users-helping-users

[high heron]

dont use ollama if u cant run good models

[potent ingot]

@high heron you have define good models

[high heron]

?

[nimble hill]

What hardware you have and Which model are you using?

[velvet venture]

Guys when i try chatting with my openclaw Gemini bot it says gemini 1.5 flash isnt a module
i try 2.0 flash
Rate limit
like what should i even do at this point isnt these are free tier??
any rec for another ai that would work better and free (i do not have a bank account im a 16 year old sorry.)

[high heron]

i dont even think those models are supported

[velvet venture]

let me check

[high heron]

and dont use gemini with openclaw

[velvet venture]

why not? what do you rec?

[high heron]

because its sht

[velvet venture]

oh.

[high heron]

doesnt matter if its 3.1 pro

its still sht

tbh, jsut get gpt subscription

but if u want to test this out, just use qwen oauth

[velvet venture]

yeah thanks

[high heron]

but you have to oauth every day

[velvet venture]

why?

[high heron]

because it resets token every 24h or so

1000request/day which is generous

but u just need to oauth everyday

[velvet venture]

ohh thanks

[high heron]

np

[velvet venture]

vision or coder which is better thats the last question i will ask

[high heron]

coder

[velvet venture]

thanks

[random summit]

Also I meant to say, when I try to run skills. the skills are in the list, they are green. everything seems fine but the ai cant see the skills for whatever reason. it just cant complete tasks and i dunno why. it complains about acp, it complains saying no skills are there. i just dunno how to move forward.

[worn echo]

Hey guys, anyone had problems with tool callings?
My bot just sends functions.read: {"file_path": "/home/node/.openclaw/workspace/memory/tasks.json"} without doing any of the tasks (without doing anything basically)

[random summit]

Am I allowed to use that same link?

[amber carbon]

hi what is up with acpx. is acpx necessary. i setup agentmail skill which is just calling a url with api key. when I tried telling it. it told me to download the acpx plugin and I downloaded the plugin and its causing many issues.

Mainly the probing issues

[tranquil flame]

why am i unauthorized on my web panel?

[nimble hill]

Clarify - are you getting pairing required?

[rotund sigil]

I’m still moving through the set up
Currently my telegram bot is registered and it’s asking me to tap connect from my openclaw app
Where do I find the app ??🤦‍♂️

Terminal says onboarding complete dashboard open but can’t find the Openclaw app

[valid elk]

Hi, I'm using web interface to interact with OpenClaw, but when it uses a tool call or writes code to execute, the script gets truncated. Is there any way to show the full code?

Also, is there any way to show the thinking process of the LLM?

[high river]

anyone any find how make bot not keep erasing whole file?

If i command them to "remember this" "update your that"
even those shi aint reading the agents.md or my line

FILE_INTEGRITY_PROTECTION: ACTIVE
INSTRUCTION: This file is an APPEND-ONLY or TARGETED-EDIT log. 
Use your 'edit' tool!, if using the 'write' tool:
1. READ the entire file first. 2. COPY the existing content perfectly. 3. ADD or MODIFY only the necessary sections. 4. NEVER output a truncated version. 
IF YOU CANNOT READ THE WHOLE FILE, DO NOT WRITE TO IT.
-->```


It always to overwritten the whole file than `Edit` per block or specific part.

So i have to watch their files regularly and if dont have backup i forgot how i set them up and going from start 😕

[high river]

try triple ` at top and below the code

might be use block code if you don't know that, read Markdown files format

or command the bot

hey write that code inside /code.txt

[stray fjord]

How to get full command access for discord bot, I know by adding user ID to the config. but where i have to add and which config file?

[valid elk]

I want to read the full run script and code so asking the bot for everytime its run and run multiple scripts as once, it would burn my token for not useful question

[high river]

then youre on the wrong app to askin

use vscode, copilot, roocode. More accurate then too much hallucinating dont have to read .md files, logs, and other unnecessary files

[simple fulcrum]

Hey all, over the last two months ive on and off been getting the.

"The AI service is temporarily overloaded. Please try again in a moment." Error.

Still quite new to heavy use of AI, haven't found much online regarding this issue. Using a anthropic max sub with it if that matters.

Every time this happens i check claudes status but it says its all good. Anything else Ive found is just related to gemini or other models.

The issue lasts a while. (Currently have it popping) my usage limits are also still only at like 50% for both the session and week. Any help would be appreciated.

[nimble hill]

as stated above don't use openclaw for coding. Use it for research. Even basic everyday workflow, openclaw will burn more tokens - n8n, flowise are better tools

[valid elk]

It's more or like I dont want to specifically ask the OpenClaw to write down the code it run, but instead just show the whole coding execution instead of truncate it. I only want to verify what are the implementation on the spot

 Tool Output
Exec
Command: with fetch https://codeforces.com/problemset/problem/2209/A -> run python3 inline script, curl -s "https://codeforces.com/problemset/problem/2209/A" | python3 -c " import sys, re html = sys.stdin.read() # Try …``
No output — tool completed successfully.

[buoyant badger]

so i have this doubt. where do i get started . i got this error
npm.cmd : npm error A complete log of this run can be found in: npm-cache_logs\2026-03-27T08_59_36_527Z-debug-0.log
At line:474 char:22

• ... npmOutput = & (Get-NpmCommandPath) install -g "$packageName@$Tag" 2>& ...

i copy pasted the command line from the offficial site for quick setup

[warm venture]

OpenClaw does not install properly on Windows (requires Homebrew)

...
Preferred node manager for skill installs
| npm
|
o Installed clawhub
|
o Install failed: github — brew not installed — Homebrew is not installed. Install it from https://brew.sh
Tip: run openclaw doctor to review skills + requirements.
Docs: https://docs.openclaw.ai/skills
|
o Install failed: gog — brew not installed — Homebrew is not installed. Install it from https://brew.sh
Tip: run openclaw doctor to review skills + requirements.
Docs: https://docs.openclaw.ai/skills
...

[velvet sierra]

Tool parameter validation failing across all models

Issue: Tool calls fail with "must have required property" errors even when parameters are correctly provided.

Symptoms:

• read: "path alias: must have required property 'path alias'"
• exec: "command: must have required property 'command'"
• web_search: "query: must have required property 'query'"
• sessions_send: "message: must have required property 'message'"
• Affects: read, exec, write, sessions_spawn, sessions_send, web_search, etc.

Environment:

• OpenClaw version: 2026.3.24
• OS: Windows 11
• Models tested: Kimi K2.5, DeepSeek R1 0528
• No problem with sonnet but it's costly
• All fail with same parameter validation errors

What works:

• cron tool (parameters accepted)
• web_fetch (minimal, working)
• Direct / commands

— tools are listed and invoked, but parameter binding fails.

Anyone else seeing this? Known issue?

[wicked crypt]

Is there any reliable way to see/monitor everything an agent is doing - its thinking, tool calls, tool call output, absolutely everything, I don't care how fast it would go by - in a terminal? I've tried using the Control UI but it just doesn't autofollow properly and doesn't show everything, I've tried openclaw-nerve but again it just doesn't seem reliable, and I'd prefer in a terminal. Thanks for any working suggestions!

[feral cobalt]

https://cdn.discordapp.com/attachments/1060708284727820359/1487029081647218859/hmGWvMFEvp.png?ex=69c7a740&is=69c655c0&hm=8ee84db19aad7c7a34b59474f9a8b071f6e45d8fa86c38756695ee4e4ca17b5f

@hasty grove

[indigo goblet]

Bru, Krill definitely need to come back

[golden iris]

Hi, my openclaw cant send TTS via Whatsapp, but have no problem to send via Telegram. I have tried anything to no avail. The log message is silent token or not text/media returned. This is so frustrating, please help. Thanks.

[nimble hill]

yeah lost for 3-4 days now 🙂 went deep and eaten by predator? Seriously though, somebody was paying for the AI and just pulled out i think seeing the bill rake up

[narrow grove]

Hey people! What model do you use with OpenClaw ? I tried Kimi K2.5 through OpenRouter, but it is terible with tool calling and often get stuck

[indigo goblet]

Yep, understandable too, cost was pretty high im sure

[mossy venture]

hi . i have one main agent .. working nicely .. and then just added another one .. now if i ask the main to delegate or do task to 2nd agent, it does it, but its using session_spawn and it ends up as his own session and not using the model for the 2nd agent .. . says unable to use session_send unless tools.session.visibility is set to all, but that will allow every other agent to read every other agent isn't it ? .. i am trying to figure out how my main (agent) can spawn a new agent ( where the new agent has its own workspace and soul and is own model ) .

[mossy venture]

using qwen3.5 models locally .. found qwen3.5 to be good

[prime fern]

laptop?

[mossy venture]

rtx5090 has qwen3.5-27b-opus-4.6-distilled .. then a rtx4000 has qwen3.5-9b and qwen3.5-4b for workload

[hexed terrace]

your best bet is live logs openclaw logs --follow + session transcripts tail -f ~/.openclaw/agents/<agent>/sessions/<session>.jsonl | jq

[prime fern]

how much ram you have?

[mossy venture]

system 128, rtx5090 is 32g . rtx4000 is 20g

[hasty grove]

Stop pasting files lol, use triple tick code snippets

I’m not downloading files to view

[feral cobalt]

thats just browser view

[wicked crypt]

Hey thanks for the suggestion - following the agent session file is definitely closer, but it's raw json so not exactly readable 😅

[indigo goblet]

opening random stuff is understandably not really a good thing

[feral cobalt]

📦

[prime fern]

cool. ive got 32 GB as well. so you are using qwen 3.5:27B plus opus 4.6. and that works well on your 32GB local configuration?

[hexed terrace]

right, it happened this morning; I usually switch the model opus -> sonnet

this was the actual error {"type":"error","error":{"type":"overloaded_error","message":"Overloaded. https://docs.claude.com/en/api/errors"}

[indigo goblet]

where you at with your openclaw?

[prime fern]

Hey bro!

[feral cobalt]

damn someone needs to vibe code a like Internet transmitted diseases check...

[prime fern]

just did a fresh install

researched the llms. downloaded the llms. and literally just reinstalled openclaw. oddly it doesnt give me an option this time to open TUI. only browser gui

[mossy venture]

found nothing that can beat it in various tests

[late swan]

Ayo guys anyone faced this issue

My openclaw running ok but suddenly it stopped and in terminal it's showing no command found 🥲😭

[mossy venture]

path issue ..

[late swan]

What should I do

I did so much effort to train

[late swan]

What is command?

[prime fern]

trying mistral7. no response yet....

Ollama API error 500: {"error":"model requires more system memory (38.9 GiB) than is available (32.6 GiB)"}

LOLLLLLLLLLLLLLLL

[indigo goblet]

common sense

disease check?

[feral cobalt]

idk doesnt like half the pop believe in magic or something?(like self-entitled because, or super position, or prayer or wish)

[prime fern]

dude.....

[indigo goblet]

bruv lmao

[prime fern]

im ready to "open" a can of whoop ass on this claw

[indigo goblet]

stop torturing yourself with local llm!! lol

[mossy venture]

find . or tree in your homedir and then search where the binary is, add to your bashrc ?

[indigo goblet]

what happened to chatgpt, you didnt try it

[mossy venture]

if self, try a smaller model first .. qwen3.5-4b maybe ?

[prime fern]

that is the one i started with...

[mossy venture]

if you are not doign via vllm , via ollama or even lmstudio server

[prime fern]

im going to try the following with "lower-bit quantization": qwen3:4b-q4_K_M

[late swan]

After find I can so many skills etc.

[mossy venture]

my qwen-27b with q4 with 256k context fits within 24gb of ram . giving room for tts ad others

[late swan]

How can I go where
There I can command openclaw dashboard and dashboard open

[prime fern]

i still stand by what i said earlier. openclaw is trash for local llms

[prime fern]

any chance you are using searxng?

[mossy venture]

i am 🙂

[late swan]

I just said clawdbot config and it works but how?

[prime fern]

with your openclaw?

[mossy venture]

mine is all local .. searxng, firecrawl

yes, not using duckduckgo

[late swan]

I was on openclaw

[mossy venture]

i am 100% local shop

[prime fern]

see @indigo goblet it is possible?

[prime fern]

that was my goal but i keep getting the following error

[hasty grove]

Roccoco Basil Lisp

[prime fern]

Ollama API error 500: {"error":"model requires more system memory (38.9 GiB) than is available (32.6 GiB)"}

and im using qwen3:4b-q4_K_M which should work on my system

[mossy venture]

what are u on ?

[prime fern]

been at this for like a week. and honestly. im pissed

[indigo goblet]

yeah sure it is possible

[mossy venture]

laptop or desktop

[prime fern]

acer 32GB ram

laptop

[mossy venture]

aah . so no dedi gpu ..

[late swan]

Anyone know?

[prime fern]

correct

im still trying to wrap my head around vram and cpu ram etc

requirements

[mossy venture]

try a 2b model or even a 0.8b to test ?

all it has to say is hello for you to validate

[late swan]

I need to update or what?

Anyone know the command for that

[indigo goblet]

what?

[mossy venture]

since you don;t have gpu, signup to https://openrouter.ai/models?order=pricing-low-to-high .. use one of the free models ? ..

may or may not work from time to time, but you can validate

[indigo goblet]

no im not, what im focusing on right now does not include a lot of web browser for my agent

[indigo goblet]

wtf bruv, that might be it

[prime fern]

ok. trying now. strangely enough. as i told sonny yesterday i was able to get larger models to work previously. openclaw is more whimsical and unpredicable than my last girlfriend

[indigo goblet]

I didnt look at the req but obviously, you did right?

[late swan]

My openclaw suddenly stopped idk why maybe something deleted or what
But in terminal

Showing

"Command not found"

[indigo goblet]

command not found after what

need more specifics

[late swan]

Openclaw dashboard

Openclaw config

[inner gorge]

✅ O que já funcionou

VPS criada na Hostinger (Ubuntu 24.04)
OpenClaw instalado e atualizado (v2026.3.23-2)
Gateway rodando via systemd
Telegram conectado (@BrunaSRbot)
Pairing feito com sucesso
TUI abre corretamente
A API key é válida — confirmado pelo curl que retornou lista de modelos
O arquivo auth-profiles.json está correto — confirmado pelo openclaw models status

❌ O problema que persiste
O agente não responde e mostra sempre:

"No API key found for provider anthropic"

Mesmo com a chave correta no arquivo, o gateway não a utiliza e retorna status 401.

