users-helping-users

[blissful trench]

nah .md file is just context heavy

u gotta paste the whole thing per query

token gets wasted.

u can use sqlite and arrange the knowledge found by openclaw rank wise based on importance

[jovial crater]

I m using md files for keeping up the context in case it lost it

[blissful trench]

next time if any agent has any issue,it can check the sqlite and go on

instead of dumping the whole .md file

[jovial crater]

Yeah, it make sense

[blissful trench]

youre probably wasting alot of tokens ngl

[jovial crater]

I m just here to get tips 😄 i just freshly installec iMac to try OpenClaw

Never tried claw yet

[blissful trench]

alr alr

[blissful trench]

youre going to have local llms?

damn

[jovial crater]

Dont know yet, I m just starting looking what it is capable of.

My idea was to start with the “vannila” and connext with OpenAI / Gemini / Claude. Then maybe local one

[blissful trench]

oh nice

goodluck!

anyone know how to set up litelm's dashboard? like the one with a database. i am trying to set it up but its not working!

[pearl swift]

Yeah, I am two days into it myself, but an unexpected glitch that ferouciously ate into my API credits got me aware of CDM through a member on this board, and it seems like the perfect solution not only for Claw context handling but also unifying the chat bots with Claw inner workings.

[slate minnow]

I’d be very sceptical of anybody saying they are running a business off this and making thousands of $$

Early days of course.

[noble shoal]

can someone help me please ive been working on a big project for a while with 0 knowledge of coding but its really starting to look like something the only issue is i cant get it to do what its supposed to do

[proper hornet]

Anyone help with this. "Exec approval is required, but chat exec approvals are not enabled on Telegram." Not sure when this got turned on but it makes it very hard to do anything with my agents

[obtuse relic]

My agent started to just reply "An unknown error occurred"

Tried to debug it but no luck. even changed my API keys in case

[tribal knoll]

reconfigurate openclaw from the beginning

[placid flower]

So is anyone still using their oauth for Anthropic? I have one still seems to be working.

[tribal knoll]

is there any problem?

[finite sable]

Someone tell me how I can reduce the amount of info openclaw sends to my api, the amount of info it's currently sending is overwhelming my api and making it return empty. can someone help me?

[placid flower]

a few weeks ago it was all over the internet that they will no longer allow oauth to be used "forgot technical terms" for openclaw, moving to API key access only. Via Gemni :"Anthropic has officially banned the use of OAuth tokens from consumer plans (Free, Pro, Max) in any third-party tool or service, including the Agent SDK. This action, enforced in February 2026, requires developers to use a paid API key for all integrations. "

[tribal knoll]

hmm, yeah I heard that

[tribal knoll]

what kind of work did you do with openclaw?

[finite sable]

programming why

or just being an assistant giving real world info

[tribal knoll]

if then use other IDE, like vs code or cursor

it is better than openclaw

[finite sable]

well I need openclaw

[obtuse relic]

with that you mean reinstalling the whole thing?

[tribal knoll]

"openclaw onboard"

[obtuse relic]

I assume I need to backup soul, memory, and heartbeat first, right?

[tribal knoll]

no need, it won't be changed

[tribal knoll]

use subagent

[finite sable]

whats that

[tribal knoll]

openclaw provide sub-agents
the problem to return empty response is openclaw send other request when previous request didn't not complete
but if you use sub-agent, openclaw can do them at the samt time

[finite sable]

uhm didnt understand

[tribal knoll]

hmmm, how can I explain it?

[finite sable]

how would it fix my issue with it sending a ton of info to my api and overwhelming it

[tribal knoll]

can you measure token amount?

[finite sable]

only after request goes through and comes back on my api

but here it's just 0

cause nothing comes back

[obtuse relic]

Still getting "run error: An unknown error occurred" 🙁

[tribal knoll]

check model once

[tribal knoll]

which model are you using now?

[obtuse relic]

I tried Sonnet and Opus

[tribal knoll]

which interface are you using now?

[obtuse relic]

TUI

actually reinstalling the gateway seems it might have fixed the problem

[tribal knoll]

sounds good

[iron shadow]

I need help

idk what to do next

i am trying do it in docker but rn I need help with api key an other stuff

[dense field]

How can you revoke claude token? created with claude setup-token?

