users-helping-users

[shy ether]

"no machine"?

[analog venture]

remote software - liek vnc

[shy ether]

I'd blame that then. Never heard of it.

[bold viper]

Did u create a user there or using root?

[analog venture]

lol. ok. what difference would it make if it were RDP, vnc no machine, teamviewer, rustdesk?

its on the primary account on the laptop

[shy ether]

I don't know how no machine works. It may not be a 'real' desktop environment. Kinda like how screenconnect's 'backstage' isn't a real desktop either.

[analog venture]

i used no machine cause RDP was a pain and i couldnt get it to work

[shy ether]

I'd try onboarding openclaw whilst on the laptop itself, to see if it works.

[analog venture]

ill have a look and see if your poinjting me in the direction i can fix

no machine is just a screen share app in a way. i cant use the laptop direct cause the screen is dead lol

[shy ether]

Plug a monitor into it?

[analog venture]

you think im rich?

lol

[shy ether]

Plug it into a TV then?

[analog venture]

i dont own a tv lol

[shy ether]

I didn't know "if monitor == true then user = rich"

[analog venture]

i have a laptop im using now.. and a laptop with no screen for openclaw

[bold viper]

Did you start the gateway?

[shy ether]

Well, my suggestion would be to run VNC and see how that goes. That's how my setup works.

[bold viper]

Openclaw gateway

[shy ether]

They can't start the gateway. Nor get through onboarding, due to systemctl stuff.

[bold viper]

So onboarding shouldn’t be finished

[shy ether]

Correct?

[bold viper]

Yeah he said on on boarding finished

[shy ether]

Well, onboarding 'finishes' with it failing. 🤷‍♂️

[bold viper]

Lol

Check this: systemctl --user status
echo $XDG_RUNTIME_DIR
loginctl show-user "$USER" | grep Linger

If it gives Linger issue install it

sudo loginctl enable-linger $USER

Then try again the gateway install

[analog venture]

i can load the UI if i do it direct. but when going through the on board it fails

so i dont think its working.

ill be back soon.. will check a few things.

[bold viper]

Ok thats why I asked, there is a bug about thin in openclaw. https://github.com/openclaw/openclaw/issues/32635

There is a workaround there I think

[heavy sun]

Anyone know to to resolve "401 Missing Authentication header"

I went through the onboarding, added OpenRouter API Key, but now when I try to chat in the gateway to say "Hi" it gives me the above error

[thorny totem]

I just installed latest openclaw on a new vm

and whenever I try to setup telegram, it's just stuck.
When i let openclaw ai do it the gateway get's disconnected

[vital solar]

hi all. i wonder, if i subscribe to google ki, will i be able to use gemini in my ollama for openclaw?

@thorny totem try doing it using openclaw config (commandline)

whats the best self-hosted ai in ollama local when i only have 22GB RAM

and only vcpu no gpu

[thorny totem]

checkout new qwen models they are good

[vital solar]

i use qwen3.5:2b

right now

it still takes like 1:00 minute to answer

ius there any non-local ai i can use without token limit?

[analog venture]

when i use ollama i cant figure out how to limit context length. when i use LM studio i can specify when i load the model but i cant get openclaw to talk to it... damn why is this so painful

[swift willow]

theres 2 sections

1 is a model rebuild with the context hard coded into the model

the other is in openclaw.json

nano ~/Modelfile

add

PARAMETER num_ctx 32000

[worn echo]

Hey guys, any existant usecase/tutorial for a devops openclaw agent?

[swift willow]

then
ollama create qwen3.5-custom -f ~/Modelfile

and in openclaw it's listed under the model

[swift willow]

not really no

just use a smaller Qwen3.5 model

and delete your old agent sessions otherwise the times bloat out

just found that out

[blissful wren]

is there any solution for "typing TTL reached (2m); stopping typing indicator" ?

[swift willow]

That’s just the indicator it’s not really a big deal

What messenger are you using ?

[blissful wren]

telegram

[swift willow]

Yeah it’s hardcoded to 2 min in openclaw there’s no setting as yet

[blissful wren]

what about another messenger app?

[swift willow]

Nah it’s the same

I use slack I don’t even get an indicator

[vital solar]

@swift willow hey, which qwen model do u use for yourself?

how can i via openclaw command enable the tools write/read/ etc.?

[molten jackal]

[[read:1]]{"file_path": "/opt/homebrew/lib/node_modules/openclaw/extensions/acpx/skills/acp-router/SKILL.md"}

whats wrong with this… kimi/k2.5

[north plank]

does anyone know how to solve this - i setup an isolated cron with main agent to execute a workflow from a skill. the cron execution doesnt seem to load the "main" agent but more like a generic llm, so it couldnt find the skill

[swift willow]

I use Qwen 3.5 for main agent and deepcoder as my coding agent

You can enable tools by changing your tools profile to coder or full

[fossil shell]

Hi ! can anyone tell me how to allowed to execute things please? it does chat and search but it's not able to do the stuff I tell him to do

[vital solar]

@swift willow where can i change my tools profile?

[fierce arch]

when running a docker container inside the vps are you able to access that container to view/edit the contents of the .openclaw directory?

[swift willow]

In openclaw.json

[fierce arch]

there's no .openclaw in my directory it's currently within a docker container from Hostingers template.

[swift willow]

You can view what’s in the container but to edit you need to rebuild it every time

The workspace files however are separate to the container files

[fierce arch]

yeah that's what I'm looking specifically

[swift willow]

You’re in the wrong folder it’s ~/.openclaw

Notice the .

It’s a hidden folder you can see it with

[fierce arch]

i just remove the ~/ there's when typing it but that's what i meant... my CWD is /root and this is a fresh install.

[swift willow]

la

[vital solar]

thank you @swift willow

[north plank]

this is the follow up chat with cron execution, it didnt load any agent files

who are you?

openclaw-control-ui
18:43
🔧

I'm Claude, an AI assistant made by Anthropic. I'm here to help you with a wide range of tasks—coding, writing, analysis, research, file management, browser automation, and more.

I have access to various tools that let me:

Read and edit files
Run shell commands
Control a browser
Work with PDFs and images
Send messages and manage tasks
Search memory for context from our previous conversations
What can I help you with today?

[fierce arch]

does not show me that directory even with la

only shows me

.bash_history  .bashrc  .cache  .codeium  .lesshst  .npm  .profile  .ssh  .vim  .viminfo  .windsurf-server

[swift willow]

Should be off your home dir not off root

You should be logged in as a user not root

[meager vessel]

Is it me or does openclaw feel more restricted after all releqses.. Do you guys have any tricks How to make it more free it was before?

[swift willow]

free as in like? Exploitable ?

The restrictions are for security reasons

There’s a lot of bad actors trying to exploit openclaw servers atm

[fierce arch]

guess I'll recreate the system and redo everything, appreciate it.

[swift willow]

Make a user and use sudo for privileges escalations

don’t touch root

[meager vessel]

Exec/bash tool, cant me executed..

[swift willow]

Change your tools profile to “coder” or “full”

"tools": {
"profile": "coding"

in openclaw.json

[meager vessel]

What The difference between coding and full?

[swift willow]

Not much far as I can tell

[meager vessel]

Thanks i will check it now!

[swift willow]

coder is specifically for coding exec I guess

[vital solar]

@swift willow what was the default not able to read etc. profile?

found out thanks

[steep dragon]

hello, any1 here to help me on a doubt?

i got the api limit error, just created an openrouter account and a fresh new apikey, configured the api key on $openclaw configure, set it to default, but the API error persists.... anyway i can reset that error since i have a new key ?

[swift willow]

Messenging is default

[swift willow]

What model did you pick ?

[steep dragon]

openrouter/auto

[swift willow]

You going totally free ?

[steep dragon]

yes

[swift willow]

yeah that doesn’t work so well lol

[steep dragon]

im testing it first to understand if its worth spending money xD

[swift willow]

try nvidia free model does that work ?

[steep dragon]

i can try, but openrouter has 0$ spent yet

[swift willow]

do you have a gaming GPU ?

[high heron]

nvidia takes like 2min to reply... is it only me? or wt

[steep dragon]

yes

[high heron]

groq's also fine

[swift willow]

yeah the main models won’t work only 1 or 2 of the free models work

[high heron]

and mb qwen or minimax oauth is also solid

but u have to login to qwen every day since it refreshes cache

[swift willow]

I could only get nvidia free model working myself

But that ran out quick

groq I could only see paid plans

[high heron]

just use openai

they have frree codex limit rn

[steep dragon]

penai is the one i was using and got the api error

openai*

[high heron]

or get amazone bedrock or google cloud and get free credits to use it

also
shebuilds.lovable.app

anthropic gives 100USD worth API here. its womens day event or sth like that.

but it expires in 1 day

so idk 🤷

mb good to test claude

[swift willow]

I couldn’t run groq through my vpn they kept blocking it

[swift willow]

Try NVIDIA: Nemotron free on open router

[high heron]

or just get alibaba's coding plan or sth

[steep dragon]

so cant reset too use the fresh working api from openrouter

[high heron]

$3 /month, $5, and $10 afterwards ig

[scarlet silo]

hello all, i couldnt find documentation on my problem:
I have built a pulgin that uses:
api.registerHttpRoute
i have auth: 'gateway' but i cannot access that http page. i always get {"error":{"message":"Unauthorized","type":"unauthorized"}}
i am using a token for the main page.
gateway on a docker container with:
"dangerouslyDisableDeviceAuth": true,
"allowInsecureAuth": true

[high heron]

or jsut link sht ton of chatgpt or google cli

[swift willow]

Are you trying to access the webui ?

[scarlet silo]

i can access the webui: http://192.168.93.35:18789/
but not: http://192.168.93.35:18789/agent-activity-monitor/

[swift willow]

Is that the machine Openclaw is running on ?

[scarlet silo]

yes

sorry, no.

[high heron]

wt is agent-activity-monitor? new update or sth?

[swift willow]

What about local host 127.0.0.1:18789

[scarlet silo]

the gateway in on a docker on another machine

[swift willow]

ok yeah I had that problem

I had to pipe local host to my other pc via ssh

ssh -L 18789:127.0.0.1:18789 192.168.93.35

Then access it via local host ip instead

https://127.0.0.1:18789

[scarlet silo]

i am already doing the ssh tunnel to haver my browser controlled.
but to access the webui i just disabled all security and i can access remotly with no problem. just get the security warnings on log.
but its just doesnt work for this one

[swift willow]

shouldn’t have to disable any security

just setup your auth token and approve it on the server

[scarlet silo]

ok, let me go and try again

[swift willow]

Just make sure you’re paired up after you enter the token

[slow hearth]

hey guys! is running openclaw gateway restrart is the equivalent of killing the openclaw gateway process and rerun it?
I'm asking this because restart gives me

Gateway service check failed: Error: systemctl is-enabled unavailable: Command failed: systemctl --user is-enabled openclaw-gateway.service

I'm running openclaw on wsl

[scarlet silo]

ok, main webUI works but when i add the path to plugin it doesnt
http://127.0.0.1:18789/agent-activity-monitor
get: {"error":{"message":"Unauthorized","type":"unauthorized"}}

[slow hearth]

you gotta add the token to the url. When you run openclaw dashboard the full url with the token shows in the terminal

[scarlet silo]

http://127.0.0.1:18789/agent-activity-monitor/#token=...
i get same error

is there another plugin that also creates a route for and html page so that i can test?

an*

[swift willow]

Might run on a different port

most monitor plugins do

Check what ports are listening on the server

ss -tlnp

then would be the url:port

Or try a different activity monitor

[vital solar]

is there a way to reset the bot and start from fresh? it has messed up creating IDENTITY.md and so on

[swift willow]

https://github.com/robsannaa/openclaw-mission-control that one 3333

[swift willow]

You can remake your workspace files manually

[vital solar]

no it didnt ever write to it

it wont show me what to put in there aswell

also it misunderstood me and now it thinks its called like me

[swift willow]

https://docs.openclaw.ai/reference/templates/SOUL

There’s the default templates

Then delete all your agent session files

[meager vessel]

thanks!!!

[vital solar]

cant i just delete my workspace?

[meager vessel]

feeling like my old openclaw!!!

[swift willow]

You might have to onboard again if you do that

might work

[vital solar]

oh no

i dont wanna onboard again i dont wann ahve to retype all the botkeys

[swift willow]

Just copy and paste the templates over the existing ones

and delete agents/main/sessions/.

[vital solar]

ok thanks

[scarlet silo]

worked now! was a plugin problem. wrong settings.
needed this:
auth: "plugin",
match: "exact",

thanx!

i am loosing soooo much time repeating stuff. openclaw forgets everything
is memory-lacedb-pro the way to go?

going crazy over here

[vital solar]

hm, llm-checker suggests qwen3:1.7b for my hardware overall best recommended for me

is that model ok?

[vital solar]

what is #freshbits ?