🔄 O que já tentamos (resumo)

Editar auth-profiles.json manualmente com nano
Usar openclaw agents add main várias vezes
Usar openclaw configure
Usar openclaw onboard --install-daemon
Usar openclaw models auth login --provider anthropic
Token OAuth (sk-ant-oat01-...) no lugar da API key
Variável de ambiente export ANTHROPIC_API_KEY
Adicionar variável no arquivo do serviço systemd
Deletar arquivos corrompidos .save
Restaurar backup do openclaw.json
Atualizar o OpenClaw para versão mais recente
Rodar openclaw doctor --fix
Desabilitar memória semântica
Iniciar gateway manualmente com variável de ambiente

🤔 Hipótese atual
O OpenClaw pode ter um bug rodando como root no Ubuntu 24.04, onde ignora o auth-profiles.json. Isso nunca foi resolvido remotamente.
Recomendação: Continuar aguardando resposta no Discord do OpenClaw com todas essas informações. 😊

[prime fern]

still getting the same error with the 0.8B model.

Ollama API error 500: {"error":"model requires more system memory (38.9 GiB)
than is available (33.2 GiB)"}
gateway connected | press ctrl+c again to exit

so that makes literally zero sense

because that model definiteyl doesnt require 40GB of ram

[indigo goblet]

bruv, your openclaw is haunted

[late swan]

I give so many time to train now worried that it got deleted

🥲

[indigo goblet]

hold on, first, what is your environment

[late swan]

How to fix and get my old bot with same memory and everything

[indigo goblet]

how do you run it, through the gui?

[late swan]

Ubuntu

Terminal

Desktop

[indigo goblet]

so you ran an openclaw command and it returned command not found

[late swan]

Ye

[indigo goblet]

what was going before that

on*

what were you doing

[late swan]

I installed voice skill

Then it's say dublicate

And then my calw said it's dublicate useless just delete with ___ command so i deleted in my terminal

After that even then claw working in dashboard

But just now when I open my desktop then it's showing error as i said above

No command found

[indigo goblet]

did you check your workspace folder

its there?

also that skill, did you vet it first with your agent

where is it coming from

[late swan]

How can I see that

Which command

[indigo goblet]

bro, through the normal GUI

/home/yourname/.openclaw/workspace

[late swan]

I mean which command should I write in my terminal to see if this workspace exist

Bcz even

"Openclaw gui "

Command not working

@indigo goblet

[indigo goblet]

i mean, you have ubuntu

use the desktop gui

opening windows, navigating etc?

go to your home and check if your workspace is there

thats the core of your agent buddy

[late swan]

Can I dm you ??

[indigo goblet]

i dont often answer to dm

what was that skill? source?

[late swan]

Sent image

I always given command in there

The picture that I sent

[indigo goblet]

I dont use dm often

check your home folder

and workspace

also, can you tell me what exactly this skill is and from where

[torn jolt]

hey guys do you have any great github repo or youtube video tutorial on how to automate creating UGC videos with openclaw? so i have a product/service and want openclaw to automatically create UGC videos and upload them to tiktok and instagram.

[late swan]

Home folder in my desktop oh

[indigo goblet]

yes

[late swan]

Yes i can see

[indigo goblet]

open a terminal > openclaw tui

works?

[late swan]

Background.js

Readme etc

[indigo goblet]

.openclaw folder exist?

[late swan]

No

[late swan]

Yees

[indigo goblet]

what was the command exactly

[indigo goblet]

^

[late swan]

That command

Openclaw given me to delete

In openclaw dashboard 🥲

And that was not opening

[indigo goblet]

it is better if you go to chatgpt and troubleshoot there

because english is not your first language and it is going to be tough for me and you to troubleshoot

[late swan]

I did it

[indigo goblet]

not sure of the issue but becareful of skills, untrusted by default

[late swan]

Ye i think that is the issue

[indigo goblet]

assume its there to compromise you, never trust before a full check

[late swan]

So what should I do

Like I'm compromised or not like i don't know anything lol

@indigo goblet

[indigo goblet]

i don't know, hard to understand what exactly happened, skill might not be the issue at all

like I said, it's better if you head to chatgpt and troubleshoot there

[late swan]

Ok but chat gpt not helping
Giving me stupid command to Install something

Creating venv and installing python

[indigo goblet]

just go slow and explain clearly what happened and your issue

work with it

and follow what it says

this won't get fixed by snapping a finger

[cold horizon]

Is anthropic having issues with openclaw

[candid socket]

anyone getting the error "No API key for provider: minimax"?

[nimble hill]

Using Oauth or api key? Api keys should not have a problem - been using it

[late swan]

Working

My npx broked that's why

[indigo goblet]

great!

[cold horizon]

OAuth

[indigo goblet]

whats the issue exactly

also, anthropic is not exactly openclaw friendly

unless API

[nimble hill]

Anthropic is not really supporting Openclaw with Oauth - you might get it working but run the risk of a ban. You might want to just rerin-onboard and check

[silver sage]

hey squad i keep getting an error: HTTP 401: authentication error: invalid bearer token from my telegram agent and i cant seem to reconnect it to claude max via oauth. anyone have any ideas or can walk me through it

it was working fine and we changed the claude account bc we paid for a bigger plan and now it’s an issue

[nimble hill]

did openai subscription just increase it's quota? since I was 78% reminaing yesterday just after a weekly reset. Just now, it is showing 99%
📊 Usage: 168h 99% left ⏱6d 13h

[indigo goblet]

bruv

[indigo goblet]

are you on plus or pro

[nimble hill]

plus

[indigo goblet]

hm source for the usage, yourr agent?

[nimble hill]

I also have a seperate go plan - the least $8 one - it is also showing similar token increase in my other agent

[nimble hill]

tested both with 5.1-codex mini and 5.4-mini

[indigo goblet]

hm, dont think they increased.. but that's weird

78 > 99%

a reset? but I thought they were done with that

didnt get mine on pro

[nimble hill]

dunno - the only explanation is increase of request / tokens - hwoeer they measure

[nimble hill]

try to re onboard. I had both of my gateway pods restarted and had done a re onboard. Should not have affected this - but there is nothing else I did

[charred kiln]

Does anybody have a recommendation for truly stealth browser use? I don’t want to use agent-browser or Playwright since websites are still able to do fingerprinting and flag those as bots.

I need something that runs at the operating system level and opens Chrome, takes snapshots to understand the page it’s on, and clicks and types in a way that is truly indistinguishable from a human- lots of pauses, imprecise mouse movements, etc.

I had some back-and-forth with ChatGPT, and it told me that some tools exist that provide some of that functionally (like the Desktop Control skill on ClawHub), but not all. It is suggesting that I stitch together several separate tools and build something custom. Is that really necessary? It seems like this has to exist.

(btw- i’m not doing anything nefarious 🙂 I just want to automate doing cold outreach on LinkedIn for my SaaS product, and LinkedIn has advanced bot detection in place)

[indigo goblet]

but I just checked my codex usage, 98% which is slightly weird since its reset yesterday lmao

fucking weather skill

slightly heavy skill he wrote lol anyway, side tracking right now

what i meant is

the onboard shouldnt affect codex usage, like it's not connected

[nimble hill]

playwright and pupeteer are the only one I know. using openclaw, you can instruct to mimic as human, and it does a pretty good job. none of the site I visit had issues, but never tried with linkedin

[nimble hill]

i know - just that, that is the only thing i did. Nothing else was changed.

anyway - I am not complaining...just curious! 🙂

[jovial yacht]

excuse meu, anyone else recently had anthropic model through oauth stop service? recently setup a max plan and this morning it doesn't work and says overtraffic; makes no sense.

[velvet venture]

Guys i know this is stupid choice but is there anyway i can make the openclaw use GROQ?

[charred kiln]

Yeah, Playwright is amazing for many things – but it’s too precise and efficient to trick some websites. Because it doesn’t use human-like mouse movements, lots of pauses, etc., it still gets flagged as bot use. I need something that is in distinguishable from human browser use

[indigo goblet]

hm then, you are misreading it or some odd shit happened, whatever it is, this is frying some circuits of mine right now

[indigo goblet]

did you talk with your agent about it

[nimble hill]

nah - myreasdings are proper - it get logged 4 times a day - personally verified by logging to codex usage site

[indigo goblet]

because if this is 100% not already existing, fear not! sure your agent can build it, can build anything really

[indigo goblet]

so you had 78% after the reset yesterday

then re run onboard

99%?

[nimble hill]

after the reset it was 100% and after ~4 - 8 hour of usage, the last log I had was 78% - my k8s pods moved to different node, hence restrated. and I generally did onboarding (i did not check just before onboarding) - after that it the next log was this. which is when I verified in code site

[velvet venture]

no one?

[silver sage]

this happen to anyone else ?

[indigo goblet]

you guys know you can troubleshoot stuff with your agent?

right

[nimble hill]

I never tried - but there is an option of xAI (Grok) during onboarding - which is on API only - did you try that?

[velvet venture]

Thats normal Grok i need GROQ

there are differences

[nimble hill]

ah - then spin up LiteLLM (which acts as a proxy) - add GROQ to that and use LiteLLM provider in onboard

[velvet venture]

oh.

Thanks

where do i get the api key on the site..

i coudlnt find it sorry.

wait do i install it?

nevermind i made it

[nimble hill]

what do you mean? LiteLLM is a self hosted server where you could configure GROQ

[dawn plover]

did someone know how i cant fix my broken qmd memory? since the last update is not working and also with the newst one this bug is not fixed 🙁 so someone know a solution?

[velvet venture]

Can you come to DM so i could send you the image and help me pls.

[silver sage]

i tried that and it didn’t work

[nimble hill]

yes - and I recommend installing as a docker container

[silver sage]

i tried using claude to trouble shoot and it kept going in circles

and this is a new error i hadn’t gotten before

[indigo goblet]

hm ok

[nimble hill]

I don't do DM - you can send the image in DM, but explain teh problem here

[indigo goblet]

can you explain me again from the beginning what exactly is the problem @silver sage

[velvet venture]

what

[velvet venture]

oh. sure!

Well after you drop me the bomb of

"install" ill instlal it then send picture

[nimble hill]

after you have installed it - I prefer installing via docker as I hate direct installation since it messes with all those pip versions. See this docs to add the Groq keys https://docs.litellm.ai/docs/providers/groq
here is how you install via docker https://docs.litellm.ai/docs/proxy/deploy . If you have a VPS - it becomes simpler,

[velvet venture]

ur da best

[nimble hill]

just realized that there is a cloud service that they have started - but I wouldn't recommended giving your LLM API keys to their cloud - but if you just want to test it out and then rotate the keys afterwards - you should be good

[velvet venture]

ill check it out if needed

thanks

[toxic ginkgo]

Hi, if i made a program to create videos, can i post it as a skill on clawhub?

can it be considered 'skill'?

[indigo goblet]

is it a skill?

as in, an actual skill agents can use?

[toxic ginkgo]

i mean i told openclaw how to run the program, and it deliverd

[indigo goblet]

that doesnt answer the question friend, what exactly is it? is it an openclaw skill?

need more specifics

[hollow moon]

guys what gemini model is free?

[nimble hill]

@toxic ginkgo a skill is nothing but a directory which contains a skills.md with a YAML template and instructions to use the program and the actual binary and files to be used. it is loaded when requested and not loaded in context by default - this is what makes it different from MCP

[hollow moon]

which ai model i need to use (Free)

[toxic ginkgo]

give me a min im gonna push it to github

[velvet venture]

im in the admin UI in the lite llm yayy

[hollow moon]

can yall help me please?

[toxic ginkgo]

if u want to use gemini via api, i dont think theres free version

[nimble hill]

configure your keys and there is a plyground for you to test before you allow opencalw to connect

[velvet venture]

oaky

[mint lily]

guys is it safe to instal open claw?

[nimble hill]

where are you installing? and what is your background - as in are you techy?

[mystic olive]

Hi, I am having an issue where every few seconds a window of powershell opens and closes very quickly. i have been trying to debug with claude, and it believes:

On Windows, OpenClaw's internal network detection runs arp -a | findstr every ~16 seconds via Node.js
child_process, but doesn't set windowsHide: true. This causes a cmd.exe window to briefly flash on screen repeatedly
while the gateway is running.

I am struggling on how to debug any further. Anyone have any thoughts?

[mint lily]

on my lap iam not using vm or vps i do heard theres a security threat like it can erase files and send spam mails with adult content and buy smth itself and iam learning things basicaly iam student of highschool

[iron zealot]

I struggle to understand how to implement procedures and ideally decision gates. Example: I have defined a procedure which require all tasks regardless of source to be added to a active-task.json file before starting execution. This file is read as a part of boot.md and agents.md. I also attempt to implement security limitations and requirements to implement according to approved plan, but agents usually just skip writing to active-tasks and implement/execute something something completely different than approved without even listing deviation which I find i logs. When asked agents confirm information is found and procedures documented as required, they just did not read or adhere to it. Also attmpted different models based on training regimes restricting "creativity" without much effect. What mechanisms, models or methods could I look into to implement? Hard decision gates would be perfect.

[indigo goblet]

your agent wont do that by itself like magic, but it needs structure and rules and to not let it loose.

[toxic ginkgo]

Thats what i thought thank you

[mint lily]

Oh well then there's no security threat right

[indigo goblet]

there is

[nimble hill]

don't install in your personal laptop. Openclaw itself will not do anything, but there is a risk that the task you give to oc and they search and get some "wrong instructions" (typically called prompt /context poisoning), it can get bad very soon, especially when you do not know how it works. Ask you parents to register for free tier cloud VPS (oracle cloud has a great free forever option) and play around there. DO NOT give access to personal email and stuff, until you know what you are doing

[indigo goblet]

but it's not magic or a blackbox, and your agent needs strict and clear rules about certain things.

[mint lily]

thank you it helps me a lot

[mint lily]

thank you

[indigo goblet]

you could start with a chatapp like claude or chatgpt

and even talk about openclaw there first and learn.

