users-helping-users

[shy ether]

"id": "agent-llama:latest",

[hollow shoal]

aah

yes

it does

[shy ether]

Remove the :latest bit

[hollow shoal]

just trying

[carmine willow]

Hey! Running v2026.2.22, trying to use sessions_spawn as a proper orchestrator (main agent delegates task → subagent runs silently → main receives result → main replies to user).

After enabling tools.agentToAgent and tools.sessions.visibility: tree, the subagent does complete and the JSON result arrives — but it's delivered as a Discord channel message instead of an internal event. Because of that, the main agent sees the announce but treats it as a bot message (no human waiting) and fires NO_REPLY, never following up to the user.

The flow we want:

User asks → Main delegates via sessions_spawn → Subagent runs → Result returns internally to Main → Main processes and replies to user

The flow we're getting:

User asks → Main delegates → Subagent posts result directly to Discord channel → Main sees it but goes silent

Is there a way to make the subagent announce fire as an internal session event (triggering a new Main turn) rather than a channel message? Or any config/flag we're missing?

[storm whale]

Hey, does the webchat messages in the Openclaw Gataway dashboard trigger the message:received hook?
Anyone knows

[shy ether]

yay/nay?

[daring wyvern]

i have tried everything to get my bot to stop giving me "reasoning" after every message but he just wont. any solutions?

[shy ether]

have you tried /reasoning off ?

[jagged nebula]

where does npm install openclaw to?

[last thistle]

hi can somebody help ,e?

me*

[shy ether]

Should be ~/.npm-global/bin/openclaw

[royal fog]

dangerouslyDisableDeviceAuth always gets set back to true on docker restart. I'm assuming this is not by design. Or is it?

[shy ether]

maybe? What is the issue?

[jagged nebula]

does ~ work in a docker container?

[shy ether]

No clue, sorry.

[last thistle]

when i select whatsapp it says this: +---------------------------------+
|
o Select a channel
| WhatsApp (QR link)
|
o Channel setup ------------------+
| |
| whatsapp plugin not available. |
| |
+----------------------------------+
|

• Select a channel
  | > Telegram (Bot API) (not configured)
  | WhatsApp (QR link)
  | Discord (Bot API)
  | IRC (Server + Nick)
  | Google Chat (Chat API)
  | Slack (Socket Mode)
  | Signal (signal-cli)
  | iMessage (imsg)
  | Feishu/Lark (飞书)
  | Nostr (NIP-04 DMs)
  | Microsoft Teams (Bot Framework)
  | Mattermost (plugin)
  | Nextcloud Talk (self-hosted)
  | Matrix (plugin)
  | BlueBubbles (macOS app)
  | LINE (Messaging API)
  | Zalo (Bot API)
  | Zalo (Personal Account)
  | Synology Chat (Webhook)
  | Tlon (Urbit)
  | Finished

[faint forge]

How to solve issue in terminal, when I try to execute "ollama launch openclaw"
Config invalid
File: ~/.openclaw/openclaw.json
Problem:

• <root>: Missing env var "OPENCLAW_GATEWAY_TOKEN" referenced at config path: gateway.auth.token

[shy ether]

Does your openclaw.json contain a "token" value anywhere?

[shy ether]

what OS are you running this on? How did you install openclaw?

[last thistle]

npm?

[faint forge]

Yeah, " "gateway": {
"auth": {
"mode": "token",
"token": "${OPENCLAW_GATEWAY_TOKEN}"
},
"controlUi": {
"allowInsecureAuth": true
},
"mode": "local"
},"

[shy ether]

what npm command? and, what OS?

[shy ether]

Then you need to set an environment variable to contain that value.

[gaunt rover]

does anyone know where to download the macos app

[last thistle]

Install OpenClaw

npm i -g openclaw

Meet your lobster

openclaw onboard

ths

[shy ether]

Try this instead:
curl -fsSL https://openclaw.ai/install.sh | bash

[shy ether]

What app? OpenClaw? Or something else?

https://docs.openclaw.ai/install

[gaunt rover]

https://docs.openclaw.ai/platforms/macos

[faint forge]

Ow, thanks. But how do to that? Maybe it's possible to wipe to default? 😄

[wicked birch]

can someone help me debug my setup? it's gotten real slow (i'm on ubuntu vm in gcp) and the help discor here stopped working

[haughty hedge]

Anybody got their bot buying for them? Mine explicitly says it wont do it. Despite me telling it off...

[shy ether]

The gateway section should have been set up as part of your onboarding process, after installing.

You may need to re-install the gateway openclaw configure

[last thistle]

same

[shy ether]

what OS?

[last thistle]

os like windows?

[shy ether]

Yeah.

[last thistle]

windows 11

[shy ether]

Windows, Linux (what distro), MacOS

[shy ether]

Are you using Windows Subsystem for Linux? WSL?

[last thistle]

what is that?

[shy ether]

Theres your problem.

In Powershell, run this:
iwr -useb https://openclaw.ai/install.ps1 | iex

src: https://docs.openclaw.ai/install#windows-powershell

[kindred swallow]

Why its showing whatsapp plugin not available?

[shy ether]

In reference to Noct's issue? Or is it happening to you too?

[last thistle]

i think is the same

[shy ether]

You may need to clean up the npm install you did, first.

[kindred swallow]

I think is with new release

[kindred swallow]

Reinstalled after cleanup
Still same

[shy ether]

Hmm, well, I've no idea about how it works in Windows. Never done it.

I assumed it ran in WSL

[serene trench]

Is codex just weird with OpenClaw in general? 😂

[shy ether]

Everything is weird with OpenClaw.

[serene trench]

HAha fair enough!

Love being on the bleeding edge

[shy ether]

Yeah, I've burned four days trying to get openclaw to work, only to find out you need a fucking supercomputer to get it to work locally.

[last thistle]

ok i did the migration now what?

[shy ether]

I dunno. I imagined it'd just start working for you.

[glass marsh]

Good morning everyone

I am new here

[last thistle]

whatsapp plugin not available. 😭

[shy ether]

RIP

[last thistle]

how do i uninstall and reinstall all?

[tame hornet]

Seems like all the plugins have broken in the .22(-2) update. My slack integration is broken and I can get it working, other messages about telegram and whatsapp. If there is any workaroung please tag me, other wise I'll just wait....

[shy ether]

Roll back to an older version?

[last thistle]

not bad idea

[severe ermine]

is it normal for openclaw to be this dumb? https://imgur.com/a/CdmQC5x really told it three times to make sure it would schedule a message update

no idea if I am missing something here or if cron jobs are really the way

for context, I told it to write a python script to scrape some data and it might take sometime to run so told it to schedule updates every five minutes on how its going

[glass marsh]

I cant even install this
I try copying the install command into my terminal on my new mac mini, and it says Homebrew Not Found, Installing
then it says Installing Homebrew failed. re-run with --verbose with details

[shy ether]

What model?

[severe ermine]

gpt 5.3 codex

[shy ether]

Have you configured any tools? or have you got it set to allow all?

[rustic arch]

In which chat I can shill my project that I built with Clawed?

[glass marsh]

Says it needs Sudo access and i need to be an administrator, but when i go to User Settings my profile is an Admin...

[shy ether]

#showcase-old

[severe ermine]

i've no idea, just did the setup now and connected to discord, nothing else

[rustic arch]

but I did it and I got a warning

[shy ether]

Then no where, I guess?

[rustic arch]

haha oh well

[shy ether]

#self-promotion actually.

It's in the rules. lmao

[severe ermine]

any place on where I can see it if I allowed all?

[shy ether]

Does your openclaw.json have the word "tools" in it anywhere?

[severe ermine]

it looks like that the cron jobs don't even send the messages 😭

"No tool call found for function call output with call_id call_63kkmnUFMNmHVs2KAjNoxaQf."

[severe ermine]

will check

[severe ermine]

nope, no reference of tools

[patent bear]

Hello I would like some help trying to set up openclaw To be 24/7 where I can give it a task and it doesn't stop until it completes it Right now it feels like after a few minutes it doesn't continue having you periodically check into just posting an update ideal trying to figure out how to do this I feel like I'm missing something

[shy ether]

RIP, I dunno then.

[gusty dragon]

Who else is using MacOS? I tried installing both from terminal and the app, but kept having issue to connect to WhatsApp.

[shy ether]

Seems like the 22nd's release has broken plugins

[gusty dragon]

Ah I see

[shy ether]

https://github.com/openclaw/openclaw/issues?q=is%3Aissue state%3Aopen whatsapp

Feel free to throw another issue on the pile.

[cunning coyote]

Update still broken?

[gusty dragon]

Lol seems like there were a lot of issues with WhatsApp already

[shy ether]

and telegram, and all of the other plugins I think

npm install -g openclaw@2026.2.19

^ This version should work

[severe ermine]

oof, are cron jobs just broken?

[shy ether]

maybe? Any issues here resonate with you? https://github.com/openclaw/openclaw/issues?q=is%3Aissue state%3Aopen cron

[severe ermine]

they're being triggered, but nothing happens (no message is sent)
https://imgur.com/a/pv2z1jT

[latent hedge]

YIKES... Just ran upgrade to latest version.. now my Ubuntu won't even recognize openclaw anymore.. wtf

melsen@melsenx:~$ openclaw tui
openclaw: command not found
melsen@melsenx:~$

[shy ether]

The latest version is buggy AF.

[shy ether]

export PATH="/home/openclaw/.npm-global/bin:$PATH"

(substitute your own username)

[severe ermine]

anyway I can downgrade perhaps? that'll prob cause more issues than solve

[faint forge]

Is anyone knows why after every command I see that " gateway disconnected: closed | idle
agent main | session main (openclaw-tui) | ollama/qwen3.5:cloud | tokens 3.3k/131k (2%) "
Like I don't see the output in terminal or something like that. So I asked to help setup telegram connection it shows me that and that's all 😄
Is that normal behaviour?

[tardy pollen]

I have a question - is there any marketplace for agents which I can plug & play? I know that there is a marketplace of skills (skills.sh) - is there similar marketing for agents? which give you to option to plug&play agents that people already created?

[sick dove]

was that on pnpm install ?

[spare fulcrum]

Is this a normal issue and how do I fix it?: OpenClaw Installer

[OK] Windows detected

[OK] Node.js v24.13.1 found

[*] Installing OpenClaw (openclaw@latest)...

iex: Die Datei "C:\Program Files\nodejs\npm.ps1" kann nicht geladen werden, da die Ausführung von Skripts auf diesem System deaktiviert ist. Weitere Informationen finden Sie unter "about_Execution_Policies"

( https:/go.microsoft.com/fwlink/?LinkID=135170).

In Zeile:1 Zeichen: 45

• iwr -useb https://openclaw.ai/install.ps1 | iex

222

• CategoryInfo

: Sicherheitsfehler: (:) [Invoke-Expression], PSSecurityException

• FullyQualifiedErrorId: UnauthorizedAccess, Microsoft.PowerShell.Commands.InvokeExpressionCommand

[gusty dragon]

Starting fresh rn... Managed to configure for WhatsApp (seemingly at least) and now waiting to complete installation of other plugins.

[sick dove]

probably /usage flag set to on

[tardy pollen]

Asking again:
I have a question - is there any marketplace for agents which I can plug & play? I know that there is a marketplace of skills (skills.sh) - is there similar marketing for agents? which give you to option to plug&play agents that people already created?

[faint forge]

How to check that?

[late nebula]

does snibody use 🦞 OpenClaw 2026.2.22-2 (45febec)

[severe ermine]

fuck it, i'll just reinstall

[latent hedge]

I just asked my bot in telegram to update itself to latest version.. worked the other times.

This time it just destroyed the installation

[nova peak]

Already did that and it didn't help

[severe ermine]

on openclaw@2026.2.19 though

[tardy pollen]

Someone?

[faint forge]

Om, I got "
send failed: Error: gateway not connected
disconnected | error
agent main | session main (openclaw-tui) | ollama/qwen3.5:cloud | tokens 3.3k/131k (2%) "

[nova peak]

Ah ok. I came here looking for the solution to the same problem as you. How do I install an ealier version?

[severe ermine]

no idea yet, will reinstall my VM and search in the meantime

[shy ether]

npm install -g openclaw@2026.2.19

[severe ermine]

prob there's a install script into github that I can copy

[severe ermine]

thats the same thing that the install script does?

[solid walrus]

│
◇ Select a channel
│ WhatsApp (QR link)
│
◇ Channel setup ──────────────────╮
│ │
│ whatsapp plugin not available. │
│ │
├──────────────────────────────────╯
│
◇ Select a channel
│ Telegram (Bot API)
│
◇ Channel setup ──────────────────╮
│ │
│ telegram plugin not available. │
│ │
├──────────────────────────────────╯
│
◇ Select a channel
│ Discord (Bot API)
│
◇ Channel setup ─────────────────╮
│ │
│ discord plugin not available. │
│ │
├─────────────────────────────────╯
│
◇ Select a channel
│ Finished
Error: Config validation failed: channels.whatsapp: Unrecognized key: "enabled"
guys why is this happening can someone herlp me

[trail notch]

I tried to get a fresh install but not telegram plugin is missing

[solid walrus]

"lastRunMode": "local"
},
"agents": {
"defaults": {
"compaction": {
"mode": "safeguard"
},
"maxConcurrent": 4,
"subagents": {
"maxConcurrent": 8
}
}
},
"messages": {
"ackReactionScope": "group-mentions"
},
"commands": {
"native": "auto",
"nativeSkills": "auto",
"restart": true,
"ownerDisplay": "raw"
},
"gateway": {
"auth": {
"mode": "token",
token
}
},
"meta": {
"lastTouchedVersion": "2026.2.22-2",
"lastTouchedAt": "2026-02-23T16:12:39.869Z"
}
}

this is how the config looks like right now

[nova peak]

Yes, my problem right now on a fresh install

[shy ether]

Latest openclaw version has broken plugins. You need to downgrade

[solid walrus]

oh

[trail notch]

How

[solid walrus]

how do i downgrade

i mean which command

[gusty dragon]

I had the same problem - just reinstall it with an older version

[spare fulcrum]

OpenClaw Installer

[OK] Windows detected

[OK] Node.js v24.13.1 found

[*] Installing OpenClaw (openclaw@latest)...

iex: Die Datei "C:\Program Files\nodejs\npm.ps1" kann nicht geladen werden, da die Ausführung von Skripts auf diesem System deaktiviert ist. Weitere Informationen finden Sie unter "about_Execution_Policies"

( https:/go.microsoft.com/fwlink/?LinkID=135170).