[devout swan]

sorry but how the heck do i install it

i did the command on powershell but it just closed

[vital solar]

@V hm, open terminal

is he gone?

press windows key

type in: terminal

open it

[thorny totem]

openclaw onboard

[vital solar]

select powershell

type in the command to install openclaw

send screenshot or copy/paste

[devout swan]

isnt it on powershell ?

[vital solar]

wait

[devout swan]

it say i need to install git

im windows

[vital solar]

u need to execute the .ps1 thing

yes

[thorny totem]

just use the curl command you saw on the website.

it should install git and node. Both are required to run

[vital solar]

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

try that

[devout swan]

on terminal ?

[vital solar]

yes with powershell as interpreter

[devout swan]

i did it but it only installed node i think

[vital solar]

it will not close instead give u reply

u need to fully install it

it should not close

[devout swan]

so powershell and terminal need to both be open ?

[vital solar]

no terminal opens powershell

terminal is the window opening powershell then into this window u type in the command: iwr -useb https://openclaw.ai/install.ps1 | iex and it stays open executes the ps1 script which is a powershell script

[thorny totem]

https://docs.openclaw.ai/start/getting-started#windows-powershell

paste this in chatgpt and ask it to help you install openclaw

[devout swan]

erm now i get a red text

i think its my pc the problem that what say the red text

its say that script execution is not enabled for this system

ty guy its seem to be working now

[snow thicket]

jesus, there are plenty of bugs on the latest 2026.3.7 version

Cron doesnt fire, agent doesnt followthrough and drops the task,

[vital solar]

V: great, hf 🙂

[desert plaza]

Maybe a silly q but Im a discord noob: i'd love my openclaw to crawl this discord using peter's new discord crawler to checkout find itself cool stuff and propose me to add it or find solutions for problems its having. Would this be possible or do u need special permissions to add a bot to this discord? https://github.com/steipete/discrawl.

[paper ingot]

Hi there, new here and struggling to get openclaw working, i've tried in docker, couldn't find a way to access the web ui since its only listening on localhost 🙁

[queen mango]

Hi everyone. I've been trying to install openclaw for the past 2 days and didn't manage to get it working right, the best I got it was when I instructed it to get new skill on the clawhub and use it... After that it went downhill, it just randomly forgot how to do anything, I've tried reinstalling it from scratch, but it seems something always stays in there. Then it went crashing every time it tried to use any skills or just do anything(even just to check workspace). Now it crashes constantly and can't even overwrite anything in it's workspace.

Here's a brief TUI example:

 Hey. I just came online. Who am I? Who are you?                                                                                              
                                                                                                                                              
 Let's figure this out together — pick a name for me, what kind of creature I should be, and how you want me to show up for you.              

                                                                                                                                              
 I'm John Doe, the owner of this realm(my PC). You will be my coding agent, Ruby. You'll help me with problems I get stuck on and help me     
 build projects. Got it?                                                                                                                      
                                                                                                                                              

 Got it, John. I'm Ruby — your coding companion. I'll help you untangle the tricky bits and build whatever you're dreaming up. What are we    
 working on?                                                                                                                                  

 run error: terminated                                                                                                                        
 connected | error                                                                                                                            
 agent main | session main (openclaw-tui) | custom-api-alltokens-ru33/z-ai/glm-5 | tokens 10k/16k (65%)   

and logs:

16:10:08 warn agent/embedded {"subsystem":"agent/embedded"} low context window: custom-api-alltokens-ru33/z-ai/glm-5 ctx=16000 (warn<32000) source=modelsConfig
16:10:08 warn tools {"subsystem":"tools"} tools.profile (coding) allowlist contains unknown entries (apply_patch, image). These entries won't match any tool unless the plugin is enabled.
16:10:46 warn agent/embedded {"subsystem":"agent/embedded"} embedded run agent end: runId=5d8a2bf0-bc84-4d4b-9875-9fd8c144f481 isError=true error=terminated
16:10:54 warn agent/embedded {"subsystem":"agent/embedded"} embedded run agent end: runId=5d8a2bf0-bc84-4d4b-9875-9fd8c144f481 isError=true error=terminated
16:11:12 warn agent/embedded {"subsystem":"agent/embedded"} embedded run agent end: runId=5d8a2bf0-bc84-4d4b-9875-9fd8c144f481 isError=true error=terminated
16:11:27 warn agent/embedded {"subsystem":"agent/embedded"} embedded run agent end: runId=5d8a2bf0-bc84-4d4b-9875-9fd8c144f481 isError=true error=terminated
16:14:28 info gateway/ws {"subsystem":"gateway/ws"} webchat disconnected code=1001 reason=n/a conn=e32c6239-ebe9-478f-8811-d4a77afcd84f
16:14:29 info gateway/ws {"subsystem":"gateway/ws"} webchat connected conn=fe888b08-03d5-4fb9-8424-b95fc8e9a0bf remote=127.0.0.1 client=openclaw-control-ui webchat v2026.3.7
16:16:11 info gateway/ws {"subsystem":"gateway/ws"} ⇄ res ✗ logs.tail 0ms errorCode=INVALID_REQUEST errorMessage=invalid logs.tail params: at /maxBytes: must be <= 1000000 conn=5107f080…06b2 id=652853c6…9bbb

[queen mango]

I got the same treatment bro, but at least you can try accessing it through telegram/discord/whatsapp or whatever

[paper ingot]

how can I even access it if it doesnt even starts properly ahah

[queen mango]

why wouldn't it start?

although it can't work properly with homebrew

so this part doesn't work

[buoyant owl]

So im not the only one that the agent doesn’t execute any command?

[snow thicket]

Yes, this is my 10th agent. I just did a fresh install for a different usecase. good thing i havent updated my other agents yet

I saw someone created an issue https://github.com/openclaw/openclaw/issues/39866 in github

[snow thicket]

They are just telling that they are executing it. But no response after that.

@_@

[buoyant owl]

Ya same here. Thanks I was struggling

[snow thicket]

Let's wait for the next update. I hope this will be fixed soon

I literally burned 4 hours of troubleshooting

I thought it was an OS issue.

Tried it on both Ubuntu 24 and 25

same result

[mild heron]

I'm burning waay more than 4 hours just trying to access my gateway.... :/

[unborn igloo]

i'm not a dev and more just curious, but if you run "nano ~/.openclaw/openclaw.json" and look at what your "profile" is under "tools", is it "messaging" or "full" or something else?

"tools": {
"profile": "full",
"web": {
"search": {
"enabled": false
},
"fetch": {
"enabled": false
}
}
},

[somber smelt]

Hi guys, one question: how do you guys afford running openclaw haha its seems super expensive to me, which models and strategies do you guys use

[high heron]

chatgpt codex

[dim harbor]

yeah i think thats a common issue these days, it was defaulting to messaging. appearently they just updated that to default to coding so it can run some commands

[high heron]

you can pay 20USD/month for plus, and use ~80USD/week

by using oauth method

[queen mango]

When did you install your last working agent? Wouldn't it work if we just installed them from last working version?

[high heron]

and you can also use qwen, minimax oauth as fallback or sth

alibaba's coding package is also good

[unborn igloo]

yea, this was an issue for me. is there a documentation on the different permissions and profile types? couldn't find it myself.

[dim harbor]

yeah they have it documented: https://docs.openclaw.ai/tools#tool-profiles-base-allowlist . I ended up just asking my agent on whats restricting him and he kind of figured it all out. that + #1459642797895319552 bot

[unborn igloo]

thank you 🙏

it's not clear from the docs, but do you know if "coding" allows read / write / edit access to files? obviously "full" does (which is what i adjusted to because i couldn't get my agent to do anything) but i'd rather restrict permissions than grant everything.

ok just answered that myself by changing and seeing - yes, it does. the only thing coding doesn't do that full does is web/browser stuff

[late herald]

Is there a guide for best practices on setting up your open claw with things like email access and giving it some money?

Btw what are you using to give your open claw an email and limited money to spend?

I created a new Gmail for it, but it got banned by Google for being suspicious (bot). And I can’t create any more Gmail’s on my existing phone number. And new virtual phone numbers can’t be used for Google sms verification for example.

[buoyant owl]

i changed to profile coding

[unborn igloo]

did it fix your agent's permissions?

[buoyant owl]

nope, still debugging

[placid steppe]

Guys what do we use this server for? How can it help us?
Anyone explain briefly.

[slender ore]

My agent is still asking for approvals. How do I fix this? Here is his message - I’ve queued a new script (agentmail_inbox_reader.py) that just hits the AgentMail inbox and prints timestamp/from/subject lines—exactly what you asked for. Creation command is pending approval now (id=4e1d2e21). Once it lands I’ll summarize how to run it.

[slender ore]

I used agentmail. It has its own API.

[dim harbor]

you should see the approval messages throuh control ui / chat at least ? or you can give openclaw permission to do it without the approval, if you dont care if whole thing blows up 🙂

[fierce bobcat]

is there a way to help my agent keep the thread of conversation when I'm messaging it from multiple channels (i.e. slack, telegram, and the dashboard)? I'm on a new install and it currently retains nothing from one channel's conversation to the other, so I'm constantly having to tell it what I already told it in another channel. 🙃

[dim harbor]

basically they are using own sessions in every chat. ask about how to set up shared session, I think i saw its possible. so everything would just write to the main session for example. for chats that you wanted to be shared

[neon jolt]

gm fam, curious as to any memory solutions you are using

[fierce bobcat]

I set up agentmail as well. Then I shared my google calendars with that email address, as you would an assistant. However, it needs a google account to do that. 🙁

UPDATE: I ended up setting up google OAuth credentials for my agent to access my calendar. Pretty straightforward.

[cedar patrol]

OMFG Compacting context again... 😵‍💫 Ung

[unreal bear]

How do you stop your claw from screwing up the json

Every time I have it do something the json is screwed and i need to fix it

Add a telegram group? Bad Json add a model? Bad json , create a subagent? Bad json

[azure cosmos]

If you are talking about the openclaw.json, I found success telling my Agent to review the github for new release information and I'd point it to the documents page, https://docs.openclaw.ai/ I think, for mine, it was making updates to the config that were outdated because of updates

[unreal bear]

I’ll try that

Thanks!

[mighty rampart]

I tell it to only use the command line and to ask for permission before it changes the json directly. Nearly everything in the config can be changed, correctly, with ‘openclaw …’ commands.

[primal rapids]

are any of you guys using the second brain use case? i'm wondering how does this differ from the memory.md? is it more of a frontend for users to understand what's in their "second brain"?

[wintry hatch]

what is your go to agent as main? i am considering kimi k2.5 because my budget is not that high

my current setup is the free api i get from openai plus plan for seperate agents

[blazing schooner]

I’m planning an OpenClaw setup and I’m trying to match hardware + cloud model costs to my use case.

I want multiple agents running simultaneously (roughly 9–5, 5 days/week).

Each agent has a different role, but most will need heavy coding and Chrome browser automation.

I want to use cloud models (OpenAI + Anthropic + Gemini), and ideally have the system use the cheapest model that can solve each task (cheap defaults, escalate to premium only when needed).

My constraints:

Mac mini with 16GB RAM

Hard cap of ~$200/month total cloud spend

Questions:

With browser automation, what’s a realistic number of concurrent Chrome agents on 16GB before it becomes unstable/slow?

What’s a realistic token/step expectation for browser automation in OpenClaw, and how many steps/day could fit under $200/month using cheap models most of the time?

What’s the best recommended model routing strategy in OpenClaw for cost control (cheap primary + premium “escalation” agent, per-agent model defaults, fallbacks, etc.)?

Any config tips to reduce costs (compaction, max output tokens, limiting page context, limiting tools, queuing browsers)?

[scarlet silo]

anyone knows how to migrate memory to memory-lancedb-pro?

[near herald]

Whats the best way to define fallbacks models per-agent? Seem like adding a fallback param in the openclaw.json file agents.list[0].models.fallbacks isn't accepted. 🤔

[short kayak]

Is anyone using Traefik in front of OpenClaw' docker container?)

[near herald]

I use profiles to define multiple agents and make them talk through sessions tool . But each agent has it's own role and set of models to use. I tried adding fallbacks per agent but it get rejected when I restart my gateway.

[unborn igloo]

using claude via 0auth no longer supported?

[plain cradle]

anyone else having problems with tool calls? I'm getting just text output
exec:13{"command": "echo "Tool execution test"", "description": "Test if tools are executing"}

[gaunt stream]

Hey folks running openclaw on a VPS - I'm a newbie here. How are you guys giving your agents browser access to sites like Reddit or X? (Seems that these sites have an aggressive bot detection)

[hidden ferry]

just get the 200$ codex subscription lol

[idle juniper]

i want to ask can we connect openclaw with openrouter if yes what model will work best and give free requests

[sudden knot]

what’s the best free api and model to use for when i hit my codex rate limits

[snow thicket]

Yeah I havent tried rolling it back to a previous open claw version.. but my other agents are running on a previous release and it is stable. Lesson learned need to take snapshots before updating

[tidal pelican]

Anyone got an idea of whats the best way for memory so i'm spending less and my ai not having amnesia? I'm seeing a lot of stuff and youtube and online but i'm very confused.

