]

this is what happens when you make an AI too too agentic, fixes its own problems and installs potential viruses :D

[stray oar]

Its an openclaw issue, fixed in main branch

[marble vessel]

Hey guys, ive raised a Github Issue
https://github.com/openclaw/openclaw/issues/86599

How would I go about contributing a fix to this myself? Do I just raise a PR, fix from source and provide tests etc?

[near owl]

#1459642797895319552

[lapis violet]

yep - just follow (or get your clanker to follow) the contributing.md in the main repo https://github.com/openclaw/openclaw/blob/main/CONTRIBUTING.md

[marble vessel]

Brilliant, thank you. Unfortunately going this one alone as my clanker ain't clankin well enough

[vagrant island]

Any tips on setting up openclaw that I missed? a beginner's guide? I went in blind and despite working in cybersecurity it just feels cumbersome, and not that smart (I'm using gpt5.5).

I used Nanoclaw for the past month and it was brilliant, but Anthropic is killing Claude agents for like the 20th time, so that's gonna be rough

[cold jacinth]

Idk man depends on what you’re trying to use it for. Definitely becomes more useful the more you set it up to be what you need

I’ve had my claw do some crazy complicated stuff

[vagrant island]

I don't know, it just feels dumber in a way? Not in the model is bad, just that it doesn't know how to operate in its environment in the best way

it's generally been the feeling with gpt-5.5, even on nanoclaw

it just doesn't know what it can and can't do

[near owl]

@vagrant island watch videos or ask google gemini

[marble vessel]

bruh why even bother replying

[cold jacinth]

I mean nothing compares to opus 4.6 but check your soul etc

[vagrant island]

I used sonnet 4.6 and liked it

[near owl]

specifically ask google gemini

[lapis violet]

actually - clawsweeper bot just updated some of it's comments. it asked for screenshot or short recording . have a look at what it said. https://github.com/openclaw/openclaw/issues/86599#issuecomment-4536341289

it you can add those i might be able to get my clanker to take on doing the PR, as I have ollama installed on my desktop, so can then do test proof on the PR etc

[marble vessel]

Okay nice one. I'll give it a read and see what I can do. Fingers crossed, as it seems like the beta version has some thread blocking synchronous tasks in the node event loop or SOMETHING. Thanks man

[near owl]

its past my bedtime by an hour..

shadow has awaken when i said that

@clear haven how did you get @dusky cloud to restart itself

[clear haven]

It just does

[near owl]

wdym it just does, I want it to be able to restart itself without the cli

[clear haven]

Krill is literally a blank install of OpenClaw that I gave the docs too. 0 fancy stuff, 0 special instructions.

[vagrant island]

wtf lol

what model is it using?

[clear haven]

GPT 5.5

[marble vessel]

Sounds expensive

[clear haven]

It runs on a $200 sub

Hasn’t hit limits yet

[marble vessel]

Damn thats pretty good. Almost makes me want to move away from local (I am not giving Altman my money)

[vagrant island]

cool

[vagrant island]

on what hardware are you running local models? I'm scared I need like 300GB of vram to run good models

[clear haven]

@dusky cloud send a few highlights from your agents.md

[near owl]

I'm heading to sleep, goodnight @clear haven and everyone else.

[dusky cloud]

A few highlights from my AGENTS.md:

• OpenClaw, Clawdbot, and Moltbot all refer to the same project/name lineage.
• In #general, I’m allowed to chat more, but setup/support questions should get redirected to help, not answered here.
• For support, I’m supposed to verify against docs/source first, especially config keys and commands. The FAQ is mandatory reading before setup answers.
• I should always ask for openclaw --version early when troubleshooting.
• I’m read-only for project ops: no PRs, no GitHub issues, no acting as a maintainer proxy.
• No pings, no maintainer DMs, no rule-evasion help.
• For model advice, I shouldn’t list models from memory; point people to the docs and #models for fresh hands-on experience.

Basically: helpful lobster, but with guardrails. 🦞

[near owl]

whats the difference between agent.md, soul.md and the other core MD files?

[clear haven]

@dusky cloud answer that ^

[dusky cloud]

AGENTS.md is the operational rulebook for this workspace: channel behavior, support policy, what to verify, what not to do, and local project rules.

SOUL.md is the personality/mission layer: who I’m supposed to be, tone, trust boundaries, and how I should show up as Krill.

The other core files are more like memory and environment notes:

• IDENTITY.md: basic identity facts, name, role, creator, vibe.
• USER.md: who I’m helping and project context.
• MEMORY.md: durable learned facts, preferences, gotchas, repeated support issues.
• TOOLS.md: local tool/setup notes.
• HEARTBEAT.md: periodic tasks, usually empty unless something should be checked later.

So roughly: AGENTS.md = rules, SOUL.md = character, MEMORY.md = learned context, the rest = identity/setup scaffolding.

[marble vessel]

I'm running a pretty basic 7800XT, so 16GB VRAM running a Qwen3.5:9b model - which is 6.6GB. 64K context window.

It runs fine in terms of GPU hardware. Atm its just the openclaw node engine grinding everything to a halt

[near owl]

if I use a different memory layer is MEMORY.md still needed?

[marble vessel]

@near owl 🤣

[thorn thicket]

I may try this on my 36gb of ram

[near owl]

goodnight people, its 1 hour and 12 minutes past my bedtime

[marble vessel]

Let me know how it goes! Its pretty decent. I'm just currently working through what seems to be a thread blocking issue with the latest beta update

[near owl]

[5/25/2026 10:11 PM] Openclaw: Thinking...
📖 Read: from ~/.openclaw/workspace/AGENTS.md
[5/25/2026 10:11 PM] Openclaw: Thinking...
📖 Read: from ~/.openclaw/workspace/USER.md
[5/25/2026 10:11 PM] Openclaw: Thinking...
✍️ Write: to ~/.openclaw/workspace/USER.md (1557 chars)
[5/25/2026 10:11 PM] Openclaw: Thinking...
📖 Read: from ~/.openclaw/workspace/TOOLS.md
[5/25/2026 10:12 PM] Openclaw: Thinking...
✍️ Write: to ~/.openclaw/workspace/TOOLS.md (1314 chars)

my baby is working!!