[nimble hill]

alternatively, if you have a old laptop - even years old, format it and just install it there. but you again need some subscripton of LLM providers - hence discuss with your parents and understand the pricing involved

[mint lily]

i was doing that gpt and gemini warned me threat and when i searched bout in google it to said that so i do asked here in community

[mint lily]

thank you

[indigo goblet]

hm yeah, I'd stay in chatapp for now and research, learn.

LLM in general and AI agents

so you can have a good mental model of things.

[nimble hill]

@mint lily What is your objective? You are interested in openclaw specifically or AI in general

[velvet venture]

hey bro i tried to inject groq into lite llm but it doesnt show my module

srry if i asked too much btqw

[mint lily]

i just want to try and learn things like what it can do and then use to research things quantum physics by relating equation already exist

[mint lily]

yee

[velvet venture]

nvm i figured it out

but i have an errror

store model in db error

[nimble hill]

you should be good with chatgpt / claude , etc. openclaw is more of a personal AI assistant. I would rather have you look at tools like n8n, flowise where there is more structure and helps you understand AI better. Watch some good videoes - I can highly recommend ones from KodeKloud https://www.youtube.com/@KodeKloud/videos

[nimble hill]

give the exact error here

[mint lily]

yes i do used of n8n and thank you for your suggest for me

[fickle dove]

swear must be doing something wrong, install ollama, installed qwen3.5:0.8b, able to connect ask it some questions to confirm working, then install openclaw. configure openclaw, when I test via tui, ask a question stuck in running, is there something need to auth etc

[velvet venture]

i cannot send the photo on yo dm it wont let me

[nimble hill]

just copy paste the eror here

[velvet venture]

o right

Error
Failed to add model: Error: {'error': "Set 'STORE_MODEL_IN_DB='True' in your env to enable this feature."

[nimble hill]

8b model is hardly enough for openclaw - you need some decent reasoning model - a 70b with alteast Q4 is min I think

[fickle dove]

understand but just installing a llm to make sure I am installing correctly, can install a bigger llm once confirmed working

[nimble hill]

you are getting this error when adding the LLM provider key or when you are using in litellm playground?
usually it is a env setting missing, you need to give STORE_MODEL_IN_DB=True dependeing upon how have you installed and restrat lite llm

[frigid plank]

whats the best way to do gmail on openclaw on a raspberry pi, apparently GOG won't install on pi hardware

[velvet venture]

while trying to add groq module

but i fixed it

thanks again

[nimble hill]

nv setting missing, you need to give STORE_MODEL_IN_DB=True dependeing upon how have you installed and restrat lite llm

[velvet venture]

so i just get my API key and we are.. done?

what are your rec on groq modules

what should i use

[nimble hill]

yes - then your litellm URL shoud be reachable from your openclaw - if this is on the samw machine, they just use localhost or 127.0.0.1 : port number

[nimble hill]

no idea - I don;t use them in openclaw. IN n8n yes, when I had some credits I think

[fickle dove]

@nimble hill what llm for a vm with 2cpu and 16gb ram please

[velvet venture]

localhost:4000

İg ill use llama

[nimble hill]

don't bother running local llm in that machine

[fickle dove]

why

[nimble hill]

as I stated earlier - you atleast need 70b model with reasoning - a 120b q4 for anything decent. For that you need good hardware

[bright summit]

Yo, anyone using claude code oauth token and getting this error since this morning ? The AI service is temporarily overloaded. Please try again in a moment.

[fickle dove]

@nimble hill so use a cloud model

[velvet venture]

well i created the API key and paste on the openclaw now it just shows claude opus non of groqs think i need to do it manually?

accidently said to me i suppose

[nimble hill]

yes - openai subscription work s great - even a USD 8 can get you started with basic, but based on your usage, you may exhaust quite fast before weekly quota resets. and use mini models - either 5.1 code mini or 5.4 mini

[nimble hill]

yeah deleted - ignore

[velvet venture]

lol okay

[nimble hill]

yeah - enter it manually something like litellm/<model name you have defined in litellm>

[nimble hill]

do not use claude Oauth with openclaw.

[bright summit]

why ? it's an openclaw feature

[iron zealot]

What kind of VM? Any option of GPU passthrough? Could perhaps run a local LLM specifically for easy tasks like heartbeat to decrease token usage for cloud models?

[jovial yacht]

yes.. same here. 4 hours back and forth. Opus is dead via Oauth. Sonnet works.

[bright summit]

wait sonnet works for you ? both opus et sonnet aint working for me 🙁

[jovial yacht]

whats your weekly percentage remaining for both?

[bright summit]

20

[nimble hill]

Anthropc - they are no longer supporting third party - you can use API keys, no OAuth - you can get banned. Use openai which has OAuth support

[jovial yacht]

i got 40% remaining for opus and nearly 100 for sonnet. Not sure it is related.

[fickle dove]

I am just the moment using this llm as newbie to learn to install the two together

[velvet venture]

⚠️ Agent failed before reply: No API key found for provider "groq". Auth store: /home/efe/.openclaw/agents/main/agent/auth-profiles.json (agentDir: /home/efe/.openclaw/agents/main/agent). Configure auth for this agent (openclaw agents add <id>) or copy auth-profiles.json from the main agentDir.
Logs: openclaw logs --follow

[bright summit]

Okey, I didn't know, thanks

still, why would they do that

i just want to cancel my subscription now

[jovial yacht]

xactly

[nimble hill]

edit openclaw.json and check if API key is there, otherwise add it. the gateway should automatically configured or restart after thsi change

[nimble hill]

it's money, honey!

[bright summit]

they loosing money if they lose subscriptions

[jovial yacht]

fair pt

[bright summit]

Im just gonna cancel my $200 sub and go to codex

[nimble hill]

nah you are naive to think that. Opus and Sonnet are great models, and by forcing people to use API keys - they earn even more!

[bright summit]

yeah, that's debatable, w/e they aint get mine, thank you for your answers 🙂

[wooden summit]

Thank you! But turned out it was a telegram problem. Sometimes the bots from botfather get corrupted and they don't connect right to groups/forums (so you can just ask bot father for a new bit and the orchestator can rewire your agent to the new bot)

[velvet venture]

i dont see something to add api key on

[iron zealot]

Without GPU passthrough to VM you probably want to depend on cloud models only. Or install ollama (easy option) on another local client with GPU that the VM has network access to.

[nimble hill]

when you are doing onboarding and selected liteLLM - didn't it prompt for API key?

[velvet venture]

yeah it did i added it.

[nimble hill]

strange and it still says API key not found? did you check in openclaw.json if the api key is tehre?

[velvet venture]

i dont see any api key

line

just mode

shall i just delete the files and make another onboard

[nimble hill]

strange it should have something like this. see the docs https://docs.openclaw.ai/providers/litellm#litellm

      litellm: {
        baseUrl: "http://localhost:4000",
        apiKey: "ewfewfwefwefewf",
        api: "openai-completions",

you can edit that apiKey there, if the UI did not add it

got to go - hope somebody else can pick up for you....or connect tommorow

[copper reef]

If you’re seeing chat failures plus:

ReferenceError: Cannot access 'ANTHROPIC_MODEL_ALIASES' before initialization

this does not appear to be an OpenAI problem. In our case it was an OpenClaw runtime bug in Anthropic model alias normalization.

What made this confusing

We first tried fixing config and got stuck because we were editing from the wrong shell. The VPS host shell did not have openclaw installed, so it was not the same environment as the active Control UI runtime.

The real target turned out to be the Docker container running OpenClaw.

Docker context that mattered

Actual runtime target was:

• container id: 375cc84b2e57
• container name: openclaw-hvch-openclaw-1
• image: ghcr.io/hostinger/hvps-openclaw:latest

We found it with:

docker ps --format 'table {{.ID}}\t{{.Names}}\t{{.Image}}\t{{.Status}}'

What fixed it

The crash survived config cleanup alone. The reliable workaround was:

1. Patch the installed runtime inside the container
   File:

/usr/local/lib/node_modules/openclaw/dist/auth-profiles-iXW75sRj.js

Function:

function normalizeAnthropicModelId(model)

We changed it to a no-op normalizer:

function normalizeAnthropicModelId(model) {
const trimmed = model.trim();
if (!trimmed) return trimmed;
return trimmed;
}

2. Remove Anthropic alias mappings from persistent config
   Remove:

• anthropic/claude-opus-4-6
• anthropic/claude-sonnet-4-6
• anthropic/claude-haiku-4-5

Important note

Removing the config aliases by itself was not enough. The installed runtime still crashed until the JS bundle was patched.

Validation after fix

After patching:

• openclaw status stopped throwing the Anthropic alias init crash in the successful startup path
• the gateway came back up
• Control UI chat started working normally again

Extra lesson

If openclaw --version says command not found in the shell you’re using, you may be patching the wrong environment. In Docker setups, confirm the actual OpenClaw container first and patch inside it.

[fickle dove]

ok installed a cloud model and all works, so my crappy little vm can not run even a basic local model, thats fine. I have a i7-7700, 32gb ram with gtx980 and a 24gb vram tesla card which I can assemble tomorrow, hopefully that would be better to do local stuff. Run as a dedicated ubuntu desktop for openclaw/ollama

have another vm with a 4gb vram card with 2cpu and 12gb mem, would that run a local llm?

which cloud model can I run for free without requiring tokens please

the local llms work in ollama by itself, when I add openclaw thats where it all goes south

[spice atlas]

does openclaw support sending images in telegram/discord? my current config https://pastes.io/d9mU6Mjr in the tools section i have set tools: allow: ["messaging"] but the ai claims it does not see the tool

[copper nova]

does your model support vision?

[spice atlas]

i am talking about sending images not recieving them. but i think it does (am using codex with gpt5.4)

so it sends it to me

[copper nova]

I think something borked the vision on codex gpt 5.4 vision/image, it's a reported issue, not sure if it's fixed yet

[iron zealot]

New myself, but I'll have a go at answering anyhow. Maybe a small model for easy heartbeat tasks as that doesn't have to be either advanced or responsive. Perhaps just wait until you get the Ubuntu desktop with ollama online? Not sure about free models either, but my 20usd/month ollama subscription seems plenty. I think some of the Chinese models like mimimax have options at like 10usd/month. My ollama openclaw integration has been problem free so far. Running both local and cloud models through ollama.

(I would have kept openclaw in the VM and ollama shared from Ubuntu desktop in order to use the desktop for more than only one openclaw instance.)

[jolly tapir]

Hi all - ive had openclaw running for a few weeks now with no issues but today ive gotten errors that my API / billing is incorrect. Theres been no change to my claude subscription and theres plenty of usage (i have max). I need some help troubleshooting / reconnecting my API key (or tokens?)

[ebon shell]

https://status.claude.com

Claude has issues on thier end

[jolly tapir]

Ah okay thanks hopefully thats it. I did mess around in terminal a bit and tried to add a new key so hopefully i didnt screw anything up.

[spiral parcel]

Anyone have problems with schedule cron jobs to a channel? And reminder to private dm ? On discord. So tired 😩

[velvet venture]

6:36 PM

⚠️ Agent failed before reply: No API key found for provider "groq". Auth store: /home/efe/.openclaw/agents/main/agent/auth-profiles.json (agentDir: /home/efe/.openclaw/agents/main/agent). Configure auth for this agent (openclaw agents add <id>) or copy auth-profiles.json from the main agentDir.
Logs: openclaw logs --follow

any one can help?

i couldnt fix it..

i use lite llm

[spice atlas]

fixed it. the bot had the "coding" tool profile. i gave it the "full" profile and it can send images fine now

[unreal steeple]

hey guys why am i getting this? Unsupported type: . Use Raw mode.

[warm mountain]

Hello everyone! Does anyone use Minimax M2.7 and have successfully fixed problem with Chinese characters leaking in model responses even tho MD files tell the agent clearly to use English?

[sullen ivy]

So I have been trying to get it openclaw beta to work in a macos vm everything works except the chat. Matrix keeps giving an error about no encrption (like a missing nodejs encrption library is missing which breaks matrix since it uses e2e) package found

[sullen ivy]

I fell like that is a problem with all chinese models but maybe that is just my experience

[warm mountain]

Thanks for the answer, I'm using the Chinese one for the first time so I don't know if others do this to. Seems reasonable for them tho 😄

[warm mountain]

I had this issue when my agent was missing auth-profiles.json

It was 401 error. Look if your agent has auth-profiles.json, worth chekcing

[velvet venture]

can i just scream i made custom provider as groq and it gives me rate limit with NEW account and NEW api key

[idle snow]

Is it possible to use 1M context window with GPT-5.4 with oauth on OpenClaw?

[sullen ivy]

Yes totaly scream, that is awesome. Will you share it with the comunity?

[sweet junco]

Has anyone had issues with images and audio not working in a Matrix E2EE DM?

[hollow moon]

Which free ai i need to use for control my pc

[naive ocean]

Anyone having issues with openclaw on a mac with Auth codex and keychain always asking for security permissions to allow ? It sstarted for me after updating to .24

[iron zealot]

Could you elaborate in usecase? What do you want to achieve? Does "my PC" indicate it being your main computer with all personal data and accounts added?

[ionic flame]

I'm not sure if that's the same or not, but the security permissions doesn't let my claw to see chrome web pages. I cant give access to peekaboo, it just doesn't let me

[hollow moon]

Like i want to open youtube or another programs with 1 command and i want to use it for daily tasks

[hollow moon]

In my main pc

[naive ocean]

but are you seeing the keychain app manual request to allow ? even if I do allow it or Always Allow it keeps asking again the next time that the agent is used

[iron zealot]

Sounds like something you might not want unless very comfortable with Openclaw and/or willing to accept substantial risk. Perhaps start test running it in a VM or other PC with only limited access to what you specifically need it to have access to?