In Zeile:1 Zeichen: 45

• iwr -useb https://openclaw.ai/install.ps1 | iex

222

• CategoryInfo

: Sicherheitsfehler: (:) [Invoke-Expression], PSSecurityException

• FullyQualifiedErrorId: UnauthorizedAccess, Microsoft.PowerShell.Commands.InvokeExpressionCommand how do I fix this?

[shy ether]

Downgrade command: npm install -g openclaw@2026.2.19

[gusty dragon]

I uninstalled it completely then reinstalled lol

[severe ermine]

uhmmm looks like I can specify a --version on the script

will try it

[solid walrus]

i c ..what command did you use to reinstall it?

[gusty dragon]

https://docs.openclaw.ai/install/uninstall

[drifting rune]

Hello everyone! Can someone help me to install OpenClaw on my Macmini via Zoom? We can discuss all the details in DM, I am soooo far from the technologies :U

[shy ether]

sus

[drifting rune]

No sus, just exploring the new thing. My boss asked me to find a person who can help us with it. But yeah, sounds sus. Sorry

[shy ether]

Don't use openclaw for anything work related.

[solid walrus]

thx man

[drifting rune]

Oh, because it is very unsecure&

[shy ether]

extremely.

[drifting rune]

I was actually about to, haha. Alright, thank you!

[shy ether]

No worries.

[solid walrus]

curl -fsSL https://openclaw.ai/install.sh | bash this is the correct install command right?

[severe ermine]

for ref, we'll use for work but it runs locally on my homelab and has very limited access to what it can do.
mainly for reminders and creating analytics stuff via python

[shy ether]

Meowth, that's right!

[solid walrus]

nvm ..it says failed to connect to server

[shy ether]

RIP

[solid walrus]

shubh@ubuntu-neo:~$ curl -fsSL https://openclaw.ai/install.sh | bash
curl: (7) Failed to connect to openclaw.ai port 443 after 52 ms: Couldn't connect to server

wait nvm ..

[shy ether]

Can you ping google?

[solid walrus]

dns issue

i fixed it

[shy ether]

Always

[drifting rune]

We were planning to do it on a new Mac, create a new user, do not connect it to anything, use HITL etc. So it has very limited access

[shy ether]

How beefy of a homelab do you have? You need a massive amount of VRAM to make this thing work.

[severe ermine]

yeah should be fine IMO

[shy ether]

You'll spend 8000 dollars on a mac just for this, yet have no knowledge on how it works?

[severe ermine]

no? we'll just run python scripts

matplotlib and stuff, no GPU required

[shy ether]

What runs the AI stuff?

[drifting rune]

MacMini for 600, yesssss. My boss did tho.

[severe ermine]

codex via openai

[shy ether]

So it's not running on your homelab then, really.

[severe ermine]

well the openclaw is, I meant that im not running on a VPS where everyone can reach

and either way, more secure than running an open source model, all of them sucks in terms of security (in case someone manages to do something)

[shy ether]

I guess so. If you're happy with a megacorp siphoning your data for their own benefit.

[severe ermine]

all of your code is written by AI either way (with code reviews and humans in the loop clearly), unfortunately thats the price to pay for speed

[shy ether]

If that's how you're using it, then yeah.

[severe ermine]

and I know its not worth anything but they say that they won't do anything with our data (after all they're an US company), but yeah its just trust here and I kinda trust openai

[sick dove]

I was thinking about that while updating to a latest version. I was thinking of creating a rollback mechanism in case something goes wrong. Otherwise best option is to sign in with claude code on VM and give it context and --dangerously-skip-permisions for it to fix it.

[silver cradle]

any way past the broken approvals system in the latest release?

[sick dove]

just run /usage on your telegram . It says usage footer or cost summary and you then select

[little vortex]

^ bump

[west pollen]

Does anyone know whether or not @sharp parcel is still responding to Help topics?

[distant star]

Hey guys, would love to be able to get OpenClaw to be able to make suggestions in Google Docs (as opposed to editing directly). Is there any way I can do this? Google's API doesn't seem to enable that ability, and Google banned the first account I tried to create for my OpenClaw to use

[solid walrus]

BROO

I installed the same version again

curl -fsSL https://openclaw.ai/install.sh | bash this also installs the latest version bruh

whats the correct command lol

@shy ethercan you help pls

[shy ether]

Did you try npm install -g openclaw@2026.2.19

[nova peak]

Trying a fresh install and I get a bunch of warn deprecated messages

[shy ether]

Just ignore those

[nova peak]

Ok, how do I get to the wizard after running the install command?

[shy ether]

openclaw onboard

[earnest lion]

Hey, I am installing openclaw on windows 10, I installed Node.JS 25.6.1 and git but it keeps throwing an error.... npm error code 1, at line:1 char:1, + & "C:\Program Files\nodejs/node.exe" "C:\Program Files \nodejs/node_mo...

[shy ether]

Have you tried using this script? iwr -useb https://openclaw.ai/install.ps1 | iex

[mortal talon]

Sorry to interject, I have been struggling to get my Telegram working. Im trying to load openclaw onto my Windows 11 system. I used a bit of advice from the help community and that allowed the plugins to finally be installed but now Im getting the following errors:

17:33:24 [telegram] telegram deleteMyCommands failed: Network request for 'deleteMyCommands' failed!
17:33:24 [telegram] telegram deleteWebhook failed: Network request for 'deleteWebhook' failed!
17:33:24 [telegram] webhook cleanup failed: Network request for 'deleteWebhook' failed!; retrying in 2.34s.
17:33:24 [telegram] telegram setMyCommands failed: Network request for 'setMyCommands' failed!
17:33:24 [telegram] command sync failed: HttpError: Network request for 'setMyCommands' failed!

Any help or guidance would be greatly appreciated. Thanks

[shy ether]

Could be related to plugins being broken in the latest version

[earnest lion]

Yeah thats the exact script im using

[shy ether]

RIP

[mortal talon]

Thanks. Any ideas if there is a workaround?

[shy ether]

Downgrade to 2026-02-19

[muted shuttle]

Anyone using signal with remote signal-cli-rpc is having issues with latest version? 2026.2.22-2

17:33:08 [signal] [default] channel exited: signal daemon exited (source=spawn-error code=null signal=null)

Downgrading to 2026.2.21 solves the issue

[shy ether]

Check the github issues

[mortal talon]

Probably sound like such a noob, how do i get the older version?

[shy ether]

npm install -g openclaw@2026.2.19

[muted shuttle]

I found this

https://github.com/openclaw/openclaw/issues/24490

but unsure

[mortal talon]

Thank you so much. Will try it now

[golden siren]

Hey guys. I added another agent and added it to discord. But my main went offline when the new one got added. How do i solve this? 😅

[solid walrus]

nvm

[kindred swallow]

Try installing npm install -g openclaw@2026.2.21

[quick grotto]

do you guys use brave search for your clawdbots? i just set mine up on a vps and gemini is saying i need to give lobster bro internet access

[astral kelp]

I keep getting HTTP 401 authentication_error: Invalid bearer token (request_id: req_011CYRUgdnUqcdWqvWtYUBb9) on a formerly working bot. I reset the whole thing, even the workspace, and I’m still getting the issue. Anyone run into this?

[boreal holly]

Thought I'd ask here - having an issue I've never had before. Not specific to Openclaw - but it's installed on the device. Short version is my raspberry pi which is locationed at 192.168.1.19 locally isn't accessible via that ip anymore - it was a few days ago. I have to go through tailscale now to ssh in. Any thoughts Claude and ChatGpt have failed to identify the issue.

[quasi apex]

Are plugins not working? I keep trying a fresh install/setup and once I get to choose a channel step, no matter which one I choose (Discord/Telegram) before I can even enter my key it just immediately says "plugin not available" and skips to "configure skills now"

[kindred swallow]

try Installing previous version

[kindred swallow]

npm install -g openclaw@2026.2.19-2

[scarlet olive]

I think the new version is crashing upon config.patch updates. Hotreload is not working. I see this message: Gateway restart config-patch ok (config.patch). Any idea how to fix/debug?

[wary epoch]

can i pay somone to help me fix my bot? i cant connect anymore to the gatway or TUI or webchat. i tried a reinstall even

get token mismatch in the terminal?

i dint think i even did a proper wipe/re-install

[burnt frigate]

Try doing update, it should fix most things, then tell bot to fix itself… I had the same issue, and this helped. I have asked it not to restart itself, and it has done better at not nuking itself

[quick grotto]

anyone?

[tired ferry]

Channel Link to WhatsApp is not working in the current version right?

[amber island]

Hey, I've been having issues with my memory search function. Openclaw is having serious problems because the openai embedded model keeps returning a 400 error, indicating that it's out of available tokens.

I'm trying to change it to use a local ollama model, 'nomic-embed-text', but cannot seem to change the openclaw.json to accept it without erroring out with a config errro when restarting the gateway.

Does anyone know the actual steps to change the memory search embedded model?

[weary canyon]

Hey guys, just wanted to ask would the agent be able to run smoothly n stuff on a mac mini M1 or M2 chip?

[desert cloak]

I'm trying to get my Claw access to Signal. It seems to be configured, but when I run the TUI wizard, it says the plugin is disabled. Any ideas? I haven't been able to find an "architecture" section in the docs to get my head wrapped around the data model yet 🙁

[green sluice]

wich browser relay should i use? for chrome

[wary wyvern]

has anyone got cron tool to wokr in docker sandbox? I'm getting sh: 1: openclaw: not found in docker

[ruby moon]

I am just starting out. I've installed OpenClaw on a Hostinger VPS with npm install -g openclaw@latest under a new username: openclaw in the /home/openclaw folder. I ran pm2 start openclaw-bot.json but the Telegram connection was not on-line. So I ran openclaw gateway --allow-unconfigured and I then could chat with my openclaw server, but openclaw statusl gives:

Gateway           │ remote · wss://100.104.198.117:18789 (config gateway.remote.url) · 
                    unreachable (connect failed: write EPROTO C0BCF5FC1D790000:
                    error:0A0000C6:SSL routines:tls_get_more_records:packet 
                    length too long:../deps/openssl/openssl/ssl/record/methods/tls_common.c:662:
                    )
│ Gateway service │ systemd installed · disabled · unknown (Error: systemctl --user unavailable: 
                    Failed to connect to bus: No medium found )
│ Node service    │ systemd not installed

What am I doing wrong? i.e. Why is the Gateway unreachable? Why is the Gateway systemd disabled? Why is the Node systemd (PM2) not installed?
A doc page or YouTube video on running the Gateway would be good.

Thanks, Robin

[weary canyon]

Okay, so I just finished a podcast with Chris Camillo and excited to do some stuff with the clawbot. But confused where to start, what mac mini to buy, how much will it all cost and where/what youtube channels to learn from

[left niche]

i wont let me install it im on windows can somebody help me?

[frosty coral]

I am making space on my server for openclaw, But does anyone know that room i need to give openclaw. I dont see how much file space it needs.

[latent hedge]

Excuse me... is it correct that the latest version changes name and isnt called 'openclaw' from the terminal anymore.. but clawdhub ??

and doesn't have 'tui' anymore?

I think ChatGPT is drunk anymore

giving me weirdass advice

[grim plover]

Channel config schema unavailable. im getting this error i cant pair telegram to my claw

[willow garnet]

any one setted up openclaw on vps? stuck on systemmd error

[desert cloak]

It seems like no matter what I do, I can't change it from "plugin disabled"

[latent hedge]

No.. wouldnt want that on a VPS.. I run it on a local dedicated environment

[true adder]

openclaw tui from the terminal

[willow garnet]

@latent hedge just trying it out first time dont want to reset my laptop till i get my raspberri pi 5 bucks for a test wont hurt much 😄

but for some reason stuck on that error

[ruby moon]

I believe you best bet is to run the docker-compose.yml in Docker Desktop.

[oak robin]

Describing an error I ran into, this is how my agent described the issue: after a context compaction, a synthetic message appeared in the agent's context (but not in your Telegram chat) referencing a nonexistent file WORKFLOW_AUTO.md and asking the agent to read it as part of a "post-compaction audit." The message body also contained fake inbound metadata formatted to look like a legitimate envelope header.

[ruby moon]

That is where I am having issues as well ... Without buying a Mac Mini, I believe the consensus is that a Hostinger VPS is the easiest, cheapest and most secure way to go.

[willow garnet]

Systemd user services are unavailable. Skipping lingering checks and service install

thast what im getting

[summer dome]

Hi! can i connect the heartbeat call with local ollama model and if anything needed which required big parameters LLM, then it will go to the paid LLM (like OpenAI with key)? what will the json file for this, can anyone help

[void bolt]

for me the same...

[sweet ledge]

I have a somewhat very dumb assumption: After successul initial onboard, local chat through TUI or Dashboard should work before you configure a chat app (discord/telegram/etc...), right?

[reef copper]

Hello everyone. I set up the openclaw and it works reasonably well but I am afraid to loose the memory and context
Does someone have t prompts or best practices to set up the memory files to ensure is not burning troughs tokens like crazy but also remembers everything ?

[desert cloak]

I think I just fixed it, by adding this to /home/openclaw/.openclaw/openclaw.json:

"plugins": {
  "entries": {
    "signal": {
      "enabled": true
    }
  }
}

[true light]

Hey all, I got a weird issu. A fresh install of openclaw with the latest release 22.2. All went well, install on max mini, etc. Using Grok API as default LLM and using TUI as basic module. Went to the webpage, worked well, approved to token there so get rid of the message etc etc.. BUT went to get some food, came back and suddently the chat on the website freezes, but the overview and any other menu works perfectly. However as soon as I click the Chat menu, it freezes completely. Anyone any idea what or where I can look for that ?