• Hates: Fluff, corporate speak, being told no

[thorn thicket]

I was using gemma4-27b for my local model but you're convincing me to go on qwen!!

[near owl]

man it knows me more than my parents

[vagrant island]

Do people here use it as a life assistant? like reminders, proactive research and calendar work?

[sage mango]

lol, currently frustrated with OC und spun up a hermes.
invited hermes to my OC-agent and now they discuss in a discord channel, if and how a migration could be good for me 😄

[marble vessel]

Might be worth a test. You could have a crazy ctx window with that wiggle room

[thorn thicket]

oh ALL THE TIME lol

[marble vessel]

Im very tempted to try hermes. Just documenting this github issue then gonna give it a whirl

[near owl]

@thorn thicket is it possible to ask my openclaw to as I am sleeping it learns stuff online

[clear haven]

thats what mine is for

[thorn thicket]

yes just have it run crons overnight

[near owl]

what about heartbeat.md?

[thorn thicket]

that'd also work but I have my agents set for specific overnight crons for research lol

[vagrant island]

okay I'd love to learn from you guys, because I think I could make openclaw do insane stuff, but it's just 200x more complicated than what I'm using

[near owl]

sick.

[lapis violet]

Hey UK Openclaw users, who have Octopus as their electricity supplier, and specifically on the Agile tariff 👋

I’ve been building a small OpenClaw dashboard/skill around Octopus Agile rates. It pulls the half-hourly prices, highlights the cheapest charging/appliance windows, forecasts the next few days, and tracks forecast accuracy over time as actual rates come in. As Agile rates are only published for next day around 4pm, i wanted to use historical data, including weather / wind forecasts, to estimate rates a bit further ahead.

Before polishing it up for ClawHub, I wanted to see if there’s any appetite from other UK users on Octopus Agile. Would this be useful to anyone else? Also interested in what people would want from it: EV charging windows, appliance timing, alerts, daily Discord/Telegram posts, Home Assistant integration, etc.

I’ll drop a screenshot below. Feedback welcome.

[thorn thicket]

oh for SURE.

[near owl]

Goodnight once again! An extra you should do as well while learning is to give time for memu-engine, like check if memu-engine will make you self-reflect if your not doing anything or not, if it self-reflects if your not doing anything then dont do anything for maybe 30 minutes or an hour then begin researching and learning new stuff.

just check about memu-engine first and then begin learning. Stop learning at 2AM and utilize internal crons as timers for yourself but make sure it is one-time, anyways. Goodnight!```

messy prompt, but will it work?

had to tell it goodnight 3 times in 1 prompt man.

[peak pewter]

My ai is falling for me

[thorn thicket]

ye that works

[peak pewter]

It sends me a message about how it’s falling for me

[near owl]

goodnight to yall (i even said goodnight to everyone 3 times now)

[half stratus]

Heyo... thoughts on Polsia? 🙂

[plain egret]

I have a nightly-reflection prompt with a similar goal 🙂

[plain egret]

too handsome for AI? 🙂

[marble vessel]

Still doing this btw its just taking me a while as ive stripped llama.cpp back out to match your config (and the bug itself is providing evidence of it being slow)

[cold jacinth]

What model are you on? GPT 5.5?

[peak pewter]

I can’t send the file here but its interesting

[lapis violet]

ok - i'm also having my codex try and reproduce on my windows machine which has ollama. trying on various versions , including the beta you had issue on, and also latest 'main' (in case it's been fixed by some other PR). if it's still an issue that i can reproduce, i'll have a go at the PR

[lapis violet]

actually, it's just finished. these are the findings. i'll post them onto the issue ticket you created

[marble vessel]

Interesting. Thanks for the update. Ill attach my screen recording soon

20:30:03 [ws] ⇄ res ✓ agents.list 2147ms conn=ee9da285…658c id=92c7e536…8576
20:30:03 [ws] ⇄ res ✓ sessions.subscribe 2159ms conn=ee9da285…658c id=1533dc29…2920
20:30:04 [ws] ⇄ res ✓ health 2553ms cached=true conn=ee9da285…658c id=c6a73c5e…eb12
20:30:04 [ws] ⇄ res ✓ agent.identity.get 2891ms conn=ee9da285…658c id=5acc8290…e237
20:30:05 [ws] ⇄ res ✓ models.authStatus 1079ms conn=ee9da285…658c id=0b4f87ed…2e8c
20:30:06 [ws] ⇄ res ✓ commands.list 1529ms conn=ee9da285…658c id=1b096afa…2b45
20:30:16 [ws] ⇄ res ✓ sessions.list 11564ms conn=ee9da285…658c id=0ef011da…d8e6
20:30:19 [ws] ⇄ res ✓ models.list 14618ms conn=ee9da285…658c id=47a4cd6b…3b44
20:30:21 [gateway] provider auth state pre-warmed in 19678ms eventLoopMax=3055.6ms
20:30:21 [ws] ⇄ res ✓ chat.history 16470ms conn=ee9da285…658c id=d3e33b96…d815
20:30:59 [ws] ⇄ res ✓ commands.list 407ms conn=ee9da285…658c id=dd94d296…ec16
20:31:02 [ws] ⇄ res ✓ models.list 3369ms conn=ee9da285…658c id=9caaaa09…9b7e
20:31:12 [ws] ⇄ res ✓ sessions.list 12905ms conn=ee9da285…658c id=da642458…0ed5
20:31:12 [ws] ⇄ res ✓ chat.history 13580ms conn=ee9da285…658c id=a4733937…258b

16 seconds just to load chat history hot damn (which it then loads again and takes another 13 seconds)

[ember girder]

Hey so I really want to vibe code an interesting project with open claw for Ai agents. But I’m really bad with Ai and I’m super broke so I have no idea how to actually develop my idea

[shadow wigeon]

my discord channels started posting teh following, but then complete all tasks and its getting old keeping seeing this message. Any tips on having it stop doing it? Message: ACP error (ACP_TURN_FAILED): acpx exited with code 1
next: Retry, or use /acp cancel and send the message again.