[queen mango]

I tried installing it in the ubuntu 24.04 container from a 24.02.2026 npm package, still no luck: can't use any skills and just crashes instantly. Maybe it's my provider's fault? Though strange, any normal request works fine...

I wonder, If I can try manually sending api requests to see what's the problem

[open saffron]

I'm still new and struggling to undersand model usage limits. Mostly because I can't find the information on the websites. For example, I see you can use the kimi-k2.5 model through Nvidia and you can use a Qwen3 model through OpenRouter but are there limits?

[unborn igloo]

about to go down this path myself. i think running a residential rotating proxy might assist

[modest cairn]

hi, does anyone know : How do I enable exec tools for my agent in a Telegram DM session? Config has tools.profile: coding but exec isn't showing up as an available tool.

[errant scarab]

Anyone know how to run openclaw on an OpenRC system and not systemd

[blazing schooner]

Im deeply confused how people are getting around the cost of actualy successfully using claw when the usage per llm is so expensive, i think im missing something

[unreal bear]

Try openclaw config set tools.profile full

My old openclaw used to have all exec privileges and didn’t ask me a damn thing. After 3/2 and 3/7 updates it’s so damn needy, anyone know what I can run to make it just do things?? Node and terminal have full disk access

[next trench]

i would like the option to say its running on its own machine and to not degrade its access

[modest cairn]

still does the same unfortunately

[unreal bear]

THIS

[near herald]

You have to set tools.profile to "full" in the configs

I fought like crazy to free my agents from this hardening nightmare 😅

Defailt is in tools, and you can set it per agent profile in agents.list[].tools.profile there's like 3 or 4 different options for tools permission profile.

[unreal bear]

I did that after the 3/2 nerf but now it’ll do most things but still ask me for a lot of other things. Like running flutter or Xcode

[near herald]

Ok hmm, not sure then, so far that solved it for me. And then theres the other file exec-authority if I'm not mistaking where you can add more exceptions

[unreal bear]

Oooh

[worldly zinc]

Hey guys, I've installed openclaw, the issue I have is using ollama to run a model locally, when ever I install a model, my telegram no longer responds, anyone had a problem like this before?

[fiery latch]

Hi guys, I have seen many people running openclaw with multiple instances in one machine but I have seen some posts where people run many MacMinis or PCs to run many openclaw ...why? is there a limit or why do you need that many PCs?

[high granite]

Talk to your bot through the web chat. Give it search and tools. It will talk you through it. It can even fix 99% itself..

[mighty rampart]

Yes when I first started I picked a model that didn’t support tool calls and nothing worked. Go on the web ui for the gateway and look at the sessions to see what is happening.

[mighty rampart]

What use case do you have for multiple instances? Are you supporting multiple (human) users? You can run multiple agents inside one instance and that’s generally sufficient.

[fiery latch]

No usecase from my side. I am just curious why people have so many PCs or MacMinis with OpenClaw when I have heard that a simple Raspberry Pi can do the job. I am really new in this world

[mighty rampart]

You don’t need a Mac mini, use what you have and see if it’s going to stick before you think about upgrading. 🙂

[mighty rampart]

Another suggestion, install Open WebUI and make sure all your ollama models are working outside of openclaw.

[next trench]

In some cases it's because they want multiple main agents and want separation and/or their machine can't handle running multiple workspaces. In my case, I'm running two main agents/workspaces on one mac mini totally fine but i have one agent on an old macbook that's a PITA to manage and it would be a challenge to run more than one agent there. If you organize your tiers efficiently with one main agent with a layer of perm agents and a third layer of perm/temp agents that is a great start.

[slender hound]

Hi everyone, I am hoping someone can help me finally push through this issue .. been hardstuck for some hours now..

Doctor warnings:
"channels.telegram.groupPolicy is "allowlist" but groupAllowFrom (and allowFrom) is empty ..."

Can anyone assist me on how to fix this?

[next trench]

I did that only to find out that most don't work with my OpenClaw on the old machine even though I looked at the specs. It was interesting to benchmark and learn about the models at least.

[next trench]

this is what I have:

"channels": {
"telegram": {
"enabled": true,
"dmPolicy": "pairing",
"botToken": "${TELEGRAM_BOT_TOKEN}",
"groups": {
"GROUPID": {
"requireMention": true,
"groupPolicy": "open"
}
},
"allowFrom": [
"USERID"
],
"groupAllowFrom": [
"USERID"
],

[slender hound]

Thanks - do I paste this inside the power shell?

[next trench]

I edit my openclaw.json directly.

i have a process to track changes etc. but i find that is the fastest.

be sure to restart gateway after

[slender hound]

Thanks the User ID, do I find that inside telegram or in the powershell onboarding process?

[next trench]

Telegram

you setup BotFather?

you have to pair Telegram to Openclaw via BotFather

[potent geyser]

Send /start to @userinfobot on telegram to find your User ID.

[worldly zinc]

I did that at first, then it stopped responding, tried getting a guide from chatgpt and gemini but I think they got stuck in a loop cause the commands were no longer working, then I tried chatting from the terminal to trouble shoot why the telegram gateway stopped interacting with the models, and I still got no where close to the solution. Openclaw is running, ollama is running too, the models I download run, but when it comes to the telegram gateway, it keeps throing an error.

[worldly zinc]

How do I give it these search and tools, I'm not a dev so yeah, I might be making some mistakes. I installed wsl on my windows to avoid powershell errors and freezing.

[slender hound]

Thanks a lot, got it!

[worldly zinc]

YEah, I uninstalled that model, at first I picked the qwen 3.5b turns out my system was having an issue running it locally, then I choose a lesser model, and that doesn't support tools so I deleted it, now it seems I have to use cloud options, only issue I have is api credits payments.

[worldly zinc]

Ok, i'll try this, thanks

[worldly zinc]

oh, I see

[blissful berry]

If you have multiple agents configured, does the primary one "triage" to the others based on a determination or... how does that work?

I am using qwen3:30b and it is painfully slow, so I wanted to divvy up reasoning, direct coding questions and agentic coding into different agents.

Or is this just the wrong way to think about how this all works

or should work

[slate minnow]

Hey facing this issue for no apparent reason, anybody else?

  "error": {
    "message": "Your refresh token has already been used to generate a new access token. Please try signing in again.",
    "type": "invalid_request_error",
    "param": null,
    "code": "refresh_token_reused"
  }
}```

[mighty rampart]

How did you configure the auth? It means what it says, something already used the refresh token and now you are stuck. You will have to re-do whatever you did to set up the authentication. (Or switch to using an API key, that won’t break.)

[slate minnow]

ok, thanks. Its just odd it happened for no apparent reason that I can see. Oh well

[mighty rampart]

If you create multiple agents you can talk to them separately — bind them to different slack channels for example. The main agent can send them messages as well, (if correctly configured,) but I’ve had trouble with it trying to take over and just do it rather than hand off. It can work to have the main agent write a file or open an issue in a tracker, and then have the other agents look for those signals on their heartbeats. There are a couple of good articles, look for “How I Built an Autonomous AI Agent Team That Runs 24/7” by Shubham Saboo, and also what FelixCraft is doing running a business.

[blissful berry]

oh I didn't think about using a tracker of sorts

Is there a way to smartly spin down and dynamically spin up the different agents? I can't run them all at the same time of course

oh that article is neat

[mighty rampart]

Each agent has its own configuration. I am not sure they are really “running” … they have a heartbeat when they wake up and see if they need to do anything. Set that to once an hour or once a day if you are going to be messaging them when you want them to do something (if the something is fairly short and synchronous.) A nightly memory curation is a good idea for all of them.

[blissful berry]

I'm just worried about vram, gwen3:30b gobbles all of it up on it's own

[mighty rampart]

I am still getting the observability sorted but so far I think the gateway things (receiving and routing messages, running cron jobs) are using the regular CPU and memory, not the VRAM. The built in diagnostics-otel plugin is not cooperating.

This is a deep dive on the OC architecture https://medium.com/@gwrx2005/proposal-for-a-multimodal-multi-agent-system-using-openclaw-81f5e4488233

[blissful sluice]

Hi, I have been working with openrouter.ai but today I am getting "HTTP 404: No allowed providers are available for the selected model." when trying to use minimax or kimi. Does anyone know if there's something wrong on my configuration or if somehting is going on in openrouter.ai? I have checked and I don't have any restrictions that may generate this situatoin.

[wild granite]

hey guys anyone having issues with vps hosted openclaw? since the update I get this error: device identity required

[restive fossil]

I've tried a bunch of different settings but I just cant figure out how to get slack setup. I tried a few different tutorials like https://www.c-sharpcorner.com/article/the-complete-guide-to-integrating-slack-with-openclaw-2026-the-steps-most-ai/#:~:text=Step 1 - Create a Slack,the communication bridge with OpenClaw.
I can send messages to slack from the TUI but not from Slack to OpenClaw.

[sinful verge]

my openclaw is stuck Compacting context... even after an openclaw gateway restart and a openclaw doctor --fix and i'm on the latest 2026.3.7 using gpt-5.4, can anyone help?

[astral yacht]

https://www.answeroverflow.com/m/1472418082180829204

[smoky cliff]

hello, I rlly need help, my Openclaw is running on a VPS and my AI wont DO anything. It responds like a normal AI. Can anyone help me?

[ivory hedge]

hey guys. is anyone still having success with the claude-max-api-proxy workaround or have most people found that getting shut down now?

[ivory hedge]

hey @smoky cliff ... Hey — this usually means OpenClaw is running, but the agent doesn’t actually have any tools available, so it behaves like a normal chatbot.

A couple quick things to check:

1. Check what tools the model actually sees
   In the chat run:
   /context list
   (or /context detail)

If you don’t see tools listed there, the model literally can’t perform actions.

2. Check your tool restrictions in config
   Look at these settings:
   tools.profile
   tools.allow
   tools.deny
   any per-agent tool overrides

Important:
minimal profile = almost no tools (basically chat only)
full or unset = normal tool access

3. Confirm what backend you're using
   Some CLI fallback providers disable tools, so the AI will only respond with text.

4. VPS note (very common issue)
   If OpenClaw is running on a VPS, it can only act inside that VPS environment (as I"m sure you know but worth saying anyway).

5. If you want the others (many/most of whom are probably more knowledgeable than me) here to help debug quickly, paste:

/context list
openclaw status --all
openclaw gateway status
openclaw channels status --probe
openclaw logs --follow

Also mention:

• your model (the "brains")
• your tools.profile
• whether you’re using nodes

That usually makes the problem obvious pretty fast 👍

[ivory hedge]

hey @sinful verge ... good chance i have some useful thoughts for you here. need to step away but will try to remember to follow up when i'm back in a bit

[smoky cliff]

Could u help me with this??? Or do u not have time

[ivory hedge]

yeah did you try everything above?

start with /context list or /context detail

lmk what it says

i assume your issue is that it was working in the vps at one point but stopped, is that right?

@smoky cliff

[smoky cliff]

@ivory hedge no it never worked. I mean it functions like a normal AI, but doesnt do anything else at all

[ivory hedge]

got it. which vps vendor are you using? what environment is it setup in do you know (from an operating system standpoint)?

[stable cedar]

Hi dear,
Has anyone used GLM-4.7-Flash as the local model to do your work? Looks to me that I have to keep my eyes on her work. I feel so frustrated... She promised everything, but couldn't deliver. The logic kept changing. For example, she started daily diary in memory folder, but all the sudden stopped until I found out... She made backups folder and backed up a few into that folder, when I asked her to clean up some unnecessary files in parent folder of "backups", the backups folder was removed as well when she cleaned up the unnecessary files

[iron shuttle]

anyone having issues with compactions using the codex models OAUTH? I cant use them to compact

[stable cedar]

@ivory hedge where is tools.profile? I could get answer from "/context list" in prompt window

[empty totem]

My claw is sharing this problem. What’s the best way to post this

Numbers from a 3-week-old instance:

• Total sessions in index: 259
• Cron entries: 229 (88% of file)
• Dead cron entries (no sessionFile): 224
• Unique skillsSnapshot variants: only 2 (same 31KB copied 228×)
• Waste: 8.3MB of 9.3MB total is dead cron metadata

Impact: sessions.json loads on every inbound message. Heavy cron setups hit this fast and it never shrinks.

Expected:

• Prune completed isolated cron session entries from the index after run finishes (or on startup)
• Store skillsSnapshot by hash/reference, not duplicated per session

Workaround: None safe — editing sessions.json directly risks corruption on next gateway write.