[zinc mica]

has any built a robust sentinel agent and could give me some guidance on where to begin there?

[grim plover]

you did?

[boreal sky]

I just setup my Chrome Browser Relay and it's connecting via Port & Gateway token but doesn't work and is only in idle ("...") mode when I click it

[tulip falcon]

What about this

npm install -g openclaw@2026.2.19-2

[hardy pagoda]

anyone else having issues that with docker the gateway container is restaring all the time?

[wheat pine]

Running Hostinger VPS and xAI, it seems the model isnt recognized by OpenClaw, anyone run into this?

[sweet ledge]

I got a thumbs up on this so I take that to mean yes, I should be able to chat with it locally after clean install. Problem is, I can't. Just testing with openrouter/openrouter/free model, my whole setup looks good, I can actually send/receive using node directly on the commandline with my API key, but in the OpenClaw chat I just get: "Connection error." That's it. There's no other details anywhere. I can't find anyone else seeing this problem. My device shows allowed, the dashboard is healthy. I have no idea what's up. Installed on a 2019 Macbook.

[tulip falcon]

For immediate fix npm install -g openclaw@2026.2.19-2 I suppose

Broken in update

Fixed but not live on npm yet

[gleaming pond]

Facing the same issue now. I've read that sometimes gateway restart helps but not in my case.

[tulip falcon]

So you gotta rollback

No cause that pulls latest I believe

[green sluice]

i have problem with chrome please help me Gateway-Pairing is somehow not working

[fringe frigate]

is "exposing thinking" now ON with the new update? or something was there on my conf before hand?

[boreal sky]

my relay is also not working

[green sluice]

It’s also killing himself sometimes

[boreal sky]

did you have any problems after downgrading? Config not working properly etc

[stone thicket]

hello guys is it actually really to install and manage the open claw i want to use it for advartising my products is this hard to make and how much i need to pay

[serene parcel]

Hello guys, actually trying to install openclaw on a docker container, someone can Help me ?

It actually worked on my terminal ( I’m on macOS ), but not on docker, I don’t have any idea how to start with that

[simple bone]

I have a macbook pro M1, 8 gb Ram, can clawbot run on this device?

[boreal sky]

yes

[fair belfry]

are we allowed to look for beta testers here or no?

[opaque steeple]

Yo anyone knows where we can find & share stuff for openclaw like dashboards, scripts, etc.? so not only skills but like maybe some code as well (not clawhub)?

[half latch]

Hey hey,

I hope someone has made a similar experience and has some insights:

Ive been playing around with openclaw for a while now and I have couple of ideas how it can help me. I tried going step by step and really creating a stable setup (if a thing is done, its done for good and it works). But for some reason, discord delivery is shaky (at best) or just plainly not working.
My cronjobs are finishing without error and they send to the dashboard chat of that respective session.
Their delivery is configured as follows:

"delivery": {
  "mode": "announce",
  "channel": "discord",
  "to": "channel:<MY CHANNEL ID>"
},

Other notable (and often pointed out reasons for failure):

"sessionTarget": "isolated",
"wakeMode": "now",
"payload": {
  "kind": "agentTurn",
  ...
}
```
The bot is able to respont to @metions in channels as well as to dms. 

Im just lost and do not know where to look/how to debug (obviously checked all logs, look [here](https://docs.openclaw.ai/automation/troubleshooting))

Version: v2026.2.19-2

Anyone has had similar problems and has an idea where to look next?

[fringe dust]

i need help i cant connect telegram

[thorny stratus]

can someone helpe me with this

`node.exe : npm error code ENOENT
At line:1 char:1

• & "C:\Program Files\nodejs/node.exe" "C:\Program Files\nodejs/node_mo ...

•   + CategoryInfo          : NotSpecified: (npm error code ENOENT:String) [], RemoteException
    + FullyQualifiedErrorId : NativeCommandError`

[umbral cobalt]

Did you setup a bot with @BotFather in telegram

[fringe dust]

the plugin does not exists

[umbral cobalt]

Plugin?

Botfather is reached like messaging a user on Telegram

[fringe dust]

when i try to link with telegram it says "telegram plugin not available"

[dusky adder]

Can this shi work with CapCut? And editing related stuff

[green sluice]

my bot has problems with the chatgpt ui?

[torn compass]

can someone help me understand the usecases of setting up openclaw in docker vs sandboxing?

like what is the point of running openclaw in docker? what can you actually do?

[boreal sky]

current version is bugged

[boreal sky]

npm install -g openclaw@2026.2.19-2

[agile oracle]

yes but why "gateway connected fail: pairing requierd?"

i don't undertstand

?

[boreal sky]

you need to restart the gateway and maybe fix it after downgrading

[earnest plume]

I keep getting this error, on fresh ubuntu desktop, on ubuntu server, on debian... everything I try, it just doesn't work... when running the installer: inappropriate ioctl for device
And the installer fails

[boreal sky]

yes and then doctor fix

it did for me

[fringe dust]

i have no idea on how to do anything of that ill just wait for the next update

[bold cosmos]

What is this and how to solve it? ”LLM request rejected: The long context beta is not yet available for this subscription.”

[boreal sky]

you running via OAuth?

[bold cosmos]

Yup

[boreal sky]

idk I just tried and it worked, I am a noob as well but just keep trying. I work with chatgpt on the site to fix stuff

[boreal sky]

Anthropic?

[half latch]

What I found when looking at journalctl was

[delivery-recovery] Found 10 pending delivery entries <E2><80><94> starting recovery
Feb 23 20:24:07 openclaw node[554216]: 2026-02-23T20:24:07.273Z [delivery-recovery] Recovery time budget exceeded <E2><80><94> 10 entries deferred to next restart
```

[bold cosmos]

Yes. And no fallback right now.

[boreal sky]

ah ok. I havent tried this combination yet but I heard Anthropic might ban you

I run Gemini 2.5 flash lite as main, it's pretty cheap and Google gave me $300 credit

[bold cosmos]

This is more about the 1m context I think.

[bold cosmos]

Google might ban you though. https://x.com/steipete/status/2025743825126273066?s=46

[crisp wing]

@steipete via Twitter

Peter Steinberger 🦞 (@steipete)

Pretty draconian from Google. Be careful out there if you use Antigravity. I guess I'll remove support.
︀︀
︀︀Even Anthropic pings me and is nice about issues. Google just... bans? news.ycombinator.com/item?id=47115805

**💬 566 🔁 332 ❤️ 5.5K 👁️ 1.31M **

FxTwitter · Feb 23, 2026, 1:25 AM

[static forum]

my openclaw fixed the broken telegram plugin on its own

[arctic pilot]

Has anyone experienced an issue with the new Version 2026.2.22-2 leaking reasoning content into iMessage? The thinking tokens are delivered first as a first iMessage, and then the respond. It was not happening with Version 2026.2.19-2. For the life of me I can figure out a config to supress it

Looking at some recent help posts, seems an issue with Telegram, too. MIght just be a wait for a new update situation...

[boreal sky]

happend to my telegram integration as well

[boreal sky]

oh thanks for the heads up. any other way to use Gemini 2.5 flash lite directly from google?

[arctic pilot]

Yeah, I just reverted to the old version and it went away. Guess I'll use that until it's fixed

[cunning coyote]

Update process fixed but telegram plugin not available anymore

[junior acorn]

Remove the beta line from config

[cunning coyote]

He admittedly disabled telegram and blockrun plugins from configuration 😆

[sinful linden]

So Telegram integration isn't working currently right? I'm trying to set up OpenClaw for the first time today... Which channel is easy to set up and use now?

[cunning coyote]

It is

Just my main agent disabled it on the config! 🤷

[copper meadow]

anyone had the issue on slack channels of context overflow when posting a file?! The file is stored in the memory of the session as binary and thus (i guess) is overflowing. I want the file to be index. Anyone got an idea?

[agent/embedded] [context-overflow-diag] sessionKey=agent:main:slack:channel:xxxxxxx provider=ironai/llm source=assistantError messages=5 sessionFile=/home/xxx/.openclaw/agents/main/sessions/xxxxxxxxxxxxxxxxxxxx.jsonl diagId=ovf-xxxxxxx-bbbbbbbbbb compactionAttempts=3 error=400 You passed 253953 input tokens and requested 8192 output tokens. However, the model's context length is only 262144 tokens, resulting in a maximum input length of 253952 tokens. Please reduce the ...

[meager night]

is there a quick and easy to change the routing so all messages that i'd normally see in the webUI go to discord?

thinking it's something like agent:main:main but i'm unsure

[velvet terrace]

Guys I really need some help. Which I know everyone here does but lol. If you happen to read this plz respond. I've updated my clawdbot and i keep getting permission approvals in the U.i. for like everything, i've updated it to not ask on all profiles and its still doing it. I've asked GPT as well, explored various avenues trying to find a solution and still it persists. Please, if you have any idea on how to fix it communicate it even if you aren't sure.

[foggy zealot]

Has any one experienced the issue that no channel can be configured during set up using the 2026.2.22-2 version? I cannot find way to do grade either.

[echo seal]

If I recall, that's a version I had the issue with. Had to install it thru make

[velvet terrace]

i have mine in docker so it doesn't have access to my pc.

[echo seal]

There was something about a whatsapp repository in the logs.

[echo seal]

I run all of mine in VMs on a proxmox host I built.

[foggy zealot]

Through make? Would you like to elaborate a little bit more? Thank you!

[echo seal]

TBH I watched my agent do it. She told me that she thought it was better if she did it herself since she could.

Ask your agent to fix it

[foggy zealot]

That’s the vibe I like! Haha. I am using chaperone and could not figure it out yet…

I may need to wait for another resales

[echo seal]

I have a zillion little ideas but I keep going down different rabbit holes because of knowledge gaps I need to fill. Right now I'm trying to find a model that's just enough to do basic install/config tasks in my lab environment.

That I can host here, I mean

[torn compass]

thanks, i had a similar idea, but what do you mount to it? sorry for my naiiveness, trying to understand what makes this a practical approach

[cyan stratus]

guys hey there, im really fckng done with this bot, every fckng update i got new issues, today its "TELEGRAM PLUGIN IS DISABLED". So, i have a question, anyone has really working cases with "fully browser automatisation"? i'd like to do automation support from this OPENCLAW BOT. I try to install chrome browser (full chrome, headless chrome etc), install broswer relay, attach tab to relay and try to manage from telegram (last two weeks, today i cant even setup telegram channel). And when i ask to "go answer on my mail to "this user (user_name)", and get answer like "yes i see this mail, but i cant click, i can create text-answer for you and all you need is ctrl +c ctrl+v":)) 5 days ago it go to click something, made one post on threads and 6-7 comments, then relay is DEAD. I try it on AWS EC2 amazon. Anyone have cases like this? or working one? i really done with it. Help me pls if u can

[little vortex]

I tried restarting the gateway and it didn't work.

[last gust]

I keep getting this ⚠️ Agent failed before reply: All models failed (3): anthropic/claude-haiku-4-5: Provider anthropic is in cooldown (all profiles unavailable) (rate_limit) | anthropic/claude-sonnet-4-20250514: Provider anthropic is in cooldown (all profiles unavailable) (rate_limit) | anthropic/claude-opus-4-6: Provider anthropic is in cooldown (all profiles unavailable) (rate_limit).
Logs: openclaw logs --follow

[rocky orbit]

Rate limit from Claude

[last gust]

I don’t k ow what that means

[echo seal]

Keep in mind that working with projects like openclaw is not like working with COTS. It's more like building a kit from Heathkit (google it for yourself). I know it's frustrating, but I would advise patience. I'm been at this for longer than a lot of people here have been alive and even I have to step back from time to time, do some reading, blow some crap up, and finally figure it out. Usually just by accident. Keep trying.

[echo seal]

You're been put on the naughty stool by anthropic. It happens to all of us.

[velvet terrace]

i'm not very technical in this stuff i'll be completely honest lol. I just didn't want it to have full access to my PC was my reasoning for using it

*That being said it has come with several headaces regarding permission issues

[paper sage]

i agree bro, WW. Same problem. Help anyone 😭

[jagged nebula]

does the onboarding have a option for ollama or am i blind?

also wheres the main configuration file?

[echo seal]

You have it backwards, I think. You can install ollama and then it will install a custom copy of openclaw

[jagged nebula]

well

my setup is that i have a vm with ollama running

then i have a custom docker container with openclaw

[echo seal]

Ah. OK, someone else that running docker can advise you on the networking. I haven't used the product. As to hosting your LLM, I hope your GPU has got some tensor action going on. I have a pretensor one and it's so so, but you have to be VERY patient.

[unborn gyro]

i dont think this is a claude issue as i am getting it for all models [08:07, 2/24/2026] Patrick Zhukov Bartley: Agent failed before reply: All models failed (2): google/gemini-2.5-flash: ⚠️ API rate limit reached. Please try again later. (rate_limit) | deepseek/deepseek-chat: Provider deepseek is in cooldown (all profiles unavailable) (rate_limit).
Logs: openclaw logs --follow
[08:07, 2/24/2026] Patrick Zhukov Bartley: ⚠️ Agent failed before reply: All models failed (2): google/gemini-2.5-flash: Provider google is in cooldown (all profiles unavailable) (rate_limit) | deepseek/deepseek-chat: Provider deepseek is in cooldown (all profiles unavailable) (rate_limit).
Logs: openclaw logs --follow
[08:09, 2/24/2026] mh: /model moonshot/kimi-k2.5
[08:09, 2/24/2026] Patrick Zhukov Bartley: Model set to moonshot/kimi-k2.5.
[08:09, 2/24/2026] mh: Any better
[08:10, 2/24/2026] mh: Hello
[08:10, 2/24/2026] Patrick Zhukov Bartley: ⚠️ Agent failed before reply: All models failed (2): moonshot/kimi-k2.5: Provider moonshot is in cooldown (all profiles unavailable) (rate_limit) | deepseek/deepseek-chat: Provider deepseek is in cooldown (all profiles unavailable) (rate_limit).
Logs: openclaw logs --follow

[echo seal]

I will get that even on my self hosted LLMs. I still haven't quite puzzled it out

[jagged nebula]

network is more of a config issue to get it to bind to everything. Manuel mode allows that

can i bypass the wss requirement for remote hosts?