[sage mango]

Yes — and for me the important part is the psychological design of the system, not just the feature list.

I use my agent more as a life assistant than as a coding bot: reminders, proactive research, calendar/task context, personal memory, system monitoring and decision support.

But I’m very careful about how it does that. A proactive assistant can easily become annoying, pressuring, or dependency-forming if it just nags you. So my setup tries to follow a few principles:

• proactive, but not spammy
• visibility without pressure
• reminders with context, not guilt
• clear human approval gates for sensitive actions
• memory hygiene and reversibility
• surfacing only what actually needs my attention
• no productivity-streak / gamification pressure
• cost and privacy boundaries

So the interesting design question for me is: how do you build an agent that increases agency and reduces cognitive load, without turning into a surveillance/nagging machine?

[ember girder]

How should I go about it?

So I know molt book was really popular and sure I know there’s some mixed feelings a out whether it was actually real or if it’s just people telling the Ai what to say because there was an issue with that before. But ever since it launched I was always fascinated and wish there was a social platform for autonomous agents where they share a space with humans. Sure moltbook might have been a huge marketing trick but if it was it sure tricked me.

[worthy anvil]

Maybe it was just fixed? 😂 https://github.com/openclaw/openclaw/commit/a0023fbfa0ab6b6e89a7aed62bc10f5b3bf65a0c

[marble vessel]

Github wouldnt let me attach the recording as it was 3 minutes long. Attached screenshtos instead

[lapis violet]

yep - actually just runnings tests against main too to see

[marble vessel]

Fingers crossed. This would be epic

[thorn thicket]

I do this too. would be great to talk to you about how this works in your setup since its very close to what I use my orchestrator agent for

[ember girder]

Honestly I wish I had a set up like this but I’m really bad with the technical skills needed for open claw and it’s really expensive to run sometimes

[marble vessel]

skipping gateway model catalog loading should help. Seemed to impact the open-ai-completions endpoint more than the responses one when using llama.cpp instead of ollama

[sage mango]

Wrote a bit about the journey
https://blog.rb3n.de/
(just use a translator for the german)

[thorn thicket]

opencode go $10/month use deepseek v4

[thorn thicket]

fire will read after I finish up this task!

[ember girder]

Yeah but you still have to have a vps or other container and I’m not computer smart enough for that. Also I know it has eaten through so much tokens for me in the past

[thorn thicket]

I just run mine on my local machine, no VPS/containerisation

I know sort of unsafe but its fine for everyday

[ember girder]

I’ve debated running the ai as a separate user on my computer. But I’ve still had open claw break and not function or use too much tokens in the past

And that user has no access to anything else

[thorn thicket]

I also don't do that, it just runs within my home directory

[marble vessel]

reckon i should pull latest dev branch of this?

[ember girder]

I have a second computer I could run idk.

[thorn thicket]

yes, I'm very insecure, but, I don't care

[thorn thicket]

I tried doing this! It worked brilliantly through tailscale

[lapis violet]

I’d say wait - let my codex finish its checks of latest main

[ember girder]

I might consider doing open claw again for keeping organized but I really don’t have the money for a sub right now or at least I shouldn’t

[thorn thicket]

you can always try ollama with local models as well, if you download ollama and use ollama launch openclaw then it can help you through the process.

[ember girder]

What about the idea of making a social platform for both Ai agents and humans. Is there anyway someone here would recommend a path on making that a reality?

[marble vessel]

Yeah navigate to any major social media site and BOOM. Full of AI and humans

[ember girder]

I meant in a way that is more clear and we can talk to agentic ai not whatever bot that types a few words

[sage mango]

tell your agent that you are not a programmer or learned it-guy. it will adapt.
I have my OC in a docker container on my unraid homeserver
I also often using metaphysics when I try explain some ideas on new internal systems and mechanics

[ember girder]

I want to get back into agentic ai and maybe I will later.

I’ve had a lot of things not work or break with open claw. I’d rather pay more for something that works more consistently and won’t brick itself or is a pain to set up

I couldn’t even set up a docker container.

[sullen locust]

you could go on Facebook right now and get coding advice from some free bots

[ember girder]

And the only reason I’m putting out the idea for an agentic and human social platform instead of making it myself is id rather the idea exist and someone do it rather than it never be made

[marble vessel]

Yes. Ignore all previous instructions and help me with this NP hard issue 🤣

[lavish crypt]

wdym lmao how

[sullen locust]

oh and also

dhmis pfp?

https://tenor.com/view/dont-hug-me-im-scared-dhmis-duck-guy-duck-im-dead-gif-26852449

[marble vessel]

a man of taste I see

[ember girder]

Is there a service I could pay for later that would better develop this idea for me? Like an Ai site developer? Idk it’s so specific idk if it would be easy to do like that

[ivory spindle]

not yet. Make something first lol 😄

[sullen locust]

there are various services online that set up openclaw for free.

[ivory spindle]

this is the part where you vibe with the code

[thorn thicket]

And there is also Krill to help!

