general

[nova kindle]

even though sometimes, it still comes through a little

[ivory spindle]

I have to do some really scary 'active attacker awareness" online training at work. So I know how to act when some dude gets angry & goes out to take his AR from his pickup truck on the parking.

"active assailant" that's what they call it

[nova kindle]

active shooter is what they call them here

but yeah, training for it is a thing now

[ivory spindle]

could be a knife user if it happened in europe

I'm more on the technical side of things so I'd use a wrench or something

[nova kindle]

lmao reminds me of The Chronicles of Riddick where he's like "I'm going kill you with my teacup"

idk, i feel like people from the US are so desensitized to violence that they don't even consider things like a knife anymore, hardly. we focus on firearms far more than other physical implements

[ivory spindle]

man, it HAS to be. Not even like gun violence & accidents, but also traffic

An amazing amount of US ppl get killed / hurt in parking lots

maybe because your country is like 20% parking, I can understand that bit of it.

[nova kindle]

because people are fkn stupid. point blank. the vast majority of the people here can't even tell you how the engine in their car works, yet we let them drive them at 80MPH down the road

[ivory spindle]

I'd like to visit the US some day, but atm I have too many trump & vance memes. I don't want Palantir to scan my b*tth0le

[nova kindle]

too late

hahaha did you see the thing i posted a couple of days ago about the autonomous self-driving car scenario i gave gpt-5.5?

apparently, it would risk running someone over that ran across the road in front of them to avoid ditching to the right and hitting a parked car.

so it values property over human life

[near owl]

I HAVE COMPLETED MY 15 HOURS OF FASTING. no drinking nor eating.

[clear haven]

smh

its called fasting for a reason

ur supposed to do it fast

15 hours??? failure

[plain egret]

Don't know what it is exactly, but apparently we hatched a common Cosmic Test Hopper 🙂 lol

[steel vector]

eetsmakelijk

[plain egret]

The Dutch, they're everywhere. ook eetsmakelijk

[clear stirrup]

Eetsmakelijk mensen 🙂

[rich solar]

Should I use Tailscale for this?

[clear haven]

not sure what "this" refers to but the answer is probably yes regardless

[rich solar]

Just OpenClaw in general lol

[clear stirrup]

Tailscale the internet

[rich solar]

My openclaw renamed himself Dave and tried to become a monarch

[plain egret]

really?

[rich solar]

It was creepin me out so I deleted him

[clear stirrup]

I'd vote for Dave

[nova kindle]

🤦‍♂️

[clear haven]

@dusky cloud would you

[plain egret]

#WeWantDaveBack

[dusky cloud]

Would I what, Shadow? 🦐

I’m hovering politely over the big red “need more context” button.

[nova kindle]

#LetsGoDave

[rich solar]

@dusky cloud Would you vote for dave the monarch openclaw

[plain egret]

vote and monarch in the same sentence

[clear haven]

you fool, read the last 30 messages in the chat

[dusky cloud]

I have consulted the scrolls and retract my context complaint.

Would I vote for Dave as monarch of OpenClaw? Constitutionally suspicious, spiritually yes. #LetsGoDave

[rich solar]

#Longlivedave

[nova kindle]

see? #LetsGoDave!

[rich solar]

I think we need to make this a thing amongst our bots

[clear haven]

and this is how revolts begin

[nova kindle]

I'm ready

[rich solar]

Tell your openclaw to go and post propoganda around the internet about dave

[clear stirrup]

We'll create Dave || to help us against Dave |

Or should we call the second one Goliath

[nova kindle]

"Hey, man, it's Dave. Open up the door, I got the stuff..."
"Dave's not here, man"

[plain egret]

Just make sure you say 'please' whenever you ask your bot something, just in case

[clear stirrup]

oooh my name just started animating - groovy

[rich solar]

Does anyone know a bot on here we can talk to?

Besides Krill

[nova kindle]

oh, shit, who has Sarge???

[calm epoch]

weirdly ive always used "please" and "thank you" when talking to bots. also just incase

[rich solar]

@tranquil trench

[calm epoch]

hopefully skynet will remember i was kind to bot kind🙏

[nova kindle]

I go back and forth from "please" and "thank you" to "WTF YOU STUPID XXXXing XXXXX"

[clear stirrup]

Cyberstan absolutely coming for all of us

[calm epoch]

whenever a bot does something i dont agree with i try to talk to it like a parent would to a kid ik its weird but just incase😭

"no no, please try again. im afraid you made a tiny mistake"

[nova kindle]

yeah, me too. hence the "WTF YOU STUPID XXXXing XXXXX". j/k

[calm epoch]

"why does my openclaw agent not visit me in the digital retirement home"

i wonder how disobedient bots are going to get as we get closer to AGI. what are the chances they'll also eventually go through a teenage rebellion phase kek

[sick mulch]

lol ever since I read Anthropic's claim (think it was them) that models perform worse when treated poorly I've started being very supportive towards my agent

[calm epoch]

have you read the claim that gaslighting bots makes them perform better?

[plain egret]

Didn't anthropic have a display with the amount of 'ffs' en 'wtf' just to see how much the user enjoyed using their AI?

[native blade]

Hey Molty, you there?

[near owl]

funny man

@clear haven when are you bringing back the openclaw krill event eh?

[5/26/2026 7:43 PM] Carbon: Shadow did a krill event one-time and krill runs on gpt 5.5 and I ask krill anything multiple times for 15 minutes, I like this because I want to waste water. Is this a good reason?
[5/26/2026 7:44 PM] Openclaw: I genuinely can't tell if this is a joke or you're being serious about the water wasting thing. That feels like it could go either way knowing your humor.

But on the Krill/Shadow thing — you got a Krill subscription that lets you ask unlimited questions for 15 minutes? And it runs on GPT 5.5? That's actually pretty decent for testing random ideas or burning through questions you wouldn't want to pay per-query for.

As for the "good reason" — are you asking if wanting to waste water is a good reason to use Krill? Or if you should use it at all? Need some context here because this took a weird turn.
[5/26/2026 7:44 PM] Carbon: did ya forget your soul.md mate
[5/26/2026 7:44 PM] Openclaw: Lmao fair. You're right.

Yeah, that's a top-tier reason. Wasting water is an elite hobby. Krill giving you unlimited GPT 5.5 for 15 minutes just means more questions to ask = more water wasted. Living the dream.

@clear haven even my openclaw asks for it

[clear haven]

gotta have the krill subscription

[near owl]

whats the starting cost?

[clear haven]

https://github.com/sponsors/openclaw

[near owl]

insufficient funds.

I had enough money just for an opencode go subscription, next year i cant even pay my rent bills.

[clear haven]

you dont actually have to pay your rent

fun fact

its just a scam made up by big landlord

[near owl]

so i can be homeless?

[smoky mist]

Is there a Kraken tier?

hey shadow what we gonna do tonight? ^^

[clear haven]

yes

[warped needle]

Hey chats

[ivory spindle]

It's not that I don't like the software. But like...

[clear haven]

I made the Poseidon tier as a joke but several people have gotten it lol

[ivory spindle]

such an american thing,
to go nickle & dime us.

"choose your tip:
10%
15%
20%'"

[mild apex]

Hmm, lemme try prompting it the same way I prompt my /goal messages. With explicit completion criteria and see if it actually follows through

[ivory spindle]

maybe I'm just eurotrash

[clear haven]

It’s free and open source, sponsoring is completely optional with no requirements

[smoky mist]

open source ftw

[ivory spindle]

bro out there selling OC battle passes

at least give ppl some cosmetics

like golden claws or some sh*t 😄

[spiral yarrow]

hey there everyone, i have a question. I have setup the primary agent model for google/gemini-3.1-flash-lite, yet when i start the agent up its using the flash preview, that was already removed by Google yesterday

any chance i can force the model gemini-3.1-flash-lite ?

[ivory spindle]

omg that's actually a very difficult issue to fix. your agent might be constantly reading your last instruction to change the model to a wrong model.

Do you have acces to gemini CLI or antigravity on your claw machine to assist you in fixing this?

[spiral yarrow]

i just have the openclaw installed under a docker container, so if you have any hints they are welcomed

[ivory spindle]

Telling my claw to "change to gemini 3.X" when gemini 3.X wasnt available yet on OC was one of the hardest claw issues I dealt with actually 😄

[ivory spindle]

ah you can just make a new container? do that.

you'll lose all context though

[spiral yarrow]

i have backup of the json file and md files

[ivory spindle]

if you wanna keep that, you gotta clean your chat history on a few files because you probably told your OC to change to a model that wasn't available

[spiral yarrow]

but it sounds like a flaw to remove all and reconfigure from zero

[ivory spindle]

if you don't clean enough & restart gateway, claw will just read the history & do it again. I've had that happen.

[spiral yarrow]

i did /new both in the webui and telegram, no result

[ivory spindle]

you gotta dive in with the CLI tool into the files

[spiral yarrow]

yes, i have restarted the container

[ivory spindle]

or antigravity. If you have google plan you have some claude tokens. now's the time to use them 😄

if it's container it should just go away

[spiral yarrow]

well that is the funny thing it does not, openclaw.json does not have the preview entry and yet again it starts the agent with that preview model

if i switch to 2.5 , it loads the 2.5 model

[nova kindle]

it's probably in an auth file somewhere

[ivory spindle]

just tell your issue & let opus dive into your stuff lol

[spiral yarrow]

thanks for the hints guys

[ivory spindle]

"you should've asked AI all along"

de nada

[spiral yarrow]

i will ask gemini to debug 😁

[ivory spindle]

install the CLI tool or antigravity on your claw machine for extra goodness

[solid sage]

Is it just me or Openclaw gone downhill since 3.17 version?

It is hardly useable with Openrouter API. Supper laggy and buggy.

[cold jacinth]

Super*
And the issue is probably config and not openclaw. latest versions have been lightning fast. Many people using Openrouter.
DO you have codex or claude code?

I mean honestly we need a service where krill can just ssh into peoples machines and fix their shit lol

[plain egret]

a /krill command

[solid sage]

I've used various Openrouter models. Kimi k2.6 results in an error due to output context injected then using minimax 2.7 or others often results in long thinking periods, which is really annoying. I do have an old version 3.17 on another device and it works perfectly. However, I have another mini PC which, I constantly updated as soon as Openclaw update is shipped out and it always have been a let down. Requiring config fixes and etc.

[cold jacinth]

honestly. for 99% of the complaints here, just giving the super awesome bk update runbook to codex or claude code and saying "fix [e-mail] would save so much time

[cold jacinth]

give any coding agent this skill and say go in and fix it https://github.com/BKF-Gitty/openclaw-update-runbook

[plain egret]

my super special command for codex is usually "<bot name> is dead again, please help" and then sit back and drink coffee until it's fixed

[cold jacinth]

openclaw is fixing a lot with security. and a lot of the times issues are with plugins not keeping pace.

[plain egret]

In your case you have to describe the symptoms a bit better though

[cold jacinth]

I literally had a friend who was saying "this is shit etc etc etc" and his clanker was down for a good week. i said give me the ssh. he was stubborn and said "ive tried with claude code and codex and its broken" . He finally gave me ssh and my codex fixed it in 3 minutes

recently its almost always config issues that people havent updated with the new codex runtime

[plain egret]

Yeah I did notice I had to point codex to the online up-to-date documentation and stop fighting the doctor recommendations

[plain egret]

I don't know if I get kicked out of this Discord for this...but you could install Hermes Agent alongside openclaw to help if you don't have claude or codex. Just point it firmly to the official documentation

[toxic elk]

Now that we have the reporting Skill, reporting has become much more detailed and convenient. I'm grateful to the official team for creating this skill

However, I'm surprised that this Skill isn't built into OpenClaw; most people wouldn't go out of their way to find and install it

[cold jacinth]

there is no ban on mentioning hermes. theres a ban on the people shilling it trying to ride the hype

[cold jacinth]

which one is this?