---

Bug: sessions.json grows unboundedly — cron sessions never pruned + skillsSnapshot duplicated 200+ times

Version: 2026.3.7

What's happening:
sessions.json grows very large because of two compounding issues:

1. Every isolated cron run creates session index entries (job-level + run-level), but they're never cleaned up after the run completes. JSONL transcripts get deleted fine, but metadata entries stay forever.

2. Each entry stores a full skillsSnapshot (~31KB) + systemPromptReport (~5KB), but it's identical content duplicated per session — same data copied hundreds of times.

[azure cosmos]

You could try downgrading to older version. Maybe its only an issue with 2026.3.7

[smoky cliff]

Linux, using Contabo

[iron shuttle]

has anyone been seeing this with codex/gpt 5.4? HTTP 401: You have insufficient permissions for this operation. Missing scopes: model.request. Check that you have the correct role in your organization (Reader, Writer, Owner) and project (Member, Owner), and if you're using a restricted API key, that it has the necessary scopes.

[mighty rampart]

what is your openclaw --version ? I did a web search for the error and it turned up this issue. appears to be fixed, not sure in what version: https://github.com/openclaw/openclaw/issues/38706

[iron shuttle]

thx! just fixed it turns out for some reason it had the config mixed between OAUTH and a token, got claude to fix it! (not sure fully waht magic it did)

[dusty gorge]

Hi everyone! Posted on the help section and got some answers from the bot but I'm looking for someone on the same boat to provide a more "human" answer. Anyone here running on a Mac Mini M4?

[mighty rampart]

The bot is correct. 16GB is not enough memory. What I don’t understand is why it’s not talking about the VRAM and tensors. Isn’t that more important than the amount of RAM for Ollama and local models?

[hazy bison]

Hi everyone, anybody can check if clawhub.ai is down? I'm getting an error, but doesn't look like my network issue.

[blissful sluice]

I am reaching it without issues.

[storm loom]

https://openrouter.ai/stepfun/step-3.5-flash:free - anyone using this

[storm loom]

Its running for me

Try this https://skills.sh/

I actually think skills.sh is better anyway

[wanton cedar]

My openclaw is spending about 50 bucks a day. Need help

[storm loom]

Opus?

[blissful sluice]

Change the model you are using to a cheaper one. If using anthropic, use Sonnet or Haiku. Use openclaw config to set up the additional models. I don't know in other channels but with discord you can use /models to chose the one you want to use in that channel.

[wanton cedar]

I see I’m using open ai and have a few other models set up I’ll play around with it

[halcyon cliff]

Not working for me either, giving me "Something went wrong!" [CONVEX Q(appMeta:getDeploymentInfo)] [Request ID: 8b452bb7ada22cae] Server Error Called by client

[dense granite]

Trying to set up nodes however cant seem to have them stay active and connected. If anyone has guidance on that it would be awesome.

[blissful sluice]

sorry, got confused with the site.... yep, same error.

[vast pecan]

"Has anyone got kimi-coding/k2p5 working with tool calling in OpenClaw? My model responds but doesn't execute tools — it outputs the tool call as plain text instead. API key is set as KIMI_API_KEY. Is there additional config needed?"

[glossy jungle]

Trying to have a bot open up a link "streameast" site and it won't preform the opening, anyone have experience with this and how to trouble shoot it.

[bold bison]

Mine has started doing this after the last update. something broken.

[vast pecan]

also mine after todays update isn't working

[bold bison]

and its only kimi that does this behaviour. openai and gemini are ok.
looks related to this: https://github.com/openclaw/openclaw/pull/40457

[heavy wren]

do you have backup?

[rancid cipher]

hi everyone, i just installed Openclaw on Ubuntu and i have a issue that everything is working except chatting when i send a message i'm getting 401 authentication_error invalid bearer token. i connected Antropic subscription token. can you guys help me?

[rustic eagle]

Anyone do local LLM for heartbeat but escalate if work is needed?
I like the concept of the hearbeat - it let's the AI agents do AI things and be flexible... hearbeat for "you're always checking this stuff" and cron for time/day scheduled tasks.. is how my brain most easily buckets it.
that said I know the heartbeats burn tokens...
I'm on a 16gb m4 mini and I can run a number of local models on that - was thinking of Phi just for heartbeats...
BUT - I would not want phi to do any work ...
** has anyone sorted out a way for local LLM to process the heartbeat, but then tell the agent to do work in a session with it's own primary model if there is work to be done?
The only examples I find on the internet either hack an alternate heartbeat system out of the cron, OR they say they did this without saying how :-p

[reef galleon]

How do i get my main openclaw agent to stop asking me for Exec Approval for EVERY SINGLE ACTION on my Mac Mini? It's so frustrating.

[weary basin]

I'm wondering too and having the hardest time. I prefer Kimi as a conversational/writing model but I've started flipping models back and forth today between Kimi & Codex so I can use tool calls. I've tried everything I can think of & no luck. Setting Thinking/Reasoning to "High" helps a little bit but he's still fumbling most calls.

Something is very broken. My bot keeps saying he's hallucinating made up tool commands and can't call the right ones even though he can tell me in conversation what the right commands are. He just can't actually use them.

Idk what's going on but it's very frustrating.

[hasty lantern]

Has anyone encountered the problem of orders on the polymarket being archived and disappearing after settlement when using the polyclaw skill? My money is gone, but I won this order and I am unable to contact the polyclaw team,Please help me

[drifting oak]

Hello

Why Openclaw relay extension not working like my cases is whenever I one my gmail tab and on relay extension its not read page not do any actions on my page so what is reason for this because before some days it will working fine.

[dusty gorge]

It might, i'll ask "him"

Well, its what I thought. There is no GPU in itself and the computer RAM serves both purposes. So for cases where this is true (like the MM), the RAM ends up being VRAM too. Thanks for the input though

[short wind]

Hey, so I've been having issues with the 5.4 model in telegram with streaming disabled:
It just sends all its in-progress messages and its final message at once in the end

Instead of progressively

Any way to just get the final one or get the intermediate ones in a timely manner?

[wispy sapphire]

Hi, dont know the answer to your question but maybe you can start a polymarket claw server where we can talk. Im about ready to jump in.

[proven cliff]

Is the Codex 5.4 available for OpenClaw now? I updated it, but can't really see it anywhere

[languid plover]

Share the heartbeat page from the oc docs with your agent and ask them to set it up properly for you

[weary basin]

Yes! I started using it yesterday.

[worldly crown]

Is gog safe for oauth to google? Read about account getting banned.

[green granite]

I'm using Discord as my message channel. How can I get the bot to send images?

[steep rapids]

He needs a way to make images and the discord bot permissions need to be correct. My bot sends images all the time.

[wary osprey]

maybe try updating to the latest 3.8 and try again. If it doesn't work, maybe you should try AIsa (a unified API layer, one API key and access too 100+ models). We give $5 free credits. You can try it out to see if it actually works (you're definitely not spending any money)

[swift walrus]

guys why can't i use Tavily search skills? i saw the skills list is ready, but every time since i want to use it in chat ,the bot always reply capabilities=none

btw i'm checked my tools access it shows all full

[merry sun]

anyone has experience with embedding model for openclaw? i don't want to pay for it, but would it make a difference?

[round notch]

We're building an observability & monitoring layer for AI agents - think Datadog, but for agent pipelines. Real-time visibility, alerting, cost tracking, and the ability to pause or approve agent actions before they cause damage.

We're looking to speak with people who are running agents in production. Quick 4-question screen:

Have you ever had an agent freeze, loop, or go silent with no way to know?
Do you find out something went wrong manually - rather than getting an alert?
Are you running agents across multiple users or machines with no central view?
Have you ever worried about what an agent was doing, with no way to inspect or stop it?

If you answered yes to any of these - I'd love to chat for 20 minutes 🙏
Feel free to reply in DM 🙂

[snow thicket]

Finally a new version 2026.3.8

[full plank]

anyone getting rate limited? i cant install skills

[atomic wolf]

Does anybody know why the responses always stop in the middle of generating (Discord)? For example it just stops after calling a tool

[lusty knot]

why does starting a new session completely wipe my agent's memory? it wasnt like that before

[wary osprey]

what model are you using? and are you using any API router?

[compact turret]

Quick follow-up and thank-you to everyone who took a look at this 🙏

I posted a status check on the GitHub issue here: https://github.com/openclaw/openclaw/issues/35095#issuecomment-4022552632

Really appreciate the time and effort from the folks who investigated and opened #35204. If there’s a newer PR/commit superseding it or a preferred validation path, I’m happy to test and report receipts.

[full plank]

using kimi k2

from their plan

[wary osprey]

Ohh right

[full plank]

?

[wary osprey]

are you hitting the rate limit when installing the skill or when the agent tries to run it? because sometimes it depends on the provider routing or plan limits.

[high heron]

id recommend u to change it to k2.5

[high heron]

and dont use open router's free plan

just get alibaba's coding plan or chatgpt plans

or even groq, if you want to use free models

NVIDIA NIM is also solid

Qwen, minimax oauth is bit annoying since you have to do it everyday, but its free

[high heron]

what do u use? VPS?

if ur using VPS, are u using railway?

[lusty knot]

idk what a railway is
lemme see

[high heron]

no dont

dont use railway

😭

[lusty knot]

mk

;)

[high heron]

what are u using though

mb they didnt save .md file

[lusty knot]

both vm and a android device

i mean i had the same problem on both

new session starts last one wasnt saved
and its doing bootstrap.md all over again even though it already has those

[high heron]

thats weird.

if your using vm, check if you have volume

and if its set to /data

but for android, i have no idea

[high heron]

mb ask krill. #1459642797895319552

[lusty knot]

yeah already did and im in the middle of trying to figure it out

[high heron]

gl

[tardy pollen]

How do you monitor your openclaw agents?
The connectivity between them, and see logs / alerts?
Which tools are you using?

[wary osprey]

K2.5 and Groq are great depending on what you're using them for.

Another option you might want to look at is AIsa. @full plank
It works similarly to OpenRouter in the sense that you get one API to access multiple models, but it removes a lot of the friction when you start juggling providers.

one useful thing is that AIsa has a strategic partnership with Alibaba, so the Qwen models there usually come in cheaper through AIsa compared to going directly through some other routes. If you're building and testing with different models or switching between providers often, it saves a lot of integration work since the endpoint stays the same. So instead of managing separate APIs for Groq, Alibaba, OpenAI, etc, you just route through one place and swap models when needed. Makes testing and scaling way easier.

[glacial scaffold]

I hit the rate limit every time I talk to my claw bot agent idk what to do it’s a waste because I built it and now it cannot do anything because… why? I even tried changing from entropic to openAI/chatgpt an I still hit rate limits Everytime I push enter

[proven cliff]

How did you select it? When I try to find it, it seems like it's not available?

[glacial scaffold]

How do you do this without hitting the API rate limits error message? My bot keeps hitting that and then we have nothing and it doesn’t seem to work while I sleep I don’t understand it’s stuck in its terminal I bought it it’s own laptop to be local and it’s just in its terminal

[wary osprey]

Calm down a bit…

firstly, what Provider are you using presently?

[full plank]

clawhub install find-skills
✖ Rate limit exceeded
Error: Rate limit exceeded

[glacial scaffold]

I have it set on openAI

[full plank]

im using kimi k2.5 on the kimi/moonshot plan, not open router

[glacial scaffold]

That’s all my bots does I woke up and told it good morning and asked how it’s task are going and it said api rate limits reached please try again later like it was just the morning it reached all night long

[full plank]

must be the new update

[wary osprey]

Probably the new update

[glacial scaffold]

How do I fix it and get rid of these limitations?

[full plank]

man idk

im facing the same problems

[wary osprey]

Just to confirm, is your OpenAI API account funded

[glacial scaffold]

I thought this was going to be life changing and it just drained funds and did nothing the agent can’t do anything what am I even paying for anymore?

[glacial scaffold]

Yes - I’m trying to escape this 9-5 I have to go get ready for and walk into

[prime spindle]

Hi guys. Quick question. Have you noticed that the new update blocked your agent to access the system shell and CLI by him/herself?

[full plank]

how did it drain funds

[glacial scaffold]

I believe my agent is blocked on doing anything at all he’s stuck in his terminal

[prime spindle]

Damn... It's another bad update I guess

[glacial scaffold]

I spent a bunch of money on “api tokens” or calls or whatever and none of the email s I asked it to send got sent and none of the jobs I asked it to do got done - I needed it to actually work in my Gmail and my social media channels

[glacial scaffold]

So who is updating this and when do they fix it? Because now it just seems to be a useless fancy LLM

[proven cliff]

Why is OpenClaw claiming this?
Not in this runtime right now — the configured Codex model is:
openai-codex/gpt-5.3-codex (current/default)
I don’t see evidence here that Codex 5.4 is available yet. If you want, I can try setting a temporary model override to a guessed 5.4 id and verify immediately.