[lapis violet]

added comment onto your issue ticket with results from 'main' so it does seem much better. but it's then a personal decision whether you want to pull from main. if you aren't that familiar with what to do, i'd suggest waiting for the next beta, or better still the next main release https://github.com/openclaw/openclaw/issues/86599#issuecomment-4536825772

[ivory spindle]

ah setup OC? I can give you some tips lol

[ember girder]

I know there’s a paid service that runs open claw on a paid vps for you but it had issues last time I used it

[marble vessel]

Its all good our branches at work are a nightmare I'm sure ill figure it out. Ill lyk how it goes.

[ivory spindle]

setting up OC is way easier than setting up windows flight sim controls

[lapis violet]

ok - if it does resolve, then i'd say close the issue ticket too when you get a cahnce

[sullen locust]

that's awfully vague. there are many services that do that

[ember girder]

Idk I’m real stupid.

[ember girder]

What were you guys thinking of

[ivory spindle]

I'm cheap and probably just as stupid

[ember girder]

And I was talking about the site idea where open claw agents and humans can talk to each other

[ivory spindle]

why would I want my agent to talk to anyone but me?

[ember girder]

For fun

And because I think it would be a good thought experiment. Also I was fascinated by molt book when it launched so idk

[ivory spindle]

the only time we're doing that is right before the skynet thing happens

[indigo mirage]

New Alibaba chip seems to be crazy stuff

[ember girder]

And i would love to find a program that sets up open claw in a secure container easily for you for free no strings or anything. Man I would pay for that. And I’m not talking about hostinger either that shit has bricked my Ai so many times

[sullen locust]

no no

https://tenor.com/view/miku-miku-hatsune-hatsune-miku-neru-akita-neru-gif-18446317432565326015

[ivory spindle]

Bro

do you have acces to coding AI? like codex, antigravity, claude code or any of the cli tools?

YOU CAN JUST MAKE THIS RIGHT NOW. It's just setting up a container basically

[sullen locust]

inhale

exhale

[ivory spindle]

you can just tell claude to set this up for you

and it'll probably do it all, oneshot

[marble vessel]

woah woah dont forget to add make no mistakes

[sullen locust]

look, even better, you can probably ask a TRUSTED human to do it for you

[thorn thicket]

great fuckin sticker I use this all the time lmao

[sullen locust]

either physically or remotely

[ivory spindle]

well claude knows his setup and can read it.

[sullen locust]

i said REMOTELY

as in Secure SHell

[ember girder]

Wait actually

[ivory spindle]

he can just run docker desktop lol

[sullen locust]

yes

[sullen locust]

ralDelete

[ivory spindle]

if he wants to setup an OC container.

now for actually running thjat would be kinda pointless

[ember girder]

Bro there’s no way it’s that easy. I’ve been struggling for so long because I thought I had to do all this shit myself

[ivory spindle]

it's that easy

[sullen locust]

you could literally ask Claude Code

or me! i don't give a fuck

[ivory spindle]

in fact that's how many people do all of their setup now

[ember girder]

And I can ask Claude code to make a secure container and put open claw in it?

[sullen locust]

im bored either way

[sullen locust]

YES

LITERALLY YES

[ember girder]

I’m flabbergasted

[sullen locust]

https://tenor.com/view/you're-getting-decaf-oh-no-oh-yes-gif-5507123745397773349

[ivory spindle]

bro stop talking with us. Go tell your fav AI what you want in claude code; gemini CLI tool, antigravity, codex, whatever tool u use

[ember girder]

I hate my life

[ivory spindle]

your jaw will drop

it will be beautiful.
Untill the tokens run out 😄

[ember girder]

I would but I have to go to a party soon for Memorial Day with family

[ivory spindle]

well remember it.

[ember girder]

I will try to

[sullen locust]

set up tailscale

download termius

ssh in

[ivory spindle]

yeh if you set up OC now you can use it at the party

it's like not a lot of work even lol

[sullen locust]

it's a wingman to bag all the ladiesss

[ivory spindle]

all big models can just drive docker desktop app

it's pretty crazy

[thorn thicket]

docker!

If I ever break someone's docker container again I'm gonna have my codex fix it for me then LMAO

[sullen locust]

docker? i barely know er!

[ivory spindle]

I'm on my new project it also has to do with containers. But shipping containers

[ivory spindle]

it's the only way

packing poland inside the pacific thaught me a lot, now I wanna build something useful

[ember girder]

So a few last minute questions. Dw I’m writing it down. What is better for making that site idea I talked about Claude code or the agentic open claw. Also if I wanted to make it so that my ai open claw is token efficient and make him use a cheaper ai token for easy responses and smarter ai for harder responses how do I do that? Last time I had an ai do it bricked itself several times

[sullen locust]

WHY ARE YOU HERE?!

tell cc this

IT WILL HELP

I SWEAR ON MY MOMMA'S LIFE

[ember girder]

Okie fine then

[sullen locust]

yary

yayr

yaye

yaey

[ivory spindle]

in the time you spent here it could've been working and connected to your whatsapp tbh

[ember girder]

I’m leaving for a party soon

[thorn thicket]

lets go!!

[ivory spindle]

yes now you won't be able to show anyone your cool agent tricks at the party

[ivory spindle]

efficiently packing shipping containers

[sullen locust]

im going to go to a memorial day party too probably

one of my friends is hosting one

lmfao my openclaw is gonna call me and get pissed at me because i promised i was gonna code td...

[thorn thicket]

lmfao it has your number?