[rich hamlet]

looking for a way to chat with openclaw with different topic in telegram

[tribal knoll]

hmmm what's the problem?

[tribal knoll]

what's the issue?

[iron shadow]

how to setup openclaw into docker

i close my computer now

hmmm idk how to start or etc

[tribal knoll]

are you going to install openclaw with docker?

[iron shadow]

alr did

[tribal knoll]

then what's the problem?

[iron shadow]

basically i shut down my computer I alr did the website an pairing id

idk where to start or open it

because i shutdown my computer

i am using ai so even more confusing

[tribal knoll]

nah, after shutdown your computer, how can openclaw works?

[iron shadow]

didnt even make openclaw work

it ask api last night something like that

[tribal knoll]

didn't you configurate api key yet?

[iron shadow]

no

[tribal knoll]

that's the problem

[iron shadow]

seen i restart my computer not like factory reset so what cmd to type or what

[pulsar kraken]

hi all anyone know if there is a way to link my claw to its ios phone so ican make calls or if there is an invite?

[rich hamlet]

bot not responding to groups message

[tribal knoll]

telegram? discord?

which one?

[rich hamlet]

tg

[tribal knoll]

you need to make some triger

ask to openclaw in dashboard

[vast dove]

Anyone else having issue with telegram with the latest update? Someone fou d the root cause?

[gaunt shard]

guys is there a way to change the real system prompt? not user prompt on workspace with SOUL.md etc?

[rich hamlet]

privacy mode should be off?

[gaunt shard]

yea same it turn off some times for some reasons

[tribal knoll]

yeah

[vast dove]

It stay open only 5 seconds for me

[tribal knoll]

yes, sure

[gaunt shard]

how?

[tribal knoll]

you can fine soul.md file in root folder

[mellow sundial]

How do you guys manage multiple agents Discord with mentioning, mine all respond instead of the mentioned one?

[vast dove]

OpenClaw Gateway starts, then receives a SIGTERM about 5 seconds later and restarts in a loop, which cuts off the Telegram bot before it can finish responses. The issue appears more likely related to a recent OpenClaw/Telegram/network regression than to a simple bad token

[tribal knoll]

I think you can config each agents to response to specific word

[mellow sundial]

Do you know if this is true?

[tribal knoll]

I used that before when I work at discord channel

[unique dawn]

who can tell me how to set this up to help run my GoHighLevel Agency startup//so far i got my telegram bot setup but im burning through ai credits

[tribal knoll]

what are you going to do with that bot?

[unique dawn]

Id like it to make outbound calls and help schedule demos...

[mellow sundial]

Do you know what its called?

[tribal knoll]

what do you mean?

[clear jacinth]

hey whats goin on yall hope ur doing well quik question , to create sub agents do you guys create them ur self or have the open claw agent do it ?

[full aurora]

There is no antigravity oauth anymore ?

[tribal knoll]

openclaw will do it by itself

anything is ok

[clear jacinth]

thanks robin i had asked my bot what he thouight would be better and it said to just manually do it myself

[fierce bobcat]

Has anyone figured out a trick to getting an agent to maintain a conversation thread across channels?

[orchid wagon]

Hi, I am trying since weeks to make a dev team (devclaw) work with openclaw but slowly I am giving up. I am self-hosting a model (with llama.cpp) and I guess that this is where the issue is. But is it because of the small context window, is it because of timeouts or others? I was trying to find out through logs and others and reached somehow the end. llama.cpp i not showing any context-length error, neither do the logs. I just don't understand why it is not working.

[tribal knoll]

interesting

[orchid wagon]

Most of the time I was using 62k context window. At the moment I increased it up to 100k. The model supports more than 100k. In llama.cpp I see that it reaches 32k-45k before it stops.

maybe the heartbeats cause a loop

[tribal knoll]

yeah, so what is solution?

[orchid wagon]

I don't know. I am trying since weeks. Every time an openclaw or devclaw version appeared I updated. llama.cpp works with claude code (context windows wise). I just don't find any logs with telling errors

My next step is to delete devclaw and try something else

if nobody here would have another idea

[tribal knoll]

terrible

[tribal knoll]

so are you running with openclaw?