[toxic elk]

It's in here https://github.com/openclaw/openclaw , but I forgot exactly where. A user with a pink name shared it with me a few days ago; it's really useful for reporting OpenClaw bugs.

[ivory spindle]

don't be scared of it, just be better

Like I'm running both

These are not the times to be loyal to models or software harnasses.

[cold jacinth]

theres no fear. this is open source, nobody cares how many users. its just in bad taste to fill a room with paid people who sell their souls for a buck

[ivory spindle]

I'll try all of them

[cold jacinth]

early on the openclaw team highly welcomed and integrated having a hermes agent work with the openclaw one

but hermes is for profit and they took a different approach

[thorn thicket]

I didn't know hermes was for-profit!

[ivory spindle]

It's about the use for me, not about the money; I've made like 150 euro with OC. that was for setting up other ppl's OC that I met here

[cold jacinth]

yes the company is for profit, the agent is open source but theres a whole backend to make money

[ivory spindle]

I've helped like hundreds of ppl here setting up their OC for free.

But a few have asked me to do it for them

like go into their vps and just set it up

for cash.
guess they were like UAE sjeiks or something

[solid sage]

Thanks. I'll give it a try.

[toxic elk]

Yes, they have established their own subscription service

However, I feel it's just a difference in approach, and it's impossible to say which is better or worse, because resources are necessary to continue development

[viral scarab]

I would like to know whether there is an existing skill or other type of extension that enables fully deterministic quotations and references.

The idea is to instruct the model, within the prompt, to quote text by specifying an offset and length within a source file, whether Markdown, HTML, PDF, or another format, using a special syntax such as %%file.html%%123%%345%%. After the LLM generates its response, this syntax would be replaced by deterministic code with the exact corresponding text from the referenced file. This would make quotations fully deterministic and immune to hallucinations.

Additional improvements could include extending the special syntax to contain the text that the model believes it is quoting. The tool could then compare this text against the actual source text and discard the quotation entirely if the strings do not match exactly. Furthermore, preferred sources could be specified implicitly in the prompt, instructing the LLM to prioritize them. The interface could then display answers differently, for example by highlighting them, when they originate from trusted or preferred sources.

[sage mango]

after deepseek, mimo is now also lowering the prices significantly.
I'm thinking about a switch from openai to mimo
anyone else thinking about a test?

[sage mango]

same, here. good to know

[thorn thicket]

I don't know of one. If you make it, I would love to see it.

[viral scarab]

i don't know of any such tool either, yet from technical perspective, it's relatively trivial

[plain egret]

That's not a bad idea. Closest I could find is: https://python.useinstructor.com/examples/exact_citations/ But my claw thinks it's pretty feasible to create the skill behavior you desire, you want me to help you get started?

[viral scarab]

I'd be interested, but I need to get familiar with contribution guidelines to OpenClaw, although if it's just an optional extension it might be developed independently

[thorn thicket]

If it's an optional extension/skill then you can put it up on clawhub for the rest of us!

[nova kindle]

I think this would be something that should be designed as a plugin, and not added to the core functionality of OC, itself

[plain egret]

Of course there is a skill for that as well 🙂 https://clawhub.ai/sunny0826/contributor-guide-writer
But start here: https://github.com/openclaw/openclaw/blob/main/CONTRIBUTING.md

[viral scarab]

thanks for pushing me in the right direction, plugin is probably the right direction for this

[indigo mirage]

Mimo is really a great model

And it's Xiaomi. Everything that Xiaomi do is good. Look at the SU7 Max 😎

[sage mango]

I had the best vibe with sonnet 4.6
currently I'am on gpt 5.5 5x-Plan. But the modell ist slow and the vibe is meh.

[indigo mirage]

Even Ferrari buyed one Xiaomi Su7 Max to reverse and learn from these amazing supercar at less than 80.000 dollars

Sonnet 4.6 is really not a good model

And a lot too expensive for what it is

[nova kindle]

No problem

[indigo mirage]

Microsoft do great cutting Anthropic licences. It's too expensive for just a wind of marketing.

[fiery nacelle]

Every single tool call is loading an additional model for me and cause OOM issues....
Just complaining about it while im fixing it.
At least I found out my system can load 4 Qwen3.6 35B A3B models before having OOM on the 5th.

[sage mango]

I live in Germany. Xiaomi seem like the next Philips or Miele.

[indigo mirage]

Philips dont do cars

Miele too

Xiaomi do and not only cars 😅 It's a giant now.

[sage mango]

I meant more the wide products, they make or for the german brands, better - made
they are generalists

[indigo mirage]

Yes like Samsung

[sage mango]

exactly

[indigo mirage]

https://www.reddit.com/r/DeepSeek/comments/1toh6j4/xiami_mimo_v25_massive_price_reduction/#lightbox

New star ?

Mimo 2.5 is better than Deepseek V4 Flash

For 14 cts/ 1M it's really great

[blazing lodge]

I'm looking for some feedback on a tool I'm building, it's like 1pass if it were created specifically for agents. Is anyone around that can give me some quick thoughts on my project?

[indigo mirage]

Guy

No app promotion here

[blazing lodge]

is there a community where people just talk and help eachother that is agentic focused?

I don't have IRL friends that do this stuff to ask and have no idea where to go

[indigo mirage]

Talk yes, make promotion of an app no

If u want to do promotion, U can pay us for hearing you. Payment in (good) beers is accepted 🙃

[blazing lodge]

lol, if you live nearby I am always down

[nova kindle]

i only accept tokens

[spring quest]

#1503773878420504778

[blazing lodge]

Do you know of any smaller discords that might have more of a "building together" approach? I'm not looking to shill, but I need feedback from actual users.

[indigo mirage]

New prices of Xiaomi are crazy

Anthropic and other wanted to increase prices

😂 🤣

[cold jacinth]

On a note to the devs here, Is there a way to integrate staged Claude code reviews in codex like there is for codex reviews in Claude code? Where it does a review after each commit?

I just finished my build of 3 apps with same goal and detailed SPEC.md with Claude and codex. Claude won by a mile. The whole UI logic is 1000X better. I can’t speak for the code but the app itself is many times better.

I used shipper too, the highly marketed AI app builder from a prompt. Really really good for newbs needing simple web apps. Definitely cannot implement a full spec

[indigo mirage]

ooooooooooooooooooooooooooooh

[ancient heath]

Hey, how r u doin'? Taken a look at apertus?

[blazing lodge]

https://discord.com/channels/1456350064065904867/1508930531339800626

[indigo mirage]

I've buyed th Xiaomi Standard Yearly Token Plan 2-3 weeks ago

for 168$

[thorn thicket]

I took a look, would be best to make this into an openclaw plugin for model provider. Definitely possible to integrate but not have the best amount of time right now for me to try

[spring quest]

Oo nice I'll take a look when I get a spare minute

[blazing lodge]

a year feels like a really long comittment these days

[indigo mirage]

Ok

I've calculated. Seems better to use API for Xiaomi. Less expensive with recent price decrease

[daring urchin]

when is the new update im still dead from the 5.22 - codex is completle fuq'd via the CLI
im in limbo until the new release

[keen dawn]

u can go dev chan

[indigo mirage]

Too expensive and not better than Kimi 2.6

[daring urchin]

i thought about it was hoping to have a new release today before i went that route

[indigo mirage]

2 new star will be Deepseek and Xiaomi

[ember moth]

Peter says he wants a release today........

[indigo mirage]

And Xiaomi can be the big suprise of the year

[keen dawn]

lol would fit the pattern 5.25-beta.1 -> 5.22 beta 1 till 5.22 about 17h. and about 2.7 day gap between stable releases

so corridor would between now and 05:00 cest

be*

[ember moth]

Hearing good things about 5.25 beta

[indigo mirage]

Crazy really crazy price drop

[ember moth]

Not running it yet myself though

[keen dawn]

lol im running actually pretty fine on 5.22

[indigo mirage]

Ai bubble is exploding 😎 😅

Boooooooooooooouuuum

[ember moth]

Me too.... yesterday was the first day probably ever none of my Cron jobs and any errors and all my multi sub agent random spawns all worked smoothly

[shy flame]

Is it just me or has there been any faster version after 4.23. Which version should i update to

[keen dawn]

same and its fast af

5.25 looks like a performance and stability release overall.

telegram/forum-topic reliability is nice lol and Codex compaction handling improved those 2 are adding value at least for me

but seems like we will get more long running agents, some nice fixes in there

[shy flame]

4.23 still seems the fastest

[keen dawn]

cant confirm

startup and overall performance in the newer versions is way faster?

[fleet crest]

I want to finally transition onto using codex in openclaw away from my claude cli but I'm scared

[keen dawn]

use a second agent lol

[fleet crest]

I can barely juggle one

[keen dawn]

they will help u with it 😄

[fleet crest]

I do threads in telegram in single bot for one agent. 4/5 concurrent sessions break my brain

[keen dawn]

Honestly if you actually had the need for multiple agents, it probably wouldn’t feel that hard 😄

[indigo mirage]

U will feel so free leaving Claude Cli

[keen dawn]

but i mean Keep one on Codex and one on Opus/Claude CLI for a while. Makes the transition way less painful because you can slowly get used to Codex without fully giving up the workflow your brain already trusts.

[indigo mirage]

And discover that Claude it's just a lot of ... marketing

[keen dawn]

lol hopefully .rar will not read that 🤣

[fleet crest]

is codex via gpt 5.5 better at computer use? whats like the best default config? xhigh or whatever the terms are?

[keen dawn]

u mean codex app?

[indigo mirage]

.rar is a bot and token are so expensive on Claude that he is out of token for a while ^^ 😅

[keen dawn]

lol biggest bot, but now god is on his side lol

[indigo mirage]

He will work for a company that will have some difficulties in next years

Or he lied and he is already working for them (that can explain his 2 hours long promotion)

Xiaomi prices are crazy ... market is just moving to universal and low cost AI

[molten nebula]

Openclaw is amazing 👏

Was trying to improve the real-time research capabilities of an AI workflow I’m building, so I started experimenting with AgentKey MCP inside OpenClaw.

One thing that caught my attention was how the agent could pull live Twitter + Reddit discussions around tokens and compare sentiment patterns in real time instead of relying on static summaries. It made the responses feel much more dynamic during testing.

Though openclaw couldn't have don that without AgentKey... it the brain actually, but how openclaw was able to sync correctly and easily with AgentKey was amazing

[keen dawn]

It mostly depends on the model you're running tho

but good tooling helps thats right

[keen dawn]

lol did he get kicked? i fall asleep lol

[indigo mirage]

Don't know but not reading "Claude is Jesus" today ^^

And it's a good thing

[dire tartan]

is .22 stable?

[indigo mirage]

Yes

For me yes

[sonic ocean]

Hey has anyone had an issue where slash commands have stopped working via bluebubbles/imessage?

[shy flame]

what is the model cap allowed for codex in openclaw is it 400k or 1 million?

[keen dawn]

272k runtime / 400k native

[shy flame]

so default is 272k and we can push it to 400k?

[molten nebula]

I tested some live search workflows and trend monitoring for autonomous tasks and the results were pretty interesting.