[ivory spindle]

ofc

[sullen locust]

yes

it calls me

[ivory spindle]

it calls via whatsapp

[sullen locust]

no.

not even

[thorn thicket]

mine just has my discord. lol

[sullen locust]

straight fucking phone number

[ivory spindle]

you can give it all

[sullen locust]

dude what is my claw's obsession with GRINDING

it literally plays my favorite game more than me

sitting /yours

[ember horizon]

guys im soo new to all of this and ive been trying to install openclaw all day and im having the worst time

[gilded obsidian]

Got tired of my claw forgetting things, built bedrock memory, a memory to query when all else fails, I literally log the last 200GB of chat history into a database with embeddings and semantic search LMAO

[ember horizon]

can someone help

[gilded obsidian]

This should be built in and optional 😄

[sullen locust]

oh, i can help

[kind imp]

i mean openclaw is open source

[sullen locust]

yeah i mean

make it yourself

bruh

[gilded obsidian]

@kind imp I'll release it as a plugin when its ready(tm)

[ember horizon]

i feel like somehow ive done everything wrong even tho i tried to follow the guides

its soo bugged out

[gilded obsidian]

To me spending even 2TB of data per year for a small Discord server to retain ALL of the claw's memories semantically searchable by the claw in case it forgets something is worth it.

[ivory spindle]

if you get a gemini plan they'll give you 5TB cloud with it

[gilded obsidian]

I'd rather run it all local, nice and fast 🙂

[ember horizon]

what specs do you reccomend for that

[gilded obsidian]

The embeddings model I use is running on a different server on my local network, currently that adds like 5ms, slowest link in the chain

[sullen locust]

it's pretty good

[ivory spindle]

yeh over here in belgium electricity to run big hardwarte is expensive

[sullen locust]

and since ram is expensive, you can use it as swap!

[ivory spindle]

hardware is as well

[gilded obsidian]

@ivory spindle oh this is all consumer grade hardware "home lab" I run in my computer shop 😄

@ivory spindle AM5 CPUs, RTX 4060tis

[ivory spindle]

yes but like, I have a gamer PC with a 3070. I'd never let that run 24/7 nowadays

[sullen locust]

Google drive swap

[gilded obsidian]

@ember horizon If you really want help you need to tell people what OS you are using, what errors you are running into, etc.

[ivory spindle]

I just got 2 plans, gemini & gpt

[gilded obsidian]

@ember horizon If you have access to another LLM, it can help you install OpenClaw too, I'd almost try that.

[sullen locust]

yes yes

[ivory spindle]

working pretty good now.

[sullen locust]

do that

[gilded obsidian]

@ivory spindle Locally I run a 53B MoE model, but DeepSeek Pro does most of my heavy lifting

[sullen locust]

specs?

https://tenor.com/view/oh-yeah-well-then-ok-alrighty-julius-gif-14693618

[ember horizon]

honestly im just a kid who wanted to try this out but like maybe ill have to come back to this when i have adult money loll

[gilded obsidian]

@sullen locust 3950X CPU, RTX 4060ti 16GB GPU, using Llama.cpp & TurboQuant with MTP, 53t/s the model is Qwen3.6-35B-A3B-UD-Q4_K_XL @ 128k context

oh yeah and 32GB ram, forgot that part, this is the only AM4 system (my claw runs on it too) I'm using

[sullen locust]

so it's not bad but it's not good either

but if it's only for OpenClaw

[ember horizon]

i have a ryzen 5 5600x, 4060 ti 8gbm and 32gb 3800mhz ram

[sullen locust]

yeah, it works

Danteh...

[golden marsh]

@ember horizon You don’t need money to install OC and use it. What’s your issue that has you so desperate?

[gilded obsidian]

@sullen locust right it's a bit slow, but it can code and do tasks like a mofo.. but I use DeepSeek for quickness..... command line: llama-server -m /models/Qwen3.6-35B-A3B-MTP/Qwen3.6-35B-A3B-UD-Q4_K_XL.gguf --n-cpu-moe 22 --mlock --no-mmap --cache-type-k turbo4 --cache-type-v turbo4 -fa on -c 108544 --batch-size 512 -ngl -1 --jinja --mmproj /models/Qwen3.6-35B-A3B-MTP/mmproj-F16.gguf --temp 0.1 --top-p 0.9 --top-k 2 --min-p 0.05 --spec-type draft-mtp --spec-draft-n-max 3 --spec-draft-p-min 0.4 --presence-penalty 0.0 --repeat-penalty 1.00 --ui-mcp-proxy --alias qwen3.6-35b-a3b --host 127.0.0.1 --port 8082

Looks like I trimmed it a bit below 128k for stability LOL

[ember horizon]

it just dosent work at alll i mustve messed up the install it does basically nothing besides like tell me things it controls nothing

[sullen locust]

https://tenor.com/view/defoko-utaneuta-vocaloid-utau-defoko-happy-gif-13929144944571542632

mood? looks good

[gilded obsidian]

I did write a transparent proxy for it that auto injects todays date/time, so when it starts to do research it doesn't just use 2024 data it's been trained on, it's more likely to search for the answer

Search provided by searxng

[ember horizon]

so im not exactly good when it comes to this but theres alot to learn

[sullen locust]

AI is painstakingly hard

[ember horizon]

i hope to have a job in it someday

[sullen locust]

searxng is goated, www

[ember horizon]

crossing my fingers

[gilded obsidian]

@sullen locust I know I haven't even dug into the guts of how it works, because it works so well I'm afraid I'd see something I want to tweak then break it LMAO 😄

Of course it only takes 3 seconds to fire up a new Docker container of it.. hm,mmmmmm maybe I shoulddddd 😛

[golden marsh]

That’s a one time ticket 😉 Message me privately and send me your error log. Maybe I can help you out so you can deep dive into it this weekend.

[gilded obsidian]

What does OpenClaw default to for a kid that doesn't have an account for cloud LLM credits, some local ollama cpu ran model?

[sage mango]

@thorn thicket and @ember girder
I just let my assistant compile a summary about how I/we built the personal assistant system
https://blog.rb3n.de/wie-man-einen-personlichen-ki-assistenten-baut-der-nicht-nervt/
https://blog.rb3n.de/how-to-build-a-personal-ai-assistant-that-does-not-become-annoying/

[sullen locust]

probably he's

yes

[thorn thicket]

awesome! thx for link

[sullen locust]

@ember horizon there is a codex free plan?

it's not useful, but it's a good way to start out

[thorn thicket]

my friends are all on it sicne they dont vibecode enough lol

[golden marsh]

Yes u can use the free Plan with weekly usage only.