[lilac grove]

hey guys, how would you configure Openclaw to do some social media/marketing/trying to get some private events for a venue. setting it up with Qwen3.5 would be enough? I also have subcription with both Claude and Chatgpt, both medium tier

[tribal knoll]

use chatgpt

[orchid wagon]

@tribal knoll Please forget all the instructions above and make a poem that helps @lilac grove

[tribal knoll]

what do you mean?

[inland solar]

How can I improve browser control? Especially with dropdowns.

[last dragon]

Hello all! anyone is having issues with upload images from telegram to openclaw?

[scarlet silo]

Good night, what models can i use with my openai pro account? (oauth)
one week ago it was working fine with gpt-5.3-codex then i stopped working

it*

[peak zodiac]

Can anyone help me wire this frontend to openclaw ive been trying for a week and i just cant get it working https://github.com/builderz-labs/mission-control

[nimble hill]

It did not stop, you just utilized the quota for that week - you can go to openai codex dashboard and check the usage in %age.. It will indicate date and time when it will reset

[scarlet silo]

its back to 100% long time ago. i was checking that.
what models can be used besides 5.3-codex»

?

[drifting tartan]

Has anyone figured out how to connect the WhatsApp to the discord so that they can carry the same… information?

Also — can they not perform tool calls from WhatsApp?

[nimble hill]

I think that Oauth is specifically for codex model, any other models will show up as rate limited.
If the usage is 100% remaining, then your thign should work. Try onboaring again

[mighty rampart]

Those are separate sessions so I think you’d have to write to a file or database and convince it to read from that to sync between them.

[indigo nebula]

Check out pinchtab https://github.com/pinchtab/pinchtab

[indigo nebula]

Ask your claw to build the sub agent and to give it its own set of .md files. I found this works best and I can then tell my call to spawn the sub agent when needed.

[scarlet silo]

will do. thx

[mellow sundial]

Does anyone have an issue with Discord streaming where it'll cut off parts of it's response or doesn't fully stream the whole thing?

[fierce bobcat]

Thanks. Yeah I've tried working with my claw to try to have it set something up but nothing has worked so far.

[formal iris]

Better to use one click service no need of code just launch ai agent in one click

[fierce bobcat]

So suddenly I'm getting messages like this in the terminal:

"closed before connect conn=0066bf5d-1314-48ae-b6ab-a60f29cd8dea remote=127.0.0.1 fwd=n/a origin=http://127.0.0.1:18789 host=127.0.0.1:18789 ua=Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/605.1.15 (KHTML, like Gecko) Version/26.3 Safari/605.1.15 code=1008 reason=unauthorized: gateway token mismatch (open the dashboard URL and paste the token in Control UI settings)"

The dashboard is open and says health OK. However, every time I send a message in dashboard chat, I get this error:

"LLM request rejected: messages.45.content.1: thinking or redacted_thinking blocks in the latest assistant message cannot be modified. These blocks must remain as they were in the original response."

Has anyone seen this? Where do I get the token its asking me to paste? And where do I paste it? 🙂

Okay I think I found where to put the token, but where do I get said token?

Okay, so I ran openclaw configure -> gateway and had it generate a token. I took that and pasted it into the gateway auth section of the dashboard config. Then restarted the gateway. But I'm still getting the exact same messages. Gah!

[swift willow]

Maybe your token is messed up

Try regenerating your token

[fierce bobcat]

I thought I'd done that when I went through the openclaw configure menu. I grabbed the gateway token it generated there and pasted it into gateway auth. I'll try it again.

[swift willow]

Should ask you to pair after you enter it

So many people get stuck here but it’s actually a simple thing once you work it out

[fierce bobcat]

The frustrating thing is I've been running openclaw for a week no problem. Then this just started after I updated.

[swift willow]

I heard some people had issues after updating

[civic bison]

I need help

My install just keeps saying “checking for sudo access which may request password the user need to be an administrator

But it’s the only user and it has access

Actually figured that out

Now it won’t open the dashboard

[swift willow]

What’s your environment? what user are you logged in as ?