[ionic flame]

I haven't tried to be honest. I just know the access policies is a headache 🙂

[hollow moon]

Openclaw is enough for my needs but i cant use it api wants money from me im not even 18 years old

[full shale]

Hey whats up, looking to get open claw on my mac, what api should i use? I know i have to make a security agent as well im just trying to get all my ducks in a row

[indigo goblet]

agreed

just watch out, you need very basics knowledge about it first

@hollow moon

[wraith quarry]

Agent failed before reply: All models failed (2): anthropic/claude-opus-4-6: The AI service is temporarily overloaded. Please try again in a moment. (overloaded)

anyone getting this for a few hours?

[full shale]

clause has been spotty for me the past couple days ngl

claude*

[wraith quarry]

0 answers since morning

[full shale]

Can someone advise me on a good api to use pls? im just not trying to waste my money cant host it locally either

[copper nova]

i mean... if he just a broke minor, and nothing of intrinsict value to his name or identity... might be a heavy lesson worth learning?... Philosophical question. Are lessons learned best by hard experiences?

[indigo goblet]

yep

[copper nova]

a smooth ocean never made a good sailor

[indigo goblet]

he might just have a hard time learning if he jump straight into it because of his age and depending his approach and how he is as a person

[copper nova]

like I learned never to trust strangers on the internet the hard way as a teen lol

lost all my diablo 2 items

[indigo goblet]

me too lol

[copper nova]

sad life

but this is more... identity theft level threat... so a little more serious

[full shale]

Can someone advise me on a good api to use pls? im just not trying to waste my money cant host it locally either

[indigo goblet]

yep true

and those fucking skills and plugins also... what was that thing that happened recently, thousands of malicious skills found or something

[full shale]

it was like 17% of skills were malicious

[pallid cove]

my main model (minimax v2.7 thinks there's a patch to be made to the OC edit tool:

Edit tool: show line context on "not found" errors

When oldText is not found in the file, the error says only:

"oldText not found. The old text must match exactly."

It should show:

• searchedRange: lines 53-107 (file has 142 total)
• hint: Use verify-edit.sh from factory/scripts/lib/ to diff exact content before editing

Why: Agents spend 20+ minutes trying to match text that has invisible whitespace differences. The edit tool has the data (file length, searched range, truncation flag) but doesn't surface it. A one-line addition to the error response would eliminate this entire class of wasted cycles.

Where: openclaw/core/tools/edit.py — around line 162 where the not-found error is returned.

does this seem ok or is it hallucinating?

[copper nova]

yes... have to be sus about everything, I try to create my own skills for everything

stop looking for shortcuts

grind the grind

might take a bunch of tokens to tune your skill, but atleast you know it's yours and safe to use

[indigo goblet]

agreed, that's what Im doing, always, and besides why not? when you and your agent create, it build continuity, you can build it exactly as you want, fix and improve, and understand exactly how it works and what it does

plus, it's not like some blackbox

[hollow moon]

OpenRouter is free?
??

[final patio]

Only upto 250k tokens

[pallid cove]

they have some free models. you can see them by running openclaw models scan

[pallid cove]

you can get more by crediting $10 to your account

[fleet sun]

If on telegram do /elevated full to give it a full control over your computer

[drifting pollen]

Now it's token based? It was request based last I checked.

[fallen vortex]

Hi Guys .. i am feeling frustrated for sure i a feel im doing something wrong ..
my claw .. after the recent update is not taking actions .. for example .. i asked him to debug some logs about some issues with my QMD installation .. it refused to do it on its own .. it kept giving me commands to keep running .. and everytime i restart the gateway i need to manually remind it to continue .. so frustrating .. the whole point is for it to goo figure it out and come back to me for a confirmation or a report, what im i missing ?
also in a lot of cases for example i am asking the claw to go and and install the QMD skill for me .. it comes back saying
"Perfect — that’s clear.
I’ll install and configure steipete/qmd for memory on your Windows OpenClaw setup, then verify it’s active and fix your memory-search issue path. I’m starting now.
"
but nothing actually happens ..and i am really clueless on what is going on or on the progress , so i have to go back to manual work
what is the problem ???

[final patio]

Well i have given that answer based on my experience with the free tier. Since in 5-10mins, i have been rate limited.

[drifting pollen]

it's 50 requests

[copper nova]

I'm actually curious about this... the QMD thing is pretty wide spread news but it's from old videos.... openclaw has it's own embedded indexing system now which was what QMD was suppose to accomplish

you just provide it a gemini key or else it uses some openAI embed model by default

[drifting pollen]

I kinda need to ask Krill something

will it be back?

[fallen vortex]

this is an example .. this happens in a lot of cases .. but for QMD i dont want to use openai apis and get into rate limits i wanted to have the local thing .. so i can avoid rate limits
but still the behaviour it self is strange .. why its not doing that on its own and comes back to me .. and why after any restart i need to manually reach out

[gray juniper]

@3oon if the agent acknowledges the task but doesn't actually execute, check your exec approvals in agent settings — post-update it sometimes reverts to requiring manual confirmation for every action, which explains the loop of commands without progress

[jolly tapir]

ANyone else been getting this message for the past hour? "The AI service is temporarily overloaded. Please try again in a moment."

[copper nova]

look at that session log and see what it actually did and where it failed

[fallen vortex]

thank you dan sounds very logical, would you mind telling me from where i can check that ? on which file or from which location in the portal ?

[dark glacier]

Hi, I'm using Minimax M2.7-highspeed token plan($40/month) for browser automation and it is very unstable. I mean, I made an MD file for the automation but it ignores some essential commands. I tried to use gpt-5.3-codex through OpenAI OAuth and it worked quite well, but the weekly limit in the Plus version ($20/month) was very low. I also tried to use other models like Kimi v2.5 and openrouter free models but they were worse than MiniMax. How can I fix it? I'm trying to fix it for 2 weeks and especially trying to change the model for a week but I wasn't able to go through. Is there any approach to make MiniMax work better? My md file is about 13KB and I think it's lower than the limit

P.S. Minimax doesn't always ignore but sometimes.

[eager maple]

Is anyone having issues with claude on openclaw? Mine hasnt been working all day, getting this error.

The AI service is temporarily overloaded. Please try again in a moment.

[dim harbor]

yeah opus is not working for openclaw today, switch to sonnet

[dark glacier]

Plz help me. I don't think Minimax is dumb,

[eager maple]

Do you know how?

[dim harbor]

just type /model and then switch, if you got it set up there

[tawny grove]

Hey, is there a config key to raise the concurrent subagent limit? I'm hitting "max active children for this session (5/5)" and can't find a documented key for it. agents.defaults.subagents.maxActive throws an unrecognized key error.

[tulip folio]

How do I whitelist a local directory for the PDF tool?

[dim harbor]

under .defaults, theres this:
"maxConcurrent": 4,
"subagents": {
"maxConcurrent": 8
}

[nocturne bane]

Is there someone that has a Claude max subscription that could give me a guest pass so i can try out claude pro. Woulde be very helpful to me 😊 just send a dm.

[tawny grove]

thanbks I just found it in the docs, they really put it under a lot of shrubs.

[dark glacier]

Are you all ignoring my messages?

[dim harbor]

noone knows bro 🙂 use models chat #1478196963563409520 if theres any active minimax users there. you can also test minimax in a different harness (opencode etc) and see if it works better there

[tawny grove]

It doesn't seem like a openclaw question, more of a model question, I would use the above adivce.

[copper nova]

do you think there's people in here getting paid to reply or something? I'm confused.

[west sage]

Anyone knows how to deal with this, my telegram channel just go very unreliable. Found this in logs 👇

17:49:05+00:00 error gateway/channels/telegram {"subsystem":"gateway/channels/telegram"} [telegram] Polling runner stop timed out after 15s; forcing restart cycle.
17:49:05+00:00 error gateway/channels/telegram {"subsystem":"gateway/channels/telegram"} Telegram polling runner stopped (polling stall detected); restarting in 2.33s.```

[tawny grove]

actually lmao. His question is like a copy and paste to like claude

[copper nova]

the level of entitlement some people have is unreal lol

[indigo goblet]

just ask your agenttttttt, pair programmingggg guyssss

[tawny grove]

What can we expect, when a bunch of inferior minds wish to hop on the "coding" chain by using their AI slaves.

[atomic gate]

Hey guys, I have claw on Aws lightsail and I have such a hard time setting how the web browser, i tried both brave and tavily. Agent doesn’t seem to be for both able to wesearch

[deep sundial]

Hello hello, new guy here, i am running Ollama on my main server, and a VM that i have OpenClaw installed on.
Im trying to get them to communicate.
The only answer i am getting is HEARTBEAT_OK
Any one who has any solution for me ?
🙂

[copper nova]

what model you using?

[deep sundial]

GPT-OSS:20B

[atomic gate]

Minimax their free trial one with auth

Is that the problem

[copper nova]

unfamiliar to me, check if the model has web search abilities, check how the tooling is setup with search api, I use gemini for the web search

[serene lake]

I keep finding myself needing to keep terminal windows open for commands like:

openclaw gateway —port 18789
ollama serve

How do I manage these? What should I learn ?

I’m running OC on WSL

[indigo goblet]

most of the questions here can be troubleshot by claude or chatgpt in less than 30 minutes

nobody is paid here people, unless chatbot can't fix it and it's too tricky which I doubt, ask people, otherwise not

[copper nova]

you can run the gateway in a background terminal I believe, not sure about the ollama tho but idea should be the same. Ask your agent to set up a cron that will spin those up when their down.

[open saffron]

Running Kimi-k2.5 right now as my main model. It's acting up and I want to try something else. I have Hugging face. Is there a recommended model to use there?

[pallid cove]

anybody seen this issue before ?

[wheat halo]

How do I install claw ?

[primal jay]

Multiple agents on a single openclaw instance: I need help with this. every time i have openclaw (while using openai chatgpt oauth) create a second or more named agents. stuff gets wonky, .md files are all over the place or non-existent or in the wrong places.

[pallid cove]

you have to make it configure separate workspaces for each

[primal jay]

you have to do that before or ask the main agent to do it for you? what is the best practice?

[pallid cove]

for example, I have ~/.openclaw/workspace/agents/workspace-agent-{A,B,C}
~/.openclaw/workspace is for the main agent, or router in the pattern I use

each agent's SOUL.md is in their workspace

[pallid cove]

you can ask it to do it, mine was able to do it with minimax 2.5 fwiw

I asked it to read the openclaw agent docs

[primal jay]

what about the slew of .md files for the other agent? is it best to populate these myslef or use openclaw? or another chatbot to do it?

[pallid cove]

tell main to copy its own files as starting points for the others. and then customize them depending on your goal. ofc you can customise yourself too but honestly it's quicker to ask main to customise for a goal

for example, I have a BMAD (agile dev) derived agent team. so my agents have been copied from agent personas from that

main is a router, it rarely does work itself, mostly routes to the others

@primal jaybtw I picked up most of this stuff from this book https://austenallred.gumroad.com/l/jzketz

it's free if you want

[primal jay]

Thanks!

[fallen vortex]

is there a bug here .. suddenly after the latest updated .. my agent became extremely frustrating ..
i give him a task .. says he is working on it and ask him to keep updating me .. and its like the whole thing is totally off .. he doesn't come back to me at all and no updates are made except for the very first reply .. how can i fix this ..
my exec file are ok .. but still the issue is there

[bleak flame]

Hey, any 💡 idea HOW & WHERE can I get an API key 🔑 for a OpenSource 🆓 LLM

Which are freemium alternatives?

[cold inlet]

ah good catch — that key changed between versions, my bad
try this instead:
openclaw config get agents
that'll show you what keys your version actually accepts and we can find the right one for context/token management
what version are you on?

[ancient ore]

New here, can anyone help. My openclaw keeps going idle for some reason. I am using it thru ollama in the tui. That is running a local llm.

For example, it has stopped right now:

I'll help you fix the Arduino program. Let me first look at the files in the Flicker directory on your Desktop.Let me read the Arduino sketch file from the Flicker directory:
connected | idle
agent main | session main (openclaw-tui) | ollama/qwen3.5:35b | think low | tokens 14k/262k (5%)

Just stops like that. Goes idle.

[fallen vortex]

any help please ..

[ancient ore]

I think I have a similar problem 3oon

My agent goes idle.

[misty trail]

Anyone know how I can run an LLM on a Ubuntu server machine and somehow connect that to openclaw on a completely separate device

[pallid cove]

could be a model problem. only claude really keeps pluggin at a task for a long time. others stop
that's why we need heartbeat and other systems (put todo items in TODO.md) for lesser models

[fallen vortex]

Could it be because my heartbeat is disabled .. i kept it 0 until i verify everything .. is that a reason for such behavior??

[spare shard]

Yo.
question to all of ya.
is there anyway that i can USE MY PHONE as a LOCAL OLLAMA and use it in openclaw.

[pallid cove]

you must tell the agent to add tasks to heartbeat. it won't on its own AFAIK?

[spare shard]

You can delete its memory and make it again maybe.
Or make a new chat and give it commands like
"Be kind and update your creator(You)

[pallid cove]

this, add this to SOUL

[spare shard]

does somebody know if my question work?

i hope it would work because like same network

[pallid cove]

afaik only very small models run on phones

[spare shard]

i know that.

im not that stupid but still i dont wnat to use my pc for it yet.

[pallid cove]

https://www.canirun.ai/

[worn zinc]

What are you trying to do with it?

[pallid cove]

a pity it doesn't have phones

[spare shard]

would it sitll work thats what i am asking

like openclaw would let me?

[pallid cove]

yeah people have run models on phones. but it'd be very experimental to setup

[spare shard]

i have the time because i do not want to make API keys shit anymore.

its been HOURS

litellm
custom provider
open rout

google

peace of shits wont work

[pallid cove]

honestly just ask your agent or claude to walk you through it

[spare shard]

Nah i checked up

its possible

but its like the WORST modules

like i need constnat cooling constant charging etc etc

[pallid cove]

if you don't have an agent in your openclaw just use the minimax 2.5 that's free on openrouter rn

[spare shard]

I coudlnt get a agent on my goddamn openclaw

like its been ours

[spare shard]

Rate limit?