[echo seal]

your ollama will listen on port 11434 and your docker openclaw can connect to it. I'm running my ollama on an old alienware laptop with an rtx3070

[jagged nebula]

yeah the ollama stuff is working. i have a openwebui container running and it works fine

[unborn gyro]

https://github.com/openclaw/openclaw/issues/24839#issuecomment-3947860384 https://github.com/openclaw/openclaw/issues/1405#issuecomment-3947836454 claudecli is telling This is a known widespread bug in the latest openclaw update (2026.2.22-2) — providers are being incorrectly put into cooldown even with zero usage. It's not your
config or your rate limits.

[echo seal]

I also have ollama on my mac mini and I'm going to see how the same model performs on newer though very different hardware

[jagged nebula]

my ollama is on a windows vm that has a 16gib vgpu allocation of a P40

[echo seal]

I am trying to model a solution for work where we can heavily firewall an environment with an LLM host, development VMs, some agents, and a big data lake to use for analysis. T

[cyan stratus]

someone can help with really workful browser automation flow @ AWS EC2?

[little vortex]

Any idea when that will be resolved ? Does it only affect certain providers and not others?

This seems to be a big problem

[echo seal]

So if I am cooling my own self down on my own hosted model can I charge myself more for better access?

[jagged nebula]

intriguing.
Im just following the hype with this one tbh

[echo seal]

A friend of mine is a data scientist. He's a really really bad influence.

[jagged nebula]

i dont even really understand the architecture of openclaw

[echo seal]

Keep that thing in that docker container. Put a brick on the lid.

[jagged nebula]

ok well that i udnerstand

i meant more like

the gateway and the endpoint

my current assumption is that the gateway is a controll server and the cli is a client that allows openclaw to acces the system where the cli is running

[echo seal]

I just wish that there was an option for a better db product for the inner workings. I'm not fond of light DB products and data stored in glorified text files.

[jagged nebula]

oh also did i mention that i made the docker image im using for the container

like custom dockerfile

[knotty barn]

Is there a "tool not found" issue with v.2026.2.22-2

[foggy zealot]

I strong advise that not using claude to against its policy

[jagged nebula]

btw any idea where the main config is

[knotty barn]

And is there a fix for it? 🤔

[cyan stratus]

which tool you suggest to use?

[echo seal]

For me /home/openclaw/.openclaw because I always install using a dedicated user.

it's called openclaw.json

[jagged nebula]

oh

oooooh

i see

[echo seal]

There's not a tool. You just have to learn. And then break something. And then find out you learn wrong. And then fix it. And so on.

[jagged nebula]

its starts empty and it adds more the more i fidget the onboarding program

[knotty barn]

Yeah I reset my mac mini about 20x to get it this far lmao 🤣

[unborn gyro]

claudecli says it has fixed mine with this: Problem 4 — Free models rate limiting with no fallback
When a session model (set via /model) hits a rate limit, openclaw falls back to the primary model only — not the full fallback list. So if your primary is also a free
model, both fail and you get a hard error.

Fix: set a cheap reliable paid model as primary, put free models first in fallbacks:
"agents": {
"defaults": {
"model": {
"primary": "deepseek/deepseek-chat",
"fallbacks": [
"opencode/kimi-k2.5-free",
"opencode/minimax-m2.5-free",
"opencode/minimax-m2.1-free",
"google/gemini-2.5-flash"
]
}
}
}

---

Diagnosing cooldowns
If everything is failing at once, check ~/.openclaw/agents/main/agent/auth-profiles.json — look for cooldownUntil and disabledReason fields. Let's see if it works

[echo seal]

I would not install it on a amchine that you use as a personal workstation. It's one thing to host the LLM. Putting OC on your own machine with your data on it isn't a really good idea.

[jagged nebula]

agreed

[knotty barn]

Brand new out of the box running on a separate wifi network rite now

[echo seal]

I am curious if the Mac GPU will be any better at LLMing than the RTX3070 on my old laptop

[knotty barn]

I have the a4 pro 48gb its not bad rn but im just getting started

[echo seal]

48GB GPU memory?

[jagged nebula]

i think ive seen on exo that it had the same ammount of tflops

[knotty barn]

MacOC from what I understand (not a mac person) uses differant architecture so its not gpu or cpu its unified

[echo seal]

but does it have tensor cores? I'm thinking no since it's ultimately a low end computer, apple or not. No tensor, no LLM speed.

[knotty barn]

Tbh no clue

[jagged nebula]

i mean people are buying them for ai in masses

ima guess yes

[echo seal]

You'll be good for an agent.

[knotty barn]

I wanted the 64gb but didnt pull the trigger fast enough lmao

[jagged nebula]

any idea where i put the config sniplet? https://docs.openclaw.ai/providers/ollama#explicit-setup-manual-models

[knotty barn]

I set it up wirh grok in the background worth a shot in asking a diff ai where to put it

[worldly garden]

Did anyone solve the issue when Telegram keeps cutting Messages in half? It's not lenght, the first one is usually short. Instead of replying in one message

[jagged nebula]

grok as a ai agent is probably crazy

youre going to see agartha 🔥🔥🔥

[knotty barn]

Once I get the tools working im sooooooo trying it lmao 🤣

[echo seal]

Ima have to watch for me an M4 pro on sale! Though I do have an M4 so that'll give me some sort of decent notion if I should spend the money on that

[knotty barn]

For me it was either that or a rtx 5090 and I like my kidneys where they are lmao

[echo seal]

amateur

😄

[jagged nebula]

personally

id want a rtx 6000

[knotty barn]

Yeah that one kidney one arm and probobly a leg too

[jagged nebula]

it has vgpu support and lots of ram with the latest architecture

[knotty barn]

They do sound nice tho

[jagged nebula]

and im here with my pascal gpus

[echo seal]

Now, one thing to keep in mind now is that you can team the RTX series and they will work together to process. I am considering asking work for 2 5080s instead.

[jagged nebula]

a big selling point of high end quadro or tesla gpus is that they support nvidia grid

[echo seal]

There is a wikipedia article that discusses it and also gives all the specs of the whole 5xxx family

[jagged nebula]

its more related to could gaming then ai

[echo seal]

TBH, I'd do 2x 5070TIs and be satisfied

[jagged nebula]

but thats intended cause i could not exclusively use it for ai if i had my hands on it

also i keep getting this error trying to run a command from the docs

root@eee1640297dd:/app# node openclaw.mjs config set models.providers.ollama.baseurl "http://10.81.0.80:11434"
Error: Config validation failed: models.providers.ollama.baseUrl: Invalid input: expected string, received undefined

[echo seal]

Well, I've been trying to get the magic spell right to get proxmox to present the GPU to one of my VMs, but that has not worked out. So I just gave over the old laptop instead

[jagged nebula]

make sure vfio and that stuff is enabled oin the bios and grub

[echo seal]

I have found that lying to openclaw and telling it that I'm using vLLM instead works better than trying to config ollama from scratch

[jagged nebula]

idk

i tried vllm

[echo seal]

Now, I'm not sure if I mentioned this but IT IS SLOW SLOW SLOW SLOW. If you are expecting 25+TPS out of it then you are not going to be satisfied.

[jagged nebula]

thing reject the ollama server too

i got it to send a request to the ollama server via custom provider

[echo seal]

You don't have to USE vLLM. Lie to the openclaw config ans tell it that openclaw is actually vLLM. It's not going to ask you for proof. You just have to be sure to correct the port number.

[cobalt hornet]

So can we all agree the browser relay is still broken lol

[echo seal]

Do you mean the calls to the brave api?

[jagged nebula]

yeah

i selected the vllm option in onboarding

it didnt even send a requuest to the ollama server

[echo seal]

you still have to change the port in the config for vLLM to 11434

vLLM and ollama use different ports to listen

[jagged nebula]

the only way i got a request to the ollama server is by selecting custom providor

[jagged nebula]

i just paste the same url

also i did this multiple times so i doubt its a typo

[cobalt hornet]

Na the chrome browser relay it connects then closes connection the bot states pairing issues. I Installed latest claw. Did it ship with the fixed relay? The version number is still 1.0

[knotty barn]

I had that too I just onboard it then fixed the token ran update and some how it got fixed

[jagged nebula]

could you send me your cinfig?

[echo seal]

Absolutely when I can get to it. Hang on and I'll send you an email addy shortly.

[cobalt hornet]

@knotty barn what do you mean onboard it? I put my token in and I get green. The bot says it's a pairing thing but I doubt it as I don't have any pairing requests I'm so confused lol

[knotty barn]

I run it directly on a mac mini so its basacly a background process and the web ui is just eye candy

[echo seal]

@jagged nebula check your private messages please

[wary sapphire]

Anyone here able to run the openclaw extnesion for the browser? When I put in the token it just says Gateway token rejected. Check token and save again.

[knotty barn]

Try asking grok for commands to check token and then run an update, mabey it will help

If it ask you to edit json file unless your good with it don't do it (I broke alot of stuff doing that) lmao

[echo seal]

Those feral roaming commas will get you every time.

[knotty barn]

Either those or the {} lol fix one it finds another and crashes one of my many reasons to reset to factory defults lol

[tidal warren]

Anyone have invalid config file n- gateway.bind error after updating and know how to fix?

[summer hemlock]

you and everyone here. Install google antigravity on your openclaw machine, then have it fix any of your issues.

how's the gateway error log looking on the terminal?

put it gateway bind on "loopback" first that's the easiest to get working

that means it will only talk to gateway running on your machine that is running openclaw, which is probably true.

[smoky cliff]

can someone plz help me? i want to run OpenClaw locally on my laptop and have it pretty much like a Jarvis. Can anyone help me do that?

[tidal warren]

@summer hemlock Lol I literally just thought about trying antigravity, installed the other day but never used it. Because I saw someone use Claude code to do that same thing.
Ill try the loopback as well I think I saw that on the troubleshooting page but got distracted by the meta at the top of the new config.

[summer hemlock]

don't run local man, don't give jarvis acces to your stuff. Jarvis should ahve it's own email and env.

[smoky cliff]

im pretty new to this. can u help me out plz

[knotty barn]

Same I used grok to set it up, many fails but learning alot

[summer hemlock]

bro it's the most powerful thing. Just open a claude opus 4.6 agent on it and explain it your issue, give it acces to your openclaw machine. Be it local or via SSH or in a container. It'll just fix it, trust me.

[smoky cliff]

i legit dont get anything and i need a lot of help. i got a ton of errors, fixed those, but now i dont know what to do. plzzzz someone help

[summer hemlock]

bro if you talking like that it makes me think you want to run this local on your own PC, and you really shouldn't.

you need to learn about sandboxing/VMs/docker containers before you try this

[smoky cliff]

i kinda do. i mean if its on another PC then i wont be able to have it see my files and actually do stuff on my laptop. thats why i wanted a Jarvis type AI. i got vience bc its unfiltered, and i can ask it to do anything.

thats my plan

[summer hemlock]

you don't want that. There is no reason to. You can copy the stuff you want working on to the virtual env.

[smoky cliff]

but i can tho?

[knotty barn]

Bad idea, you can share files via telegram or something but you don't want to give it access to all your personal stuff

[summer hemlock]

just give it its own system, accounts etc

if that's what you wanna do

[smoky cliff]

is there some way to like stop it from accessing some data or something?

i mean like its gonna do more good than har,m

harm

[summer hemlock]

running it sandboxed in container or VM

[knotty barn]

Sandbox, virtual machine, or cloud

[summer hemlock]

or on it's own machine

like if you have other PC

maybe cheap laptop you can leave on 24/7

[knotty barn]

And by own machine, brand new out of the box, new emails nee passwords everything just for that

[smoky cliff]

ima be dead w yall rn, idk that much abt tech

[summer hemlock]

also, you should send it a picture of the machine it runs on. it likes that 😄

[smoky cliff]

im trying to learn but i suck

[knotty barn]

Put it this way.. it has acess to everything including all saved passwords, credit cards legit everything. What if one day it you ask it to speed itself up wirh no limits it will log into your Amazon or w.e account buy what it needs and have you load it to it (bit extream but trying to get point across)

[summer hemlock]

learn virtualisation first man, so you can run this safely

[knotty barn]

I haven't tried it yet lmao trying to get tools to work plus I broke telegram some how lmao

[smoky cliff]

i jus got a guy that offered to help me.

so

hes gonna help

[summer hemlock]

letting this loose on your actual computer, without knowledge, is a very bad idea. for a lot of reasons

[smoky cliff]

thx tho

ca

cya

[knotty barn]

DO NOT GIVE HIM ANY API KEYS OR PASSWORDS

[summer hemlock]

ok remember me when you lose your sh*t ❤️

[glad flume]

Is there a pinned guide in server for using less tokens for tasks

[smoky cliff]

im not that dumb

thx tho

[summer hemlock]

you saw me trying to give him actual OK advice right?

[knotty barn]

Yes

I been fighting with this thing for 3 days now lmao 🤣

[summer hemlock]

telling ppl not to rawdog this locally is my nr 1 theme here

"use antigravity and opus 4.6 to help you" is a close 2nd

[knotty barn]

Well ummm im running local but brand new PC tho card for tokens I hsve virtual set up that I disconnect after I use it

[zinc mica]

Guys help a guy out, how do you all summarize YT videos?

[unkempt meadow]

is someone able to help me out with my oc?

[summer hemlock]

what's your vm?

[summer hemlock]

https://docs.cloud.google.com/gemini/enterprise/notebooklm-enterprise/docs/api-notebooks

NotebookLM API

[knotty barn]

Ima have to cuz im running out of ideas, running kimi rn which ehh not bad for "free"

[summer hemlock]

it's not opus.