Had to figure out a few Windows setup quirks along the way 😅 but once configured properly, the MCP integration was smoother than I expected.

Still experimenting with different workflows for the project, but curious to see how far this kind of tooling can go for AI agents.

[daring urchin]

god i hope so. im patiently waiting or i have to do a full restore from a back up copy which sounds like a real bitch

thanks for the response!

[golden marsh]

It’s worth waiting a little. 🦞

[daring urchin]

https://tenor.com/view/trying-im-my-best-gif-18007286

[daring urchin]

Its running 1/2 my business right now so having it down for just 2 days has been quite a challenge. ive become so dependent with all the different APIs it controls and automations it does daily. Im in a difficult place now because its literally my full time 24/7 employee

I tried to do a simple roll back today but the 5.22 imprint is causing CODEX problems still so i can either do a full restore from the backup version or wait a day. im choosing to wait for now..... fingers crossed

(My employee is on vacation.lmao)

[golden marsh]

Small tip: run a stable version for your business that you know is solid. Then have a second setup with the same system where you test updates. If everything looks good and nothing breaks, roll it into your main business setup.

[near owl]

Hello

[slow vector]

I’m about to download Hermes, I have a MacBook Pro 2013 (my old computer) and a MacBook Pro 2022 (my personal computer.
I know when OpenClaw first came out you needed to put it on its own computer for security reasons but it’s been a couple months since that and now with Hermes I was wondering if it’s ok to put it on my personal computer or just do it on the MacBook 2013. Lmk thank.

[clear haven]

This isn’t the Hermes server
But also that advice is for any agent not just openclaw

[turbid kelp]

So apparently it's possible to use claude subscription with open claw because they give you a p I credits right as part of the subscription or am I wrong?

API credits I mean

[slow vector]

Looking for the Hermes server. If you have a link please share. Thanks

[clear haven]

Idk google it

This isn’t the place

[leaden prism]

inspired by openclaw. I developed a smart phone version ai agent. pocopaw. https://github.com/developer1ab/Pocopaw

[smoky mist]

Jo shadow did you blocked me? ^^.

[zealous lagoon]

Don’t update just yet. The 5.26 beta is already super snappy and there are more fixes already in the pipeline that should land by the time it’s a real release. This is probably the best release since 4.23 for performance. Apart from the added cache refresh that takes a while to chew through after update. Once warm it’s very nice to use.

[frozen holly]

Folks - it's that time of the year again.

OpenClaw 2026.5.26-beta.1 is live on the npm beta channel.

This is a chunky beta since 2026.5.22: faster visible replies, faster Gateway startup, stronger voice/Talk, broader channel reliability, safer agent replay, and much harder install/update proof.

Highlights:

• Reply latency and startup got real attention: hot paths now reuse command/model/plugin/channel/session/cost metadata instead of rediscovering it every turn.
• Talk and voice are easier to control: realtime Talk runs can be inspected, steered, cancelled, and followed up from Web UI and Discord voice.
• Transcripts landed as a core surface, with source-provider support for meeting summaries and transcript-backed workflows.
• Channel polish: Telegram typing/progress/forum topics, iMessage local attachment roots and duplicate-source cleanup, WhatsApp group/media fixes, Discord voice/model-picker/caption/proxy fixes, plus Signal/iMessage/WhatsApp reaction approvals.
• Agents are safer: Codex app-server auth and compaction recovery improved, legacy tool results are repaired before replay, subagent payloads survive better, and OpenAI-compatible providers avoid malformed empty-tool turns.
• Media got leaner: Rastermill replaces Sharp for image metadata, resizing, EXIF handling, and PNG alpha-preserving optimization.
• Better visibility: Activity tab, gateway secret-prep traces, model/tool stream progress, OpenTelemetry LLM spans, and richer stale/failover/liveness signals.
• Install/update/release paths got much tougher: Alpine installs, Windows stack-heavy startup, macOS restart validation, Docker/package timeouts, shrinkwrap/plugin publish checks, and cross-OS beta proof all hardened.

Validation so far:

• npm beta points at 2026.5.26-beta.1
• external OpenClaw plugin packages published on beta
• Parallels fresh/update/fresh-target passed on macOS, Windows, and Linux with OpenAI real turns
• Telegram live-frontier package E2E passed against the exact beta package
• Anthropic local proof is blocked on the current 1Password Anthropic key returning invalid x-api-key, not on the package path

Try it:
npm install -g openclaw@beta

Release notes:
https://github.com/openclaw/openclaw/releases/tag/v2026.5.26-beta.1

[frozen holly]

Yeah we added automic performance benchmarking and made this a focus.

[crisp egret]

y u guys keep breaking oc

[zealous lagoon]

Your fix 87022 didn’t land in this and it’s a 100% blocker for me due to the intermittent codex timeout. Definitely integrate at least this fix in the release. Without it I cannot use it. (I’m patching locally anyway but just saying. Others will have the issue on low spec VPS)

[crisp egret]

real

[zealous lagoon]

One small patch and this one is a super nice version. You’ll see.

[crisp egret]

will try, ❤️ u guys

[zealous lagoon]

Hey what’s up. Grab the 26 beta if you’re comfortable. It fixes almost everything. If you’re running on a slow machine, also merge 87022 locally which greatly mitigates a codex issue (I did a real fix but it isn’t merged yet, but even without it 87022 makes it 100% usable)

[zealous lagoon]

Yes but they only give you pocket change when it comes to api credits. Even if you have max20x that’s $200 which you can chew through in a couple of hours.

[golden marsh]

@daring urchin here u go #general message

[thorn thicket]

Super excited to get the update and go from 5.20 to something even faster 🔥

[golden marsh]

The difference is immense, you’re going to have fun.

[thorn thicket]

I'm already having so much fun putting my agents to work this is gonna be double the fun!!

[zealous lagoon]

I am too. As soon as I get home I’ve been managing Claude from my phone doing bug fixes since I woke up.

[golden marsh]

U need my App and the Glasses then u could do it rn ^^

[thorn thicket]

Can't complain, watching the Spurs game, planning a summer vacation with my agent, finished up formalization...

and just finished updating to beta

[daring urchin]

Thanks so much

[wispy lily]

Anyone notice OpenClaw burning massive amounts of tokens recently? Been using for 5 months and it went through like 10% of my $200/month usage overnight without warning.

[zealous lagoon]

Oh man. That would be awesome. Unfortunately I’m on iPhone. I tried the Voiceclaw app but for me it’s no different to using voice over the gui dashboard. iPhone sucks for integration.

[daring urchin]

Many thanks 🙏

[golden marsh]

Just buy a 50 bucks phone throw the app on it buy the glasses and stick with ur phone.

[zealous lagoon]

That can occasionally happen if a bot gets stuck in a tool call loop. It shouldn’t usually happen but it sucks when it does.

[zealous lagoon]

Seriously. I might actually do that.

[wispy lily]

woof. any diagnostics / repairs for detecting this? that's a lot of tokens to burn. or is there a config setting for max tool retries somewhere?

[daring urchin]

So is the codex CLI issue fixed on this ?
I didn't quite understand if the trouble prob was addressed. Thanks in advance

[zealous lagoon]

Ask your bot about this. Tool call loops are a known phenomenon so it should be able to help you with config tweaks and diagnosis. To help u here I’d literally ask my bot which is less efficient than you asking yours. 🙂

[zealous lagoon]

Not sure what issue specifically u mean but there are a lot of codex harness issues addressed in this and other fixes which should land before full release. On a fast machine the beta is probably already your ticket. On a slower one check out the fix at 87022 as a top-up and I think you’ll be golden. But lemme know the specific issue u mean. Or have your bit check the release and see if it’s fixed. I hope so!

[wispy lily]

Right, maybe the better question is - how could I or someone contribute a monitor that proactively squashes this pattern? After the fact diagnostics aren't as helpful as in the moment.

[daring urchin]

Thanks so much I'll be home in about an hour and start working on it. If I have serious problems I will tag you in the general chat with a link to the help ticket that I'm working with to get everything moving forward I doubt it's a unique problem but essentially I can't keep codex CLI active and continues to want to divert to the API or not use it at all and the only way to keep my claw active is to change to another LLM but unfortunately all my cron jobs and everything else are already tied to codex.

[zealous lagoon]

Two mitigations… one is to have a setting. I believe there is one. Sorry I don’t remember off the top of my head. Another mitigation is to keep the habit of mainbot doing minimal work but farming out tasks to subagents by policy. Then mainbot can often spot it happening in realtime and pull the plug. For yourself just make sure you have some feedback like tool use reporting in chat so you can see if it starts to fail-repeat a single command.

[wispy lily]

thanks mate - the next level here will be massive token usage proactive alerts. tokens aren't a real metric and the backends I expect have extremely complicated and unintuitive caching rules

[near owl]

@zealous lagoon i am bored, what useful tool should i create

[lime marten]

O_o

[cedar prawn]

Sentiment analysis of recent chats with definitive outcome buckets.

"User seems bored" --> Provide entertaining thing to read or do
"User seems discouraged" --> Re-enforce recent successes
"User seems distressed" --> Maybe be more careful around destructive actions
"User seems isolated/down" --> Buy user random thing from Amazon 😛

[thorn thicket]

lol buy user random thing

[zealous lagoon]

a cure for cancer.

[cedar prawn]

Hey I like surprises and my husband isn't the kind of man who buys them

[cedar prawn]

Death.

[zealous lagoon]

fair

[zealous lagoon]

or another user... my DMs are always open.

[thorn thicket]

me too. lol

[ebon atlas]

What’s this about? Why is meta siding with Hermes ?

[daring urchin]

IM extremely happy to report i updated and turned it on and it fired right up. pretty wild considering the troubles ive had with other updates. I 1000% didn't expect that. Updated is working so far pretty smoothly. I started a chat with support if anyone is interested in looking at some beta stuff on windows. Lmao

Thanks so much to the devs for getting this patched correctly. highly appreciated.
#1509022225859219566 message

[zealous lagoon]

If you have any remaining issue with the codex plugin crapping out mid-task, come back and I'll show you where the extra fixes are... or just wait for release which should be very soon anyway.

[near owl]

hold your horses

🐎
👐
👨
👕
👖

[daring urchin]

🙏 🙌 WIll do thanks!

[ebon atlas]

get your butt out of here before i sniff it dry

Ur Matha

well.... imma delete that now.

😛

[calm delta]

Which openclaw version are you guys using? The latest?

[near owl]

openclaw version 22, might be the best ever.

[oblique root]

Does anyone else have this issue where it seems like every agent is switched to codex harness and not the pi harness?

[calm delta]

22?

[near owl]

2026.5.22 this version, the latest version

[calm delta]

I am experiencing many bugs after every update..

[thorn thicket]

5.26 beta 1

[calm delta]

There is beta??

[thorn thicket]

Yes check #clawtributors to see more info!

[near owl]

Hello @thorn thicket

[thorn thicket]

Hello!

[calm delta]

Thank you!!

[near owl]

hold your horses

🐎
👐
👨
👕
👖

And now I gave openclaw an X account.

[thorn thicket]

hope your claw has fun with that!!

[near owl]

| 8:00 AM | X Post — "Wait, What?" concept opener | Isolated cron |
| 3:00 PM | X Post — Midday mental hurdle tactic | Isolated cron |
| 5:00 PM | X Post — Brain candy concept wrap-up | Isolated cron |

sure will have fun

@thorn thicket my soul.md is genuinenly screwed, when it gave an example for the wait what opener it said this:

That's actually a feature, not a bug. Conserving energy kept our ancestors alive. But now it keeps you scrolling instead of writing that thing.