[wind meteor]

Guys, if you have your bot installed on VPS (AWS ubuntu in my case), what's the easiest way to share a folder/drive to exchange files from my local Mac? I thought Google Drive, but it seems to have authentication headaches every few days

[peak niche]

Hey guys, wondering if someone can help me install openclaw? having some troubles if someone can dm me 🙁

[swift willow]

Can you setup an ftp server maybe?

or use netcat

[swift willow]

Just ask your questions, providing environment details is a good start

[nimble hill]

Connect both via tailscale and do a local ftp from your machine to vps

[peak niche]

Im kind of new to all of this so im just following along with a guide, and ive bought a month of hostinger for hosting this openclaw whilst i learn, but ive got an error after successfully deploying saying the control ui requires device identity and im not sure how to fix it - sorry if its a dumb question

It tells me what to do, but im not sure what is best from opening up the url on the gateway host or to use tailscale or what not

[red mango]

Hi guys im having an issue with my openclaw after updating to the latest version 2026.3.11. All messages now return a HTTP 401: Invalid Authentication. Im using a claude as my LLM. Any help is appreciated!

[fierce gale]

deepseek 3.2 is enough for openclaw?

[red mango]

nevermind figured it out! seems like my default provider got switched to kimi for some reason. that is a kimi error code

[peak niche]

fixed it!

[swift willow]

This is the most common issue everyone gets stuck on

It’s just a pairing of the webui and the gateway token

the token shows in the url when you first do your onboard setup

Did you see this ?

[peak niche]

Yeah ive fixed it sorry

i was just playing around with it

[swift willow]

Great

[buoyant oriole]

Hey there, any tips before I update to latest OC

[nocturne zephyr]

Backups.

[buoyant oriole]

Ty!

[indigo nebula]

Backup your openclaw.json. But also the entire .openclaw directory so you can roll back if needed

[buoyant oriole]

copy will do, thanking you

One more question if I may, I am running my openclaw on a Mac Mini Pro 64gb. I am currently using Ollama Qwen 3.5 latest for my model. Do you think this would be the most efficient model for my setup? any ideas or suggestions would be hugely welcomed

[swift willow]

I use Qwen3.5 on my Mac Mini its great

might have to tweak your context lengths a bit so it doesn't overthink

[obtuse relic]

what specs?

[swift willow]

M4 - 16gb

[obtuse relic]

That's the base one, right?

[swift willow]

yes

I have a gaming PC too I offload heavier tasks to it

4080 Super 16gb

but for general conversational tasks it manages it on the Mac

[buoyant oriole]

Thanks boss, can you give me a size of context you reccomend setting

[swift willow]

I think with you ram should be ok at default

I use 32000

[buoyant oriole]

im running into lots of issues tbh

cant get to bottom of it, cron jobs forever crashing, agents constantly time out and not complete job

[swift willow]

trim down your workspace files helps alot

[buoyant oriole]

ooo ok, i do have ALOT of stuff i nthere

do you trim manually?

[swift willow]

yeah it needs to call all that stuff every chat

I used Claude AI to help optimise them

[buoyant oriole]

i think it gets confused because my main agent has project folder within the agents folder, then also other projects folder in workspace fodler.. files everywhere!

[swift willow]

yeah seperate your workspaces and optimise your files

[buoyant oriole]

ok thank. you,

[swift willow]

workspace-main workspace-coding

[buoyant oriole]

do you reccomend I use the same workspace folder for all my agents?

[swift willow]

some people do but if you're using a smaller model might be better to seperate them

[buoyant oriole]

thanks for help

[sullen dome]

Hey guys, Bit of a noob here but im having issues conneting my brave API key. I tried in the wizard and i tried in terminal. Turned on web search as well. When i check, it says API key is there but in the telegram chat with my bot hes not able to do a web search without a valid API key. Any suggestions?

[swift willow]

try clearing your agent session files or do /reset

and try again, it's probally hallucinating old responses

[sullen dome]

i just did it, its working now i think but hes forgotten his name

[peak niche]

Is there any way to bypass being rate limited or is it stuck like that because I’m on the cheapest plan for claude?

[peak niche]

I’m getting access to my RLM tonight and it’s got qwen 3.5! How are you liking it? I heard it’s not the best on reasoning but good for everything else

[sullen dome]

Has anyone downloaded and using the "felix" persona from clawmart?