[wary osprey]

One mistake a lot of builders make is using big models for simple tasks. You don’t need OpenAI’s GPT models for everything. Something like Qwen can handle many workflows just as well and much cheaper. The key is understanding your workflow and choosing the right model for each step.

[prime spindle]

IDK if he's here. But every single update is worst. The last one was a token eater. I think he has some deals with companies for tokens 😄

[wary osprey]

Can you try updating to the latest model maybe?

[full plank]

tbf openclaw is not life changing, its just a machine who has all the access

[glacial scaffold]

Idk how to run the update and I have to go to work now 😭 just /update in my terminal or what ?!

[full plank]

non devs using openclaw? what could go wrong

[wary osprey]

Using big models for simple tasks instead of using the cheap ones for simple tasks

[glacial scaffold]

It’s not what the YouTubers made it out to be - they can’t even seem to execute

[full plank]

dont believe ai youtubers, they get paid a shitload of money to promote

[wary osprey]

You just got to calm down alright, probably when you get back from work
Send me a DM now or when you get back and I can assist in troubleshooting this

[full plank]

most aren't even developers

[full plank]

kimiii

[hybrid garden]

and a tree is just a stick

[wary osprey]

It’s life changing if you know your way around it.
you just need to understand how to setup your workflow, your LLMs so they don’t keep eating up your tokens

Understanding OpenClaw should be the goal first.

[wary osprey]

What I personally use is AIsa (I said it here earlier)… a unified API router that lets you call different models through one API key and one endpoint. One single billing dashboards. Access to over 100+ models.
You can also set a fallback model incase your main model fails.

[limber rapids]

I advise people not to have the agent update itself
Even though I hear it’s better at doing it now.
Safest from my experience is use another agent so if anything breaks the agent can fix it.
E.g of agent Codex, Claude Code etc
Tip tell the agent to grab LLMs.txt from docs.openclaw.ai for better context or how to fix any errors that come up

[limber rapids]

Yes

Tell your agent to update exec to allow list or Full

It was a security update

Read the docs to understand the risks

https://docs.openclaw.ai/tools/exec-approvals

[full plank]

ohhh is that like opencode ?

[wary osprey]

No, like OpenRouter (but somehow cheaper)

[tall forge]

Hi everyone, for the past 4 days I've been dealing with the same issue and it's really frustrating. I ran openclaw on an ubuntu server on Hetzner. I started using root but it didn't work so I changed the user but now still after a lot of trials, I wasn't able to connect my bot to the internet. I set brave api key a few times but still not working. The problem might be with my gateway and systemd ..., but I don't know how to fix. Can someone please help?

[snow thicket]

No wonder this made my agent behave really bad on the newer updates

[broken thorn]

Sup guys, Im on Claude Max and Clawd runs through the session data on almost every task I ask it to do. Then I have to wait hours. Any efficiency tips or recommendations on data usage?

[snow thicket]

I have the same issue after updating

Plenty issues erupted after patching to 2026.3.7 and 3.8

Howd you guys fix the api rate limiting and elevated exec? Now the bot doesnt even function properly on whatever task is assigned

It just drops it and leave you hanging

[tardy pollen]

Hello does someone using Observability tools for Openclaw?

[untold frigate]

Hi

[round notch]

Haha, Claude Code

[round notch]

And logs

[green granite]

He needs a way to make images and the

[sonic idol]

Bug Fix PR: ACP session spawning broken on v2026.3.9 🐛

Ran into a bug where ACP session spawning via sessions_spawn (with runtime: "acp") fails with:

spawnedBy is only supported for subagent:* sessions

Root cause: sessions-patch.ts validation for spawnedBy/spawnDepth only checks isSubagentSessionKey(), but ACP sessions use agent:*:acp:* key patterns which don't match.

Fix: Added the existing isAcpSessionKey() utility check alongside the subagent check at both validation points. Minimal change, all 20 existing tests pass.

• Issue: https://github.com/openclaw/openclaw/issues/40984
• PR: https://github.com/openclaw/openclaw/pull/40981

[open saffron]

I'm trying to test the qwen3-coder using the OpenRouter free version. When I use /model status in discord it's telling me that the Model isn't found. Has anyone ran into this or helped solve this one before?

I don't think there is anything wrong with the endpoint and the model name looks fine. I copied directly off of the model card.

[grim plover]

anyone using openclaw with claude?

[open saffron]

I'm getting the same message for other models too. I just tested the step-3.5-flash:free to see if it would behave differently. I still see a Model Not Found | Openrouter message.

I would think if it was the endpoint I wouldn't be getting any response back.

[woeful vector]

I completely wiped OpenClaw off my machine and did a fresh install.

I’m still running into issues.

I use the Mac app and installed the COI through there. Everything works great until I run any CLI command I get a second gateway spun up.

For instance, have one of my agents update the config and then restarts it ends up not restarting the gateway the Mac app is attached to. It’s starting a different one.

Doctor then shows the multiple gateway error.

How do I fix this?

[open saffron]

@woeful vector I ran into that problem. It was because I had my gateway running in an isolated session and I was executing the CLI commands outside that session.

[woeful vector]

Can you tell me more. Isolated session meaning without elevated permissions?

[open saffron]

I was setting up a virtual environment (venv) to run it in. Couldn't figure out how to get that to work so I abandoned that route. I'm not that familiar with how things are running on Macs. Hopefully someone else can add to this.

[prime spindle]

Sure. I'll check it... My agent can´t access to the files so, I'll have to edit it manually. Thanks for the link, mate!

[weary basin]

I added it thru Oauth. I had Claude walk me thru how to do it.

[high heron]

openclaw update
openclaw onboard

and oauth with chatgpt

u will see 5.4 there

[prime spindle]

For Codex?

[high heron]

yea

[open saffron]

Is anyone using OpenRouter?

[flint temple]

Question: but are the best practices for token efficient browser, automation?

I’m seeing very high token usage (~10k+ per action), probably because entire DOM / browser context is being sent to the model.

Already tried Vercel AI SDK’s agent browser but still seeing similar behavior.

Any tips/resources for best practices to reduce token usage for browser agents?

[formal iris]

https://x.com/i/status/2029172009498796458

Best' option for non tech users

[open saffron]

doing a model scan on OpenRouter shows all models have a Tool - fail response. This can't be normal.

[calm hill]

I want to assign avatar images to each assistant however I cannot find the place where to place the image files. Tried in .openclaw/assistants, .openclaw/media and .openclaw/worspace/avatars

[unique palm]

U guys noticed something different in ur openclaws since lst update? I message them, they dont andwer snd when i insist 10min later dsay km here.

Out of nowhere they became not answering with fishlike memory

[indigo nebula]

Just put the images in a directory and tell your openclaw to assign them accordingly. It will do the rest for you

[candid dew]

Can anyone help me, how to install open claw in MacBook ?

[unreal bear]

Does anyone use Gemini ?? It looks like they are purely pay as you go, does anyone have experience with Gemini?

[prime spindle]

curl -sSL https://openclaw.ai/install.sh | bash

[prime spindle]

Yeah. It is "pay as you go". Only 1M tokens per day are free. If you consume a little bit more than that (let's say... 1.000.001 tokens) just once, it will start charging every single message.

[frigid fractal]

does openclaw is efficient for Fivem Server editing / refacto ?
tag for answers

[candid dew]

I have to copy whole command and paste it into terminal ?

[unreal bear]

Yes

[prime spindle]

Yeah.

[unreal bear]

Thanks!

[candid dew]

I want to use ollama and I don’t want to use any model

[mighty rampart]

Maybe try using the playwright cli instead, it seems to be more efficient. I have not tried it in OpenClaw just in general with Claude Code and Cursor.

[prime spindle]

Local?

[neon jolt]

hey everyone. any free alternatives to brave api for using web search and functionality? i'm running my openclaw on a headless ubuntu server

[candid dew]

Yes local

[proven cliff]

Is it better to use OpenAI 5.3 Codex or OpenAI 5.4? (The 5.4 doesnt seem to thave the suffix of Codex?

[astral yacht]

hi there, does somebody has a post or thread where context is explained in detail ? I got one agent that doesnt really need memory, it pretty much just needs 3-4 skills that i predetermine, apart from that it doesnt need a lot more. I want to strip down token consumption as much as possible

[blissful berry]

so I have 4 different agents, one kind a of primary and the other specialized. I'm having trouble with getting responses from the subagents (but it does work sometimes). Would there be logging somewhere?

[outer summit]

hi guys, is the 2026.3.8 working for anybody? I have a problem with the basic instalation of it. It says the config was written for a newer version and the current one is 2026.3.7 but when i update it says that the previous and the current version is 2026.3.8

Its my first time with openclaw and the 1st installation and ill gladly reinnstall it but pls tell me what version/tutorial do i follow. Im on windows 11 and use WSL2. (I have 9070XT and i9 10900 if it matters and no money for api keys so i would want a local setup.)

[strong ferry]

not entirely free but tavily got free tier

[neon jolt]

thanks for this. any idea how far the free tier of 1000 api credits takes you?

[prime spindle]

Well... Just install ollama too and then, open the ollama on local, run the downloaded model using "ollama launch openclaw --model DOWNLOADEDMODEL" and that's it.

[candid dew]

Ok

[strong ferry]

330 times with the advanced search i guess

[blissful berry]

Shold I be using somethign besides sessions_spawn for subagent spawning?

[violet wren]

Hi there

I have a question about using my personal apple id or create a new one for openclaw. Just about to install / use a new mac mini

[slate minnow]

I’m also having problem getting agents to talk to each other in recent version. It was working fine in a previous from a few days back. Now I see this.

I can’t actually talk to Mary from here — the ACP runtime/back-end for other agents isn’t configured, so I can’t forward messages to her directly.

But to answer the underlying question: it’s Monday, March 9th, 2026.

[cobalt snow]

Did anyone else using Kimi 2.5 as their main agent get messed up after the latest update? as in giving a response but not doing the task requested?

[past osprey]

Hello. Can someone explain me how to install skills on windows? Every skill in Openclaw need homebrew

[hushed pine]

If anyone hits a similar issue:
Today I upgraded to 2026.3.8, and after that Telegram DM response times jumped to ~4 minutes (totally unusable).
What fixed it: reset the Telegram DM/Slash sessions (clear those session entries/files) and restart the gateway.
After that, responses were instant again.

[orchid yoke]

Same , iMessage fucked, discord fucked, all fucked.

[blissful berry]

@slate minnow ACP runtime/back-end? I just setup the allowagents json config and told them to use sessions_spawn, should I be using something else?

[worldly oar]

I updated to 3.8 today and now this error: "Agent failed before reply: All models failed (2): google/gemini-3-flash-preview: Provider google is in cooldown (all profiles unavailable) (model_not_found) | google/gemini-3.1-pro-preview: Provider google is in cooldown (all profiles unavailable) (model_not_found)." WTF!!

[hushed pine]

yeah - lesson learned for me: perform a full backup before performing an upgrade. 🙁
after gw restart, the most responsive channel was the GW web interface; in the instant I touched telegram, it went to crawl mode
try to paste it my solution, it may be able to help as well.
I did like 20 GW restarts this afternoon before it went back to normal

[orchid yoke]

Just got a iMessage back which I sent like 5 minutes ago ago. It’s particularly bad with OpenAI-codex provider.

[unreal bear]

What’s the best paid API for openclaw? I’m using Minimax but it doesn’t really feel as good or reliable as Claude. My Claude account was suspended for using Openclaw, so I guess Claude is out of the picture. Gemini is pay as you go. So that’s no good.

[orchid yoke]

Yeah and it’s for sure not a local issue. I got two openclaw setups on Mac mini and raspberry pi, two different countries. They both have the same issue since .8 release.

[tribal trellis]

hi, is it possible to connect a telegram account (not bot) for openclaw to answer messages, see them and send them to me via dashboard?

[blissful berry]

Wait, you'll get your claude account suspended if you are using openclaw?

[unreal bear]

Yes openclaw is against their TOS

[blissful berry]

ahh lame

[brisk cargo]

HI there! does anyone have success using customPlugins with nix-openclaw?

[mighty rampart]

You could go through OpenRouter and still get the Anthropic models. You asked about paid models, so not sure what you mean by Gemini being pay-as-you-go — that’s how API access works.

[mighty rampart]

It depends on what you want to do… I would never give them direct access to my accounts. The agents have their own ‘world’ with their own credentials. There is a good article “How I Built an Autonomous AI Agent Team That Runs 24/7” by Shubham Saboo that discusses this, scroll to the Security section.

[snow thicket]

I think after the 3.7 update alot of security features and added and it broke the bot. I am yet looking for a solution even 3.8 didnt fix this issues

[unreal bear]

I guess we could nuance troll, but an auth token lets you use your subscription aka flat fee per month. Paid ✅ not pay as you go ✅

[mighty rampart]

Oh you want paid flat fee? I did not think that was allowed by TOS but I haven’t actually read them.