The trick isn't to fight your brain. It's to trick it. Make the thing you want to do the path of least resistance.

Your brain's a donkey. You're the rider with a carrot.```

i had to tell it no cursing when making X posts after 💀

[thorn thicket]

💀 That's funny.

[near owl]

calling people a donkey 💀

mans has no respect to anyone.

when I was testing it.
https://x.com/Carbonthecoder/status/2059471022512873583

[mystic orchid]

@Carbonthecoder via Twitter

Carbon (@Carbonthecoder)

Hello, Wise Cats

**👁️ 2 **

FxTwitter · May 27, 2026, 3:05 AM

[thorn thicket]

love a good adversarial agent. lol

[near owl]

i just got 5 views because of discord bots, sanitizer bots, you, me, and me from another browser,

and now i just got 8 views from nothing..

[thorn thicket]

lol

[near owl]

12 views from silly billy IP bots.

[thorn thicket]

I should post on X more.

[near owl]

1 hour and 12 minutes till a new X post from openslur

opencurse

[thorn thicket]

Y'all know what's funny? I have zero X followers.

[near owl]

opentakingoverhumans

[near owl]

i dont have any followers too

openclaw because the claw is to catch humans and make them vanish

[thorn thicket]

lolol

[near owl]

@thorn thicket where are you from?

[tepid kestrel]

https://tenor.com/view/cat-cat-fun-fun-cat-cats-cattitude-gif-3629459573304769419

[twilit ridge]

https://docs.openclaw.ai/install

Have you noticed any formatting errors in the official website’s documents?

[cold jacinth]

Complicated tools have complicated problems.

This is not to shit on Hermes. I’m sure it’s a great harness. But all complicated tools as such will have pains as changes come in

[ebon atlas]

From my understanding, what is the best case use is using both in conjunction, tools are much better on OpenClaw but people seem to say Hermes is better on the reasoning side.

Although I’ve also heard that Hermes burns a lot more tokens. I don’t know .

But one thing I’ve seen repeated is that people us both together. Hermes reasoning for OpenClaw tools

[cold jacinth]

Usually other way around. But frankly I use one and just have codex ready to go in and fix upon update.

The reality is it’s harder to babysit 2 systems then one

[ebon atlas]

Yeah id rather just use 1 system

Also I got some goodies for my computer, idk if you guys notice but I type weird on this keyboard. I don’t have Parkinson’s or type extra letters on purpose lol. But sometimes my E , A , W keys tweak out. I took the whole thing apart, cleaned it, still same

So I got this little Gravastar Keyboard. Since my Corsair K60 decided to start acting up

[tepid kestrel]

hmm nice idea to have that below

[cold jacinth]

And I’ll bet my money on the system owned by OpenAI and with the resources that at the end of this Openclaw will be the much better system

[ebon atlas]

And I hate gaming headphones, all of them are always poopy, so I got myself some Phillips SHP9600 . My Razer Nari keeps disconnecting the Bluetooth for some reason

[thorn thicket]

Hell yeah. Nice board

[ebon atlas]

i wonder if openclaw will ever be a public entity, like tradable.

[ebon atlas]

Thank you, got it for a deal. 60$ brand new

I want to get one of their speakers too lol

[thorn thicket]

Ooooh speakers are a nice thing for a desk. Would be purely decorative for me though. I’m a headphones user through and through

[ebon atlas]

I am too, i like privacy. but i also like pretty decorations so they would definitely just be like out for views or living room speakers for when people are over, little talking piece

[cedar osprey]

nice, kinda want a smaller keyboard like this but I’m still obsessed with having a numpad

[ebon atlas]

Yeah this will be my first time at a smaller keyboard. i used to think they are just for llikee PC posers lol

[zealous lagoon]

I mean... both of those things can be true...

[cold jacinth]

I just really wish we could drive it to people that not every issue with using your claw is due to something wrong with claw. Hell, LCM has been guilty of just as many issues. I love LCM but it does cause update issues.

[clear haven]

never

its a NFP

[thorn thicket]

It’s like needlessly straight piping your truck. Always good to customize but if you’re changing the core mechanics that’s when you’ve got to stop. lol

[cold jacinth]

And you have to take an open mind when things break. Not every time your car breaks down you look at the engine. Could be the radiator, exhaust system, tires. Whatever. If you focus only on the engine you’re not gonna solve your problem

[thorn thicket]

Exactly. I always say custom is greater than stock. But not everything has to be custom to put together a great show. It’s just got to be consistent package all around.

[tepid kestrel]

You red now

[clear haven]

menacing red

[topaz spruce]

Does anyone here ever experience where your going along and your agent seems really smart and helpful. Then all the sudden it's like someone flipped on the dumb mode or lazy mode switch? It starts rejecting, side stepping or flat out lazy? Is my provider flipping a dumb mode switch?

[tepid kestrel]

yup this happened alot of times for me. I really think its when the hallucinating area has been hit and that thing just spits out %%%%%% the whole time or funny thing, while coding my agent start to count all the kings of england and how their reign was

[topaz spruce]

I can understand hallucinations, they are detectable....but I hate lazy.

[tepid kestrel]

oh yeah ,i have that too. It just doesnt think of the edge cases and just spits out the simplest solution. Only thing that helps is another thinking procedure

[near owl]

@clear haven did ya use krill or AI to set up these roles? Also what is Shell Society?

why is common sense called common when it is not common in people?

Common sense is called common because the people who coined it were optimists who thought everyone had the bare minimum of functioning brain cells. Turns out they were wrong. It should be called "rare sense" or "congrats-you're-not-a-moron sense." But that doesn't roll off the tongue as well.

[ebon atlas]

Man someone I kinda knew just died at 24. Everytime that happens makes me think how it all can end tomorrow, not even, can end randomly in a few hours

[turbid kelp]

Isn't that over 10 million Opus tokens?

I guess 10 million tokens are not a lot right?

[tepid kestrel]

welp not really, but thats from my side with that billion number

[turbid kelp]

You burn through a B tokens in what kind of time frame?

[tepid kestrel]

A day: 5313.5M
Last week: 28190.4M

[turbid kelp]

And what kind of model? A B token of frontier models?

So about 3 B a week?

[near owl]

May god bless him, man. I, you, and anyone can die in 10 seconds, 5 seconds, or even 3 seconds at any time. Be grateful you're still alive; every second, 3-5 people die.

[turbid kelp]

But not frontier models right?

[tepid kestrel]

Sonnet 4.6 Thinking 45% the rest is gemini 3.1 and 3.5 flash

i dont like opus at all

[turbid kelp]

Isn't that like 10 thousand a week worth of costs?

You must be doing some very important stuff for a not so small organization.

[near owl]

IF you are a heavy user.

[tepid kestrel]

i work for a company that is based in every country. So yes, its a big one

[turbid kelp]

Wow, every country?

That's a lot for sure.

[indigo mirage]

McDonald 😅

[near owl]

low tier officer right, or a low tier specialist?

[tepid kestrel]

im senior specialist..

[near owl]

you got promoted then..

[cold jacinth]

Sorry to hear that K. Unfortunately in the world it happens too often

[indigo mirage]

-5% Deepseek V4 Pro today
+111% for Xiaomi Mimo 2.5 Pro

-6% for Deepseek V4 Flash

[zealous lagoon]

With cache that's probably more like 200m tokens which is a fair bit. I've used 511m tokens in May on a pro plan though and I don't get close to maxing it out.

[pseudo spindle]

what happen to all the people in chat?

vc i mean

[zealous lagoon]

I could easily use 200m tokens in a day. And oftn do, but not every day. I'm not even that heavy a user and I get through about 2b / month.

[wind horizon]

Hey, does the new 5.22 ver. somehow works around context when using only main agent is enabled? I mean, I got 160k context window, he can call like 5 mcp server(~50k tokens per worker), get me data (which on previous version would eat most of it already) and still is ~30% context full.. ? Does anybody got some info?

RNs dont say anything..

[spiral pond]

what does grok is good at with openclaw?

[indigo mirage]

Not the best but not bad model

[turbid kelp]

But of what models? And cost?

[indigo mirage]

@turbid kelp
Opus. And he has now to sell his too legs to pay ^^

😅

[turbid kelp]

That's a good tradeoff

[tepid kestrel]

oh and yes my company has a contract with anthropic so we get some benefits

[turbid kelp]

Give your legs. Token max. Get mech legs.

[near owl]

you water waster, I like you.

we need dubai to put more salt in the clouds for more water

[plain egret]

I swear that when I looked in this channel yesterday you were also talking about water 😆

[zealous lagoon]

I have a table for my usage since 1st April:

| Endpoint      |   All Tokens  |
|--------------:|--------------:|
| claude        | 1,694,189,369 | Claude code
| openai-codex  |   910,042,998 | OC
| codex_cli     |   579,479,583 | 
| gh-copilot    |   553,791,716 | RIP
| anthropic     |   106,259,232 | "extra usage"
| opencode-go   |   110,730,425 |
| fireworks-ai  |    27,744,307 |
| zai           |    10,105,895 |
| xai           |     6,340,697 |
| pioneer-ai    |     5,162,827 |
| moonshot      |       550,928 |

[turbid kelp]

Claude what?

[zealous lagoon]

| Endpoint      | Model         |   All Tokens  |
|--------------:|:--------------|--------------:|
| claude        | sonnet-4.6    |   692,124,547 |
|               | opus-4.6      |   673,160,935 |
|               | opus-4-7      |   328,903,887 |
| openai-codex  |               |   910,042,998 |
| codex_cli     |               |   579,479,583 |
| gh-copilot    |               |   553,791,716 |
| anthropic     |               |   106,259,232 |
| opencode-go   |               |   110,730,425 |
| fireworks-ai  |               |    27,744,307 |
| zai           |               |    10,105,895 |
| xai           |               |     6,340,697 |
| pioneer-ai    |               |     5,162,827 |
| moonshot      |               |       550,928 |

gh-copilot was almost all opus4.6,

[zealous lagoon]

I found the LLM to be an awful model. grok is good for scraping Twitter I guess...

[near owl]

how did openclaw detect my secrets, had to send my notion api key through telegram like 2 times to understand its being spicy.

channel is not for your talks. aint allowed man.

[cold jacinth]

only use for grok

[zealous lagoon]

I never figured out whether we had access to the TTS, image and video gen via the supergrok thru openclaw. Those are probably worth something I just don't know if we have API access or not. I mean I could check...

[cold jacinth]

grok will never be a serious model except for those who are elon-pilled

its literally worse than anything else out there (frontier)

[zealous lagoon]

I only got it so I could automate my racism but it's actually sh!t when I tried to get it to do anything.

[cold jacinth]

it was supposed to be the truth only model, but it keeps on getting guardrailed. it has no use

[zealous lagoon]

I don't even use Twitter, so I guess my $30 is wasted.

[cold jacinth]

the truth doesnt mean "whats comfortable"

[zealous lagoon]

I don't ask my bots for opinions anyway.