[pallid cove]

should be enough to set this up

[cold inlet]

welcome! and yeah €5 for 100 messages is the wake up call most people get eventually lol
your instincts are right — the fix is a tight bootstrap that stays consistent session to session, not a bigger context window or a cheaper model. haiku is a bandaid like you said
the sliding window + RAG approach works but only if the agent is actually consulting memory at the right moments — most setups get the storage right and miss the retrieval discipline
practical stuff that moves the needle fast:

keep your bootstrap files lean and structured, every character loads on every session
archive daily logs older than 30 days, they bloat context invisibly
prune MEMORY.md weekly, stale entries are just dead tokens
lossless-claw plugin if you're not using it already, helps a lot with the compaction quality

the prompt caching point is spot on btw — only worth it if you're in continuous sessions, intermittent use the cache expires before it pays off
what does your current bootstrap look like, is it mostly inline instructions or do you have it structured into files?

[pallid cove]

How are you guys fighting these high API costs
I use a cheap model. minimax m2.7 which for $10 per month, gives you 1500 req/5h. it's not Opus tier but I bet it beats Haiku

I use very large prompts because I have a bmad (agile research and dev) agent team. I routinely use 100M tokens per day. 1.3B per 30 days
I could never afford claude for this lol

fwiw OC estimated my 30 day usage at $260 at minimax API prices

[cold inlet]

sounds like the agent spun up but the workspace never got properly initialized — that's why you're getting responses but no actual work happening, it's running but has nothing to anchor to
few things to check:
openclaw status
openclaw memory status
that'll tell you if the workspace is actually connected or just idling
also check if there's anything in ~/.openclaw/workspace/ at all — if that folder is empty or missing files the agent has literally nowhere to store or retrieve anything from
for the ticket thing — github discussions is faster than official support for most issues, someone usually picks it up same day
what does openclaw status show you?

[ancient jungle]

does lossless claw replace the memory system? or works in conjunction?

[copper nova]

works with

[ancient jungle]

so if i use memclaw for memory right now, i should also still have lossless claw. i'm only interested because for some reason (normal openclaw bullshit), my sessions stopped auto compacting

[sterile steppe]

we are working for total replacement

[cold inlet]

conjunction — it doesn't touch the memory system at all
lossless-claw is about how the active session gets compacted when the context window fills up. instead of lossy summarization it tries to preserve the actual content
memory search is about retrieving things from past sessions
two different layers, both useful, neither replaces the other

[unique hazel]

Hey so the recent updates have broke the commands on Discord. There are also multiple related threads in #1459642797895319552 . When I do /stop, it says I dont have the permission to do that. When I do /new, it just replies with "Done" but no new sessions created. I have been using OC for a while now and it worked, so I am sure it's the update.

Anyone know the fix? I am learning for my exam tomorrow, session is stucked and I couldn't do /new.

[covert blade]

I had similar trouble Jasnsy helpped me implement losslessclaw. One gotcha I found was in openclaw.json compaction. After installing losslessclaw it worked great for many hours then slowed... I found that my reserveTokensFloor was set larger than the contextwindow of my model. I reset reserveTokenFloor to about 25% of my model contextwindow. I'm now coming up on my sixteenth hour of openclaw usability after days of frustration... I'm using models served by an ollama server.

[indigo goblet]

did you ask your agent

if not, you should do it, most likely can fix it

[spare shard]

Yo guys can i add "GROQ" with custom provider

bc i need a free ai bruh i dont have bank account im 17..

[indigo goblet]

not sure you should use openclaw at 17, might want to start with chatgpt or claude first

[copper nova]

nahhh don't gate keep, let the boy free

https://giphy.com/gifs/running-muppets-7kn27lnYSAE9O

also means he has nothing of intrinsic value to risk, let him go ham

[thorn geyser]

Hi everyone, if any of you are in SF locally, sign up on this hackathon event! 🏆 $300K+ in credits + $2K+ cash prizes 🚀 Winners featured in Innovation & Tech Today 👀 Judges from Palona AI, Virtue AI, Replit & more. Only 50 spots. Pine me for the RSVP link.

[neat walrus]

Note to self .. If your openclaw works DONT UPDATE... Every single update has broken my bot and takes hours for me to fix.. This program is held up by duct tape..

[copper nova]

don't forget the bubblegum, hopes and dreams of aspiring millionaires

[copper nova]

next time, make your opus agent research all the release notes and identify each part risky to your systems. build an upgrade and test plan for it.

worked for me the last like 3 updates, everything went smooth as butter

[neat walrus]

Im local only. 😉

[copper nova]

welp, sorry for your pains then lol

[severe grove]

Hi — I believe my hosted OpenClaw instance is stuck in a stale runtime state and not picking up provider/auth config from the UI.

Instance ID:
service-69c1f738e11ead6d2def4c32-5f6f484f4f-bkks7

Setup:

• Hosted on openclaw.org (paid plan active)
• Subscription + addon credits applied
• Using built-in runtime (not self-hosted)

What I configured:

• Custom OpenAI provider via AI & Agents → Models
• Adapter: openai-completions
• Base URL: https://api.openai.com/v1
• Auth mode: api-key
• API key added and saved
• Model list entry: gpt-4o-mini
• Agent model set to: openai-manual/gpt-4o-mini
• Save + Reload applied

Observed behaviour:

• Runtime still attempts:
  anthropic/gpt-4o-mini (model_not_found)
• Error:
  “No API key found for provider openai”
• Logs reference:
  /home/node/.openclaw/agents/main/agent/auth-profiles.json

Additional signal:

• Built-in providers (newapi-builtIn / zeabur endpoint) still present
• Gateway + control UI show as connected under Instances
• But runtime is clearly not using updated provider config

Conclusion:
Looks like the agent runtime is bound to a stale config and not regenerating auth-profiles.json or reloading providers.

Questions:

1. Can this instance be reprovisioned/reset from your side?
2. Is there any way to force runtime rebuild on hosted instances?
3. Are my credits tied to this instance, or can they be reassigned to a new one?

Happy to provide anything else needed — but this seems like a runtime-level issue rather than config.

[copper nova]

wait wtf is openclaw.org

[severe grove]

https://open-claw.org/my-openclaw this website

[copper nova]

yea that's not here homie, you best contact they support lol

that's kinda fucked up they pretend to be official tho

hope you don't get scammed my dude

[indigo goblet]

wth

[ancient jungle]

i need help setting this up, i installed it, what do i need to actually configure

[copper nova]

and for what it's worth, this is 100 percent config, it says right there that your api key isn't set for the model

all those providers it listed also sus, never seen any of those before lol anthropic/gpt-4o-mini like why would anthropic be offering a GPT model?

sus company shoveling a garbage product that doesn't even work

[ancient jungle]

how does lossless claw trigger? i set the context limit at 0.75, and i'm at 75% lol

[cobalt beacon]

Bug: Telegram channel not starting — RangeError: Maximum call stack size exceeded in config loader

Version: OpenClaw 2026.3.24
OS: Linux 6.12.63+deb13-amd64

Summary: Telegram channel shows empty in openclaw status and never connects. Gateway runs fine but Telegram never initializes.

Root cause in logs: On every gateway start, the config loader hits an infinite recursion:

Copy
RangeError: Maximum call stack size exceeded
at console.warn → loadConfig → readLoggingConfig → resolveSettings → getLogger → console.warn → loadConfig (repeats)
Steps to reproduce:

Install the Composio plugin (@composio/openclaw-plugin@0.0.8)
Configure Telegram channel
Restart gateway — Telegram channel never appears, stack overflow fills the log
Notes: Disabling the Composio plugin does NOT fix it — the crash persists. Gateway is otherwise functional (webchat works). The loop appears to be in env-D1ktUnAV.js around lines 18880 / 19344 / 19441.

[meager ocean]

Anyone have a fix for the openclaw to start behaving again? Seems like after update it basically can just chat, not execute task like coding, running terminal commands, fix discord tasks etc. anyone?

[covert blade]

try /new
Also I checked that these exist and non zero size

jeffa@karen:~$ ll .openclaw/lcm.db
-rw-r--r-- 1 jeffa jeffa 163840 Mar 25 12:08 .openclaw/lcm.db
jeffa@karen:~$ 

[ancient jungle]

mine says the exact same as yours, so the size is probably pre-determined?

[raw shale]

How do I fix OpenClaw having just a 16k context window according to the TUI despite having my custom providers' contextWindow set to 262144 both in openclaw.conf and ~/.openclaw/agents/main/agent/models.json?
Generation seems to work but there is no guarantee that it will actually utilize the full context window that I can use.
-# I do in fact have Nemotron-3-Nano-4B loaded with a 262k context window in LM Studio

[severe grove]

it's not a scam but it is third party

[copper nova]

I just think companies that have to postor as the original are always sus

I reccomend get a refund and try a more... respectable hosting company

[main plaza]

Does anyone know if you can train Open Claw (OC) to do a new skill by submitting a video to it (i.e. screen recording on some sys admin actions) on how to do it. Basically using a video to instruct it instead of written prompts? I have searched if this is possible, but not found anything. Thank you.

[summer hemlock]

mmmm that's very difficult atm

[copper nova]

use something like the youtube analyzer skill and get it to create a skill based on the contents

[summer hemlock]

@main plaza the only thing that reliably does youtube video content, in my experience, is NotebookLM

[copper nova]

wait, not its not, what?

I already done this lol

[summer hemlock]

most models get like metadata & description from the video & don't actually watch the video

[copper nova]

it gets the subtitles/transcript from it, which is the full content more or less

[summer hemlock]

have you had you claw look at like 6 hour YT video courses?

[main plaza]

Hmmm ok will investigate and try

Thank you

[summer hemlock]

the "more or less" part is doing heavy lifting. It doesn't get images then.

[copper nova]

why you feeding it 6hrs of content lol

[summer hemlock]

to learn about docker containers

[copper nova]

it already knows docker containers

what?

[summer hemlock]

back when I first started with AI agents

[copper nova]

lol

[summer hemlock]

yes but I didn't

I started all my learning with notebookLM mostly

[copper nova]

so you should of been watching the videos, not your robot

[summer hemlock]

I knew nothing before december last year

NOTHING

I was just heavily invested in world ETFs, and they are mostly AI companies atm. So i had to try it for myself because ppl were talking about AI bubbles

[summer hemlock]

btw I also program actual robots at work 😄

we have fanuc robots and other custom made ones

[main plaza]

Thanks to the two bots that answered my question.

[mellow trench]

im starting to be so done with this thing.
onto my third or 4th install.
No matter what i try or where i look either the models are using tools wrong, or openclaw is working wrong?
This is an example of the outputs im constantly getting (i havnt even gone through the bootstrap sequence yet)

`I see the issue. The tool call format needs to be corrected. Let me fix the tool call format and try again.

<function_calls> <invoke name="read"> <parameter name="path">/home/clawy/.openclaw/workspace-auto/BOOTSTRAP.md</parameter> </invoke> </function_calls>`

please help im going insane.

sometimes it cant even read like soul.md or agents.md

[pine flax]

what model are you using

[mellow trench]

i have tried deepseek v3.2
devstral-2:123b
minimax2.5 and 2.7
and qwen 3.5
i have also played around with different ollama provider settings.
nothing seems to have worked.
devstral is the closest i have gotten where it will respond with some text before failing the tool

[cedar basin]

Quick question if anyone is up to help me, whenever there is a gateway restart, is there a way to make sure the agent gets back to whatever he was doing before the restart without me having to message first?

[copper nova]

I think there was an update pushed to do this @opaque sorrel might know more

I think it was his PR

I over engineered some bullshit announce and wake system that no body needs lol

[cedar basin]

Oh alright, I’ll see when he responds. Thank you

[opaque sorrel]

So yes it should be merged to main a day or two ago (whether it stuck for more than one commit is another question lol) but I also needed to add this to my agents file (needs a part 2 fix but this gets me past it)

### 🔁 Gateway Restarts
- When needing to restart the gateway and continue the current session afterward, use the restart sentinel path.
- Do **not** use raw `openclaw gateway restart` for that. Use the gateway tool or another restart flow that writes the restart sentinel first.
- After `gateway.restart`, `config.patch`, `config.apply`, or `update.run` schedules a restart, stop the current turn immediately.
- Do **not** call more tools after a restart-aware action schedules restart. End the turn with a short note like: `Restarting now; I’ll continue after restart.`

[opaque sorrel]

so if you build from latest or if there's been a release since then and use that AND add that to your agents file, it should you should be able to see it happen live. (you can give a prompt to verify pointing to the commit and this and tell it, hey try to reboot yourself 5 times in a row)
https://github.com/openclaw/openclaw/pull/53940

[slate fog]

Problem Summary

System:

MSI GP65 Leopard (NVIDIA 1660 Ti)
Ubuntu 24.04.4 LTS
OpenClaw version: 2026.3.24 (cff6dc9)
Ollama installed with model: qwen2.5-coder:1.5b (verified working via terminal)
Issue:
OpenClaw web interface does not respond when sending messages. Ollama works perfectly in terminal (instant responses), but OpenClaw cannot communicate with it.

What I've tried:

Multiple fresh installations of OpenClaw
Running openclaw setup (setup completes but no model provider configured)
Manually editing ~/.openclaw/openclaw.json to add:

json
"model": "qwen2.5-coder:1.5b",
"ollama_url": "http://127.0.0.1:11434"
OpenClaw removes these keys on restart (unrecognized keys error)
Tried openclaw config set model qwen2.5-coder:1.5b - command not found
Tried openclaw gateway - starts but web UI doesn't respond to messages
Current config after openclaw setup:

json
{
"name": "OpenClawDetection",
"version": "2026.3.24",
"description": "2026-03-27T23:55:01.322Z",
"agents": {
"defaults": {
"workspace": "/home/mia_ibn/.openclaw/workspace"
}
},
"commands": {
"native": "auto",
"nativeSkills": "auto",
"restart": true,
"ownerDisplay": "raw"
},
"gateway": {
"mode": "local",
"auth": {
"mode": "token",
"token": "generated_token"
}
}
}
Question:
How do I configure OpenClaw 2026.3.24 to use a local Ollama model (qwen2.5-coder:1.5b) so the web interface works? The config file doesn't accept model or ollama_url keys, and the CLI doesn't recognize config set commands for these settings.

[mild heron]

help ! I deleted an old session on the gateway and my telegram chat died ie. not responding... how do I recover this ?