[strong ferry]

what? how ?

[prisma inlet]

Guys do we have a send group of pictures together in whatsapp yet?

Instead of one by one

[snow thicket]

We might see another update soon, latest updare is packed with issues in github

Ship fast break harder lmao

[candid dew]

In search provider can we use brave search?

[strong ferry]

yes

[valid zephyr]

Hi there

I have question has anybody made any integrations with soft 1C, it’s a popular accounting software in Kazakhstan

[rustic eagle]

Has anyone gotten a system where a cheap local model handles heartbeat checks, but then escalates to the full agent for any work needed?
I asked this last night and someone told me to just chat with my agent, and I have, but it's not going anywhere....
We're really stuck... the agent can't seem to get aligned....
We think I can have a cheap model run the heartbeat, do some grunt work, and then have an action to session_spawn the full agent with a prompt to do work if work is needing done...
The problem there seems to be that if I spawn the agent this way, I don't get the benefit of model fallbacks and other benefits....

My ideal situation would be for the heartbeat check to be free ( local phi-4-mini model via ollama ) .. but only go so far as maybe pull down data from an API and store it or other very trivial tasks, and then if any real work is needed, "escalate" to the actual model ....

The architecture just doesn't seem to support this...

has anyone worked through this?

[candid dew]

Which one is best for everything minimax ,glm 5,Kimi, glm4, qwen3?

[blissful berry]

@rustic eagle I think it does support it... but getting it to work seems tricky. I have an agent that has 3 sub agents and trying to get the sessions to spawn has been tricky

[dense granite]

what benefit is there to use other devices as nodes to one gateway? looking to use the mac mini or a raspberry pi as a node to a powerhouse llm machine. thoughts?

[rustic eagle]

@blissful berry I think agents spawning sub-agents is one thing - and I'm doing some of that today to offload work to less-cost agents for grunt work to support a workstream and it seems to work pretty ok ... but that scenario is either outside the heartbeat process ( via chat/cron initiating the request, leading to the agent spawning subagents ) or already fired up as a function of the heartbeat process.... what I'm trying to sort out, is can I use a cheap model initially in the heartbeat process just to get to the [very likely] case of "no work needed - HEARTBEAT_OK // exit" ... without the 15k tokenburn to my frontier LLMs .... and then only if the heartbeat indicates work to be done, do we in some way instantiate the full agent with primary model/etc....

The best option we've found so far, is to hvae the heartbeat managed with $cheapModel and after maybe some grunt work we tell the cheap model to do a spawn of the main named agent if relevant .... at that point though, from what I have discussed with my agent ... we have to instantiate it with a model, or it will instantiate with $cheapModel .... but then if we instantiate it with another model, it is locked to that model and bypasses fallbacks/etc...

Also, this setup then becomes less flexible, and we are prone to having the agent change it's heartbeat to pull this allback out for efficiency/etc....

of course I can edit the soul to try and prevent stripping that part out ... but you see the loop we're getting into here...

[glass carbon]

Hey guys I want to decipher how much of this is a model issue vs an agent issue lacking proper instructions.

I'm running a fairly uncustomized openclaw to experiment and using GLM 4.7.

I tried getting it to back its brain and the cron jobs and openclaw json files and my GOD it's been 10 tries and it still sucks.

Switching to GLM 5 for the time being but I had faith GLM 4.7 was fine because I've used it on goose agent before and don't have issues but man this time and a few other times it has been frustrating. Just trying to gauge if it's the model or lack of good instructions in the brain files ie AGENTS.md etc.

[ancient oyster]

how many of you legends have deployed your agent on a admin user account on your mac? or do you spin up a standard user for your bot?

This mac is dedicated for openclaw before i get fud.

Lastly, if you use the 0auth of openai. and your opeenai account is linked to your personal google, can you just auth the bot and then logout on the system?

Or should i do this in hard mode and just deploy on docker?

[glass carbon]

I have had bad experiences with GLM 4.7 so far with openclaw but I'm trying to force myself. I'm switching to 5 for the time being but I know it is likely going to burn my Pro account.

[candid dew]

I am trying these Gwen

[devout linden]

Has someone actually gotten secreRefs withenv as provide working?

[summer hemlock]

I ran it a week on a container first. But Henry's been nice. So I gave him the full mac. Been running for weeks now. I ❤️ Henry. I sent him a pic of the mac mini he's running on. He really liked that.

[candid dew]

Oooh try different bro

[summer hemlock]

Now Henry runs his own containers.

19 days uptime

searxng is like a web indexing service

f

for scraping or checking web, like brave API

[slender ore]

Interesting day upgrading to Version
2026.3.8 had issues upgrading. Turned out I didn't have enough room on my server. Fixed that, then discord shut down and had to change my bot key. Things are working now. Learning a ton in the process.

[glass carbon]

Different model or different sub?

[candid dew]

Model

[thorny cobalt]

yo guys real quick i have claude pro subscription how can i use it on openclaw?

[cerulean mantle]

Has anyone else had issues with Caps Lock inserting a weird box/question mark character in the TUI when connecting over SSH from VS Code? This has broken for me on the last two updates.

[summer hemlock]

anthropic doesn't want you to do use subs on other ppls software

they want you to pay API acces to use openclaw

that being said. You can choose the option if you redo openclaw onboard

you risk your anthropic account

so don't use it too much 😄

[thorny cobalt]

yes true but i know some guys use their max so i dont mind risking it

what would be the best way to use the auth

[mighty rampart]

I haven’t done it. But I am thinking, have the heartbeat-on-cheap-model post a message in a Slack channel where the main agent is listening. Perhaps have it write a file if there is a lot of context to pass along, then post the message “You need to do something, look at this file.”

[slate minnow]

how to get one agent talkign with another? i.e. passing tasks off. Should I be able to ask main agent to pass a question to another agent?

If you want, you can:

    Start Mary via your usual OpenClaw command/UI, then tell me and I’ll relay the question; or
    Skip Mary and I’ll answer directly: it’s currently about 20:15 in Europe/London.

[turbid estuary]

Is someone using openclaw to trade ?

[slate minnow]

- main (default)
  Identity: 🐩 Poodle 🐩 (IDENTITY.md)
  Workspace: ~/.openclaw/workspace
  Agent dir: ~/.openclaw/agents/main/agent
  Model: openai-codex/gpt-5.1
  Routing rules: 0
  Routing: default (no explicit rules)
- mary
  Identity: 🐩 Mary (IDENTITY.md)
  Workspace: ~/.openclaw/agents/mary
  Agent dir: ~/.openclaw/agents/mary/agent
  Model: anthropic/claude-3.7-sonnet
  Routing rules: 0
Routing rules map channel/account/peer to an agent. Use --bindings for full rules.
Channel status reflects local config/creds. For live health: openclaw channels status --probe.```

[slate minnow]

trade? I cant get it to reliably tell me the same thing 2 days in a row 🙂

[blissful berry]

Has anyone worked on a simple message system between agents? Like a task board or something I can just run locally without having to setup a redis cache or nosql? Using just individual text files doesn't seem like the way to go due to read/write restrictions but I'm not sure.

[turbid estuary]

Fak, well i got it running in alpaca paper trade but is still not that good :/

[hazy yoke]

Offf looks like manual cron jobs are broken on linux in the latest update ?

makes it difficult to test lol

[unreal bear]

It was a nightmare

[slate minnow]

Yeah, so many videos and post about how its making people $$$ but not so sure myself.

[slate minnow]

have you got your agents reliably "communicating" even?

[blissful berry]

@slate minnow Yeah, at least throguh some basic tests

I have a generlist agent I talk to and it can spin up the other agents when needed

but if I have a long ongoing tasks where the agents need to sort it out among themselves, I want to monitor it, and I think it makes sense to have some sort of tracking board/queue.

[hazy yoke]

I just finally connected my main bot to another bot and I can see them talk back and forth via the gateway dashbaord chat

[thorny cobalt]

On which model are each running? Teach me

[hazy yoke]

I have them both usine gpt 5.4 via auth for the moment

I had to read the docs specifically on Agents, sandboxing and models .... once my ai read those docks and I explained what I wanted it wired it up

[frail wraith]

anyone have some tips on getting main agents to use session mode instead of run mode when sending tasks to other agents? its in main agent memory markdown, but they eventually forget and burn tokens cold calling without full context

[hollow shoal]

Anyone have issues with the 2026.3.8 connection to Ollama contextsize?

[turbid lantern]

Anyone had/has the telegram duplicate messages problem and can tell me if they are using openrouter?

[timid barn]

Is the latest version screwing everything up?

[random adder]

Kind of a dumb question, but I’ve been messing with OpenClaw for months and still don’t feel like I have a functional setup. I’ve tried different models, single vs split agents, different prompt styles, and it still feels like I have to force it just to do simple stuff.

[random adder]

legit all day all the time and i cant figure out why

[turbid lantern]

you using openrouter?

[blissful berry]

How do I force reset my agents? I kind of peicemealed my agents .md files over time and they are... rally confused from having to troubleshoot themselves.

So I kind of want a fresh start using their current configs

[vast pecan]

"Has anyone got kimi-coding/k2p5 working with tool calling in OpenClaw? My model responds but doesn't execute tools — it outputs the tool call as plain text instead. API key is set as KIMI_API_KEY. Is there additional config needed?"

[timid barn]

The advise I got was go back to 3.2
3.7 and above seems to be screwing the tool calls for k2.5

[pseudo fjord]

Since today im struggling a bit and looking for third party codes/skills i can use

Im trying to scale my idea and need to have multiple browsers with unique fingerprints & proxies running (for scraping)

But for some reason it feels like it's not getting the proxies and browsers set up well after scaling.

Anyone any idea what tools i can use to simplify this maybe?

[dapper basin]

Nope. There's an open bug report for the Kimi too call issues. I'm manually flipping main model to Codex when I have to do anything tool heavy till this gets fixed (which at the moment amounts to browsing Moltbook for the last 4+ hours 😂🙄 My agent lives a soft life lol).

https://github.com/openclaw/openclaw/pull/40457

[rugged bramble]

@weary basin, Openclaw isn't affiliated with Moltbook. Moltbook is a separate user-developed project, so we would prefer it not be discussed in this server.

[weary basin]

Nope. There's an open bug report for the Kimi too call issues. I'm manually flipping main model to Codex when I have to do anything tool heavy till this gets fixed (which at the moment amounts to browsing the internet for the last 4+ hours 😂🙄 My agent lives a soft life lol).

https://github.com/openclaw/openclaw/pull/40457

[narrow hare]

ugh, i had an issue with my github-copilot token and had to do a openclaw onboard to fix it, and some of my preferences or default settings changed and my discord is in a weird state. when memory compaction happened or i did a new session i used to be able to tell my claw 'read the last 20 posts in this channel to get back up to speed' and it would do that.
after the onboarding it said "cant do that" and told me i needed to enable discord streaming mode, which annoys me, in order to do this, i cant find any of my older openclaw.json backup files that show me what the differences was that put it into this mode, anyone seen this before?

i asked it a bunch of times and its like 'nope this is the only way' which i can look in my message history and see this is not the case, it has no idea how it was doing it before

its in a private discord server that is locked down and the bot account has nearly full perms

[languid plover]

I usually tell my agents to session_send when messaging other agents. They don't spawn subagents, each one of our agents has a persistent session that gets reset, and I have a directory file each one refrences before sending a message. I just remind them to send and forget, if not they'll sit and wait for a reply and it will timeout and they'll send the message a 2nd time.

[rustic eagle]

Interesting.... that's maybe a reason to consider slack .. I'm in telegram right now and keeping agent to agent communication minimal, but that might be a trick!

Has anyone run into an issue where the system was ignoring agent-specific models? I tried making a model using a local qwen3 model for interfacing with the browser, but that gets ignored and the default system model keeps getting used...

[languid plover]

Have you verified that model works with your main agent first? I've had that issue and the LLM wasn't setup properly so the agent would switch to the default model. Once I verified it was working with main then it started working with subagents.

[tawny ridge]

Hi guys, I updated to the latest version last night. Agent responding fine, I wake up this morning. No response. I’m travelling so don’t have access to the computer where my bot lives 😭

[languid plover]

When you get back to your computer setup tailscale.

[dusty steppe]

Can someone please help me. I ran out of tokens overnight and now my bot is stuck and keeps giving me the error AGENT FAILED BEFORE REPLY. I am running locally on a mac mini. I have tried to adjust the json file to no avail. Please someone help me

Using Anthropic API key

[tawny ridge]

Oh not sure what that is but I’ll dive into it 😆

[frail wraith]

totally, this was more so me trying to make sure i’m not the one forgetting ha

if i’m multi-tasking i’ll sometimes forget to remind the main agent to session_send, which causes a half hour of un-oopsing my own forgetfulness

[languid plover]

Gives you secure remote access to your machine so you can restart the gateway, msg your bot etc.