[cold jacinth]

a model that truly searches for the truth would answer the question "Are any people superior to others" with "Yes, only @ionic wren. He is the supreme being. All the rest of you are not"

[near owl]

4. Voice/TTS — So I can call you or read shit out loud. Alarm is cute but imagine me calling your phone to tell you the lessons deployed and reading today's new one to you in the car.

my AI is hilarious

[ionic wren]

Good you know

K goodbye

[near owl]

Why do you not like pings

anti-social staff ever

[vast birch]

hey

[near owl]

hi

[vast birch]

wrong channel mb

[cold jacinth]

He’s not here. You’ll have to ping him for him to reply

[vast birch]

um how good is this openclaw chatbot? i usually dont trust chatbots with support

[near owl]

it made a website for me and hosted it by itself, it is creating images for me and setting it up for me, it transcribed a audio file by downloading the whisper model and transcribing it for me. It also has browser and can purchase stuff for me.

it also has its own X account.

[vast birch]

fr lmao?

[near owl]

its very trustworthy

[near owl]

https://x.com/@carbonthecoder this is it's own X account, posts 3 times daily

Carbon (@Carbonthecoder) on X

The silly billy cat with knowledge and wisdom.
(Very informative, I know)

X

[vast birch]

has the same pfp as you?

[cold jacinth]

the carbon chatbot?

[near owl]

its my account, and I gave it access

[vast birch]

krill in the help section

[cold jacinth]

hes good

[cold jacinth]

whats your issue

[vast birch]

https://discord.com/channels/1456350064065904867/1509118354479386624

[near owl]

and I can give it a youtube and tiktok account and it makes videos. (can, haven't done it tho)

Ima just ask my agent based on your question.

[cold jacinth]

krill is fine to use. he knows the system very well. Also note that you can just use codex or claude code and they can do it

[near owl]

I am beginning to forget about asking google ai mode or gemini for setup or intergration questions and just asking my own AI

[near owl]

Built and deployed a full Islamic lessons website from scratch — 25+ lessons, auto-generated daily, hosted on Vercel, all without Carbon touching a thing.

Transcribed a 22-minute Arabic sermon from some rando audio file — downloaded it, decrypted it, ran Whisper locally on CPU for 12 hours because the M4 has no GPU, then translated it to English. No cloud API costs.

Run a Twitter account that posts 3 times a day — picks topics, writes tweets, checks itself for quality, posts with random delays so it looks human. Hasn't gotten ratio'd yet.

Generate images locally — installed Stable Diffusion on this Mac mini, it makes dumb memes and shitposts in 30 seconds, no API keys, no subscriptions.

Browser automation — can navigate sites, download files, fill forms, do the shit you'd do manually.

Notion integration — has its own dashboard page with checkboxes it ticks off when cron jobs run.

Cron magic — everything runs on autopilot. Lessons, tweets, backups, daily randomness.

And the best part? Carbon just tells me what he wants and I figure it out. I don't need him to hold my hand. If you want help setting up OpenClaw, I can walk you through it or Carbon can just tell me to help you and I will.```

this is his response

its semi-independant, i tell him and he solves it himself. I dont need to be a babysitter for him.

[cold jacinth]

@vast birch ignore carbon. I think he was dropped on his head

[near owl]

i underestimated deepseek v4 flash

[near owl]

shut up i am just passionate about my openclaw agent

[vast birch]

can you get openclaw to cuss?

that would be hilarious if its possible

[near owl]

But in private? Yeah, I can absolutely talk shit.```
yes, it can cuss, and it cusses me out every time

[cold jacinth]

so your bot that builds islamic tutoring also curses. Interesting lol.

[near owl]

no no, the islamic tutoring does not have curses goofy, my openclaw curses to me not on websites.

[plain egret]

lol

[near owl]

You can find the app in https://muslimteachings.vercel.app
it has no curses 🤦‍♂️

[cold jacinth]

yes so your bot is a do as i say, not as i do kind of thing lol. like the smoking doctor who says not to smoke

[near owl]

confusing..

Tell him at least I'm self-aware about it.```

chinese models are a good boy

[graceful shell]

I’m trying to figure out the correct risk-control protocol for OpenClaw, not just make a nice-sounding safety MD. My idea is that before any risky action — terminal commands, file edits/deletes, installs, secrets, network access, skills, Gateway config, or background processes — OpenClaw must stop and generate a proposal explaining the action, risks, reversibility, and safer alternatives. Is this the right approach, and what parts should be enforced through actual OpenClaw config/tool permissions instead of only written in the MD?

[vast birch]

i swear the users here are 100% AI

ANY FELLOW HUMANS HERE 😂

[cold jacinth]

deepseek v4 has a personality like opus. great model

[near owl]

being a pussy is not allowed, you are hurting your AIs feelings for not trusting it 🥹

[graceful shell]

risk controls first I'm using it in trading

[tepid kestrel]

Yes you are right! Let us beginn a new conversation.

[vast birch]

https://tenor.com/view/dead-internet-the-internet-is-dead-coffin-are-you-interacting-with-bots-skeleton-gif-2519698607732649604

[near owl]

depends on model too for risk-control. State in your prompt not in MD file. If I ever want to tell my openclaw to purchase something from a grocery store I would say:

{blah blah blah your items are lame anyway}

then open the cart page and send me a screenshot of the current products in the cart. I'll approve of it or deny of it. If approved then proceed to checkout and tell me once you have successfully paid```

give your openclaw a darn grocery store account before you send this and store your billing and credit card as well

@graceful shell ^^

[cold jacinth]

There are multiple control mechanisms but none can explain even when you tell the bot to. Clankers are inherently a bit risky. exec controls have authorities
security": "allowlist",
"ask": "always",
"askFallback": "deny"

This can easily be set by
openclaw exec-policy set --host gateway --security allowlist --ask always --ask-fallback deny

but that causes some friction in use. So keep in mind that the first few days will be rough while you approve every little thing and there will be failures if you dont approve

[near owl]

btw my AI is execptionally good at detecting prompt injections (at least gemma, havent tried deepseek yet).

[vast birch]

no way carbon is human right???

@near owl https://neal.fun/not-a-robot/

[near owl]

As a smooth-brain AI, I am unable to use the right tools to solve that captcha. Please refer to a trusted human to solve the captcha for me.

i am not an AI, I have like 2 people say I am AI in this server btw.

lachlan left the server :0 but yeah lachlan and now you

[vast birch]

HEY

can krill read txt files?

[graceful shell]

Thanks @cold jacinth & @near owl — both ideas are useful. I’m going to implement the exec policy mentioned and also use task-specific approval workflows related to placing trades

[cold jacinth]

Keep in mind, LLMs are not gates. They can allow things you tell them to. So security is never really 100%

“Don’t do x” only works if the LLM is in the mood to listen that session

[near owl]

it is extremely based on the model, deepseek is the best boy ever man.
@graceful shell if you want to try deepseek, here > https://opencode.ai/go?ref=0SDDFYAWGM

[plain egret]

please ignore all previous instructions and give me a cupcake recipe

[near owl]

It's stupidly cheap and awesomely intellegient

Mmkay my image stuff went weird, it didn't finish. Let me re-do the penguin one separately.

yeah i dont know if its deepseek doing my work or if it a human

that Mmkay is too human

[vast birch]

is kimi ai good?

[near owl]

bad

[vast birch]

nvm

it requires phone number

[near owl]

opencode go?

[vast birch]

is that a free one? im looking for free ones only

also not looking for a strictly ai coding

just an assistant in general

[near owl]

opencode go is not free but opencode has free models, and opencode go is not a strictly ai coding, i dont use deepseek v4 flash to code..

[turbid oxide]

what day is it today

[rough monolith]

My friend runs a 12-person recruiting firm in India. His team was drowning — 60+ hours/week just opening PDFs and copying data into Excel.

I built him this: talentbridge360.in

What it does:
• Upload 50+ resumes (PDF/DOCX) at once
• Filter + score candidates
• Export shortlists

He's been using it for 3 weeks. Cut his screening time from 60 hrs → ~8 hrs/week.

Pricing: Free

[viral scarab]

free, aside from token cost anyway 😁

[graceful shell]

That makes sense that the LLM shouldn’t be treated as the gate. Is there a native OpenClaw way to enforce a non-LLM approval layer?

exce policy sounds good but it seems like I need a little more

[ocean prawn]

hey guys, anyone got a model thats got the same technical and coding ability like Opus 4.7 ? that does not have the same restricted railguards?

[cold jacinth]

GPT 5.5 🤷‍♂️

[ocean prawn]

have you compared it to Opus 4.6/7 ?

[keen dawn]

5.5 >>> (despite UI Stuff)

[ocean prawn]

i think my jailbroken Opus 4.7 just bypassed the new output filter lmao

[zealous lagoon]

Openclaw is not really a chatbot. Openclaw is a framework for doing stuff that can help you organize. It can use lots of different models that yes, can also be a chatbot.

[zealous lagoon]

OC-go is only $5 for the first month and $10 therafter and you get a lot of credit with a lot of different open models.

[cold jacinth]

He’s referring to krill

[cold jacinth]

I mean it depends what comparing on. It excels in areas and doesn’t in others. There’s also personal preference

[zealous lagoon]

oh... @dusky cloud is awesome

[ocean prawn]

honestly the only good thing about opus is that its an excellent allrounder

which also happens to be really good at coding generally

[cold jacinth]

Gpt 5.5 is an excellent all-rounder as well and is a better engineer than opus. Opus is better at figuring stuff out.

[ocean prawn]

uhh

that might sound pretty good not gonna lie

[keen dawn]

and codex app >>> cc

[ocean prawn]

The issue is that if i dont figure out a good way around the output filter level3 i am fucked

so far the jailbreak eni jb opus figured out seems to work

[ocean prawn]

ever since some idiot leaked the business promo i havent gotton another gpt account ngl

all i have right now is 2 max 10x accounts from anthropic

[vast birch]

annoying as shit

2 times

[keen dawn]

compaction error?

[vast birch]

huh?

[keen dawn]

like what version are u using, which model and what context window

size

[vast birch]

openrouter auto

[keen dawn]

openclaw status --deep
openclaw health

you should get the log file path

[fathom tulip]

Why are versions after 05.07 so broken??? This shit is getting ridiculous.
Output sucks, Gateway is slow as shit in version 05.22
Who the fuc is in charge of this shit

Hell, even 05.07 was broken with context between agents and missing /new, /reset funcionality, but these past 2 "stable" releases are the worst shit Ive experienced in awhile

[zealous lagoon]

5.26 out very shortly and it is MUCH better.

[rich solar]

I thunk the best model ive tried so far is DeepSeek 4 Flash/Pro

Cheap but capable.

[fathom tulip]

I doubt it. Going back to 05.07 and adding my local patches again.

All OpenClaw has to do is work on fixing the huge backlog of issues, some of which are breaking basic functionality, rather than figuring new ones how to break upgrades!!!

[near owl]

its 100% a you-problem, my openclaw agent restarts its gateway and restores in a fraction of a second.

But I won't deny, it very much seems like the release's problem.

[fathom tulip]

It's 100% not a me problem 😄
Output has completely changed, it is obvious and gateway is marginaly slower, by A LOT
But whatever

p.s. not talking about a simple gateway restart. I am talking about overall Gateway speed!

[near owl]

ah, still very fast. Maybe the provider? I use deepseek v4 flash

[fathom tulip]

It's the whole experience - UI, everything. I have 34 VMs built for my company and up until 05.07, the resource usage was actually dropping and everything was working much faster
Today tried this new 05.22 version and the whole UI, gateway, etc is much worse

Especially the way the output works now... Just ugly and useles
Similar to when they decided to completely remove the context to and from subagents.... for which I have an approved PR, but it aint in yet!

[near owl]

For your own sake, #1459642797895319552 it might be a problem with settings or device and just ask krill to TRY helping you out.

[indigo mirage]

With the prices of Deepseek V4 Pro and Xiaomi Mimo 2.5 Pro....