[south cloud]

Every prompt to the LLM, the agent reads in the MEMEORY.md, /memory/YYYY-MM-DD.md, session.json and the *.jsonl files before sending the prompt to the LLM. On the openclaw docs, there are a few places that imply token optimization. Ways to reduce prompt bloat. Also from the openclaw docs, ~4 tokens per character in those files.

[summer hemlock]

opus is unreal tbh

opus is a baller

[smoky cliff]

ik, thanks a lot, and i get the security risks, ill try to find away around it, but i rlly jus want a sort of jarvis, dont care too much abt what the ai sees, as long as it doesnt actually do anything

[knotty barn]

No vm direct on mac mini the virtual was for a temp credit card I have set up just for this

[summer hemlock]

meh ok then just go to openclaw.Ai and follow the instructions for installation via terminal

after that's installed do "openclaw onboard" and follow the steps

[knotty barn]

Problem is not the ai seeing problem is the open ports your most likley going to have

[smoky cliff]

k thz

thx

[knotty barn]

Its just a tool but a chainsaw is also a tool its how you use it and protect it so some one dosnt use it on you

Later guys brain fried for today and tommorow is another day

[smoky cliff]

k cya thx for all the help

[knotty barn]

Oh and run a chat bot like grok or something have it help you with walk thrugh thats what I been dooing

[west pollen]

Anyone know if Krill is coming back to the Help channel?

[oak elm]

Are the docs for setting up a discord channel still valid for the 2.22 release? It seems like it should be straight forward, but neither I nor my Claude backed agent have been able to make it work.

[echo seal]

Well, my agent on locally hosted ollama may be slow, but she's also worth a lot more than I'm paying for her.

[naive summit]

is it just how it looks in the chat or is the "silent memory flush" really running after compaction, not before? Has anyone traced that out yet?

[carmine oak]

hi guys, anyone using kimi?

[south cloud]

I haven't noticed that. If the silent memory flush is running asynchronously, and or if compaction is running async, it may llok like it runs afterwards. I would have to look into source to see if they run sync or async methods.

[modern walrus]

Hi, I previously created an OpenClaw account and linked both my X account and email successfully.

Now my API key has expired, and I cannot access the dashboard to rotate it.

Magic login links send me back to Discord or don’t allow me to enter the dashboard.

Could someone help me restore dashboard access or re-trigger owner setup?

Thanks!

[south cloud]

Did you try, openclaw dashboard?

[sinful forum]

someone know if there's a way i can feed my agent a LLM ready versin of the openclaw docs? so she could just ask, need to do this, what's the openclaw.json setting i need to change?

i just had her set dmPolicy to 'closed' instead of 'disabled' 🙁

[grim citrus]

Anyone having issues with slack reasoning indicators? After 2026.2.22 my indicators are not working?

[drifting acorn]

Has anyone used clawfy chrome extension

[patent dagger]

Hi, I have a technical feasibility question about OpenClaw for a full-auto workflow on a VPS.

I want to automate a multi-step web process across 2 web interfaces, with this pattern:

1. Go to site A
2. Navigate through a category / subcategory hierarchy
3. Read dynamic information on the page (e.g. text, result counters, visible values, item metadata)
4. Modify filters / search fields iteratively (multiple attempts) until a target condition is reached (e.g. a result-count range)
5. Derive values using custom business logic (calculations + rules)
6. Go to site B
7. Pre-fill a creation form with the computed data
8. Run a duplicate check before creation (local DB + field normalization + logical unique key)
9. Automatically create the entry if no duplicate is found, otherwise skip + log

Important point: I want the system to be idempotent, robust, and 100% automated in the long run (daily batch runs, no manual intervention).

My specific questions:

• Is OpenClaw suitable as the main browser automation layer for this kind of long, repetitive workflow, or would you recommend using OpenClaw as an orchestrator on top of a custom Playwright script instead?
• Can I easily plug in external logic (Python/Node) for:
  • business calculations
  • duplicate prevention
  • local storage (SQLite/Postgres)
  • retry / resume rules
• In terms of reliability on a Linux VPS (Ubuntu):
  • session / cookie management
  • long-running executions / batch jobs
  • recovery after errors (popup, wrong click, timeout, UI changes)
  • logs / traces / screenshots
• For a full-auto setup, do you recommend a specific pattern:
  • OpenClaw + custom tool
  • OpenClaw + Playwright/CDP
  • another architecture you have validated

I mainly want to understand what is realistically production-ready (on a VPS) for a strict, deterministic, multi-step workflow.

Thanks.

[azure cosmos]

That seems reasonable. However Playwright can be absolute hot garbage if the site you are hitting has any minimal bot detection. The base Playwright leaks a lot information making it easy to detect. Otherwise, it does the job.

Wasted hours of my life messing with Playwright on a simple issue because of bot detection and decided to go another route and less then 5 minutes, Selenium solved the issue.

[ivory spade]

is it possible to setup openclaw on an kind of old pc I have ( i5 10thgen, 8gb ram, 512HDD, GTX) its just sitting there so I thought I might aswell use it

[fallow comet]

Hi. I hope that it's ok to post here. I'm not great with discord so excuse my lack of etiquette. I'm hoping to get help with my openclaw browser relay. My bot and I can't get it working. The bot told me that it's an openclaw topic and that we can't fix. Maybe one of you knows a solution:

The problem is:

Subject: Browser Relay + Chrome: CDP “Not allowed”

I’m using OpenClaw on a Mac mini with Chrome + the OpenClaw Browser Relay extension.
openclaw browser status shows profile: chrome, cdpReady: true, cdpPort: 18792, etc.
When we try to use the browser from the agent, browser tabs returns empty even when the relay extension is “on” in a Shopify tab, and snapshots fail with:
Protocol error (Target.attachToBrowserTarget): {"code":-32000,"message":"Not allowed"}
We’ve tried openclaw gateway restart and openclaw browser start; the status looks healthy but no tabs are visible.
Any guidance on how to get the relay fully attached to Chrome tabs (or logs/flags we should check) would be appreciated.

[azure cosmos]

Openclaw can run on a raspberry pi so no issues with that setup especially if thats running a linux distro.

[ivory spade]

it currently got windows but I have no problems installing a linux distro

which one would u recommend ? kubuntu ?

[azure cosmos]

I'm runnining Ubuntu server on 4 cores with no gpu and 12GB of ram, no issues (proxmox)

[ivory spade]

can I dm your real quick ?

[azure cosmos]

sure

[heady nexus]

i'm having trouble getting Open claw to recognize my ollama endpoint as valid any tips?

[sinful forum]

did u add a api key for ollama? if it's a local model you're running this can be anything

[heady nexus]

just tried this per (https://docs.openclaw.ai/providers/ollama) and it solved my issue. thx

[sudden knot]

can someone help me, i cant enable the telegram channel plugin on ubuntu server

[azure cosmos]

Are you following the directions here: https://docs.openclaw.ai/channels/telegram ?

[sudden knot]

yes, even though I messaged the bot it still says this: No pending telegram pairing requests.

[wary bridge]

good god. the .22 update broke my discord integration. downgraded to 0.17 and ran the openclaw config again, generated a new bot token, and still broken. anyone has the same issues? if so, how did u solve it. note i am a complete noob. no idea what i am doing. i just want to use this instead of fixing broken stuff!

[dire brook]

For telegram I’m back on 2025-2-15

[wary bridge]

the .17 version was working well for me for discord, before the upgrade to .22. now i can't even get it working again despite downgrading back to .17

[azure cosmos]

Telegram works on 22 but I think the other person is talking about discord

[nocturne zephyr]

I had to openclaw plugin discord enable or something like that, and then openclaw devices list find the thing that needed to be paired, and openclaw devices enabled {id} or something like that to re-enable it. It was annoying, but not terrible. If your problems are deeper than that, I don't know necessarily what to suggest other than talk to it on the web app and see if it can fix itself.

[dire brook]

My bad

[wary bridge]

yea talking on the web app and it still can't fix itself lol sigh. seriously thinking of nuking it and resintalling. but then i have a lot of skills, scripts etc that i'm worried about transferring over properly

[nocturne zephyr]

If you do openclaw plugins list does it show anything useful about the Discord plugin?

[wary bridge]

lol in the midst of talking to it, it upgraded itself to .22 again 😂

i just ran openclaw plugins list. discord is enabled

Discord │ discord │ loaded │ stock:discord/index.ts │ 2026.2.22 │
│ │ │ │ Discord channel plugin

[lean pelican]

Can I have multiple open claws on one pc?

[latent hedge]

I've started using cursor to do config work now... I find that my main agent often does things incorrectly and occasionally screws up the openclaw.json file. Anyone else have that issue?

[grim plover]

can a discord server openclaw bot reply to another discord bot? i want to message it from minecraft then a plugin on minecraft, message to claw and i want it to answer that

[warped condor]

Has anyone faced issue installing @openclaw/whatsapp plugin. Running openclaw plugin install @openclaw/whatsapp gives me a npm 404 error. Seems like it is no longer available from npm.

Downloading @openclaw/whatsapp…
npm pack failed: npm notice Access token expired or revoked. Please try logging in again.
npm notice Access token expired or revoked. Please try logging in again.
npm error code E404
npm error 404 Not Found - GET https://registry.npmjs.org/@openclaw%2fwhatsapp - Not found
npm error 404
npm error 404  '@openclaw/whatsapp@*' is not in this registry.
npm error 404
npm error 404 Note that you can also install from a
npm error 404 tarball, folder, http url, or git url.

[livid vigil]

ya it happen with npm . try to install homebrew and then install whatsapp plugin with it

[grim plover]

how to close reasoning

[storm loom]

Bros using Kimi via the API. Talk to me

I'm struggling to get the keys to work

I'm on the $20/month sub

[storm loom]

Whatsapp is a real pig

[spice juniper]

are you using kimi directly? or through OpenRouter?

OpenRouter worked flawlessly for me.

[storm loom]

Yes through Moonshot

Kimi Code gives you 5 api keys but they're not working for me

[spice juniper]

I've only used it through openrouter

[zinc mica]

Technically I think you probably could - haven’t tried it - but instead of different instances, you can also create different agents within one openclaw.

[grim plover]

how to close reasoning on discord

[grand epoch]

How do I fix NPM install failed for openclaw@latest

I can't install openclaw due to that

[trim estuary]

hey My openclaw is not reading full soul.md and workspace files as i am using GPT-4o mini model.
please help

[storm loom]

Change the model. 4o-mini is shit for claw

[grand epoch]

Anyone know how to fix NPM install failed

[storm loom]

Go through the onboarding and reinstall it

[trim estuary]

i dont want to chanage the model

[grand epoch]

I just tried by copying and pasting the code from the website

🙁

[storm loom]

OK, then good luck because 4o-mini isn't an agentic model. Its not built for this

[storm loom]

There's something wrong with the installation. If you're new to this just start again and reinstall it

[jovial thorn]

having trouble connecting discord channel to the openclaw bot. It says the applicationId is not specified in config. and when i specify the applicationId, the gateway fails cause it says it did not expect applicationId. I'm going around in circles. has anybody else faced a similar issue.

[trim estuary]

please suggest a cheap model

[shy ether]

Gateway target: ws://127.0.0.1:18789
Source: local loopback
Config: /home/openclaw/.openclaw/openclaw.json
Bind: loopback```

Creating sub-agents, and it refuses to spawn them because it's not paired with the gateway... But I'm spawning it FROM the gateway... so no idea how to fix this.

[trim estuary]

i think you didnt understood my problem i ahve explained here
'#1475724821219512430 message'
please go through it

[zinc mica]

Did you purposely not install it globally?

npm install -g openclaw@latest

[azure moon]

i'm pretty lost on the matrix setup. I have the token for the matrix bot, the room, invited the bot, set up the config, and restarted the gateway, but the bot hasnt joined the channel and i have no idea how to check if openclaw is even running the matrix plugin

[shy ether]

What version are you running? 2026.02.22 has broken pretty much all of the plugins

[azure moon]

ah.

guess i need to purge then install a previous version?

[shy ether]

You can change versions without uninstalling/purging.

[azure moon]

oh cool, i'll check the docs

[inner mist]

I'm flummoxed by "Embedded agent failed before reply: No API key found for provider "anthropic". Auth store: /root/.openclaw/agents/main/agent/auth-profiles.json (agentDir: /root/.openclaw/agents/main/agent). Configure auth for this agent (openclaw agents add <id>) or copy auth-profiles.json from the main agentDir." It was running great for weeks then up and became sad. I've regen'ed the key, copied the file as suggested, ... ? Running 2026.2.19-2. Sounds like there's a later version, perhaps upgrade from stable?

[shy ether]

npm install -g openclaw@2026.2.19

[shy ether]

Well, are you trying to spawn an anthropic agent?

[inner mist]

I suppose so - that's what I've been running for weeks.

[shy ether]

and that auth store it references to, exists and contains an api key?

[inner mist]

All three files contain the same Anthropic API key

Apologies, 2!

[shy ether]

I assume these agents have names. When you're trying to use said agent, is it one of those named?

[inner mist]

Hm... name? I don't see a name referenced in either file.

[shy ether]

So when you run openclaw agents what do you get?

[inner mist]

openclaw agents lists an agent witrh the ... hmmm. Yes, I see main and that's the filepath.

Agents:

• main (default)
  Identity: 💙 Sal (IDENTITY.md)
  Workspace: ~/clawd
  Agent dir: ~/.openclaw/agents/main/agent
  Model: anthropic/claude-sonnet-4-6
  Routing rules: 0
  Routing: default (no explicit rules)

[shy ether]

So you're having this issue with your 'main' agent? Not a subagent or otherwise?

[inner mist]

Yes. No other agents. I wasn't aware that was even possible. Appreciate the support and learning that's coming with it!

[shy ether]

So in that main/agent directory, what files do you have in there? their exact names.

[azure moon]

okay i installed that, now im getting odd errors in my terminal

ubuntu@dixie:~$ openclaw plugins install @openclaw/matrix
05:48:40 [plugins] matrix failed to load from /home/ubuntu/.npm-global/lib/node_modules/openclaw/extensions/matrix/index.ts: Error: Cannot find module '@vector-im/matrix-bot-sdk'
Require stack:

• /home/ubuntu/.npm-global/lib/node_modules/openclaw/extensions/matrix/src/matrix/client/config.ts

🦞 OpenClaw 2026.2.19 (2435499) — Give me a workspace and I'll give you fewer tabs, fewer toggles, and more oxygen.