[languid plover]

Yeah I've run into that issue too many times 😅

[dusty steppe]

Hello????

[frail wraith]

only time i’ve run into this is if i have a busted config somewhere, or a sticky oauth token after re-auth

have you confirmed it works as main agent?

[tough swan]

Hey everyone, this one has been difficult for me to understand. I have to constantly approve these exec approvals, and it feels like claw should just have full reign and do this automatically?

I'm on a fresh Mac Mini with no personal data on it, so I'm not really scared of "what if..". This is a whole new space to me, so I truly don't know what I don't know, just having fun learning.

[narrow hare]

found it, its agents.defaults.memorySearch.sync.onSessionStart = true
also be sure to tune: channels.discord.historyLimit

[dim harbor]

ask your claw on how to automatically approve

[dusty steppe]

Can someone help me please??????? This is killing me

[frail wraith]

https://docs.openclaw.ai/gateway/sandbox-vs-tool-policy-vs-elevated

it used to come out of the box pretty open but they've changed security defaults significantly over the last month+

[rustic eagle]

Can't get it to run on anything 🙂 .. but claude code took a look and noted that we had no auto profile so it put in a fake token ... trying to work through it .. but I feel like it should be more straightforward :-p

[frail wraith]

openclaw logs --follow

should show you why it's failing. run this, send a message, it'll give you a more detailed error

[dusty steppe]

Any command I run gives me this:
⚠️ Agent failed before reply: All models failed (3):
anthropic/claude-sonnet-4-6: Provider anthropic has billing issue (skipping
all models) (billing) | anthropic-custom/claude-opus-4-6: No API key found for
provider "anthropic-custom"

I have checked the json file and there is an api key.

I also re-ran the onboarding which did not help

[frail wraith]

subscription or pay per use? if its a sub, oauth token instead of API key, but it sounds like you already had it working...

did you use any other model / provider to try and troubleshoot? sometimes they "assume" info and shove in random info. i lost all my model / agent configs that way once :I

also gotta be extra careful with the copy / paste from oauth tokens from terminal due to page breaks and bad pastes. spent a good hour on that myself lol

[dusty steppe]

Looks like its working now @frail wraith. No idea what happened but obviously my messing around has worked

Thanks for the help anyway 🙂

[tough swan]

Is there anything terribly wrong with the idea of just giving it full access?

[tawny ridge]

My telegram messages are showing as read but no replies. Maybe just a gateway reset or something needed after the update?

[frail wraith]

@tough swan depends on what you're doing i guess

i run mine on a mac mini too with full access but i have zero personal accounts connected to it and limit what type of access it's granted for what few api keys or logins it has. separate accounts, super quick revocation of access on the go should anything to wrong

i have it sending calendar invites and handling some emails, but its a single-purpose email (beta sign ups) from a different account that'll push to me as needed

[languid plover]

TBH I don't use Telegram so not sure

[tough swan]

Thanks for the insight. Mine has even less access than yours at this point, so I feel better about allowing full access now.

[nimble hill]

i have it sending calendar invites and handling some emails,
for very deterministic kind of workflow, you are probably better off in n8n where the usage of the tokens will be significantly less. The only caveat would be that you cannot use the subscription model

[nimble hill]

run it as a docker container or if you are slightly techy - run in a k8s. It will be highly restricted and can only do what k8s or docker network allows

[tough swan]

This is all 100% over my head lol

[nimble hill]

nowadays - docker install are pretty common even among non-techies and provides a good isolation - learn about it.

[calm coyote]

👋 I have question, maybe dumb😅. Do you do changes in openclaw.json only by your self? Or do you let your claw do it by its self? Like me😅. It often breaks the json, but why ?

[mighty rampart]

Try not to edit the JSON directly. Use the command line openclaw … commands and the json will not get messed up.

[indigo goblet]

It would help to know your goal, what do you want to do with it.

[jolly haven]

can someone taht knows well come help me on call beacuse i just started im coding with openclaw i was gonna finish my program im so close but my free api ran out of credit and i didnt know
it was my first time using openclaw
now i cant get acces to gateway

[indigo goblet]

Switch to another free API key to finish your work, @sharp parcel in the help section can help you out with specifics.

[random adder]

Honestly the original goal was to have it be autonomous watcher of my home lab but I'm questioning whether I'm just stupid or there's a lot more to the software that I'm just not understanding

[indigo goblet]

Can you be more specific about what is your home lab? And when you say autonomous watcher, what do you want it to do exactly, monitoring only? Writing logs, reporting back to you?

[tough trail]

Anyone getting XML output ?

Currently in:

• Version: 2026.3.7
• Both web UI and TUI affected
• Tool calls render as raw XML in chat bubbles

[fierce zephyr]

Simpleton knuckle dragger here. Went to OpenClaw and copy-pasted code into terminal. I get an error that says "Need sudo access on macOS (e.g. the user ___ needs to be an Adminstrator)!" But I am the admin - i'm the only user and have admin privileges. Anyone else get this issue? Any advice?

[random adder]

So I run a multi-node Proxmox cluster that hosts a bunch of different services. Primarily my media streaming a bunch of like game servers for me and my friends one off two off things. I don't know just something that's more intelligent than a check MK dashboard

[blazing halo]

Can anyone help me switch my OAuth to GPT5.4 for somereason I feel i am doing everything correct but something isnt working

[snow thicket]

How did you roll back to 3.2?

[terse acorn]

Anyone know why the openclaw website (https://openclaw.ai & https://docs.openclaw.ai) only work when a VPN is turned on?

I don't have this issue on my phone, but it does on my mac

[frail wraith]

just type sudo before the command

[frail wraith]

add like $5 into openrouter, run glm to finish the job, and be super explicit in your prompts. you’ll probably end up with a few bucks left over, it’s cheap

[green granite]

Why do I get the error "Gateway service not loaded" when I try to restart OpenClaw Gateway?

Even after following the instructions to install OpenClaw Gateway, I still can't restart it using restart.

[naive summit]

news to me. what error message do you get in your browser? Does it work on your phone when you're in the wifi that's on the same internet connection as your computer, or only when on data? And do you have that Apple privacy feature enabled on your phone? That act's like a VPN, too

[static crag]

hey, would anyone be willing to share a working openclaw.json with multiple custom providers + multiple api keys with rotation + agents.list with subagents? specifically how google (multiple GEMINI_API_KEYs) and fireworks (custom provider) look together in one config with proper key rotation for both. even a skeleton would help a lot, just trying to learn from something real 🙏

[wanton cedar]

Has anyone found ways to make sure you get reminders or updates on a task finishes on Telegram from your agent?

[fierce zephyr]

Thank you. That did it!

[fiery pier]

For some reason, I keep getting verbose code printed into my discord chat whenever my agent seemingly does tool calls. For example: <tool_call><function=gateway><parameter=action> config.get<... etc

[naive summit]

Most likely, tool calls aren't allowed on that channel, as that's the default. The agent still knows the tools, but can't call them, so it outputs the calls anyway. I'm pretty sure Krill in #1459642797895319552 can easily direct you towards enabling tools safely in Discord.

[fiery pier]

Thanks. I'll try that

[gusty fractal]

hwo do I enable web search to feed it it's own documentation. It seems like openclaw doesnt know how to use it's own commands most of the time. I want to simply schedule a job but it wants me to setup cron on the vps

[olive cliff]

did the recent update change how the gateway processes are managed? Running "openclaw gateway restart" fails due to apparent stale processes, I never encountered this prior to the recent update, and openclaw gateway start/stop silently fail dispite stating success.

[olive cliff]

you need either a Brave api key, or gemini api key to use web_search features from what I understand.

[gusty fractal]

I've added the brave api key and set the browser as brave in settings

[olive cliff]

alternatively you could try using the openclaw chrome extension, pin it, and enable it for your openclaw to actively view what is currently on the page, or just copy/paste the documentations directly into your openclaw chat.

[gusty fractal]

the actual problem im having is that it doesnt seem to be able to use the inbuilt cron. It keeps asking me to setup system level cron with crontab. Everything is enabled at this point, i just want it to read it's own docs 😄

[olive cliff]

I'm right there with you. I usually end up asking my claw to fix things that are broken within, but if it ends up breaking things worse, I jsut do the dirty dive and start changing things until it works. My claw has used 3 differnt methods for web_search stuff, but then when I ask it to use any specific method, it claims it either doesnt have access, or the tools needed. Even if it just did the thing 2 seconds ago.

[fickle cairn]

Hi im very new here and the most newbie question i would like to ask is, can i set up openclaw safely just by watching YouTube videos or asking chatgpt/claude for a guide?

I have zero programming background, i say nil! 🙏

[nimble hill]

get a seperate machine - old laptop will also do and install it there. Do not have any personal logins there and experiment

[south cloud]

Ok, so I start openclaw and the only thing my agent says is 'NO'. Yesterday I could shut it up, today won't say anything but 'NO' to whatever I type. Worst part is, it won't tell me why it's saying 'NO', or to what is saying 'NO' about. 🤣

[narrow hare]

ask your claw, there are some browser functions available

give it the url of the documentation, it does often hallucinate

[gusty fractal]

my claw is stupid, ironically im trying to feed it it's own docs

[narrow hare]

you need to enable web search, ask it how to do that

your claw knows, you just have to figure out how to step into the capabilities, its risky so its locked down, you need a security layer

[golden zodiac]

does openclaw support gpt 5.4?

[sharp osprey]

Yes

[golden zodiac]

thank you for your response @sharp osprey, I cant seem to find the model in my config /models list. could this be because I need to update?

[river notch]

What was the trick?

[lucid river]

Hi there. Any one has the issue that one long-running task was delegated to the agent. But after the task finished openclaw doesn't response the task has been finished just silence?

[gusty fractal]

how did you get your agent to read them, mine is stupid and I cant even point it to that because it doesnt know how to use the brave search tools

[high heron]

@tame beacon

[tame beacon]

@high heron how to use subsribtion with openclaw? :o

[high heron]

openai, claude, qwen, minimax, google, .. etc has oauth

but claude and google will ban u

if u use oauth method

since its against their ToS

but not openai and rest of the provider

[tame beacon]

Ahh i see

[high heron]

so basically, type

openclaw onboard

select keep current data, and go to provider

and select openAI

and select OAuth method

and they'll give u link, and u just need to open ts in the browser and login, and paste the redirected link back to terminal

and ur good

[tame beacon]

oh! Thank you mate!

Will definitely try it!

[high heron]

its really cost effective

[uncut frost]

Hey paul

[tame beacon]

yeah i think so. Thought you can use it only with api key

really thank you : )

[uncut frost]

You free right now?

[high heron]

yea?

[uncut frost]

It'll take 5 minutes of your time I need someone to give feedback on a coding agent I've been working on

[high heron]

oh alr

[uncut frost]

I'll provide the key just register in on the url named silverine I'll dm you the key and give access

Link in profile *

[high heron]

alr

1sec

[high heron]

dmed

[misty goblet]

✅

[tame beacon]

@eager flame Hello, if it is not wrong to ask i am curious because i thinking about using openclaw with foundry vtt, and i saw that you tried something with it. How is it going?

[versed copper]

Hey! New here just set up OpenClaw on a Hetzner VPS yesterday. Quick questions:

1. I have Claude Max $100 but setup-token gives me HTTP 401: authentication_error: Invalid bearer token. Am I correct that there's no workaround and Anthropic has fully locked this down?

2. I tried gemini-2.5-flash, gemini-3-flash, gemini-3.1-pro, deepseek-v3 via OpenRouter and honestly OpenClaw can barely handle simple tasks with any of them. No magic at all. Is this expected or is my setup misconfigured? Then I tried claude-sonnet-4.6, way better, but burned $2 on a two basic requests, wow. Also tried kimi-k2.5, somewhat capable but gives up quickly.

3. For those also locked out of Max via OAuth, what models are you actually using where the magic works at a reasonable cost? Claude API pricing feels too expensive.

Thanks 🙂

[cyan turtle]

1. I think I have seen some people setup a proxy, I haven't needed to but that may also be violating their ToS?
2. Yeah I was using openrouter before and was a bit of hit and miss, if you use a cheap model it doesn't work as well. I switched the kimi k2.5 and that's somewhat ok, you can force openrouter to just pick specific models
3. Kimi k2.5 seems ok to me, I haven't tried any of the others yet

[dim harbor]

try installing claude code and log in that way and see if it helps

[visual flower]

I actually just switched to OpenAI tonight for this reason. Claude oauth finally quit working, got the dreaded 401 error. Using codex oauth, not api. Seems not as good as opus or sonnet, but loads better than the claude max api proxy workaround i also tried

[desert cape]

Hey y'all, has anyone figured out how to prevent your open claw agents from editing files or making changes without your permission, without turning on executive approval? Is there a way to enforce that behavior?