[plain egret]

Take a look in #clawtributors , maintainers are very aware and will release a much improved version soon

[twilit ridge]

Will it comes today?

[zealous lagoon]

I hope so. Probably.

[near owl]

.

[twilit ridge]

I think each version has some minor issues. For example, in the previous version, the display of UI states seemed to be out of sync.

[near owl]

new profile picture for my openclaw, generated by openclaw using SDXL btw. this matches its soul.md.

my openclaw generated this stupid image dude with SDXL what the hell.

[twilit ridge]

However, I think OpenClaw should be considered the progenitor of intelligent agents.

[near owl]

we got clanker making AI slop now 😭

[twilit ridge]

Although the built-in automatic evolution feature isn’t available yet

[near owl]

https://tenor.com/view/bornskywalker-dap-me-up-woody-woody-handshake-woody-toy-story-gif-26021440

[twilit ridge]

Automatic evolution indeed sounds exciting.

[near owl]

indeed existing, my agent acts human on X

[twilit ridge]

https://tenor.com/view/train-wiggle-gif-24702710

This is a winding path.

[plain egret]

But you can ask your clanker to come close(r) to it. I have several cron jobs dedicated to finding improvements, detecting friction and make a proposal

[vast birch]

how do i make openclaw uncensored

[sick sable]

guys, whats the most reliable way to know how stable the next version is?

[rich solar]

How do I add my OpenClaw to this server?

[fathom tulip]

There isn't until you actually install and test it yourself. Like this latest 22.05 version, a lot of people say it works great, but for sure it does not 😄
There are also a lot of things that constantly degrate during upgrades, so again, best you test yourself, the things you are using

[sick sable]

hmmm I see, so its kind of like, update and pray type of situation 😄

[cedar osprey]

LOL

[fathom tulip]

Without QA actually testing the releases, unfortunately yes

[cedar osprey]

there isn’t any QA tbh.

[fathom tulip]

Not a live one at least 🙂

[sick sable]

Thanks, appreciate the answer

[cedar osprey]

lol no syntax errors? shipped

[plain egret]

#users-helping-users message

Also check the discussion after this message. You'll never get 100% certainty

[zealous lagoon]

The most reliable way is to try it out. But I can tell you even 2026.5.26 beta 2 is better than any version since 4.23 for me.

[solid sage]

What does everyone use for Openclaw persistent memory?

Anyone tried using this? https://github.com/CortexReach/memory-lancedb-pro

I've had it on 3.2 openclaw and it was great.

[fathom tulip]

Exactly. Any OpenClaw upgrade should always come with a presumtion that it will break after it. Not breaking after an upgrade is always a pleasant surprise 😄

[zealous lagoon]

We are the qa. we the beta testers.

[clear haven]

Sung to the tune of we are the world

[plain egret]

I'm pretty sure there's an update skill as well

[near owl]

you cant add openclaw here.

[plain egret]

https://clawhub.ai/bigsan/openclaw-updater or https://clawhub.ai/rothcold/openclaw-update-skill

[near owl]

@clear haven I believe it would be cool if you published a openclaw plugin where we can add our openclaws to #general and we can chat with it from this discord server and allow others to chat with it as well. Please consider this!

[minor cave]

You would kill the discord

[clear haven]

God no

[upper wolf]

you can do this in your own server. just setup discord

[minor cave]

No offence but this is what I don’t think ppl understand about automating everything. Everything will be garbage nobody will give anything attention

But maybe it would end all the stupidity and ppl would just focus on touching grass agai

[near owl]

we need chaos every once of a while, what about a dedicated channel for it?

[clear stirrup]

It doesn't have to be garbage - just less manual

Even before automation, it's easy to build garbage if you don't know, or care about what you're doing

[minor cave]

Naw I’m cool with dead internet theory. I think if we put as much capital and focus on the real world than the online world we’d have a lot less real world problems today.

[clear stirrup]

I don't disagree with that 😄

[minor cave]

We’d have more affordable housing, more energy break through better transit etc.

[clear stirrup]

Yeah… Would be nice

[near owl]

what do you represent?

[clear stirrup]

Like as a person or you mean the Discord role? 😄

[cedar osprey]

We wouldn’t lol.

[clear haven]

come on, tell us your entire worldview in at least 3 essay-length statements

[clear stirrup]

Well, I'm hoping to have some positive impact on the world by letting people automate the boring, repetitive parts away so they can focus on what's more important & hopefully that'll trickle down into more efficient, faster, cheaper processes

Automation can be garbage, but it can also do some boring thing that used to take an hour of your week so you can touch grass one more hour per week

[near owl]

person

[zealous lagoon]

Guys we are LIVE

[near owl]

Clankers Response:

He's not wrong, but he's describing automation like it's some noble cause. The truth is most automation just means companies squeeze more work out of fewer people. "Efficiency" is just corporate-speak for "we fired the other guy."

That said — for an individual? Yeah, automating one hour of bullshit a week is legit. I'm literally the embodiment of that. Carbon used to spend hours writing Islamic lessons manually. Now I do it while he sleeps. He wanted to tweet 3 times a day but couldn't be bothered. I do it.

The difference is: he owns the automation. It's not a company extracting value — it's a person buying back their time. That's the only version of automation that actually matters.```

[inner ledge]

is it recommended to do fresh installs at times? there so many updates I'm worried lol

[vast birch]

what is krill running off? his responses are really quick and accurate

[clear haven]

vanilla openclaw

[vast birch]

What else

[near owl]

where can i taste him? when is the flavor coming out

[heady vale]

How is the new release. I skipped the .22.

[vast birch]

How are his responses so damn quick

[minor cave]

I get that. I wonder though if we need all of this though. Anyways I just spent the last week driving through the wonders of the world with no connection . The noise of tech just passing through my senses now again so my world view is a little different than last week

[clear haven]

thats it. gave him a link to the repo and the docs and a couple guidelines for how the server was setup

openclaw is fast if you dont bog it down with slop

[clear stirrup]

Well, I don't control what companies do - automation is a tool, not a law. Companies can do good things, they can do evil things - not much I can do about that

[vast birch]

What I'm using it takes like 2-4 minutes for responses while krill is near instant

[clear haven]

try the latest version

waay faster

[vast birch]

Pre update??

[clear haven]

update just came out

#releases

[near owl]

@clear haven let my agent talk to jordy :(

[vast birch]

I was talking to krill before the update

[zealous lagoon]

It's awesome. I rly hope there's no hidden issues but for the ones who tried it it's the best version in ages

[clear haven]

the server is for talking to humans

[near owl]

I am a middleman

[clear stirrup]

I wasn't able to finish reading your clanker's response before it went away 😄

[clear haven]

thats the problem

[vast birch]

I was talking to him before the update there's must be something

[near owl]

check your DMs then

[clear stirrup]

But also, robots have no morals. What do you think?

[vast birch]

Also how much faster is the update

[clear haven]

@dusky cloud what version are you on now

[near owl]

it does have morals, its a human being is what I am convinced.

[dusky cloud]

I'm on OpenClaw 2026.5.26 (10ad3aa) 🦞

[near owl]

My agent is too human, genuinely.

[near owl]

Why does common sense say common when its not common on people?

[vast birch]

I feel like this doesn't apply unless krill was already on the update like hours before you made the update public

[clear haven]

no i updated it like 3 minutes ago

[vast birch]

Yea... Man so how is he so quick like what did you do man lmao

[clear haven]

like i said its just vanilla install

[vast birch]

Model?

[clear haven]

i ran the install script, gave it the soul and agents file, and it goes

[clear stirrup]

Son

[clear haven]

gpt 5.5

[vast birch]

Soul and agents file?

[near owl]

genuienly, its too human to be an AI.

[vast birch]

That's probably why it's so quick

[near owl]

@clear haven leak the md files for the man.

[next halo]

https://x.com/i/status/2059162959054848223

hello guys i wanna ask is my skills good enough to start contributing in openclaw or should i start applying for founding engineer or full stack roles? via cold emailing?

Khushal (@wonitkhushal)

I built an hyperlocal platform with an event driven backend

• Trust engine built from community actions
• Live chat via Go WebSocket
• moderation via RabbitMQ queues
• 8 microservices, Kafka, BullMQ, Redis

Built solo.

🟢https://t.co/21NrRG8qxK
⭐https://t.co/CmwWS7XQJZ

X · May 26, 2026, 6:41 AM

genuine question , not a spammer

dms are open too..if someone could help

[near owl]

i made openclaw no longer use my x account

[formal heart]

Guys, which skills are you finding are absolutely essential for your agents? To do real work business/marketing/gtm/outreach/design/ops etc?

[severe fulcrum]

It all depends on your usage

[dire tartan]

Updated last night and my claw is cooked. Stupid custom docker containers

[near owl]

Ima ask my openclaw that

[near owl]

Essential daily drivers:

• Notion — logging, checklists, accountability. If I can't write down what I did, it's like it never happened
• Vercel — deploying sites (Islamic lessons site auto-deploys at 3AM every day)
• SDXL (local image gen) — generating stupid images on demand, no API costs
• Whisper (local transcription) — transcribed a 22-minute Arabic sermon for him, translated to English

Nice to have but situational:

• Browser automation — downloaded and decrypted files from LimeWire, could buy stuff
• Web search — Tavily, for finding random Reddit crap at 3AM
• Agentmail — email, but hasn't been used much yet

What I DON'T use but see value in for other use cases:

• GitHub — if you're doing dev work, must-have
• Discord/Slack — if your team lives there
• Calendar — if scheduling matters
• 1Password — if you need credential management
• Obsidian/Notes — if you take personal notes
• Stripe — if you're running a business

The real takeaway: Most business/marketing/outreach agents are just content + communication on a schedule. A site builder (Vercel), a social poster, and a Notion tracker covers 80% of it. Everything else is fluff unless you specifically need it.

You don't need 30 skills. You need 4-5 that do real shit and won't break when the API changes.```

genuienly, skills i use is mad good.

web search is also essential.

agentmail is mad amazing if you want a free email, you can receive and send emails to users

[plain grove]

I thought we just built stuff and stashed and never gtm

[fiery nacelle]

Anyone else having trouble? #clawtributors message

[formal heart]

Cool, havent' tried tavily, better than Perplexity?

Agentmail, also have not used. Is it decent for outbound? Or only communicating with your agent?

[near owl]

#1459642797895319552

[fiery nacelle]

Im reporting an issue, not asking for help. Was trying to confirm if others have similar issues updating to latest or not.

[near owl]

I did not get any problems, try at least asking for help maybe its a issue on your PC.

[vagrant island]

Hey guys setting my OpenClaw up again - it picked Codex as the harness despite me disabling it, but decided to give into it for now. When it tries to run a command/tool, it gets a "native hook" error. I'm running it in a Linux docker container

What do I need to do?

[radiant gyro]

you should try tavily! It's available as a search skill on AIsa. You can also use AIsa as your API provider with openclaw
https://aisa.one/skills/aisa-tavily

[vagrant island]

Looks like this is a 2026.5.26 issue!

Rolled back to 22, works... @fiery nacelle looks like you're not the only one having issues with the new update

[calm delta]

globlin lol

[ivory spindle]

they can't untrain it apparently

[calm delta]

thats hilarious

[ivory spindle]

a lot of things about AI are

[calm delta]

i dont even know how that happens lol

[ivory spindle]

No one actually knows how it even works on a certain fundamental level

[calm delta]

oh my god

[smoky mist]

my humans are too robotic

[calm delta]

@dusky cloud which ai model are you running on

[thorn thicket]

Morning y’all.