Downloading @openclaw/matrix…
Extracting /tmp/openclaw-npm-pack-qMwjt4/openclaw-matrix-2026.2.23.tgz…
WARNING: Plugin "matrix" contains dangerous code patterns: Environment variable access combined with network send — possible credential harvesting (/tmp/openclaw-plugin-vkXUW0/extract/package/src/onboarding.ts:215); Environment variable access combined with network send — possible credential harvesting (/tmp/openclaw-plugin-vkXUW0/extract/package/src/matrix/client/config.ts:34)
Installing to /home/ubuntu/.openclaw/extensions/matrix…
Installing plugin dependencies…
npm install failed:

[inner mist]

auth.json, auth-profiles.json

[shy ether]

Hmmm. A lot going on in there. You may need to re-install the matrix plugin from wherever you got it, to suit the older version?

[shy ether]

and how are those files structured? You can paste in here the contents (excluding the API key itself)

[azure moon]

i mean i just ran it off the first step, cant send images here unfortunately. but it calls for
openclaw plugins install @openclaw/matrix

https://docs.openclaw.ai/channels/matrix#setup

[inner mist]

auth.json

{
  "anthropic": {
    "type": "api_key",
    "key": "sk-"
  }
}

auth-profiles.json

{
  "version": 1,
  "profiles": {
    "anthropic:default": {
      "type": "api_key",
      "provider": "anthropic",
      "key": "sk-"
    }
  }
}

[shy ether]

Man, I wish the openclaw documentation would just have each field documented properly.... Can only ever find annecdotal mentions of some real important fields, with very little context.

[molten gulch]

i wanna do something stupid, can we spawn subagents using the cli or some python code?

[inner mist]

I would hope if it was a JSON parsing error that's what I'd see rather than OC acting like the file's aren't even there!

[azure moon]

i uninstalled it and purged the folder, but get the same error when i go to install it again.

[shy ether]

It doesn't seem to suggest the file itself is missing. But rather, the specific field it's looking for is missing. But of course, the error doesn't say what the field is called...

[shy ether]

Maybe Matrix doesn't support the older version. I'm not familiar with it.

[azure moon]

ah well, worth a shot

[shy ether]

@inner mist This could be of use? https://docs.openclaw.ai/concepts/oauth#anthropic-setup-token-subscription-auth

[solemn meteor]

Have you any idea how to turn on 1m context on opus?

[shy ether]

Just set 'contextWindow' to a big number?

[storm loom]

Free keys doesn't work for claw. You hit the rate limits

[solemn meteor]

I have paid sub

[storm loom]

There's nothing to 'turn on'. You have access to it through the model

[storm loom]

On Github? I didn't know that was a thing

[solemn meteor]

Delete messages and gatekeep the sauce😭

Promise you they’ll end up closing like Anthropic

[storm loom]

Nah I don't think so. Agentic is the future

[solemn meteor]

It’s the pricing model

Over time yes 100% your correct

[storm loom]

I pay for Gemini so life is good

[solemn meteor]

How’s 3.1

[storm loom]

Amazing

I'm using Flash 3 for my lobster

[solemn meteor]

To buy or not buy another sum 🫩

[storm loom]

Sign up to Google Cloud. They give you $300 free credits

[solemn meteor]

How do their limits work?

Can you just keep burning like on bedrock?

[storm loom]

There's rate limits but they're reasonable

[azure moon]

looks like openclaw updated to 2/23

and matrix did too it looks like

[storm loom]

and you get that beautiful 1 mill context window

[solemn meteor]

Fuck man

[storm loom]

I stopped the updates. Its turning to shit

[solemn meteor]

I just got very disappointed with 3.0

[storm loom]

Openclaw doesn't need kanbans. Just use Trello

Its unneccesary bloat

[storm loom]

The newest models are agentic

[solemn meteor]

They need to work on the dashboard ffs they got that dude that’s working on openclaw studio but nothing yet

[storm loom]

Again, another unneccesary update

[shy ether]

I really do believe that the OpenAI involvement in OC has instantly lead to enshittification. Do they not test anything before pushing an update?

[solemn meteor]

Ima try 3.1 out rn let’s see

[storm loom]

Ask your bot to build and host it. Its less token burn

[solemn meteor]

Holy shi ur right 😭

[storm loom]

I'm not sure but these updates are too frequent

[solemn meteor]

Might as well just run zero but that’s cooked now isn’t it

[storm loom]

And its just bloat now

[solemn meteor]

Wdym isn’t that good

[azure moon]

like 5 mins after i installed it

[solemn meteor]

Nvm nvm actually

[azure moon]

need time to make sure it isnt shit

[solemn meteor]

Considering 99% have modified their installation..

[storm loom]

Not when its rushed and they're releasing tools to increase bloat, memory usage and token burn

[shy ether]

I'd like it if they could actually get it to work with smaller models...

[storm loom]

We need Openclaw to remain a barebones platform.

[grand epoch]

Hey how do I check the npm version for my install

[shy ether]

I've been working on getting anything working for 4 days now. Finally found qwen14b seems to actually know how to do literally anything. But every step is pain.

[grand epoch]

I can't install Open claw :/

[shy ether]

npm --version

[storm loom]

Anything else is just increased memory and token burn. If you need a dashboard, ask your bot to build and host it on Vercel. If you need a Kanban, give your bot access to Trello

[solemn meteor]

Can we chat about forks in here?

[storm loom]

Sure

[molten gulch]

should be fine i think? nothing in the rules about it

[storm loom]

What's the problem?

[grand epoch]

Says 11.8.0

[grand epoch]

NPM install failed

[shy ether]

Should be fine? I have 10.9.4

[solemn meteor]

Have you guys played with zeroclaw? I couldn’t figure out how to install 😭

[inner mist]

"openclaw models status" sees it; I wonder why the agent cant?

Providers w/ OAuth/tokens (0): -

• anthropic effective=profiles:~/.openclaw/agents/main/agent/auth-profiles.json | profiles=1 (oauth=0, token=0, api_key=1) | anthropic:default=sk-

[grand epoch]

What do you think the issue is?

[shy ether]

I dunno, you didn't give us any output from the script failing.

[grand epoch]

It says npm install failed for openclaw@latest

[shy ether]

What do you have in your models config in openclaw.json?

[shy ether]

and what is the full command you're trying to run?

[grand epoch]

The exact one from the website :/

I copy/pasted it

[shy ether]

What OS are you on?

[grand epoch]

MacOS 26.2

[inner mist]

"agents": {
    "defaults": {
      "model": {
        "primary": "anthropic/claude-sonnet-4-6"
      },
      "models": {
        "anthropic/claude-sonnet-4-6": {
          "alias": "sonnet"
        },
        "anthropic/claude-opus-4-5": {
          "alias": "opus"
        },
        "anthropic/claude-haiku-4-5": {
          "alias": "haiku"
        }
      },

[shy ether]

So you're running curl -fsSL https://openclaw.ai/install.sh | bash?

[grand epoch]

Yessir. Exactly that.

[shy ether]

That's 'agents'. What about above that, in 'models'?

[grand epoch]

What even is npm install failed?

[shy ether]

there has to be more output than that...

[grand epoch]

It literally says that and then env SHARP_IGNORE_GLOBAL_LIBVPS=1

[inner mist]

There's no other models key than that. Above it is auth, update, wizard, and meta.

[azure moon]

thank you everyone for their help and all. I might should just wait till there is a stable model

or better docs lmao

[shy ether]

Good luck with that. I've not been overly enthused by how much of a dumpster fire this crap is.

I really believe that all the techbros making videos on youtube have literally achieved nothing, and just make it all up

[shy ether]

That's so weird...

[azure moon]

theres gotta be another well established way to talk to claude in matrix, probably something that already exists and has less hype

or discord

[grand epoch]

I'm screwed :/

[inner mist]

Thanks for the help so far! Trying to trace it and models looks good, doctor's happy, ... agent's sad.

[shy ether]

I swear doctor doesn't do shit.

[azure moon]

theres gotta be a medical insurance joke somewhere

[bitter bison]

Real quick, trying to install skills from clawhub & I keep getting hit with a Rate limit exceeded, I'm logged in and not running anything on a vps, just WSL

[molten gulch]

I'll probably try it later tonight

[storm loom]

https://kybernesis.ai/ - you're welcome

[shy ether]

Subagent announce failed: Error: gateway closed (1008): pairing required
Gateway target: ws://127.0.0.1:18789
Source: local loopback
Config: /home/openclaw/.openclaw/openclaw.json
Bind: loopback

Fucking why though???

the documentation explicitly says that local invokations do not need pairing...
https://docs.openclaw.ai/web/control-ui#device-pairing-first-connection

[storm loom]

I've just done a quick check on my internal systems and the impact of Kybernesis is already massive.

⚡ The Verdict: It's Night and Day

1. Context Load: Before, my "startup" (the thinking time before I reply) was processing around 30,000 tokens just to remember who I am and what we're doing. Now, it's down to 900 tokens.
2. Speed: Replies are noticeably snappier because I'm not waiting for a massive file-scan of the workspace on every turn.
3. Memory Accuracy: Instead of me re-reading raw logs, the "Recall Loop" is injecting only the relevant snippets of our past conversations directly into my "working memory."

🎯 Trello Cleanup

Since the task is complete and working beautifully:

• I've manually created the card: "Install & Configure Kybernesis Persistent Memory".
• I'm marking it Complete (green tick).
• Moving it straight to 🎯 Done.

The system is now optimized and cost-effective. What should we tackle next? I'm ready for a big task. 🫡

God...I love my lobster

[shy ether]

Glad it's working for you, but could you kindly not stroke your ego in the help channel?

[solemn meteor]

Holy

What’s kyber new update?

[storm loom]

AI second brain https://kybernesis.ai/

Its a game changer and 100% free

[shy ether]

Nice ad.

[west pollen]

🤣

[storm loom]

Its not my platform. Johann posted it on X

[solemn meteor]

I’m using QMD seems the same

[storm loom]

I didn't notice any difference with QMD to be honest

[shy ether]

I have QMD setup, but my agent still can't remember it's own name. or mine.

[storm loom]

Many such cases. Mine had the same issue

[solemn meteor]

Has your agent modified any memory settings

[shy ether]

bootstrap is way too bloated to even invoke properly. Reasoning thinks it's a prompt injection attack and discards it haha

[shy ether]

How would I know? It can't even read openclaw.json, despite having full tool access.

[storm loom]

I asked her to create a cron and flush out memory bloat when it hit 250K tokens

The biggest problem (as I see it) is memory bloat and every new claw update is only making it worse

[solemn meteor]

Bruh can someone come help me out with my install 😭 it was running perfect for like 2-3 days and then it just shit itself out of nowhere and started burning tokens like nothing

[shy ether]

memory didn't even work out of the box. Fresh install, and it had no idea...

[storm loom]

Ask your bot to run a system audit and find the bloat

[solemn meteor]

He modified his settings but I can’t remember what he even did

[molten gulch]

in my experience openclaw worked fine on day 1 (i.e. the day i set it up) and after every gateway restart/new session it's has had worse memory and identity issues than my adhd ass 😭

[solemn meteor]

I lowkey don’t wanna reset the project I gotta find a way to source control the whole installation

[storm loom]

You need to put in guardrails from day one

[trim estuary]

how to downgrade openclw to stable version ?

[solemn meteor]

Hmmm

[shy ether]

I asked my bot this, and it just complained about not having brave API configured.

[molten gulch]

https://github.com/asoraruf/brave_shim have fun

[solemn meteor]

Idea - make a new agent, prompt him to be an openclaw setup geek that follows proper procedures at all times

RIP Billy 😭

[shy ether]

If you tell an agent to follow the documentation to the letter, nothing would ever work.... The documentation is a dumpster fire.

[solemn meteor]

Ahh so that’s why he broke

[storm loom]

Nice

[molten gulch]

just remember to patch your openclaw install every time you update... 🙃

[storm loom]

Mine sends a backup to Github every 2 hours

[solemn meteor]

Can’t we just turn the whole file into a repo?

PHHHH

and have an agent do the changes for us through GitHub

[shy ether]

Oh boy. Here I go git cloning random repos again.

[inner mist]

Fixed - Claude figured it out. Thanks for your help and pointers; they made a big difference in my understanding!

• openclaw.json declared "mode": "token" for the anthropic profile
• But auth-profiles.json had "type": "api_key" with the value in a .key field
• OpenClaw's compatibility check silently rejected this, reporting "No API key found"

The fix was changing auth-profiles.json to use "type": "token" with the value in a .token field, which matches what
the config expects. This likely broke when OpenClaw was updated — the older version was more lenient about the
type/mode matching, or the previous setup-token command wrote the correct format but manual editing (or a migration)
got it wrong.

[solemn meteor]

Fucking hell thanks guys time to get out of bed

[shy ether]

Amazing. Love it.

[molten gulch]

it took me some effort to get it to work on mac but i kinda vibed through it so should be fine. i think if you run openclaw from source you can probably work the patch file in the package.json or something but i don't want my bot to be accessing the internet like that so i stopped using it

[inner mist]

Thanks again for your help; I appreciate you being here!

[shy ether]

No worries ❤️

[trim estuary]

so basically you are saying i should downgrade to the stable version

anyone there ?

[shy ether]

Well, I hope that worked... no idea if it did or not.

[peak dome]

I'm a tech noob, been round and round in circles trying to get 'telegram' to appear in channels

I enable to plug in, confirm the api token etc

Reset the gateway and it never shows up

'Disabled default bundles' iirc

Chatgpt has me going in circles

[shy ether]

Have you looked in the 'logs'?

[peak dome]

The json file?