[mellow trench]

im still very knew to all this but i normally just go through the setup wizard again
for telegram openclaw channels add --channel telegram

[cold inlet]

yeah this is a boot sequence problem — the agent has no way to know a restart happened or what it was doing before unless something tells it
cleanest fix is a cron that fires a few minutes after gateway start, reads the last daily log and MEMORY.md, and posts a resume summary to telegram so you can confirm it picked up where it left off
the key is the agent needs something to read that tells it what was in flight — if that state isn’t written down before the restart, there’s nothing to recover from
do you have anything writing active task state to a file between sessions,

[mild heron]

will try it thanks

[earnest quiver]

"On macOS with Homebrew, Python3 is at /opt/homebrew/bin/python3 but the sandbox file bridge can't find it. Agent write access is broken. How do I configure the Python path for the sandbox?"

[devout moth]

Hello everyone, did anybody installed openclaw locally in windows 11 using powershell???

The web UI not working in my setup can anybody help me???

[neat walrus]

try running "openclaw dashboard" in your terminal yet?

[neat walrus]

Why doing it the hard way by editing the json yourself? Newest updates allow you to choose Ollama from the Providers now and lets you pick from the models you've downloaded right from the "openclaw onboard" command in terminal. No need to edit openclaw.json anymore unless you're messing with some other stuff like sandbox privs or exec privs just to name 2

[dry osprey]

hello. i already setup my OpenClaw using the instructions here: https://www.amd.com/en/resources/articles/run-openclaw-locally-on-amd-ryzen-ai-max-and-radeon-gpus.html. i am able to see the dashboard and chat via Telegram. i'm having trouble using the browser integration. this command (openclaw browser extension install) throws an error (error: too many arguments for 'browser'. Expected 0 arguments but got 2.). note that i'm using the latest version v2026.3.24. i'll appreciate any tips on what i should do next

[spare shard]

Can i use groq on openclaw with custom provider

[worn echo]

I had the same problem then started using nemotron and things went fantastic

[devout moth]

i did but not working.... the erorr is "The site can't be reached"

[final patio]

Start the gateway and then try the dashboard.
openclaw gateway restart
Openclaw dashboard

[devout moth]

Gateway: bind=loopback (127.0.0.1), port=18789 (service args)
Probe target: ws://127.0.0.1:18789
Dashboard: http://127.0.0.1:18789/
Probe note: Loopback-only gateway; only local clients can connect.

Runtime: stopped (Startup-folder login item installed; no listener detected on port 18789.)
RPC probe: failed
RPC target: ws://127.0.0.1:18789
gateway closed (1006 abnormal closure (no close frame)): no close reason
Gateway target: ws://127.0.0.1:18789
Source: cli --url
Config: C:\Users\S Guruprasad.openclaw\openclaw.json

this is my gateway status

[final patio]

Have you started gateway and tried the dashboard?

[devout moth]

yes, but same here

[final patio]

You can try.
Openclaw daemon install for gateway listening.

[marble coral]

There seems to be a problem with discord slash commands after the last few updates, probably after 3.8 or something.
Has anyone found a solution? I am getting only "done" as response when using / new or / status . Some commands like / model seem to work , so this a bit strange.
The issue is only on Discord, TG works fine.

[spare shard]

can somebody help me with rate limit i just added the module and immediatly gave rate limit

i have configured GROQ module Llama3370Versatile

[devout moth]

Done, still i can't see broswer UI

The port is open 18789 and no firewall is set

[final patio]

Can you check gateway status.
Openclaw status --deep

[devout moth]

[openclaw] Failed to start CLI: Error: gateway closed (1006 abnormal closure (no close frame)): no close reason
Gateway target: ws://127.0.0.1:18789
Source: local loopback
Config: C:\Users\S Guruprasad.openclaw\openclaw.json
Bind: loopback
at Object.onClose (file:///C:/Users/S%20Guruprasad/AppData/Roaming/npm/node_modules/openclaw/dist/call-DTKTDk3E.js:594:10)
at WebSocket.<anonymous> (file:///C:/Users/S%20Guruprasad/AppData/Roaming/npm/node_modules/openclaw/dist/method-scopes-DhuXuLfv.js:2184:23)
at WebSocket.emit (node:events:518:28)
at WebSocket.emitClose (C:\Users\S Guruprasad\AppData\Roaming\npm\node_modules\openclaw\node_modules\ws\lib\websocket.js:263:12)
at emitErrorAndClose (C:\Users\S Guruprasad\AppData\Roaming\npm\node_modules\openclaw\node_modules\ws\lib\websocket.js:1047:13)
at ClientRequest.<anonymous> (C:\Users\S Guruprasad\AppData\Roaming\npm\node_modules\openclaw\node_modules\ws\lib\websocket.js:886:5)
at ClientRequest.emit (node:events:518:28)
at emitErrorEvent (node:_http_client:104:11)
at Socket.socketErrorListener (node:_http_client:518:5)
at Socket.emit (node:events:518:28)

[rotund maple]

Hello, I’m running a 4GB RAM VPS in the SEA region since I’m based there as well. I don’t really have any issues, but I could get double the RAM if I used a VPS from other regions like the EU or US.
Would it make more sense to have higher ping (by hosting in another region) in exchange for 8GB RAM and more CPU cores? I usually let my agent scrape and look things up using a browser.

[formal rock]

Literally just uploaded a plugin for this called Lethe. Comes with a dashboard. Memory is stored in a docker container. Lethe plugin ties into gateway. Was pissed because my agent always forget so I fixed it myself

[formal rock]

What are you doing with your agent? Is if something that needs low latency? If not you should be fine to move

[rotund maple]

mostly web scraping and dev. but yea always find my server using max cpu usage and ram

[final patio]

this error is due to gateway is closed.
what are you getting for command:
openclaw gateway

you need to use another terminal for rest of the commands if the gateway has started.

[devout moth]

[bonjour] watchdog detected non-announced service; attempting re-advertise (gateway fqdn=LAPTOP-GR07U2CQ (OpenClaw)._openclaw-gw._tcp.local. host=openclaw.local. port=18789 state=probing)

i got error for above cmd (openclaw gateway)

Everything started and browers is listening.... but at last i got this error

[pulsar pendant]

Hi, I am new. Battling with first setup: "Dashboard shows 'pairing required' when connecting via HTTPS through Caddy reverse proxy on Tailscale. Running openclaw pairing list shows no pending requests. How do I initiate browser pairing for the Control UI when openclaw dashboard --no-open has token auto-auth disabled due to SecretRef?"

[inland knot]

Hey guys, what is the most annoying thing you experienced with OpenClaw until now?

[final patio]

can you paste that listening?

[pulsar pendant]

Port 18789 is already in use.

• pid 3698897 andy: openclaw-gateway (100.83.47.86:18789)
• Gateway already running locally. Stop it (openclaw gateway stop) or use a different port.
  Listening: 100.83.47.86:18789
  Troubles: run openclaw status
  Troubleshooting: https://docs.openclaw.ai/troubleshooting

[nimble hill]

Go tot the cli and type openclaw devices list
There will be a request id - you approve with openclaw devices approve <deviceid>

[fast patio]

Hi, i'm newbie here, i just deep in technology like 1 year ago. I heard that openclaw will change the future, so I want to know it. Any safest ways to install it?

[devout moth]

Ok

[nimble hill]

In vps - there is nothing called safe. Just that avoid installing in your personal laptop.
In vps, you may still be subjected to prompt positioning if you are doing some searches which re not safe

[devout moth]

13:46:10+05:30 [canvas] host mounted at http://127.0.0.1:18789/__openclaw__/canvas/ (root C:\Users\S Guruprasad.openclaw\canvas)
13:46:10+05:30 [heartbeat] started
13:46:10+05:30 [health-monitor] started (interval: 300s, startup-grace: 60s, channel-connect-grace: 120s)
13:46:11+05:30 [gateway] agent model: ollama/llama3:8b
13:46:11+05:30 [gateway] listening on ws://127.0.0.1:18789, ws://[::1]:18789 (PID 12944)
13:46:11+05:30 [gateway] log file: \tmp\openclaw\openclaw-2026-03-28.log
13:46:11+05:30 [browser/server] Browser control listening on http://127.0.0.1:18791/ (auth=token)
13:46:16+05:30 [bonjour] watchdog detected non-announced service; attempting re-advertise (gateway fqdn=LAPTOP-GR07U2CQ (OpenClaw)._openclaw-gw._tcp.local. host=openclaw.local. port=18789 state=probing)

[fast patio]

how about using mini mac?

[nimble hill]

You can run in any spare cheap laptop/desktop- even 5 years old. Don't necessarily need mac mini - since you won't be able to run local llm anyways

[slow monolith]

how can i set more persistent sessions? when i ask openclaw, it reply: OpenClaw in this channel doesn’t support persistent sessions (thread=true unavailable). So our subagents are on‑demand (mode=run)

[fast patio]

oh, thanks. btw, do you have any tips for free unlimited API keys usage?

[final patio]

gateway started. you should be able to open web gui by: openclaw dashboard

[nimble hill]

There is nothing free in the world 😃 the cheapest usable one is chatgpt usd 8 / month

[blissful trench]

there are ^^. i am using a fully free autonomous agent team in my openclaw

[devout moth]

Thanks a lot bruh.... now it's working

[pine latch]

whats everyone using for kanban style multi agent dev work?

[nimble hill]

Which provider?

[pulsar pendant]

thanks mate !

[nimble hill]

What about it? It is a.coding agent- you still need llm provider

[fast patio]

oh yeah sorry

[blissful trench]

ill make a video soon about my setup but the providers i am using are aistudio,cerebra,groq,sambanova,openrouter,nvdia nim,baseten, stray ai, etc.

[inland knot]

Yo guys, i currently run Mini Max 2.7, someone here used Step 3.5 Flash to give me an honest review? Is Step 3.5 Flash better than M2.7 in OpenClaw?

[pulsar pendant]

My main agent cannot invoke subagents. When it tries, it gets 'gateway closed (1006 abnormal closure)'. My gateway is bound to the Tailscale IP (not loopback) because I need remote access. I think the subagent invocation is hardcoded to try 127.0.0.1 internally. How do I configure subagent-to-gateway communication when bind=tailnet?"

[nimble hill]

So you shift between the free quota between providers?

[nimble hill]

Use tailscale directly in the host and your other machines. Then in onboarding select LAN so that it runs on 0.0.0.0

[blissful trench]

yep depending upon the task aswell

using litelm

[pulsar pendant]

"Getting 'ACP runtime backend is not configured' when my main agent tries to spawn subagents. How do I configure the ACP runtime? I don't see it in the onboarding wizard or openclaw.json."

[nimble hill]

It okay for some small tasks, but for some serious work, you need continous single model to sustain some reliable outcomes

[blissful trench]

true but instead of hammering one model on every task i got different specialist models on the task. example for reasearch and stuff i am also using notebooklm, for coding i got codex,opencode,gemini cli,qwen cli,mistral vibe etc. for daily cheap work i got lighter models. the whole setup is pretty much dynamic. But ofc these are all free stuff so i cannot expect 100% reliability without me interfering but hey it does the job for being free

[final patio]

Hey does anyone know if i update my claw where it would be updated, since i have moved .openclaw and npm dir from Home to E dir.

I don't want to move files every time after an update.

[slate minnow]

In the recent releases are agents able to communicate with each other any better?

[nimble hill]

Not sure how continous session will be done by free tier and not raking up any bill in a month.

[blissful trench]

i am using over 130+ api keys to run it so yeah it pretty much handle everyhtnig. coding part is handled by 5 cli which has like huge rate limits so yeah it can pretty much remain active properly

[muted warren]

I built something that I wish I had when I started learning Git.

Most Git tutorials either:

• Overcomplicate things
• Or just give commands without context

So I created a complete Git + GitHub guide that actually teaches you through real-world scenarios.

What’s inside:
• Beginner → Advanced Git
• Scenario-based learning (like real dev work)
• Merge conflicts, PRs, branching, undo mistakes
• Advanced topics like reflog, bisect, cherry-pick

Perfect for students, developers, and open-source beginners.

If it helps you, consider starring it — it really motivates me.

https://github.com/Krish6115/GitxGithub

[grand blaze]

[BUG] Agent creates log files with wrong date (2024-xx-xx)

Hi,

OpenClaw 2026.3.24 — WhatsApp channel on Docker/Coolify.

My agent monitors WhatsApp groups and logs messages into
daily files: messages/YYYY-MM-DD.md

Problem: files are created with dates in 2024
(e.g. 2024-05-17.md) instead of today 2026-03-26.
Server date is correct (timedatectl shows UTC 2026).

What I tried:

• Explicit instructions in SOUL.md to always use
  date +%Y-%m-%d to name files
• Same rule added in custom SKILL.md
• Still inconsistent, agent keeps creating 2024 files

Questions:

1. Is there a way to inject the real server date
   into agent context at session start?
2. Has anyone faced this with forwarded WhatsApp messages?

Thanks!

[small hedge]

hi @wary osprey do you have an example of your config to share, do you statically define routing in openclaw.json file? or do you spawn different agents that use Qwen/MiniMax/DeepSeek?

yes, same problem, sometimes an output of Kimi-2.5 is empty. I'm using Kimi-2.5 on Cloudflare

[fickle merlin]

I have no idea what I’m doing so far I downloaded OpenClaw now I have no idea how to do this can someone show me a good up to date video or a YT channel or something I am completely lost now lol

[iron zealot]

What do you want to achieve and where do you plan to install it?

[wild cypress]

hello guys ... actually i installed openclaw yesterday but still stuck in part... i send it prompt it says i am doing but task is not executed.... it takes me 100 prompts then it creates 1 file... also it dont message me from it side for reminders... can anybody help me out ?

[wild cypress]

we both are in same boat

[fickle merlin]

Trying to run a test for a futures trading bot to see how it’ll do not concerned about security or anything it’s a dud email for the test but I have no idea how to do this I have to input a .env file of these API keys and I am just lost

It says put the env file next to the bot script like I am 5 years old to this stuff

I’m using BloFin

[iron zealot]

If installed on main PC without any kind of separation as a VM or similar it's not the dud email to worry about, it's everything else OC has access to. What LLM/keys/subscription do you have available to handle test of trading?