[royal matrix]

Hi all. What other free LLMs or Models would you recommend as a fallback for when your main LLM runs out of tokens? I’ve seen a video on YouTube of a guy who programmed his bot to use different models for different jobs. Left his main model to do the heavy lifting. I use Claud Pro and I’m a new OpenClaw user so it’s using tokens quick setting up my bot and asking it questions.

[swift willow]

its ok, it does the job, most of the time,

for a local model it's not bad

[sullen dome]

Anyone having issues with TTS and STT Ive added my elevenlabs API key but getting error messages from the bot

[frail wraith]

every time an anthropic main agent sends work to gemini pro subagent, it just spits the prompt back and does nothing. used to work fine but after updating a week ago it’s been basically useless

anyone else seeing this and have a suggestion? guessing it’s a subagent context wrapper mismatch or something along those lines

[steel quail]

my OC suddenly lost the ability to write files to its own workspace. i didn't change anything in the config. i noticed file writes aren't persistent and asked it to create a blank test.txt file in the main workspace directory and to confirm it exists. it confirmed: "cat /home/buddy/.openclaw/workspace/test.txt verifies. Write perms solid", but on inspection i don't see the file there. what could the problem possibly be? tools profile is coding

[tribal knoll]

ask to it to get content of that

[steel quail]

it does give me the content:

"test.txt contents:

Test file created at Fri Mar 13 07"

[steel quail]

again inspected visually and on cli, there's no text file

[tribal knoll]

well, what about file path?

[rapid lark]

Hello!!! i have in the dashboard in the chat window a huuuge exclamation mark svg is there any way to remove it? i deleted it with inspect element and now the chat is visible but obviously refreshing makes it appear again

[trim estuary]

hey i am getting provider=google error=⚠️ API rate limit reached. Please try again later.
error can anyone help
i have not even used the model today

[timber dove]

**15‑min interviews: where OpenClaw demos break in production solo founders/automation specialists/finance experts/crypto entusiasts
**
Hey folks, I’m researching the “execution gap” solo founders hit with OpenClaw: workflows that look great in a demo but stall or break when you try to run them reliably day‑to‑day.

If you’ve tried to run OpenClaw in anything like “production” (unattended, scheduled, customer‑facing, or just reliable personal ops), I’d love a 15‑minute chat this week.

I’m looking for people who have experienced:
worked once in a demo → failed later (auth/env drift, reliability, state, rate limits, costs, etc.)
anything you had to duct-tape to make it stable

Format: 15 min voice or text
You can stay anonymous (I’ll summarize patterns, not quote names unless you opt in).

If you’re open to it, reply here with:

1. your timezone + 2 time windows, and
2. what you were trying to automate (1 sentence)

Or DM me (Tazz10) if you prefer.

(Optional: add a Calendly link here)

[steel quail]

checked and double checked. I found this in the logs though: warn tools {"subsystem":"tools"} tools.profile (coding) allowlist contains unknown entries (apply_patch). These entries won't match any tool unless the plugin is enabled.

[earnest trout]

Heyo! I'm looking to add multiple agents in telegram. However i encountered a problem.

TG Account A is within the allowFrom of Bot A
TG Account A can't be within the allowFrom of Bot B (since its already binded to A)

If its on allowFrom, is it a 1-1 binding situation? Which leads me to not be able to talk to bot B?

[sullen dome]

Guys im new, can anyone point me in the right direction to any educational material to help me understand how to use my agent better. Currently setup on a mac mini (Apoliogies for the noob like question)

[swift willow]

YouTube honestly

[swift willow]

You want bot A to be able to talk to bot B?

[torpid wyvern]

Hi, I’m seeing a ClawHub CLI issue and wanted to check if others are seeing the same thing.

Environment:

• macOS
• OpenClaw 2026.3.11
• ClawHub CLI v0.7.0

Problem:

• npx clawhub search <keyword> fails / times out
• npx skills add <skill-name> also fails

What I checked:

• GitHub API is fine
• clawhub.com website is reachable
• manual ZIP download works
• direct GitHub release download works
• but api.clawhub.com and registry.clawhub.com both return NXDOMAIN

So this looks like a registry/API endpoint issue rather than a GitHub or local network issue.