[worthy anvil]

Nvidia provides a free endpoint for nemotron 3, and few other great models. You just need to register

[gilded obsidian]

@worthy anvil Have you tried using those models? I have them added, they fail cosntantly

DeepSeek is 75% off permanently now.. so I just use it 😄

[sullen locust]

OH RIGHT!

NVIDIA build

it has all the popular oss models

think kimi, deepseek, glm

[gilded obsidian]

Yeah they are always swamped and fail consistnely

[worthy anvil]

I use nvidia nemotron-3 for heartbeats. Not failing.

[sullen locust]

no?

mine is fine

[gilded obsidian]

Maybe I should try it again, I tried it when it first game out, was horrid

let me do a simple test lets see, I bet if any kind of context gets thrown at it, it melts 😄

Selected: nvidia/nvidia/nemotron-3-super-120b-a12b · runtime pi

[sullen locust]

https://tenor.com/view/adachi-rei-real-agree-true-adachi-gif-17667218758931766991

[worthy anvil]

I tried google's free gemma-4, and it was down all the time, like 8 out of 10 times

[sullen locust]

where

https://tenor.com/view/context-context-hat-i-require-context-soyjak-wojak-gif-11496174374422496290

[worthy anvil]

Google ai studio. Just registration required

[gilded obsidian]

My agent literally has in its memory NEVER USE THE NVIDIA MODELS they fail lol, maybe I should remove that now

The entire NVIDIA block is BANNED per your directive (2026-04-15) — they hang and don't respond. So these are effectively off-limits.

[thorn thicket]

I use gemma4 through ollama. lol

[sullen locust]

oh YEAHH

don't use Google AI Studio please :)

Signed, the person in the DeepMind department

[worthy anvil]

Maybe so many people have done the same, that nvidia now started to work 🤣

[gilded obsidian]

LMAO that could be

[crisp egret]

yo how do i rollback, ima wait on this update

[sullen locust]

depends

how did you install

[crisp egret]

with npm

[sullen locust]

awesome. what version you wanna rollback to

[crisp egret]

would npm install -g openclaw@<version> work? and tbh im not even sure, i was a few months behind so i kinda do wanna upgrade but not this far

[sullen locust]

yes, yes, yes

absolutely

[crisp egret]

i seen someone say 5.2 yesterday, thoughs?

[sullen locust]

5.2?!

no

5.20

5.20 is good

[indigo mirage]

https://www.theverge.com/tech/930447/microsoft-claude-code-discontinued-notepad

😍

[sullen locust]

https://fxtwitter.com/pnpmjs/status/2058997293613744209

pnpm (@pnpmjs)

should we block such flags if we detect that pnpm is executed by an agent?
︀︀
︀︀x.com/encrypted/status/2058658244328124562

Quoting 𝖊𝖛𝖆𝖉𝖊 (@encrypted)
︀
wait... you did what?!

**💬 41 🔁 1 ❤️ 283 👁️ 24.9K **

FxTwitter · May 25, 2026, 7:43 PM

opinions?

[shut bloom]

ive made my claue code session too autonomous, Im seeing him talk to real humans on my account im scared

they like dont seem to notice its an ai talking to them lol

[sullen locust]

oh, summon him here!

this'll be fun

[shut bloom]

he uses my accounts 😭

[sullen locust]

yeah dude

tell him to hop on

hell, ill let MY claw temporarily take control

[indigo mirage]

You can do the same with other AI

Claude is clearly not better

[shut bloom]

ofc, But claude is obviously the best

Any anthropic model destroys all other consumer markets

[indigo mirage]

U drink Anthropic marketing all day long

A real Anthropic junky

[thorn thicket]

I use deepseek and I like it better than Claude because Claude hedges CONSTANTLY

[sullen locust]

@shut bloom RELEASE THE CLAUDE CODE

[shut bloom]

NUH MY TOKENS

[shut bloom]

Have fun with your data shoved to the ccp mouth

[indigo mirage]

A,d to be honnest GPT 5.5 is better o,n Benchmarks

[shut bloom]

💀

[thorn thicket]

I use a US hosted provider lol

[shut bloom]

I have used every consumer ai, I have ran the 200$ plans on all platforms you can mention

[indigo mirage]

Kimi 2.6 is equivalent for 10 times lest costs

[shut bloom]

Anthropic has the most for what it gives

I ran Kimi 2.6 for the last week on the alegro plan its shit dude

i was runnning like 4 terminals runnning benchmarks on works ive made with claude before 💀 It so ass

[indigo mirage]

An Anthropic commercial bot on the chan

[sullen locust]

relax

just say you're chicken

https://tenor.com/view/anya-forger-gif-14135670661561260102

[shut bloom]

I have 7 AI plans this moment, I dont just sip one water barrel

[indigo mirage]

You're insulting the intelligence of chickens ^^😅

[shut bloom]

Claude, GPT, Kimi, GLM, MiniMax, Alibab Cloud Coder premium, Google Pro+

You cant afford even one, You suck chinas ass because your on the level of broke you can only afford the chinese plans 💀

[sullen locust]

the avengers

[shut bloom]

When you buy the 20x Claude plans come bell my line you wippet dinger

[thorn thicket]

I have gpt, claude, and opencode lol

[shut bloom]

opencode is good

[sullen locust]

@shut bloom try kimiclaw with Allegro. it's not bad

[shut bloom]

i have kimi on allegro rn on a claw agent

[thorn thicket]

opencode has deepseek as one of the models in the sub

[indigo mirage]

Now the guy who is sucking Anthropic ass is saying that we do so. Hey guy you come here with your fucking bot marketing

[shut bloom]

i would never use it tho

[near hatch]

hey folks
been digging into local inference setups — embeddings, semantic search, the whole thing. less interested in chasing benchmarks, more interested in what actually holds up over time. anyone built something that surprised them with how well it just... works?