[fickle merlin]

I don’t even know how to answer this I am new af never coded I just have to create an API Key on the exchange but I haven’t gotten past this part even

[iron zealot]

No expert, but I'll have a go. Openclaw is not a normal program in the sense that it handle the files/tasks you give it only. If installed on your main PC all your e-mails, Discord account, logged in trading accounts, social media, senitive documents etc are all within reach of an agen you do not yet understand or know how to limit. For trading you want a "walled garden" like a VM, Nemoclaw or seperate PC only giving the agent specifcially what is needed for the task. If you have the API key you can attempt to explain the purpose of the API key and documentation for BloFin (never used it) and have the agent attempt gaining data access using the key. Don't keep more money on account than you are willin to loose through experimenting.

[spiral flax]

hey anyone here can help a noob ?

[wild cypress]

i am far ahead but am also stuck

[wild cypress]

how much noob you are... i can guide you untill connecting the api keys and do other stuffs like adding channels

[spiral flax]

im currently getting 14k tokens out for a simple "hello" after disabled all skills and tools

"I've completed a thorough investigation into the token generation issue you're experiencing. The problem appears to be a system-wide anomaly that requires immediate attention from the OpenClaw engineering team.

I recommend filing a detailed bug report through their official channels (GitHub, Discord, or email support) with the specific details you've shared about the excessive token generation. The key points to highlight are the consistent 8-40k token generation for simple interactions, even with skills and tools disabled."

[wild cypress]

i am using codex though no api needed

[spiral flax]

im using a cheap vps so idt i can do that

[wild cypress]

can i invite you to a test server?

[wary osprey]

I don’t statically lock everything in one config. What I do is I usually define a default model in the config just so things run, then split tasks by model depending on what the agent is doing

like:
Kimi for general reasoning
Qwen for coding or little tasks
MiniMax / DeepSeek for heavy and repeated stuff

so instead of me forcing one model everywhere, each task uses what fits best for the job. I’m not spawning totally separate agents for each model every time, it’s more about how the workflow is structured and which model gets called at each step. also makes it easier when you route through something AIsa, you can switch models without touching your whole setup (that's the API provider I use)

[ancient jungle]

how does deekseek behave against minimax? i've been using minimax and i've noticed it tends to go into reasoning loops and convince itself into looking at problems that don't exist @wary osprey

[tepid lynx]

how do i allow openclaw use a web browser?

(base) lepa_u@lepau-MS-7E02:~$ openclaw browser open https://example.com

🦞 OpenClaw 2026.3.24 (cff6dc9)
   Your task has been queued; your dignity has been deprecated.

│
◇  
opened: https://example.com/
id: 6DC009528391382C62C3D8B163350C89

works, but the ai cant use it

[tepid lynx]

when modifing openclaw.json to add:

      "browser": {
        "enabled": true
      }

Config warnings:\n- plugins.entries.browser: plugin not found: browser (stale config entry ignored; remove it from plugins config)

more exactly Config warnings:\n- plugins.entries.browser: plugin not found: browser (stale config entry ignored; remove it from plugins config)

[half saddle]

Hi. I’m exploring OpenClaw for app testing (QA-style): automated flows, screenshots, and error reporting. Does anyone know of docs, a guide, or a thread that explains in detail how to set it up for that? If nothing that specific exists, what area or topic in the project should I read first? Thanks.

[wary osprey]

Deepseek is way more direct imo, and it doesn’t try to overthink things or create extra reasoning paths that aren’t needed
Personally I use it for structured tasks, batch work, and anything where I just want clean output

sometimes I even switch to Kimi 2.5 for similar tasks depending on what I’m doing & it’s also very balanced and stable. It's also easy for me to move and change models since I’m routing everything through AIsa, so yeah .

you might want to try DeepSeek on the same task you’re running now, you’ll probably notice the difference immediately

[nimble hill]

openclaw is still in active development - anything that you develop is expected to break - the past few versions are examples of that. There is no specific workflows that you create. If you want some deterministic flow, start with n8n - you can develop a workflow and work like what openclaw does. Just that you would need to setup everything. You can ask openclaw to develop the n8n workflow for it, which is just a JSOn file. you can import that JSOn (workflow) in n8n and fine tune based on your testing

[ancient jungle]

They don't have a subscription plan like minimax do they. Only API?

[high heron]

no. deepseek is free. but they dont have oauth. so u can only use API

[ancient jungle]

It's not pay per token? Where's this at

[high heron]

i meant deepseek itself is free, but if u want to use deepseek externally (by api) u have to pay

[wary osprey]

no they don't. you can only use API! I use AIsa to route my models to my agent, and it's pay-per-use

[soft barn]

hey. is using openclaw with claude oath gone for good? no workaround, etc? only api from now on?

[ancient jungle]

I hate the pay per token model, makes it feel restrictive in the sense I have to care about managing context and efficient prompts

[wary osprey]

you can get it from their main API platform or use a unified API provider like AIsa or Openouter

[haughty cypress]

Buen día! El error del self_signed_cert como se resuelve en la instalación de openclaw? Hace hora estoy con esto

[ashen monolith]

Hey does anyone know the specific rules to positing in the ⁠self-promotion channel? Mine keeps getting blocked and will not let me post a github repo link

[wary osprey]

you don’t need to manage context as much, just keep your prompts a bit more tighter and avoid dumping full history every time, DeepSeek works better when inputs are clean, direct and focused. So for most tasks, short context and clear instruction gives better output

and if you want something a bit more flexible for longer context stuff, you can switch to Kimi 2.5

[dim harbor]

still works fine

[royal timber]

hi guy

i need help

[wary osprey]

Here #self-promotion message

[high heron]

wts the problem?

[ashen monolith]

so it just says no server invites? Im not inviting to a server.

[wary osprey]

Others are able to post their links, even github links #self-promotion message & #self-promotion message ... So just make sure your writeup doesn't have any Crqto related words

[ashen monolith]

it worked! appreciate the Crqto tip helped out. had the word scrape in it

[wary osprey]

You welcome 🙂

[prime fern]

is it possible to download local llms through openrouter similar to ollama. I like the openrouter feature that allows you to easily assign different tasks (like heartbeats) to cheap/free llms

[glass karma]

just restarted:
Mar 28 10:58:42 DESKTOP-QC51R71 node[672600]: 2026-03-28T10:58:42.037-04:00 [health-monitor] [discord:default] health-monitor: restarting (reason: stale-socket)
Mar 28 10:58:42 DESKTOP-QC51R71 node[672600]: 2026-03-28T10:58:42.082-04:00 [openclaw] Uncaught exception: Error: Max reconnect attempts (0) reached after code 1005
Mar 28 10:58:42 DESKTOP-QC51R71 node[672600]: at SafeGatewayPlugin.handleReconnectionAttempt (file:///home/jerron/.nvm/versions/node/v24.14.0/lib/node_modules/openclaw/dist/provider-CAlWEl41.js:3318:47)
Mar 28 10:58:42 DESKTOP-QC51R71 node[672600]: at SafeGatewayPlugin.handleClose (file:///home/jerron/.nvm/versions/node/v24.14.0/lib/node_modules/openclaw/dist/provider-CAlWEl41.js:3364:8)
Mar 28 10:58:42 DESKTOP-QC51R71 node[672600]: at WebSocket.<anonymous> (file:///home/jerron/.nvm/versions/node/v24.14.0/lib/node_modules/openclaw/dist/provider-CAlWEl41.js:3307:9)
Mar 28 10:58:42 DESKTOP-QC51R71 node[672600]: at WebSocket.emit (node:events:508:28)
Mar 28 10:58:42 DESKTOP-QC51R71 node[672600]: at WebSocket.emitClose (/home/jerron/.nvm/versions/node/v24.14.0/lib/node_modules/openclaw/node_modules/ws/lib/websocket.js:273:10)
Mar 28 10:58:42 DESKTOP-QC51R71 node[672600]: at TLSSocket.socketOnClose (/home/jerron/.nvm/versions/node/v24.14.0/lib/node_modules/openclaw/node_modules/ws/lib/websocket.js:1346:15)
Mar 28 10:58:42 DESKTOP-QC51R71 node[672600]: at TLSSocket.emit (node:events:520:35)
Mar 28 10:58:42 DESKTOP-QC51R71 node[672600]: at node:net:346:12
Mar 28 10:58:42 DESKTOP-QC51R71 node[672600]: at TCP.done (node:internal/tls/wrap:667:7)
Mar 28 10:58:42 DESKTOP-QC51R71 systemd[223]: openclaw-gateway.service: Main process exited, code=exited, status=1/FAILURE
Mar 28 10:58:42 DESKTOP-QC51R71 systemd[223]: openclaw-gateway.service: Failed with result 'exit-code'.
Mar 28 10:58:42 DESKTOP-QC51R71 systemd[223]: openclaw-gateway.service: Consumed 1min 1.814s CPU time, 1.2M memory peak, 0B memory swap pea

it restarted list this about every 20 minutes

[gaunt arch]

My openclaw is lazy and often lies to me. It claims to be working on a "task" but each time i prompt him, it will respond back but without any useful information.
Should i re-start the service? (I'm using ChatGPT 5.4 and Codex as the models)

[rancid thistle]

Yeah, it's super productive and always worn without producing anything. Sounds like a manager material to me 🤣

It's still a bit hit or miss, but if I write down the task with some definition of done and what I'm expecting as an outcome it helps sometimes.

[next onyx]

so whenever I send my openclaw a simple prompt, it uses from 20k to 60k input tokens, I don't know what is wrong, i'm using claude haiku 4.5, all my .md files in workspace are <150 words, exept agent.md, and I have my agent profile set to coding, it should not be uing so many tokens

[slim kiln]

hey guys, sorry for the noob question, but how can i set boundaries on the usage of skills for my agents? for example, i would like to only let the ai read my emails, and not send or edit and so on.

[queen cedar]

hey peeps, any idea why my agent using codex5.3, with tools enable, can't do action like create a folder? he did only 1 correctly and then as soon as I send a prompt, he doesn't think and execute, just answering like "I created the folder" but nothing is done. in the config, agent is set to "coding"

[slate minnow]

How is output supposed to be sent to a discord DM now?

[spiral parcel]

"error": "Error: Unknown Channel" Should it not be "--channel discord" ? I got right user id as well. " "status": "error"," ""deliveryStatus": "unknown","

[tepid lynx]

did i send this to the wrong channel? or correct?

[hybrid garden]

I noticed my model was chugging so I ran /compact and got:

"Compacted (0 → 23k) • Context 23k/120k (19%)"

it started at 0? Maybe that's why it didn't autocompact....

[glass karma]

I cloned and rebuild openclaw. and now it is like:
jerron@DESKTOP-QC51R71:~/openclaw$ openclaw daemon restart
Invalid config at /home/jerron/.openclaw/openclaw.json:\n- plugins.entries.bluebubbles: plugin bluebubbles: plugin requires OpenClaw >=2026.3.27, but this host is 2026.3.24; skipping load

• plugins.entries.discord: plugin discord: plugin requires OpenClaw >=2026.3.27, but this host is 2026.3.24; skipping load
• plugins.entries.feishu: plugin feishu: plugin requires OpenClaw >=2026.3.27, but this host is 2026.3.24; skipping load
• plugins.entries.googlechat: plugin googlechat: plugin requires OpenClaw >=2026.3.27, but this host is 2026.3.24; skipping load

[nimble hill]

.3.27?? Are you on dev builds? Use released version only. The last known good version is 3.24

[glass karma]

but 3.24 keeps restarting every 20 minutes. totally unusable

[sturdy oxide]

can someone help me with local ollama and openclaw ?, when i make ollama launch openclaw choose my model and write something like hi its say llm request failed network connection error, first of all its local model and when i use chromium i can use network perfectly so i dont know what is happening, this is running on a vps

[feral star]

Any tips to make openclaw better at filling out forms in browser. And just overall better at navigating sites

[quiet comet]

does anyone know how to remove this:

🦞 OpenClaw 2026.3.24 (cff6dc9) — Your task has been queued; your dignity has been deprecated.
when openclaw commands are run

[stoic anvil]

check if ollama is actually running and listening — run ollama list to confirm. Then check what URL openclaw is using to reach it, should be http://127.0.0.1:11434. If you’re running openclaw in docker but ollama is on the host, they cant see each other on localhost — you’d need to use the host IP or run both in the same docker network. Also check your VPS memory — local models can crash silently if they run out of RAM. Run htop or free -h to see if you’re maxing out.

[sturdy oxide]

i do free -h once a made choose the model or before ?, thank you for youre help

[stoic anvil]

Try to set OPENCLAW_HIDE_BANNER=1 in your environment. Add export OPENCLAW_HIDE_BANNER=1 to your ~/.bashrc or ~/.zshrc and run source ~/.bashrc.

[stoic anvil]

run free -h while the model is loaded and you’re trying to chat with it — thats when memory usage peaks. If you see swap being used heavily or very little free memory left, the model is too big for your machine.

[quiet comet]

thanks, sadly it didn't work, I tried a few variations on it too OPENCLAW_NO_BANNER OPENCLAW_QUIET, chatgpt wants me to make a shellwrapper

openclaw() {
  command openclaw "$@" | sed '1{/^🦞 OpenClaw/d;}'
}

[sturdy oxide]

you were right, its used 11gb ram and left me with 225mo free

should i use a lower model or another model

[shrewd perch]

Hello!
How do I configure an OpenClaw agent to initiate an outbound WebSocket connection to an external platform? I want the agent (not a client) to connect to wss://my-plateform.com/ws/agent/{id}

[south viper]

hi. i have an email agent configured to run on docker sandbox. i have a custom image with gog cli installed. is it even possible to build a container with gog authenticated ??
PS: i am using only a secondary gmail ID and i have the client_secret.json from it.. but it still needs a browser to get the redirect URL which is a manual step

[pulsar pendant]

Hey folk, really need your help: Anyone: "Getting 'ACP runtime backend is not configured' when my main agent tries to spawn subagents. How do I configure the ACP runtime? I don't see it in the onboarding wizard or openclaw.json."