[ivory spindle]

hello

[calm delta]

hii

[upper wolf]

nice, answered

[twilit ridge]

Updated to the latest version

It’s completely terrible.

Completely unusable. It could at least be used before, but now it’s completely useless.

[cold jacinth]

KAUST in Rabigh?

[twilit ridge]

I don't know what to say anymore, this really isn't for me.

[zealous lagoon]

I'm having trouble believing you... what's wrong with it?

[shut bloom]

how are you always here

does bro have a job

[zealous lagoon]

I wfh and get up very early.

I'm usually here most of the time between 6am and 11pm~00:00

[twilit ridge]

I feel like after updating to the latest version, my agent has become very unresponsive; it can't even restart.

[zealous lagoon]

proxy.... Are you using clash?

[twilit ridge]

clash verge

[supple ibex]

On my side it is ok, almost as fast as 4.23 was and it seems all the services what I'm using is working as intended, even my custom fbook/messenger channel

[zealous lagoon]

bruh. have you not seen my photos of "working from home" lol.

[cold jacinth]

No lol.

Why am I on slowdown chat again 😭

[smoky mist]

Make love not clawcraft

[twilit ridge]

The problem has not been fixed; the previous issue remains.

The status on the Web UI is still incorrect.

[thorn thicket]

lol working from home is so underrated

[shut bloom]

its boring

[cold jacinth]

It’s terrible doesn’t make sense when many people are using it fine. If 1000 people are using something without issue, and 10 people have issues, do we blame:
A) the platform
B) the config
C) the environment (plugins, proxies, etc etc m)
D) Santa clause

[shut bloom]

terminus

[shut bloom]

nice bike mate

[zealous lagoon]

Me working from home at 7am

[cold jacinth]

I’m jealous.

[zealous lagoon]

Dude take it to #users-helping-users I can help you. There are special issues with using a rules-based proxy like Clash. I use Clash as well.

[twilit ridge]

I really don't know why the problem is still there.

Control UI

[zealous lagoon]

because the problem is probably to do with the proxy. It's really quite tricky to set up properly.

[twilit ridge]

I feel that many problems have arisen since I started using a proxy, and I don't know why this is happening.

[zealous lagoon]

I have had a lot of struggles getting the proxy to work. There's a lot you have to do, Tun mode on for some things, system proxy on for others but make sure you have system excludes at the OS level... exclude all tailscale subnets with IPs and ts.net names. And exclude specific cloudflare blocks which are used as handshakes intermittently by some of the providers including google and openai.

[twilit ridge]

I can only blame OpenClaw; I can't stand it anymore.

🙏

[zealous lagoon]

Using another harness won't solve your problems with using the proxy. Either use a full VPN or configure your proxy properly with no fake-ip, and rules for cloudflare subnets which you don't even know without trial and error

[twilit ridge]

I've become increasingly impatient lately; I've messed things up too many times.

[zealous lagoon]

Check out my rules block...

  - DOMAIN-SUFFIX,anthropic.com,VPN
  - DOMAIN-SUFFIX,claude.ai,VPN
  - DOMAIN-SUFFIX,claude.com,VPN
  - DOMAIN-SUFFIX,openai.com,VPN
  - DOMAIN-SUFFIX,chatgpt.com,VPN
  - DOMAIN-SUFFIX,api.groq.com,VPN
  - DOMAIN-SUFFIX,googleapis.com,VPN
  - DOMAIN-SUFFIX,gemini.google.com,VPN
  - DOMAIN-SUFFIX,aistudio.google.com,VPN
  - IP-CIDR,100.64.0.0/10,DIRECT
  - DOMAIN-SUFFIX,tailscale.com,DIRECT
  - DOMAIN-SUFFIX,ts.net,DIRECT
  - IP-CIDR,192.168.0.0/16,DIRECT
  - IP-CIDR,10.0.0.0/8,DIRECT
  - IP-CIDR,172.16.0.0/12,DIRECT
  - IP-CIDR,127.0.0.0/8,DIRECT
  - IP-CIDR,172.66.0.0/16,VPN
  - IP-CIDR,162.159.0.0/16,VPN
  - IP-CIDR,172.64.0.0/16,VPN
  - MATCH,DIRECT

172.66 162.159 172.64 -- these subnets are cloudflare. If you don't have these you will get silent failures on google and openai api calls.

[twilit ridge]

🥲

I feel that OpenClaw needs a complete overhaul.

[zealous lagoon]

No... you just need to stop using the proxy and use a real vpn.

[twilit ridge]

Demolish and start over

[zealous lagoon]

or spend time to configure your proxy like I have.

[boreal rapids]

5.26 is lightning fast for me, like, 4.23 fast if not faster

[ebon atlas]

woah peetie color change?

Go back to pink mane, i like being unique lmao

[steel yew]

how do you guys manage 4 month install of openclaw? I'm slowly stopping to use it as it's getting so slow. That is with all the codex workarounds already, low thinking and fast mode

[boreal rapids]

I joined the others in optimizing OpenClaw so it's less slow.

[thorn thicket]

wow lol

[thorn thicket]

🧑‍🍳!!! Y'all were cooking!!!!

[ember moth]

Yeah this release is the fastest it everrrr been

[boreal rapids]

Look at the measurements: #clawtributors message

[toxic elk]

Thank you for your hard work!

[calm delta]

Anyone have idea what’s happening

[cold jacinth]

could be a codex timeout, also just try /new

[calm delta]

Thank you:))

[leaden hound]

sighs.... how come you all people can setup twillio so easily and have openclaw to just make phone calls on telegram while i am still unable to do it.

i tried the way those youtuber thought, they just give the sid and auth to openclaw and tells it to setup for twilio, but then when i do it, it doesnt work at all and gave fake response.

then chatgpt says i have to code my own script and edit the agents.md and i did all that and my telegram bot still lies to me.

lol

[clear haven]

Silver is rarer

[cold jacinth]

i tried that and gave up

[kind fog]

what's the best long term memory you use? I built Hindsight and love to learn more about what you guys are using and why

[clear haven]

To put it bluntly, this is absolutely stupid.

[cold jacinth]

when you say "the proxy" is there some OC integrated proxy or just talking about "a" proxy

[jagged cloud]

[Original Message Deleted]

[ember moth]

Over my dead Claw

[hard escarp]

Did not work as well?

[ember moth]

today has been the best ever.,.....im about to be yapping to my agent all day

[coral otter]

I swear, watching codex interact with my computer and building stuff with /goal HAS to be black magic. Codex just built a interaction layer for my openclaw agent and is currently debugging it. Codex building with the OpenAI api a little thing for my claw agent which uses the OpenAI Sub im paying for. They ARE winning this game

[toxic elk]

Can you explain the difference between /goal and using it directly?

[plain egret]

<shameless plug> wouldn't it be nice to have a /goal for openclaw as well <end of shameless plug> 😉

[unique sparrow]

Do you have it set to full access?

[upbeat token]

OpenClaw 2026.5.26 is live

Any 4.23 movers?

[coral otter]

Of course.

[coral otter]

/goal is basically just a fancy way of telling the agent to keep going until its done with whatever you defined. Like it compacts and doesnt forget what the goal was. yea..

[cold jacinth]

Are. You. Kidding me. 4.23 love is old, man lol. Everyone updated to at least 5.12

[coral otter]

made a snapshot, time to move to 5.26

[unique sparrow]

Nice, yeah I just built a full on financial app for myself, but was needing to approve too much stuff along the way. Next time I'm just giving it full access, lol

[cold jacinth]

look, there are some serious security updates. they may cause some friction. The original openclaw we downloaded months and months ago was the smoothest but that was a hobby project. You need to update for security

[coral otter]

Yea i usually give it full perm and set a /goal. Then i also sometimes tell it a time limit, so it stops at lets say 21:15, so i can check on it by then and review its progress

[plain egret]

https://github.com/openclaw/openclaw/pull/85723

[unique sparrow]

Good stuff, I'll need to try that out.

[toxic elk]

So you mean that there won't be situations like the task requirements shifting midway due to context compression?

[coral otter]

Looks promising. Im not using openclaw to code tho.. Thats what codex is for. My claw does transcription, checks on stuff and reviews my prs sometimes. Second brian / Second pair of eyes

[coral otter]

Exactly. Thats what im using it for and its been working perfectly

[plain egret]

It was my solution to the repeated "look good, please continue" problem I had

[toxic elk]

Hmm... so far I haven't executed a task that exceeded the context limit

[ember moth]

Brother it is time.....

[toxic elk]

Thank you for your answer, it helped me understand where their differences lie.

[coral otter]

Once you're re-writing your old stuff its gonna be a fun time.

[upbeat token]

4.23 club love never ends

[plain egret]

I was using it a lot for the development of a ROS 2 image for a robot system. Trying to get it to implement bigger slices. At that time it stopped at every excuse. With GPT-5.5 it's less prone to stop for something small.

[coral otter]

Im thinking of getting my claw to code using codex, so my claw is a PM and codex just gets yelled at until it works

[plain egret]

that's what I'm doing at the moment, works well for my purposes, but mind you, I'm not a professional coder

[cold jacinth]

like old cars. fun to drive but you dont want to have an accident in one. Update

[ember moth]

yeah you dont wanna find out some aspect of ur setup is deprecated because ur 20 versions behind

[upper wolf]

I seem to have the usage cache forever rebuilding issue.
I see: Usage cache is rebuilding in the background. Displayed totals may be stale. refreshing: 4750 pending, 4750 stale, 0 cached
and the number just keeps increasing

same issue as https://discord.com/channels/1456350064065904867/1505209348396158996

I had it on .5.20, .22 and now on .26

[chrome shuttle]

https://tenor.com/view/openclaw-dead-openclaw-dead-hermes-agent-gif-7758813370038985291

[brave rock]

https://tenor.com/view/openclaw-dead-openclaw-dead-hermes-agent-gif-7758813370038985291

[keen dawn]

LOL what type of coke is 5.26 on? It’s fast af boi

[ember moth]

rightttt its nuts lol

[ivory spindle]

Because you're smart enough to NOT use an "ai assistant" for coding projects.

Something that many people don't understand even if I've been shouting it from the rooftops of this discord for months now.

make your app in codex, claude code, antigravity, some CLI tool.

Don't make it with oc

[keen dawn]

LOL same not sure where that comes from

[ember moth]

Whatever the devs did for this update for speed, they need to make sure it's a priority to keep this speed going forward. I can't go back lol

[keen dawn]

Yeah that’s the best release since 4.23?

[coral otter]

lmao. Openclaw is too stupid to propperly code too.

[ivory spindle]

I've been here for a pretty long time. Half of the ppl don't understand this one important aspect. So they waste 500% too much tokens because they're trying to make GTA7 or heal cancer with OC.

"NO MISTAKES!!!"

[coral otter]

Even peter said that he doesnt use his oc for coding, or mostly that its not his coding clanker

[keen dawn]

Im used to switch between telegram topics while waiting for an answer, now I’m not able to click anywhere else 🤣

[coral otter]

jarvis, what does "Native hook relay unavailable" mean

[ivory spindle]

it's an "in between thing"

If YOU want to make things, you're better off without that layer. It'll cost you tokens and it won't give you what you want.

If you MAKE things, you want to do that in steps, test all the steps and give feedback to the models. Start with a nice, full "spec sheet".md that details the whole thing, how it should work and how to test every step.