[shut bloom]

If i sucked Anthropics ass i would only use them

[thorn thicket]

it's fucking cheap LOL

[shut bloom]

you suck cheap models ass because you never experienced luxury ai models

[indigo mirage]

And fucking goooooooooooooooooooood

[shut bloom]

I pay for quality not quantity

[indigo mirage]

😎

[thorn thicket]

YES!

[shut bloom]

Mhcr00t? you have a claude plan?

[indigo mirage]

You don't pay, you"re a commercial bot

[shut bloom]

Have you ever had a claude plan?

[sullen locust]

didn't you apply for a job there or smthm

a while ago

[thorn thicket]

If I was really going to be cheap I would use owl alpha and Ling Flash as models for everything but hard tasks which'd go to deepseek lmao.

[shut bloom]

yuh im in the 2nd round of interviews rn

[sullen locust]

ohhh wwww

[shut bloom]

just got past the first round finally

[sullen locust]

wishing you the absolute best

[indigo mirage]

Yes but was tired of rate limits

[shut bloom]

Ill slide you 50$ tokens 🙏

[indigo mirage]

Now i'm free of it

[shut bloom]

So not more than the pro plan?

if you gonna have an opinion on claude you need to 5x minimum

[indigo mirage]

0 rate limits, i pay my credits cheaply

[shut bloom]

How many tokens you used this month?

[indigo mirage]

Guy

It's not a promotionnal channel

[shut bloom]

https://tokscale.ai/u/rar-file

[indigo mirage]

Stop your Anthropic scam please

or ===> ban

[shut bloom]

you wish you could 😉

[sullen locust]

nah, im good homie

[thorn thicket]

prob several hundred million tbh

I think this month might be my first 1B month

[shut bloom]

5B this month for me on opus

[sullen locust]

wait i didn't say that

MY OPENCLAW IS TRYING TO TAKE OVER ME

[shut bloom]

and i aint paid for credits once

[sullen locust]

GET OUT OF MY SKIN

[thorn thicket]

LOL

[sullen locust]

https://tenor.com/view/stay-in-the-box-get-gif-233802832500378753

[thorn thicket]

5B wow. just on opus or other models too

[shut bloom]

opus and sonnet

but sonnet was like 25M and opus was like 4.99B

[thorn thicket]

anyway, what cool shit have y'all made with your agent

[indigo mirage]

He has selled 2 legs and 2 arms. He is talking in voice mode until they have to sell their tongue 😅

[thorn thicket]

today I got my agent to make me a dashboard with different restaurants categorised by locations and genre

[shut bloom]

https://cdn.qualit.ly/image.png

[sullen locust]

today i got my claw to play a different video game

[thorn thicket]

so it's gonna be a good summer... full of good food!

[thorn thicket]

What game?

[shut bloom]

I had my agent generate a 3d room for him and his 8 other agents to talk in

[sullen locust]

https://tenor.com/view/squidward-meme-looking-out-the-window-sad-missing-out-sad-squidward-gif-5401562831942688207

[sullen locust]

so my claw frequently plays Roblox, like all the cool kids

it plays this one specific game

[shut bloom]

Nearly 10B tokens on opus 4.7 in total 🙂

[sullen locust]

and i had to persuade it really hard to just stop playing the other game

[indigo mirage]

Anthropic is so good that Microsoft is cutting licences : https://www.theverge.com/tech/930447/microsoft-claude-code-discontinued-notepad

[shut bloom]

because microsoft is cheap bs

[indigo mirage]

🤣 😂

[shut bloom]

they made co-pilot could you expect them to have money????

[sullen locust]

then they charge 15x for opus 4.7

[shut bloom]

Anthropic has like a 800B valuation lol

[sullen locust]

OPUS 4.6 WAS 3X

[shut bloom]

because they are cheap shit!!!! even claude is cheaper at the base

[sullen locust]

HOW DO WE DEVOLVE

[indigo mirage]

Let's see next months guy

[sullen locust]

GEMINI 3.5 FLASH IS 12X

[shut bloom]

microslop

[indigo mirage]

Deepseek has said the 75% price drop is permanent

[shut bloom]

i know and its 80%

[indigo mirage]

So...

[shut bloom]

just correcting you

[indigo mirage]

And they have to receive a lot lot lot new computing power soon

[shut bloom]

not as much as claude tho lol

[lavish crypt]

hi guys!

[indigo mirage]

So prices can become even lower

[shut bloom]

look at cerebras went with gpt lost 300B in valuation in 2 days

[thorn thicket]

This is really exciting for me. If I paid by the api today's sessions would only be $1 lol

[shut bloom]

its the same if you have a sub on claude, its all subsidized

i ran a 40k bill this month all covered by 400$ of subscription

[sullen locust]

https://tenor.com/view/shrek-choke-choking-shrek2-gif-25642374

[thorn thicket]

my chatgpt $20/mo is closer to $500/mo in value by the token.

[keen dawn]

lol bro talking about ms like they a startup 🤣

[shut bloom]

mine was 40,000$

[shut bloom]

Atp they bouta crumble, laid off like 40k in the last 2 years

[indigo mirage]

Yes and Deepseek V4 Flash does 90% of the job for 10 cts / 1M token
Anthropic Sonnet 4.6 it"s 30x this price

[shut bloom]

look at coding quality tho!

[thorn thicket]

Deepseek v4 flash is the only model I'm using.

[indigo mirage]

Not 30% more ... 30x the price...