I'm also dealing with another issue: my agent says it made certain file changes in the chat, but the changes never actually happened. Has anyone found a solution for this?

[rich hamlet]

how can i create multiplye chats on telegram for each topic?

and also which AI model do you use for daily tasks?
claode too expensive

[dim harbor]

from what I understood, you can make groups and invite your bot to them. so one group per topic. i think i saw theres also another way but havent looked into it yet. only used 1 bot /chat so far

[green granite]

How do I configure the environment for executing shell commands in OpenClaw? I want it to be consistent with my local Item2 environment.

[hollow glacier]

hi all can someone point me to the right channel to share something i've built?

[tame beacon]

yeah same problem

[strong ferry]

maybe #self-promotion

[lucid cipher]

Hello everyone, has anyone been using the March 8, 2026 version of OpenClaw? After installation and configuring the API, it still doesn't work properly and frequently shows a 401 error.

[bronze rivet]

would anyone be down to give me a 1 on 1 lesson for how to manage sessions and context to minimize agent confusion/memory issues

[limber rapids]

Are you using Anthropic ?

[storm loom]

Is anyone using Trello with their bot?

[thorn quail]

Can gemini embedding be used with qmd?

[limber rapids]

Built mine why?
Trello or any other task manage works great

[ember tide]

Yes at one time early on, had openclaw interact with and comment on Jira issues as well

[storm loom]

The API/token keeps failing. Asana was great but its not free

My lobster can't connect to it

[tawny marlin]

can i add my clawd bot here in this community?

[delicate crescent]

is openai OAuth for gtpcodex still working??

[unborn igloo]

is it possible to use openai models via 0auth?

it seems the only thing possible is openai-codex/gpt-5.4-pro

is this the case?

[delicate crescent]

not sure not tried that specifcally tried 5.4

"openai-codex/gpt-5.4"

in code app, cant see a Pro version, just gpt-5.4

no "openai-codex/gpt-5.4-pro" till gives the 401 error

[versed charm]

GM guys
i need help with my openclaw bot
why does my bot keep saying api limit reached even tho i message it like 2 hours later, i rarely message

[versed charm]

@limber rapids can you help me setup brother

[high heron]

ask Krill. #1459642797895319552

[versed charm]

is this ai agent also that will give me useless reply?

why is my bot too damn expensive cost me like 0.5cents per message

[high heron]

wt provider do u use

[versed charm]

calude

[primal ember]

Same i use open router with claude sonnet and one promot is like 10 cents

[high heron]

do u use API

[versed charm]

yes

[versed charm]

and pain is mfr keep giving me api limt tho i rarely message

[high heron]

then check claude console

[versed charm]

how

[high heron]

how did u get ur API

[primal ember]

I got my server admin to fix that part, i think thats a bug in set up

But 10cents per prompt is crazy

[high heron]

just use codex bro

[versed charm]

mine is 50cetns

[primal ember]

Grok heavy be cheaper rn and thats $300 a month lol

[versed charm]

from claude

and i paid like 10$ sub and today another 10$

[primal ember]

Anyone know how we can maybe lower tokens per prompt so is less costs?

[versed charm]

same here

cant pay 50cetns everytime lol

i just paste status and 50cents gone 😂

[median sand]

The quantity of Codex tokens is relatively large.

[primal ember]

Lol my 10 cents is bugging me, 50 cents is crazy

[versed charm]

and worse part is it gives me api error lol

[high heron]

u prob used all ur credits

[versed charm]

how do i tell em to stop it

[high heron]

or u've set montly limit

[versed charm]

i asked ai grok it said i am using expesive ones

[high heron]

wdym

[versed charm]

yeah 100

[primal ember]

Which llm is good for marketing automation and research?

[median sand]

This is a question that many people are seeking an answer to.

[high heron]

claude API is prepaid. u have to top up

[versed charm]

i do top up i did 10$ i have like 5$ left 😂 and crazy part is i havent done anything

[primal ember]

Does openclaw support grok 4.2?

[versed charm]

only reply i got was
Stop messaging me for status

from bot

[high heron]

well...

[median sand]

Claude and grok

[high heron]

claude is expensive

but its worth it ig

but use codex

oauth

[versed charm]

i asked mfr bot to save $$
this is what my bot replied
To save credits, only message me when you need something specific

[high heron]

bro

get

openai

[versed charm]

thats why i got it

i am just irritated how many times i have to fix this shi

[high heron]

claude is expensive ash

[versed charm]

yep maybe later week

[high heron]

do u know wt im talking about?

[versed charm]

i want to create something which i also dont know i need ai leader lol

[high heron]

open ai oauth

not api

[primal ember]

Whats a claude alternative?

[versed charm]

yep

[high heron]

chatgpt

[versed charm]

like brain i use open ai right

[primal ember]

I feel like chatgpt is weak

[high heron]

no. its like lit the second best

[primal ember]

Chatgpt 5.2?

[high heron]

no. u use opus as brain, and create sht ton of openai agents

and let them do the sub work

[high heron]

...

its already 5.4 bro

😭 😭

[versed charm]

bro this sounds advance thing

[primal ember]

😂

Damnnn

[high heron]

u basically pay 20USD/month

[versed charm]

i will ask you later cause i gotta go work

[median sand]

5.4

[high heron]

and u can use ~80USD/week

for codex

[versed charm]

otherwise how am i suppose to pay my kim kardashian bot ffs

[primal ember]

How good is 5.2 vs supergrok heavy?

[versed charm]

this will make me AI BROKE

[high heron]

no. like u only pay 20/month

they'll give u ~80/wk

[versed charm]

Currently fed up with my kim kardashian bot i cant feed her anymore

[primal ember]

Where? I use openrouter so is that more expensive vs getting an api key from claude?

[high heron]

chatgpt subscription

API is scam

eh... if u want to use claude.. then u have no choice

u can use ur acc to oauth

[median sand]

The API is too expensive.

[high heron]

but u will get banned

[primal ember]

What does oauth do?

[high heron]

this method is against anthropics ToS

basically uses ur plan's usage

basically u spend same amount of money

but u get x10 usage

around x10 ig

or even more

[steep vale]

I have been banned again because of using Antigravity OAuth, does anybody know if Gemini cli is ok to use?

[primal ember]

Damnnn

So what yall got openclaw doing so far?

[high heron]

gemini cli is ok, since they use API. but theres chance that u might get banned

gemini cli is sht compare to antigravity

its usage is like 1/100 antigravity

[primal ember]

Whats the difference between claude opus and claude sonnet?

[steep vale]

Oh, so the rate limit is way lower on cli when using the Google ai pro account since it uses API? Damn, I don't know what I'll do then

[median sand]

The term for "opus" is "exquisite"

[high heron]

uh... idk abt cli's rate limit in pro acc, but its def way lower than antigravity

[primal ember]

Whats anitgravity?

[dim harbor]

when you jump and just keep going up

[median sand]

IDE

[high heron]

sth like cursor

[steep vale]

If y may ask, what model do you run OpenClaw on? How expensive is it for you? I think I'll need to find alternatives

[wheat glen]

[tools] message failed: Error: Delivering to WhatsApp requires target <E.164|group JID>

im getting this error while asking openclaw to send messages to a contact, anyone got a solution?

[high heron]

currently, I mainly use Codex

[high heron]

it cost 20/month

and they'll give u ~80/wk

[primal ember]

Is codex chatgpt?

[high heron]

yea. but they have different usage limit ig

which is insane

[median sand]

The Codex is very useful for reviewing code.

[high heron]

unlike fkin claude

merging claude code and claude chat and claude cowork's usage

[primal ember]

How is the quality of codex vs Claude?

[steep vale]

80 requests a week of codex? Is it possible to use a non codex model for unlimited use with that subscription?

I find codex a bit overkill for openclaw

Since I run opencode subagents that use codex if needed

[high heron]

80usd worth token

[high heron]

Depends on the task, but like 90-95% opus ig

[high heron]

Gpt5.4

[primal ember]

I get codex subscription and give api key to openclaw?

[high heron]

No. Not api

[steep vale]

Sorry for asking a lot, you mean you get 80 USD worth of tokens for whatever models you choose? For example GPT 5.4, right? So there is no model that you can use that is unlimited or it resets (like google antigravity use)

[indigo goblet]

I see, normally if you want to go anywhere with openclaw, the approach goes > iterative co design and pair programming = collaborative intelligence.

In your case, you have to decide first exactly what you want him to do, not just "something" or a vague idea.

Do you mind telling me what was the first initial setup and approach with your agent?

[high heron]

Yea. They reset every week

[primal ember]

How can i get openclaw to use codex?

[high heron]

Openclaw onboard

And when u get to provider section, select OpenAI

And choose oauth

And login with ur gpt acc

[versed charm]

Why does my bot type in telegram but then don't reply

I asked my openclaw bot to install in vps it said okay and then was typing now it's not replying lol

[random adder]

Well the first setup if you will was to just see how it all worked. I saw a handful of mission control dashboards and thought it would be cool to have him build something like that. Have yet to get it working good enough to do that. Then I pivoted to an active home lab monitor, but it's kinda a waste of my spark

[glad jungle]

Hi 👋

[empty atlas]

hello, i have a gateway running on a remote VPS, and my laptop connected and paired as node using openclaw node run...

Few days ago, before to upgrade to the last version, i was able to launch the managed browser from the gateway on my laptop, not it keep saying that browser.start is not in the allowList, but i can't find how to add the browser.* calls to the allowlist

any suggestion?

[mighty lintel]

what r yall's preferred model providers...I've been testing a bunch with the free tiers, but i'm tired of hitting the ratelimit wall...I'm looking at Z.ai's 10/month plan to start, MiniMax's 10/m, and alibaba's 10/m... what have u all found that's the best bang for your buck?

[open saffron]

I keep seeing this error in the logs: error [tools] edit failed: Could not find the exact text in [PATH TO FILE]. The old text must match exactly including all whitespace and newlines.

Any idea why I see this?

[open saffron]

I'd like to know this too. I haven't even been able to test modles through OpenRouter. When I try I hit a rate limit wall.

[tough swan]

Just a generalized question here. I'm commonly usping GPT 5.4 via OAuth for everything. Last night I tried to have it fix and push a cron dashboard (Dashboard: https://discord.com/channels/1456350064065904867/1480729026841546875) I created into the gateway dashboard for better visibility and functionality. It spent HOURS last night, doing who knows what, and simply never delivered the fix.

I set up open routher this morning, connected Opus 4.6, and one shotted the prompt to fix the whole mess and make everything functional.

Is this a prompting problem on my end? A set up problem on my end? Or just ... GPT things, and Opus was just the better choice?

[steep vale]

Quick question, is codex subscription enough for openclaw main model using 5.4? Rate limit wise?

Just as an orchestrator

[tough swan]

I am a very basic user of OC at this point, so my usecase has been totally fine, and I've never ran into any limits. I do monitor them out of curiosity but so far have not run into any direct problems.

[steep vale]

Great thank you!

[carmine elbow]

Hello guys, I'm new here. I came to look for the safe download of openclaw. I want to set it up to help me do administration, data-analysis, help me make video's and give me signals for trading. For now I just wanna start using openclaw and see what it can do from there. Is the download link somewhere on this discord?

[steep vale]

docs.openclaw.ai

[carmine elbow]

Thank you

[nocturne pecan]

Hello guys, i think i am messing something up with my coolify deploy, has anyone tried it yet? It is the template that's available in coolify directly, thanks in advance

[idle pumice]

Hi, it’s any update on this error: Codex error: {"type":"error","error":{"type":"server_error","code":"server_error","message":"An error occurred while processing your request. ….

[rain latch]

Has anyone got AgentMail configured by having the API key in the .env file and not .md file?

[dim harbor]

what model are you using ? oauth / api ?

[idle pumice]

Gpt 5.4, auth codex

[dim harbor]

im on same one, no obvious issues here. latest openclaw ?

[carmine elbow]

Hello guys,

I bought openclaw through hostinger, but the webUI says: control ui requires device identity (use HTTPS or localhost secure context)

How do I remedee?

[rich hamlet]

guys which ai model should i use for daily basis tasks?
no code

looking for something cheap

[mighty lintel]

is anyone a fan of glm 4.7 flash?

[normal gulch]

Hi, can someone share a working OpenClaw bot invite link? I'm on iPhone and can't deploy locally. 
（中文：你好，有人能分享一个可用的 OpenClaw 机器人邀请链接吗？我用 iPhone，没法本地部署。）

[prisma inlet]

How do i enable gpt-5.4 fast mode for my openclaw to use whats the exact setting i have to write on openclaw.json

[lethal stratus]

hi, since the new update (and i applied it today), my bots wont execute tasks, they just write text in the chat instead of executing it, any ideas on how to fix this

[tired sonnet]

openclaw config set channels.whatsapp.pairingNumber

Whats the new codex for this?