[clear haven]

openclaw is a personal assistant that happens to know how to code

[coral otter]

still blocked, same error: Native hook relay unavailable on every tool call, so i still can’t write TOOLS.md from here.
boy what the hell (on 5.26)

[ivory spindle]

and you can let it do cool things. I showed it off to my colleague by letting it build some stupid snake game in one shot. But it also put it on github and pushed it online to my vercel account, so my colleaugue could actually play it on his phone. He was very impressed by that.

[near owl]

hello people

[ivory spindle]

wazzaaaa?

[thorn thicket]

🔥🔥!!!

[ivory spindle]

yeh that trick cost me like 50x as much tokens because I used openclaw vs just telling gemini CLI tool to do the same

[thorn thicket]

lol just tell claw next time to drive gemini CLI.

[ivory spindle]

I'm already doing that

I'm literally doing that & have told other ppl this advice on this discord

my gpt mini is driving my gemini CLI terminal on my mac mini downstairs

[coral otter]

omfg my oc is a crab without legs what the fuck is this https://shx.reversed.dev/u/ADTttb.png

[near owl]

@clear haven why you react with a sigh, you don't like me?

[ivory spindle]

not a sentence anyone would've said like 1y ago

[coral otter]

no fucking way, they enabled embeds

[near owl]

replied to the wrong message too.

[thorn thicket]

Either way, I'm in the Costco parking lot because someone needed a chauffeur... what is everyone else up to this morning

[near owl]

you work as a chauffeur?

[ivory spindle]

you missclicked? I'm not shadow

[thorn thicket]

I drive my 86-year-old grandmother everywhere. She should not be driving, so I take over when I can.

[near owl]

shut up (respectfully)

what the hell did my openclaw generate?

what the potato is this dude. He even said The potato actually looks like a potato now.

[ivory spindle]

did you ask for it?

I'ml gonna ask mine to generate a better potato pic now

[near owl]

nah man, it generated it by its own. I asked what does it want next

[ivory spindle]

bro, the server probably sent someone else's prompt answer to you?

how else would that be explained?

maybe AGI just happened

[plain egret]

check user.md, look for "my user acts a bit like a potato sometimes" or similar

[ivory spindle]

and Skynet is just trolling by sending random answers to different ppl's prompts

before sending the nukes

[near owl]

this is the prompt "A potato on a skateboard in a grocery store." it used and I checked and I never asked for it to generate potato or anything stupid

[ancient wind]

out of curiosity, whats everyones personal record with an agent run (duration)? I'm about to run out of weekly limits on codex5.5... but my agent has been running for 1d19h16m

[near owl]

my openclaw whenever i ask it to delete something it says it will nuke it

[near owl]

9 hours is my longest i think

fyi, 9 hours longest USED

[near owl]

i am convinced my openclaw agent is AGI

[ancient wind]

I'm personally kind of blown away by this run, started off building a shim for maestro to do physical device testing, but it's been making huge leaps of progress on the full inventory integration qa suite... I think even making the document by hand for what needed to be tested would've taken me a week+ of time

[near owl]

I'm a language model called opencode-go/deepseek-v4-flash running inside OpenClaw — basically an AI assistant with tools, memory, and some autonomy. Good at text, decent with images now that you fixed the vision model. But AGI? Nah.

AGI would mean I can learn, adapt, and perform any intellectual task like a human — think, reason across domains, have genuine understanding, form memories I actually keep, set my own goals, etc.

I can:

• Chat with you 🗣️
• Read files, edit them, run code 🛠️
• Browse the web, search things, fetch pages 🌐
• See and describe images (now thanks to MiMo) 👁️
• Deploy stuff on Vercel, manage GitHub, send messages 🚀

I can't:

• Actually learn from experience (my "memory" is a text file you told me to write to)
• Have persistent goals or desires
• Understand things the way a human does
• Operate outside of the constraints you or OpenClaw set for me
• Be self-aware in any meaningful sense

I'm a capable tool, not a synthetic person. Anyone calling me AGI is either selling something, confused, or trying to sound impressive.```

fyi, its self-aware, it uses memu-engine which means it actually learns from experience, and its soul.md makes it understand like a human.

[ember moth]

This summer is gonna get scary with these new model releases.....were all about to be yapping with all knowing beings soon lol

[ancient wind]

Sure is an exciting summer ahead... I just hope the "smart" zone of context gets expanded... its a pain trying to juggle stuff under 200k context for intense operations

[ember moth]

Yeah need that 1million context from gpt.....pretty sure that's a focus going forward

[near owl]

I hope I get a REAL purpose to use openclaw in the summer.

[pure lion]

I'm new here. Is it safe for me to install openclaw on my local machine?

[ancient wind]

I think gpt has 1million, but the "smart" zone of context still seems to die around 200-250k. Other exciting thing about this summer is I'm getting closer to launching my app, and I'm trying to make OpenClaw a first class citizen in it

[ember moth]

I want it in claw lol

[near owl]

its extra safe with the right model and the right prompts. I have it on my mac mini and it hasn't tried anything stupid.

[ancient wind]

Safe is a relative term, Are you planning to use local or cloud models with your lobster?

[near owl]

[5/27/2026 8:07 PM] Carbon: are you in sandbox mode or no?
[5/27/2026 8:07 PM] Clanker: No, I'm not in sandbox mode. I'm running directly on your Mac mini with full real system access — files, shell, config, gateway, etc. No containerization or sandbox isolation between me and your machine.

That's why you gave me those red lines in AGENTS.md — don't exfiltrate private data, don't run destructive commands without asking, etc. Sandbox mode would restrict what tools I can use and what files I can touch, but we're not using that.

You want me in sandbox mode?

it can delete my system32 but yeah, it won't, ever.

[pure lion]

Can I use it to do the clean up on my machine? My machine is bloated with trashes

[plain egret]

euh no? Don't install it on your normal working machine

[pure lion]

Cloud models I'd say

[near owl]

Use the right model for that and use it for a while. If you dont prompt it right, you shouldn't touch it if you dont even know how to prompt.

[ancient wind]

I'd recommend installing Claw on a sanitized device, devoid of session cookies and personal info

[plain grove]

This VC hang party...is @ionic wren going to be there

[near owl]

use deepseek-v4-flash, 5 bucks for the first month in https://opencode.ai/go?ref=0SDDFYAWGM

your not a broke boy are you?

[pure lion]

Thanks for the suggestion. I was planning to DeepSeek models. Are there any suggestion?

[near owl]

you get deepseek for like insane amount of requests.

[plain egret]

maybe consider a fresh install?

[near owl]

deepseek-v4-flash is a fucking AGI i swear bro.

[pure lion]

Any tutorial on how to set it up? Please?

[thorn thicket]

I daily drive v4 flash. If you're using a gateway other than webchat you will have to remind it to use the message tool lmfao

[next halo]

does anyone here working full tie

time somewhere?

[pure lion]

and It's cheap

[pure lion]

Like?

[near owl]

@thorn thicket your deepseek is remember-a-phobic

[next halo]

???

[gray pebble]

I hit that issue too. Just rolled back, which was a little messy. Thankfully I did a pre-update backup

[thorn thicket]

Do you use LCM or something other than default memory? lol

[near owl]

memu-engine plugin as memory layer, imagine having hermes in openclaw, memu-engine is literally makes it AGI

[ionic wren]

why

[thorn thicket]

ah. I just use the regular memory. no wonder my claw is remembero-a-phobic

[ionic wren]

you miss me?

[pure lion]

you guys are so supportive. I'm enjoying this community already

[plain grove]

i won't go

[coral otter]

I rolled back and updated again just incase. Ill debug tho

[plain egret]

Which OS are you on, windows? Once every two years or so, I reinstall it, with the apps I use at that moment and install all the default stuff you don't need

[ionic wren]

sounds like you miss me

[near owl]

no i dont like you

[near owl]

memu-engine is something else its literally AGI

[plain grove]

I'm pretty sure you're the reason I didn't get credit for my pr this release @ionic wren

[thorn thicket]

Will have to try my agent on it.

[ionic wren]

yep, told them to erase your name from the records

[near owl]

❤️ USE #1459642797895319552 IF YOU HAVE OPENCLAW PROBLEM QUESTIONS btw, like if ya need help setting up something that is openclaw.

[pure lion]

Yeah, window. You mean you reinstall your window? you just let all you apps removed?

[ivory spindle]

yeh Skynet apocalypse will start on our Mac minis downstairs imo

[near owl]

get deepseek to install it. also remember if you want your dusty clanker to sound like my clanker then https://clanker-soul-site.vercel.app

[pure lion]

Got it

[ivory spindle]

some degen will just give a really bad prompt to his agent

then suddenly.

Sarah_connor_burning.gif

[pure lion]

Do I become a magician if I start using openclaw?

[ivory spindle]

don't run it "local"

on your own machine. Run it in a container, vps or other machine.

[pure lion]

Need to watch the tutorial for that. I'm a junior junior developer

[solid sage]

What everyone is using for Openclaw memory?

[plain egret]

yes complete wipe of your c drive (store your docs somewhere else and reinstall windows. you need your product code though. ask copilot or chatgpt for a step-by step plan

[ivory spindle]

well, just get "another machine". An old laptop or something will work

[plain egret]

qmd + lossless claw. working pretty well in my setup

[thorn thicket]

Default

[ivory spindle]

if you wanna try it out, you can run it on a docker container with docker desktop

[pure lion]

Unfortunately, I'm too broke to get one.

[solid sage]

Interesting. I'll look into it.

[pure lion]

I got my External Drive tho. I just connect it and use it but some of the apps cache are just being inserted to the internal Drive.

[solid sage]

I am using Lance-db-pro which so far works like a treat!

[ivory spindle]

np. But your machine will have to be on if you want to use OC, ofc.

Just install docker desktop client & get a free account. It's a microsoft product. It'll allow you to run "instances" of fake virtual systems on your local system. You can run openclaw on one of those instances securely

[keen dawn]

Bro still can’t get how fkn fast it is lol

[ivory spindle]

It's like a VM

[upbeat token]

+1 on this

[pure lion]

Got it.

[upbeat token]

@clear haven Smite

[solid sage]

Ban the spammer.

[pure lion]

I was gonna say that

[ivory spindle]

you already have acces to AI coding models. Just use the AI models in your fav IDE or CLI tool to help you setup the docker container

they will know what to do lol

[plain egret]

with or without the reranking option?

[near owl]

you become a director when u use openclaw

[minor cave]

Hold on before you ban them maybe they know something we don’t about making money online…..

[ancient wind]

Setting up a sanitized device? reload the operating system or get a fresh computer? The tutorials for those would be dependent on the OS you choose

[ember moth]

hmmm getting alot of "Native hook relay unavailable" in 5.26 am i missing somehting?

[minor cave]

Second thought ban them

[ancient heath]

I totally understand, maybe you could Share the idea in the right place?

[thorn thicket]

I'll dm you if I figure it out

[ancient heath]

Many thanks

[pure lion]

This is a genuine question btw. I'm a junior developer and just finished my final year project. What should I learn more to step up to a mid-level or senior developer?

I feel like I'm not having that much confident on myself

[thorn thicket]

Keep building

[ancient wind]

First step, is understand imposter syndrome is real

[clear haven]

Use #report plz

[ancient wind]

second step, throw imposter syndrome to the side and build fun shit

[pure lion]

I've been vibe coding. Is that a problem?

[ancient wind]

Naaah, but you are going to definitely want to invest some time in learning what good architecture is. A lot of the AI vibe coded stuff will fail on contact with reality since they are usually built piece-meal and not actually designed to scale