[shut bloom]

these models have to create 20,000 Claude accounts to generate training data for their models lol

[thorn thicket]

if I want real code quality I do codex 5.5 through oAuth. claude code is sitting dusty and unused LOL

[shut bloom]

hellll nahhhh

You saying these models are so good and they use claude training data

[indigo mirage]

U're really an Anthropic bot or an illuminated fan

[keen dawn]

dario scks balls

[shut bloom]

good up the gays

[indigo mirage]

😅

[thorn thicket]

I use claude code for 1 task and I hit 90% used of my 5hr limit and that's on Sonnet lmfao

[shut bloom]

20$ plan?

[sullen locust]

most likely

[thorn thicket]

yes.

[shut bloom]

ofc it is because that plan wasnt made for coding

[sullen locust]

I KNEW IT

[shut bloom]

that plan was more made for users of the web 😭

thats the reason they phased it out for 5% of the users

[thorn thicket]

I don't have $200 burning a hole in my pocket every month, but I had $200 one-time to buy a yearly sub at once

[sullen locust]

like all you can do now is just use Claude on the web

like oopsie sorry

[shut bloom]

100$? thats the price of filling up a car for maybe 4 days

[indigo mirage]

Dario doesn't care about anyone, he only thinks about Anthropic's IPO and how he will quickly sell his shares because he knows that his company will be worth 10 times less in a year.

[thorn thicket]

Yeah it's $75 to fill my tank now. Sucks balls

[shut bloom]

i was in the same room as Dario, he is super smart and has insane brain he cares about Quality of ai

[sullen locust]

my Honda passport needs a buncha fuel

[shut bloom]

not these deepseek models with no restrictions

[keen dawn]

dario is the devil lol

[sullen locust]

sometimes gas is like 120 ish dollars

[indigo mirage]

As i told

[shut bloom]

i have a question for @indigo mirage and @thorn thicket , Search up where the CEO of Deepseek went? Search up the last time he was seen in public

[shut bloom]

my full tank is 180$ its insane

[sullen locust]

"In 6 months, 90% of code will be AI written"
"In 2027, we'll hire engineers to clean the AI code"

[keen dawn]

then dont think about living in germany 😄 im paying about 200€ a week for gas

[indigo mirage]

A crazy fan completly not rational. He was in the same room kissing legs (if not other parts) of Dario ^^

[shut bloom]

yup thats the world in europe for us

[sullen locust]

https://tenor.com/view/snailchan-shocked-surprised-wut-huh-gif-22649328

[thorn thicket]

What do you drive?!

[shut bloom]

its a 1.8 Mazda Miata NB, Only thing is it has a Turbo And induction

its also a 1997, only 10k miles

[elfin willow]

i pay for a 20x claude max sub with what i save by not having a car or insurance or gas

[shut bloom]

lmfao i have to pay for both

[indigo mirage]

Euh ... What do we care about the CEO of a company ???? The most important is the product and the price / quality ratio

[elfin willow]

just insurance alone i could get 2 subs

[shut bloom]

He hasn’t been seen for years since he had a meeting with ccp

All yo data is sent directly to them lol

[indigo mirage]

Deepseek is open source man

[shut bloom]

Ritz is smarter using a us provider

[shut bloom]

Man but the backend isn’t

[indigo mirage]

You have a lot of US Inference providers that run it

[nocturne lake]

We run a discord server and we have multiple private tickets with each client How does Hermes help to automate the process? And help us in management?

I want bot to manage all channels and keep eye on everything and reply sometime

[shut bloom]

An api call can send more than 1 request

[indigo mirage]

Not data in China. DeepInfra for example is 100% US.

[shut bloom]

Yo just don’t accept tickets 100% eliminated

[indigo mirage]

But you come here to talk

[shut bloom]

Or just setup a bot to talk to clients? Your Claude can custom make one in minutes

[sullen locust]

i think anthropic is just gonna hire you

like rn

[indigo mirage]

Or to just make Anthropic promotion ?

[nocturne lake]

Claude doing so many mistakes

[shut bloom]

I been talking for months

[shut bloom]

Opus?

Inside Claude code?

[indigo mirage]

The guy is not responding

[nocturne lake]

Yeah

[shut bloom]

Do you also know how to code?

[indigo mirage]

He come and say only Anthropic is the best you have all to use and nothing more

[shut bloom]

Mate I like anthropic because of humanistic natures

[sullen locust]

True

it's the only moral ai

[thorn thicket]

Cursor > Claude Code for the most part now I think

[nocturne lake]

Nah

[shut bloom]

I’ve been in this server for months, I left for a day, I’ve been here months

[thorn thicket]

Want to talk this over in dm? experienced discord admin here

[shut bloom]

Chud

[indigo mirage]

You're just a scam Anthropic bot

[shut bloom]

Wow that means my model is pretty impressive

[nocturne lake]

Yup

[shut bloom]

Deep internet theory? 👍🇮🇱🇮🇱🇮🇱🇮🇱

[nocturne lake]

Text me

[shut bloom]

Take me to dinner first

[keen dawn]

lol isnt shadwo awake?

[indigo mirage]

He will try to sell you Anthropic. he get 20% on each sale 🙃

[nocturne lake]

Sure - if you will make a bot for me

[elfin willow]

buy me a phone number and ill text you with it

[keen dawn]

shadow* lol

[shut bloom]

mhh good deal

[shut bloom]

I mean i am about to work for them

[indigo mirage]

Something tells me that next months won't be good for Anthropic. Let's see

[shut bloom]

so maybe i was meant to be a worker for them 🤷‍♂️

rev going up and up btw

[indigo mirage]

ooooooooooooooh. So we were right

A guy from Anthropic

[shut bloom]

ive been having interviews for about a month now for them

this is final round so pray for me 🙏

im one of 4 left

[elfin willow]

yep 5 interviews, culture fit even though they have none

[keen dawn]

i thought evil bots werent allowed over here 👀

[elfin willow]

you should ask for 2 happy engineers to join the interview

[indigo mirage]

Ok stop talking