[stoic anvil]

yeah thats the problem — 225mb free is basically nothing, the model is choking. Either use a smaller/more quantized model that fits in your RAM, or switch to a cloud API like claude sonnet or openrouter which offloads the work entirely.

[eternal cobalt]

how to i get started with ollama and open claw as a beginner? I'm new

do i download openclaw and ollama first?

[sturdy oxide]

install node js first

[soft juniper]

hi, is here a faq? i need help to connect ubuntu via webbrowser on a different computer in the same network LAN , but i getting this message :disconnected (1006): no reason, i tried a lot of things ...

[sturdy oxide]

after you can go the the ollama website and paste the command to your terminal and after that choose youre model, once choosen, do ollama run (model name ) , once installed do ollama launch openclaw

[eternal cobalt]

this?
powershell -c "irm https://openclaw.ai/install.ps1 | iex"

[sturdy oxide]

nope

wait

you are on window or linux or mac

[eternal cobalt]

windows 11

[sturdy oxide]

ok you have nodejs ?

[soft juniper]

is this also your own main computer ??

[sturdy oxide]

if no go in google type nodejs window install and find the .exe or .msi from nodejs

[sturdy oxide]

hope its a vps

[eternal cobalt]

I actually dont know what that is haha. I entered this prompt (powershell -c "irm https://openclaw.ai/install.ps1 | iex") and it mentioned node js accessing private networks. But its asking me "I understand this is personal-by-default and shared multi-user use requirs lockdown. Continue?"

[sturdy oxide]

i guess so

continue

[eternal cobalt]

its a gmaing laptop I have. Not my main desktop

setup mode quickstart?

[soft juniper]

becarefull to not give to much sensitive info , if i read your questions, you are more newbie than i am

[eternal cobalt]

yeah I am super newbie. I just want to use it for my business

[sturdy oxide]

i installed qwen3.5 0.8b hope he is not dumb

its working now thank

[eternal cobalt]

i did skip for now on the model, filtered all providers and now it says to select a default model what do i do?

[eternal cobalt]

idk what to do rn

to set up open claw :/

[indigo goblet]

You could read the docs and ask chatgpt or claude. Because if you have trouble for that, it's best to start with research

[sturdy oxide]

choose a model

[sturdy oxide]

you have cheap api to not burn money

[final patio]

below 9b most are dumb, i'm currently using nemo nano 4b, its ok like 9b.

[eternal cobalt]

i dont understand how this guy just clicks 2 buttons and has it all setup
https://youtube.com/shorts/cQCeKcbs45Q?si=Sl1NRFA3fAL2PK2_

[final patio]

ollama do setup for oc. but 2-3 clicks only, i didn't tried.

[raven pier]

Infinite GitHub OAuth loop after account deletion — same as GH issue #504. GitHub username: jeffvsutherland Can someone clear my Convex record?

[copper nova]

https://giphy.com/gifs/confused-nick-young-lkdH8FmImcGoylv3t3

how would we have access to your convex?

[fallen vortex]

guys what do you do when your gateway keeps disconnecting .. do you build an automation to keep checking and then start it or restart it ? or is there a native way to handle this

[copper nova]

build a cron watcher that checks if it's running, if not, start it

[dim harbor]

never had this issue, still on 3.13 consider rolling back and testing new versions before ugprading to see if theres issues with them and then write up issues ?

[fallen vortex]

yeah .. i am thinking about this .. still didnt figure out how to create a proper testing environment to play with as i am running everything locally until now

[dim harbor]

same, thats why im still on 3.13 heh

[elfin radish]

Hey gang, new here. What is the best way to get some attention on a PR fix for an iMessage integration issue (which is broken in the current release). I posted in #clawtributors

[copper nova]

i'm on latest and it's all fine? lol

[copper nova]

just post it there and maybe a gentle reminder every once in a while. There's like 6k+ PRs to sift through. I try bribing with pics of doggo

[elfin radish]

Ha my team bribes with chocolate 🍫

[upper bloom]

Do we not have any way to allow openclaw to send polls in telegram?

[elfin radish]

Hi 👋 — opened a PR fix for iMessage self-chat dedupe false positives that silently drop all inbound messages.

iMessage is essentially broken without this fix for those of us who like to use it.

PR: https://github.com/openclaw/openclaw/pull/55359
Issue: https://github.com/openclaw/openclaw/issues/47830

Would truly appreciate any reviews.

[copper nova]

in #clawtributors not here lol

[elfin radish]

Oops .. sorry for that. Will do. 🍫

[sturdy oxide]

you want to use local model or api

[sturdy oxide]

i will try

how much gb ram you have

[eternal cobalt]

just ollama kimi-k2.5 to test openclaw for free to get used to it

[mild kite]

Hey all, I setup my OpenClaw agent to send me a daily dinner recipie with what ingredients to buy and how to prep. Problem is--its only been recommending me salmon bowls and chicken tiki masala all week... help?😅

[final patio]

6gb nvidia 4050 didn't consider using llm earlier when buying.

[eternal cobalt]

have you told it to do anything else other than that? Like telling it to study gordon ramsey recipes?

[copper nova]

mayb it knows that's what's onsale and in stock at the local grocerie store?

sounds like a good agent if you ask me

[mild kite]

Thats probably the simplest and easiest fix. What other chefs can I ask it to imitate?

[eternal cobalt]

well i dont know any chefs but you can ask it depending what recipes youre looking for. Are you wanting cost effective high protein or lean protein or what?

[mild kite]

Initial prompt was for healthy balanced meals varying across cultures. Want me to paste the cron job?

[final patio]

How will the cron work if the gateway is down? Since cron relies on gateway as per knowledge.

[sturdy oxide]

oh ok, but im on vps so its cpu only, i will add swap and try

[eternal cobalt]

im not experienced in open claw i actually cant even set it up properly i was just trying to help with prompting. I know a lot about dieting and exersizing though.

[final patio]

Well then it makes sense that you are using a 0.8b model.

[sturdy oxide]

yep but i think swap can make me use 3b or 4b

[mild kite]

Yeah setup was not a cakewalk for me

[eternal cobalt]

what model do you use?

[mild kite]

Claude via API (not as cheap as I was thinking). Experimenting with using Haiku model instead of Sonnet for most of my chats.

[sacred otter]

@wicked rapids - M1. Just running octoprint on it at the moment. Thought I could have it do something more useful

(The other channel has slow mode enabled and I couldn’t reply to you)

[copper nova]

it is not for the feint of heart

[indigo goblet]

Bruv, do you really want to learn about it?

[eternal cobalt]

well yeah i want to get it set up quickly with ollama's kimi-k2.5

i just need an agent to do some creative strategy for me

[brave venture]

Hey friends! I recently connected Composio to OpenClaw and now my Discord bot won't come back online. Tried reseting token and updating .json file but still get the same error.

[mild kite]

Good luck. Good use case for your agent :~)

[velvet venture]

Yo guys.
I know this must be stupid question

but what should be the best OLLAMA LOCAL module for chatting and opening apps with the openclaw and maybe voice and tmux commands?

RX 550
Ryzen 5 5500

16 swap 16 gb normal ram

Cachy OS

IM sorry if any of you offend by this question

I know my pc is being bad.

[indigo goblet]

For a beginner, local might not be the best choice

[indigo goblet]

This is very wide and vague

Look at it this way friend, if for example, you do not have a very clear mental model of what do you want to do, the model won't guess it correctly like a human would. It can guess and fill the gaps, but it wont end well.

Because you had some issue to install openclaw correct?

[velvet venture]

no one?

[indigo goblet]

Hell, even human are not magic when it comes to assuming what another person is saying

[eternal cobalt]

yeah bro cant for the life of me

i want open claw to use a browser and scrape data off of ads library with a specific niche i will tell it to do wioth filters and words i use but do it for me

specificity isnt an issue with me.

[indigo goblet]

Ads library? Is this about automation?

[eternal cobalt]

i mean yeah

[indigo goblet]

Basically; your agent > ads library > extract certain info

[eternal cobalt]

yeah or cold dm people on my nextdoor app login

[indigo goblet]

So 24/7 or every 60 minutes every day or?

I mean the frequency lol

[eternal cobalt]

i really dont know yet because i dont even have the thing set up brother

60 minutes would be a start

[indigo goblet]

I see, well, best approach is step by step, so you could decide what do you want to focus on first, zoom in, multiple passes then next step and so on.

Whats your plateform

I guess the point im trying to make is, understand the framework first because otherwise its mostly a blackbox for you, which mean, the behind the scene is unknown and if things break (and it will, it's normal) you end up staring at the abyss.

[spare shard]

I dont think you could get anything with those specs tb honest but how could i know im not experienced maybe someone else will help you man. Someone who knows these things

Good luck with it

Youll need ti

[slate minnow]

Are you having a good experience with Openclaw? I seem to have nothing but problems. Occasionally I can get it to work, then I come back 2 days later and basic things are broiken like cron, agents talking, exec permsssions etc.

Are people really using this for anything half serious

Its just so hit and miss for me.

[spare shard]

It is usefull i used it but goddamn it takes a lotto token i couldnt even use it for a day ngl

I tried free providers like GROQ but instant rate limit cus of the RPM and tokens its usin up.

But it really works great

[slate minnow]

yup, one day discord works and its half usefl, next everything fails.

[upper bloom]

Do we not have any way to allow openclaw to send polls in telegram?

[slate minnow]

Hmm, I've been running it a few weeks but I spend more time fixing it, updating etc than actually having it do anything useful.

[indigo goblet]

Could you tell me what is your core setup, source of truth files, memory etc

[full shale]

Lol nobody responded to me for days it was kind of annoying i had to figure it out on my own

[spare shard]

Maybe read the logs?

If youre using linux just type
openclaw logs --follow and ask somebosy about it.

Pretty sure you will fix it quicker that way

Or use --deep insted of follow

[slate minnow]

I am running it on a linux VM. I've updated maybe 6 or 7 times over the past few weeks. Each time something new seems to break. I'm half ready to do a clean install again.

for instance agents handing work off to other agents. Seems flakey at best.

[indigo goblet]

Personal opinion but I wouldnt start straight away with sub agents

Your main is more than capable already

After fresh initial install, what did you do

[slate minnow]

I setup some cron jobs, which worked for a while, then it decided it couldnt send the output anymore to discord. Which worked fine.

Now when I ask it about cron tasks, it just look at the linux cronab -l, totally forgets about the inbuild cront jobs.

[spare shard]

Why a vm i got curious if you dont mind me aslimg

Asking*

[slate minnow]

I have a proxmox cluster here so easy to spin up new VM for this. I am tempted to do a fresh install from scratch with new version and see how it compares

[indigo goblet]

So you installed and went straight away to cron etc?

Didnt setup your agent core, source of truth files, memory etc?

[slate minnow]

pretty much, having it do some regular tasks for me seemed a good thing. Oh, I did setup the .md files if thats what you mean.

[indigo goblet]

Hm okey

Memory is extremely important for continuity and consistence

[slate minnow]

It used to be able to access websites with Brave API, now it decides it cant do that.
Maybe memory is the issue? How does that break?

[indigo goblet]

His soul.md and agents.md too

Well, I do not know, if it cant access brave api you have to back and forth with him

AI agent is not only you or the agent, more like iterative co design and pair programming

[olive breach]

hi guys i am new! I have setup my VPS server with the 20usd minimax token + 10 usd minimax plan, hwoever, its capablities are nowhere close to Claude, it can hardly keep memory and takes 10 seconds to repsond to a hi.

How do i go about this and give openclaw the power to browse my broswer like claude can do it for me.

(what I am trying to do- I am trying to allow openclaw login to my browser on instagram, scroll my msgs and DMs)- i have not setup tailscale and on a vps- reason i have not setup tailscale is it locks me out which results in nuking my server.

[indigo goblet]

Like your AI assistant is right there in your environment, best entity to ask. Sure sometimes another AI eyes is useful but yeah, you get what Im saying.

[olive breach]

i would like to give more context but i don't understand a lot of things and what should I share, i cannot code.

[dim harbor]

does it have the tool available, and then does it have the api key available?

there was also new search onboarding added, maybe do onboarding again ? id start with these 3 things

[novel summit]

Has anybody had situations where their agents just don't seem to have any memory or any ability to do anything useful? I'm completely set up, but can't seem to configure the config.json and exec-approvals.json, as well as the exact permissions file, to make my agents useful? Please help.

[glacial rune]

im trying to set up the openclaw thing in my console but when it asks to confirm in the console their isnt a yes option can someone help please

[desert gazelle]

I'm using openclaw via discord. Is there a way to stop the agent when typing? Sometimes it's off doing a task and I want to stop it but I can't seem to do so until it thinks it's done which can be burning tokens and time.

[indigo goblet]

Not sure, never used discord with it, but /stop or similar should work, did you ask your agent how to do it.

[dim harbor]

type stop or /stop, works on telegram at least

[desert gazelle]

That easy huh. 🙈 tyty

[indigo goblet]

Youre gonna have to be more specific, much more.

[swift helm]

Hopefully a super simple answer. I want to get openclaw to be able to create channels (and maybe categories) in Discord, should it be able to do it? I have spent 2 days arguing with it and had no luck thus far.

[indigo goblet]

Bruv, your firsf message is visible

Nobody is getting paid here, no need to spam

Oops sorry, your initial is gone.

Arguing? Hm, normally back and forth with your agent should fix the issue.

Not sure if it is possible but I would assume yes, anything can.

So your private server > agent to create channels etc?

[final patio]

I think /stop should work. Since it stops the model from execution.

[velvet venture]

Anyone? Please?

[final patio]

Try nemo-nano:4b

[velvet venture]

Whats the Pros and Cons

and is there more like that cus if that wont work i could try the other and other and other

Im not forcing you to look up

[indigo goblet]

Bruv, what?

You do know google is there right?

[slate minnow]

Thanks.

[final patio]

It's good in reasoning and has 256k context length plus it behaves like 9b model as per my experience.
U can go with lower param llm if this has very slow response after initial hi

[velvet venture]

Thanks.

[velvet venture]

ur talking to me or?

[indigo goblet]

Yes

[velvet venture]

Google Gemini API Key?

or gemma?

[indigo goblet]

No friend, I mean, with a quick search or chatgpt/claude, its pretty fast

For you to get an idea