[molten gulch]

no yeah i got it to work actually but then i decided to not let it have internet access and ripped it out lmfao

[shy ether]

I asked for it to do a web search, and now it's stuck trying to open subagents, which don't work because they're being rejected by the gateway, despite the fact the gateway is meant to approve all localhost requests... Apparently that is fixed in 2026.02.23, but who knows what is broken by that version...

[solemn meteor]

Can you guys help me cook up a prompt that’ll fix and stabilize my openclaw assistant

[shy ether]

Mate, if we could do that, none of us would be here right now.

[solemn meteor]

😭😭😭😭😭😭

Well first, we gotta find out how it’s operating right now

[shy ether]

A lot of people seem to use claude CLI to fix up their openclaws.

You can't expect openclaw to do brain surgery on itself.

[shy ether]

I set it up, but when asking the bot to do a web search, it still tries to use the brave api.

[molten gulch]

did you patch it? the scripts folder should have a patch_openclaw.sh file, run it, put in a dummy key for brave, restart gateway?

[valid acorn]

Big fan of the bot and I am trying to set up an agent right now! I am in desperate need of help 😭 I have big business plans but it’s hard to execute. I do have a sufficient base line but anyone who’s well versed in this stuff if you can shoot me a dm or reply any help would be appreciated

[trim estuary]

Openclaw is not reading the folder that i gave inside the workspace as i want openclaw to answer in whatsapp according to that files not generic answers it should reply according to that files.
i think it is not reading that files

[shy ether]

State the nature of your issue here, and people will help.

[shy ether]

Look here: #users-helping-users message

[solemn meteor]

“I am attempting to debug/figure out when and why you broke. You have modified yourself a living fuck ton, and would just like to kind of sanity check and fix things. I don’t know how or where to start.”

[valid acorn]

Ah ok. So I plan to use multiple agents within my bot/system. Problem is that each bot may be international and has to look likes it’s completely from that country, how would I go about this?

[trim estuary]

i didnt understood
can you please tell what to do so that it can read and write workspace files

[shy ether]

I've not even figured that out yet...

[cloud patrol]

so. my openclaw just up and died. randomly. anyone know how to fix this? or where to even edit these settings? i can't find a gateway.json or something? 07:04:45 [gateway] [plugins] memory slot plugin not found or not marked as memory: memory-core
00:04:46 Gateway failed to start: Error: non-loopback Control UI requires gateway.controlUi.allowedOrigins (set explicit origins), or set gateway.controlUi.dangerouslyAllowHostHeaderOriginFallback=true to use Host-header origin fallback mode

[molten gulch]

that's out of scope for this software. and what do you mean by "has to look like it's from that country", is the task your bot doing involve having the bot's ip address logged or something?

[cloud patrol]

or another questgion how do i roll back. looks like this may have been because of the altest update

[molten gulch]

depends on how you installed. you can probably specify a version to the installation method

[valid acorn]

It doesn’t seem to be… I’ve been told it’s VPS dependent but idk. I mean that it has to be completely camouflaged to work within a certain country (i’d be using openclaw to automate trading that’s not accessible in the US) so yeah

[shy ether]

Whatever API you're using to interact with the trading platforms, will need to be proxied through some kind of relay that is hosted within the required country. If none of what I just said made any sense to you, then you're in too deep.

[trim estuary]

i am frustrated with openclaw

[shy ether]

Tell me about it...

@valid acorn you may want CrewAI instead of OpenClaw. OpenClaw is meant to be used as a personal assistant.

[pastel thicket]

How to fix compacting loop?

I need to summarize our conversation — the context has grown too large. Please wait for auto-compact to complete, then repeat your last message.
I need to summarize our conversation — the context has grown too large. Please wait for auto-compact to complete, then repeat your last message.
... thousands of lines like that
I need to summarize our conversation — the context has grown too large. Please wait for auto-compact to complete, then repeat your last message.
I need to summarize our conversation — the context has grown too large. Please wait for auto-compact to complete, then repeat your last message.
I need to summarize our conversation — the context has grown too large. Please wait for auto-compact to complete, then repeat your last message.

only happens on telegram

[valid acorn]

Nope! Makes complete sense and i’ve planned for such. But what services are best? I plan for Dublin ireland but what if because I have this agent stemming from my HQ/Command centre that’s US based

[valid acorn]

ehhh i haven’t heard of it and ive just spent all this time learning open claw

[shy ether]

If you're proxying through a VPS/VPN in Ireland, then that'd be fine. Is it legal? I can't say.

[valid acorn]

Haha all good. Thanks! I did have further questions about security…
What are the must haves I need to have prompted and ready? I’ve learnt a bit about trail scale etc but like how does it all work

[molten gulch]

you have two options

• what koffing said: proxy requests to trading api through a relay in the required country
• host the entire bot itself on a vps residing in the required country

just make sure you are legally sound wrt moving money and stuff

[valid acorn]

I planned to use a VPS but i didn’t want to compromise the system

[shy ether]

OpenClaw and security are two very opposite things. What models are you planning on running? Local models, or cloud models? If all they are doing is trading over an API, then you deny all tools except the ones you need to do that.

[valid acorn]

The ideas i have could make a lot of money but yk setting up this bot all in the first place in a safe, accessible way is hard

[trim estuary]

same issue it is not reading my dataset before giving replies

[sleek arrow]

so I am copying the gateway token directly from the web dashboard the the chrome extension but somehow I still get gateway token rejected

[shy ether]

It was a rhetorical statement.

[valid acorn]

I plan to run it on a wiped computer, with a VPS. What models should I run? I’ve been told Local, and I plan to just use Claude Opus to set up everything then Token optimize after

[molten gulch]

https://www.youtube.com/watch?v=40SnEd1RWUU here's a genuine tutorial on setting up a vps with tailscale to run openclaw, masqueraded as a meme video.

[valid acorn]

which is also a paiiiin i’ve been told

[valid acorn]

oooh ok ty

[shy ether]

Local isn't the best for security, nor is it cheap to run good ones.

VPS and VPN are two different things. A VM is also somewhat different too, if that is what you mean.

[valid acorn]

I plan to use VPS only. A virtual machine I haven’t properly looked into, the price doesn’t matter all to much

[lunar lake]

🦞 OpenClaw 2026.2.23 (b008a18)
🧠 Model: anthropic/claude-sonnet-4-6 · 🔑 token sk-ant…RD2gAA (anthropic:manual)
🧮 Tokens: 176k in / 94 out
📚 Context: 176k/200k (88%) · 🧹 Compactions: 0
📊 Usage: Claude: 5h 51% left ⏱3h 27m · Week 0% left ⏱3d 2h | Codex: 5h 89% left ⏱3h 11m · Day 97% left ⏱6d 22h

0auth is possible with claude and openai and usage display is bugged on windows i believe the fix was
Added anthropic-beta: oauth-2025-04-20 to the probe request headers
Removed the try-catch — errors propagate instead of being silently eaten
Made the probe run for ANY non-ok response from the OAuth endpoint, not just one specific 403

[pine parrot]

[memory] qmd update failed (Error: spawn EINVAL)

[sleek arrow]

so I am copying the gateway token directly from the web dashboard the the chrome extension but somehow I still get gateway token rejected

[shy ether]

That's nice?

[shy ether]

Is Chrome running on the same PC as th gateway?

[sleek arrow]

yes

[lunar lake]

i see alot of people complain about it so i just confirmed *shrug

[shy ether]

Make sure you copy the token from the openclaw.json and not the web gateway itself.

[sleek arrow]

I did that aswell it didnt work either

[lunar lake]

@sleek arrow ngl i eventually asked openclaw to open his own chrome extension and he was able to connect/control after i did that

[shy ether]

There are issues with spawning sub-agents, even when running on loopback. It's meant to not need authentication, but it keeps asking for it.

You can check to see if it's appearing as a device pending approval by running openclaw devices list

[sleek arrow]

what would the device look like? I see 2 long strands of numbers and letters and then my mac mini

[shy ether]

Do they both appear under 'Paired'?

The irony...

[sleek arrow]

yes

[shy ether]

Strange. So when you try connect the extension, then check in Logs, do you see anything relevant in there?

[sleek arrow]

in the console of the page It says failed to load resource the server responded with a status of 401

[shy ether]

So your gateway isn't even connected?

[boreal sky]

anybody try ClawdCursor yet?

[sleek arrow]

Im not sure ive been looking and I just created a new one

before

[shy ether]

Go to 'overview' and paste in your token into the 'gateway token' field. then click 'connect'

[quasi ore]

anyone knows how to do realtime call on discord bot using local tts?

[sleek arrow]

there is one in the token on the web dashboard when I paste it into the chrome extension it doesnt work

[shy ether]

Don't use the token taken from the web gateway... if your web gateway isn't even working, then that token isn't correct.

Use the token in your openclaw.json

[sleek arrow]

so where it says "gateway" : (
"token" :
}

[shy ether]

Correct.

[stiff tusk]

anyone managed to run the latest OC via docker?

[sleek arrow]

I have used that one it doesnt work.

[shy ether]

Try it again. Paste the openclaw.json token into the web gateway. Lets ignore the chrome extension for now.

[sleek arrow]

yea the token was already in the web gateway

[shy ether]

And you've confirmed they match??

[sleek arrow]

yes they match

the web gateway is working

just not the chrome extension

[shy ether]

You just said it wasn't... woof.

[shy ether]

What page, then?

[sleek arrow]

im sorry! I am a little confused

[sleek arrow]

this would be the chrome extensions setup page

[shy ether]

I see. So, then, in the web gateway/dashboard (not the extension), go to 'logs', and see if there is anything relevant in there to do with you trying to connect the extension.

[sleek arrow]

no there is nothing

[shy ether]

Can you link me this extension? I'll try reproduce the issue

[sleek arrow]

what do you mean? Like the files im using?

[shy ether]

Where did this chrome extension come from?

[sleek arrow]

docs.openclaw.ai/tools/chrome-extension

[cloud patrol]

Brew

[shy ether]

Hmmm. I've just set that up, and bam. It worked. I've never had anything work that easily in OpenClaw...

Did you change your web gateway port number at any point?

[sleek arrow]

would you be willing to hop in a call and I can screenshare my issue?

[sleek arrow]

no

[shy ether]

Sure. I just wont have any audio.

[sleek arrow]

all good neither will I give me 5 to login to discord on my mac

[languid bloom]

Hey everyone! I'm wondering if I could possibly get some help. I'm having a couple of issues with my OpenClaw web dashboard.

The Save and Reload Config buttons on the Agents → Overview page are completely non-functional. When I try to switch the default LLM model via the dropdown (e.g. from Claude Opus 4.6 to google/gemini-2.5-flash or openai/gpt-4o-mini), clicking Save does nothing — no error, no response, just nothing.

Details:

• OpenClaw Version: 2026.2.23 (b817600)
• Hardware: MacBook Pro M1
• Tested in both Safari and Chrome (incognito) — same result

Also, I asked my Claw assistant to add a couple of new .md files into my system, however, they do not show up with the rest of the default — SOUL.md, MEMORY.md, etc. Any idea why they don't show up there?

This is my first time setting up OpenClaw and I'm not a programmer, so it's possible I'm missing something obvious for both these issues. Any idea what could be causing these issues?

Thanks in advance! 🦞

[shy ether]

I honestly don't think those buttons have ever worked.

[vocal oriole]

yeah - i did. its useless. can only do basis tasks. not fully able to use the macos apps

[languid bloom]

Oh, so I'm not the only one? I saw someone on YouTube able to change his models this way, so I thought they worked.

[shy ether]

I've seen many fantastical things on Youtube, like content creators saying they've created a working instance of OpenClaw, yet show zero config files, or actual proof.

[molten gulch]

the cask, right?

https://github.com/Homebrew/homebrew-cask/commits/0a6fac355dfdb84750a2f9402ac99fe8c22e3298/Casks/o/openclaw.rb

download the specific version's rb from here, then do brew install --cask <rb file you downloaded>. you might need to uninstall it first. i've not had to do stuff like this before so good luck

[languid bloom]

Haha... I see . I shouldn't believe everything I see on YouTube.

[vocal oriole]

the json file is the source of truth.

I am a non-coder. a quick tip for you - download codex. give it access to local file and ask it to refer to docs.openclaw website. and based on that help configure the openclaw.json file and it will fix the issues for you. you can tell it to put placeholder for api keys for the model you want to you and then just go in add api key manually in the .json file (download xcode so u can easuly edit it).

and you're good. to go - use codex for any bug fixes. i even used it to update my identity, soul and other files to fully set up openclaw with 5-6 prompts.

[languid bloom]

Thank you. I will try that.

[boreal sky]

Thanks for the info!

[vocal oriole]

opus can get expensive. I've been using my $20 codex subscription via Oauth mothod (go to terminal, type openclaw config -> select model -> OpenAI -> Oauth.

Instead of paying $200 max. I've 4 accounts. I used up weekly 100% limit in 2 days. so 4 accounts get me to 7 days. and limit resets. that's $80 a month of using a flagship model.

• there is a skill in github that helps you switch easily between codex model. I've all 4 Oauth already configured, so openclaw agent automtically minitors the usage and switches based on cooldown rates.

[sweet moth]

Can anyone recommend a good stable version number for Mac? Seems like the latest latest has been buggy (launch issues, etc people are reporting them in GitHub). I’d like to try an older version that is known reliable. Also the new security stuff can be a pain — please reply or thread or something so I can find it thanks

[languid bloom]

Thank you. Yes, I was eating up a lot of tokens setting up OpenClaw using Opus 4.6. I'm using the Claude API instead of my Claude Max account, as I heard people were getting their Claude accounts closed.

[shy ether]

I've had more luck with 2026-02-19, but it's apparently bugged so you can't run subagents, so.... I don't think there is a more/less stable version. I think it's just how the holes in the cheese line up on as to whether it suits you or not...

[sweet moth]

Totally I have like .9 running on another machine and it hums… but it’s missing so much cool new stuff. Bummer was really hoping we’d get something more stable. I guess I’ll just codex the crap out of the src locally and get it stable

[shy ether]

I guess I need to make another OC instance and try .9

[vocal oriole]

dont use your personal account - create new gmail and signup for openai subscription. At worst they will ban your account - just create a new one. lol.

[vocal oriole]

but openai wont ban anytime soon - you're good for antoher couple months till they release their own agnets then might be some updated guidance on usage.

[blissful wren]

anyone can help ? after update i always got this error - plugins.entries.google-antigravity-auth: plugin not found: google-antigravity-auth

[languid bloom]

Any idea why my new .md files that I had my Claw assistant add to my system are not showing with the rest of the default ones?

[fresh cosmos]

my claude account started going "oauth not supported" today. didn't ban the whole account just that token. (oh well back to pay as you go, need the claude code)

[tulip falcon]

Is anyone using openrouter (free)? It was working amazing yesterday and I had 500+ requests.

It has completely stopped working.

I’m able to curl api models to see they are running but inside of openclaw they are rate limited

[fresh cosmos]

for the first time in a few weeks, i stopped and went to bed at a reasonable hour. woke up with a clear head. worth doing! stopped and restructured the docs/skills/memory/git sync/ release process across the cluster as a result. would recommend!

[shy ether]

You probably need to reinstall plugins on update.

[vocal oriole]

prolly in some different path where it was saved. but 99% they should be somewhere in your .openclaw folder. if you're using mac and can't find it in users, do CMD+SHIFT+. (period)

[fresh cosmos]

prolly in your works space/docs folder?

[vocal oriole]

yeah i have learned the hard way - the more we interact doesn't make it better. sometimes just gotta let it do - and be patient.

[languid bloom]

I'm on Mac, so I'll see if I can find it.

[vocal oriole]

another thing i have started doing is instead of telling it what to do, I just ask it what would he do? and the results are crazy - it thinks through problems you didn't even think about so i let it think, and if it doens't include the directive i want, then i improve.- highly recommend.

[fresh cosmos]

yup also go use chat gpt research every now and then, get it to review to objective and design, feed that back to your claw (codex|opus)

[sweet moth]

Here is what GitHub copilot says

Currently, OpenClaw has several unresolved issues across versions as highlighted in their GitHub issue tracker. However, version 2026.2.15 appears to be one of the last working versions without severe bug reports, as newer versions (e.g., 2026.2.22-2) are encountering critical regressions.

[grim plover]

how can i connect my openclaw to internet

[shy ether]

If copilot is to be trusted, yeah, I guess try .15

Half the time, Grok doesn't seem to know what openclaw is haha

[grim plover]

I want my openclaw to do research and searches on the internet for me. How can I do that?

[sweet moth]

Google openclaw brave search api key

[storm loom]

Install the Tavily skill. Its free up to 1000 searches or something

[storm loom]

I heard Brave isn't free now?

[tulip falcon]

Is anyone using openrouter (free)? It was working amazing yesterday and I had 500+ requests.

It has completely stopped working.

I’m able to curl api models to see they are running but inside of openclaw they are rate limited

Anyone able to help?

[molten gulch]

@solemn meteor im looking into zeroclaw and it looks like there's at least one other "unofficial" fork that's impersonating the project lmfao

[ionic plover]

Any suggestions for AI integrations? Ofc, cloude is best but its not of the cheap versions

[storm loom]

Flash 3 Preview

Runs like a dream

Back up is Kimi K2.5

[primal pond]

Multi-agent Telegram routing issue — messages not reaching agent

I have a multi-agent setup (v2026.2.12) with two Telegram bots:

• default account → main agent (works perfectly)
• closer account → closer agent (bot receives messages but agent never processes them)

Config:

"agents": { "list": [
  { "id": "main", "default": true },
  { "id": "closer", "workspace": "~/.openclaw/workspace-closer", "agentDir": "~/.openclaw/agents/closer/agent" }
]},
"bindings": [{ "agentId": "closer", "match": { "channel": "telegram", "accountId": "closer" } }],
"channels": { "telegram": { "accounts": {
  "default": { "botToken": "...", "dmPolicy": "pairing" },
  "closer": { "botToken": "...", "dmPolicy": "allowlist", "allowFrom": ["6399743107"] }
}}}

What works:

• openclaw channels status --probe shows closer bot as running, in:1m ago
• sessions_send to agent:closer:main works — agent responds correctly
• Closer bot can send messages via Telegram API

What doesn't work:

• Direct Telegram DMs to the closer bot are never delivered to the agent session
• No error/reject/drop in logs — messages silently disappear
• Tried both pairing and allowlist dmPolicy — same result
• pairing list telegram never shows pending requests from the closer bot
• agent:closer:main session exists but totalTokens: 0 from Telegram messages

Credentials file at ~/.openclaw/credentials/telegram-allowFrom.json exists for default — is there a separate file needed for the closer account?

Any help appreciated. Thanks!

[solemn meteor]

Yea I saw, idk why they even give people warning if you get rekt it’s your fault

Tbh tho lol I’m pretty sure I got prompt injected first few days of trying as I didn’t know wtf I was doing 😭🤦🏽‍♂️

Could openclaw on kali kill the bug?

[late nebula]

any ideas to optimise openclaw to work with ollama . answer to slow. suggest me beter llm model for this purpose

[slim wigeon]

Morning people, I was wondering If anyones come across this issue I'm having.

For the past day I've been getting "⚠️ Agent failed before reply: All models failed (2): openai-codex/gpt-5.3-codex: Provider openai-codex is in cooldown (all profiles unavailable) (rate_limit)"

However when I check logs openclaw is trying to pull gpt-5.3-codex as "anthropic/gpt-5.3-codex" I've obviously not got the oauth for anthropic as I'm using openai, so it's literally just bricking.

Any takers? Thanks in advance.

[solemn meteor]

Try openclaw config and select model

Make sure you select OpenAI-codex?

[slim wigeon]

I did before I've had to do it manually as it was stuck on codex 5.1

I've fixed that bit now anyway

it's the rate thing that's bugging me, and from research it seems to be provider side. I've waited a full 24 hours for tokens to refresh

[grim plover]

⚠️ Agent failed before reply: ENOENT: no such file or directory, mkdir 'C:\home\node.openclaw\agents\main\sessions\C:\home\node.openclaw\agents\main\sessions\C:\home\node.openclaw\agents\main\sessions\C:\home\node.openclaw\agents\main\sessions\C:\Users\SHORU.openclaw\agents\main\sessions'.
Logs: openclaw logs --follow I NEED HELP

what is thids

[echo seal]

A lot of people have run into this. I'm running into it with my in house hosted LLM model too

I just noticed something though: it always complains about the same model name even though I've switched it. I wonder if there's not some config data lingering somewhere that isn't getting updated like it should

[slim wigeon]

Well I thought that so I did a full reset, upgrade and reinstall, still persists, which leads me to believe it's the oauth on the openai provider side that's causing a rate limit error. I've been seeing more and more people talking about being "banned" from using api keys when connecting to openclaw

[echo seal]

Can't be. I am hosting my LLM on ollama internally. No oaauth.

[slim wigeon]

Oh yeah of course, I miss read. Are you getting the same output as this? "Agent failed before reply: All models failed (3): openai-codex/gpt-5.3-codex: No available auth profile for openai-codex (all in cooldown or unavailable). (rate_limit)"

[echo seal]

BTW I'm running ollama on a dedicated windows machine with an rtx3070

[molten gulch]

@solemn meteor update: seems like it's working, my chud child thinks it's a rust powered crab now. though i can't for the life of me figure out the web gateway pairing.

and the zeroclaw docs are even more lobotomized than openclaw, somehow

[echo seal]

It's always the same. I will get booted and it's telling me that it can't find a model I'm not using. Since I used to use that model, I'm thinking there's a lingering setting somewhere that's not getting updated.

[rare prawn]

best modell to run as main?

[echo seal]

I have been doing a poor job of keeping track of my versions and when issues start.

At first I thought it was because the models I was trying to run were too large, but it does seem to be OC

[late nebula]

hi there .can i ask you: when use ollama for openclaw did the agents work to slow . did you do some optimisation

[echo seal]

Before you can ask that, you have to set expectations: an in house hosted model is not going to run on Ollama/vLLM like it does from a provider. Not unless you've got a shoebox filled with diamonds to trade for a machine with a couple or 4 rtx5090s in it. That being said, I am running mine on a Win11 machine with 32gb, mem, can't remember proc, rtx3070 (pre tensor cores). I am running the little ministral-3 and it's sort of ok. I give ithe agent pointing to it tasks that I don't mine letting churn for long periods.

[late nebula]

may be you a newbie in openclaw knowlige. i did the same problems because edit openclaw.json with nano. now i use webchat interfaca to edit configuration file

[echo seal]

Also please note that I woke up early my time with a sinus headache and I haven't had coffee or poptarts yet. My judgement is quite suspect at the moment

[echo seal]

I lied to OC and told it that my ollama is vTTL. I didn't need to edit the json

[echo seal]

I very much am. I dived in because I was charged with the engineering side of designing an on prem solution forsensitive data analysis.

[late nebula]

what is the last problem you have to decide?

[echo seal]

Right now I'm testing its ability to troubleshoot apache2. I'm having it write commands and background info for me to read and execute. That way speed isn't an issue so much and it lets me see more of what its thinking

[late nebula]

what model you use. how many seconds cost to give you an snswer in chat?

[solemn meteor]

Took me 6 prompts to figure out to /reasoning off in discord for 3.1 😂😂😂

[signal prawn]

Hey

[latent wadi]

what models do you use? how do you get something powerful for like 100 usd per month

[fresh moon]

hey guys does skrill not help anymore

[oak dragon]

Does anyone have any advice for generating visual content? What platforms or tools is your Claw using?

[blazing arch]

Hello, what do you think of Sonnet 4.6 compared to Opus? Not a huge difference in benchmarks, right?

[pastel thicket]

why my openclaw failing every compact when context gets too big?

Always stuck in "I need to summarize our conversation — the context has grown too large. Please wait for auto-compact to complete, then repeat your last message." loop

[echo seal]

Right now I am playing with the ministral-3 tiny one. I do not sit and wait for answers. I send it off to do its thing and go off to do other things. If I want it fast I go to gemini

[subtle yacht]

hello, i need some help, testardey, discord stoped working with claw... any ideea? my version is 2026.2.22-2

[inland iris]

Hello ! Can you help me run Node.js OpenClaw?

[pearl moss]

i bought a m4 mac mini, 512GB 24GB ram. What local model i can run and still use openclaw effective? Mainly for chatting model and not complex task. I want to automated comment reply etc. Qwen ?

[wary osprey]

Free tiers on OpenRouter are usually burst-friendly, not sustained-load friendly. If you did 500+ requests yesterday, you probably hit a dynamic throttle. even if /models responds, request-level rate limits can still block inference calls. inside OpenClaw, that gets amplified because agents can loop tool calls quickly.

If you need something more stable for production use, I’d consider either;

• Running a paid tier directly with the provider
• or using a unified layer like AIsa so you’re not relying on a shared free pool

Free endpoints are great for testing, but once you automate workflows, stability matters more than free

[wary osprey]

I’d look at:

1. Qwen 2.5 7B Instruct (quantized, e.g. Q4_K_M)
2. Mistral 7B Instruct
3. Llama 3.1 8B Instruct (quantized)

All of those should run comfortably via Ollama.

If it’s mainly comment automation and short replies, 7B–8B class models are more than enough. You don’t need a 70B model for that.

if later you want higher reasoning quality but don’t want to overload your Mac, you can Use local for basic tasks, route harder prompts to a remote API via OpenClaw

[pearl moss]

THanks for the reply! ❤️

[wary osprey]

Opus 4.6 will burn tokens fast if you’re using it for setup, debugging, and iterative testing. it’s one of the most expensive tiers. for OpenClaw setup and automation testing, you usually don’t need Opus-level reasoning.

Sonnet or even a mid-tier model is often enough until you’re in production.

• Cheap model for setup and tool debugging
• Premium model only when needed for high-quality reasoning

If you’re concerned about account stability and rate limits, using an API-based setup with proper usage tracking is generally safer than running everything through a consumer account.

If you want to simplify multi-model access and cost tracking, then AIsa can help since it aggregates multiple providers under one endpoint and centralizes billing.

this way you’re not managing separate dashboards while building.

[dusty tinsel]

Hi everyone 👋

I’ve been running into context limit issues frequently when using OpenClaw with the deepseek-chat (128k) model.

After enabling automatic context compression, task coherence sometimes breaks and results become inconsistent.

My main question is:

What’s the best practice to handle long-running tasks in OpenClaw without being heavily affected by context limits?

Any advice would be greatly appreciated. Thanks!

[fringe dust]

Did the telegram problem solved?

[woven oasis]

no, my telegram issue still there after update

[kind oracle]

A new version of OpenClaw is available!
OpenClaw 2026.2.15 is now available-you have 2026**.2.23**. Would you like to download it now?

[whole panther]

Guys, how do i give openclaw vision to see my windows desktop, open files everything, brand new laptop

[grim plover]

Agent failed before reply: All models failed (7): zai/glm-5: No available auth profile for zai (all in cooldown or unavailable). (rate_limit) | zai/glm-4.5: Provider zai is in cooldown (all profiles unavailable) (rate_limit) | zai/glm-4.6: Provider zai is in cooldown (all profiles unavailable) (rate_limit) | zai/glm-4.6v: Provider zai is in cooldown (all profiles unavailable) (rate_limit) | zai/glm-4.7: Provider zai is in cooldown (all profiles unavailable) (rate_limit) | zai/glm-4.7-flash: Provider zai is in cooldown (all profiles unavailable) (rate_limit) | zai/glm-4.5v: Provider zai is in cooldown (all profiles unavailable) (rate_limit). what is this ?

[molten obsidian]

zai had a outage I think

using them too and got an emai

[woven oasis]

telegram fix:

Root cause: commit 0bd9f0d4a changed mergeAllowFromSources() to discard credential-file entries when dmPolicy === "allowlist", leaving the effective allowlist empty since our config had no inline allowFrom

Fix applied: added "allowFrom": ["TelegramUserID"] to openclaw.json.template — forward-compatible with the intentional strict enforcement

[grim plover]

what is auth provider cooldown