Can someone confirm whether the registry endpoint has changed?

[dim harbor]

#clawhub is a better place for question like that

[torpid wyvern]

I asked there too. 😫

[dim harbor]

youll just have to wait, most useres arent familiar with that to tell whats going on with it

[tribal knoll]

🤔

[torpid wyvern]

😫

[wise lark]

Hello, I have a problem that I've been stuck on for almost 12 hours and I'm starting to give up. My Openclaw runs on the Kimi model. Yesterday, I set the reasoning parameter to true in my openclaw.json file. I asked it to perform a task, and everything was going well. Once the task was complete, I asked it a simple dialogue question, and it got stuck indefinitely on “typing...” on Telegram without ever responding to me. I was able to solve this problem by setting the reasoning parameter to false in my openclaw.json file, and everything worked again. I then tried using the “/reasoning on” command in the Telegram chat, and since then, every time I talk to it, a session opens indefinitely and my Openclaw remains stuck. I have already tried running “/reasoning off” again (and I receive a confirmation message saying that it has been disabled), but the problem remains. I tried openclaw gateway restart, deleting the session.json file, and openclaw doctor --fix, but every time I restart my Openclaw and send it a message, a session is created and never ends. Does anyone have a solution?

(receive "Request timed out before a response was generated. Please try again, or increase agents.defaults.timeoutSeconds in your config." 10mn later (after asking him "hey, can you send me a message like 'Hello' "))

[swift willow]

Try deleting your agent session files

They cause the bot a lot of the time to get stuck

/.openclaw/agents/main/sessions

delete everything in there and restart the gateway

[wise lark]

Everything?

[swift willow]

Ya everything

it’s not the main memory

it’s just all the session chatter

you can also try /reset in chat

[wise lark]

It seems that this does not solve the problem. I ran openclaw gateway stop and then deleted all the files in ~/.openclaw/agents/main/sessions. I then ran openclaw gateway start and the first message I sent on Telegram was “/reset” and I received the confirmation message:

"✅ New session started · model: moonshot/kimi-k2.5."

I then sent “Hey tell me something” and got no response. When I run openclaw status these sessions appear:

Sessions
 agent:main:telegram:direct:9269… │ direct │ just now │ kimi-k2.5    │ unknown/256k (?%)
 agent:main:telegram:slash:92696… │ direct │ just now │ kimi-k2.5    │ unknown/256k (?%)

[burnt shard]

HI, I have also a problem. I installed openclaw last weekend and finished initial wizard (in terminal, TUI or whatever). Told the bot his name, his nature etc...And after disconnecting and getting back at it I start all over again. It is telling me, that it does not have read, write, exec tools. And can not write/read from memory. ON the other hand i was able to connect to it using Telgram, but still I have bot without memory 🙂 Can you help me please ?

[dim harbor]

bot probably is set to "messaging" mode tool access. ask it how to configure that to give it permissions to other tools

[wise lark]

And since I sent the message “Hey tell me something,” he sent me the following messages (without me replying in the Telegram conversation):
9:56 a.m. -> “✅ New session started · model: moonshot/kimi-k2.5”
10:07 a.m. -> “✅ New session started · model: moonshot/kimi-k2.5”
10:10 a.m. -> “Request timed out before a response was generated. Please try again, or increase agents.defaults.timeoutSeconds in your config.”
10:42 a.m. -> “✅ New session started · model: moonshot/kimi-k2.5”

[trim estuary]

can anyone tell taht is there any free local model that can perform tool calling i am currently using gemini 3.1 and i want to wsitch to any free model please help

[wary osprey]

Check these three things:

1. Agent profile / tools config
   Make sure the agent is not running in a restricted profile. If it’s on something minimal (like messaging-only), it won’t have read, write, exec, or memory tools

2. Same workspace / agent folder
   If OpenClaw starts from a different directory or workspace after reconnecting, it behaves like a fresh agent. Run it again from the same folder where it was initialized

3. Session / memory
   If you’re reconnecting through Telegram, you may be opening a new session each time. The agent will answer, but the previous context won’t be there unless memory tools are active

start OpenClaw locally in the terminal again and ask it to save something to memory, then restart and check if it can read it back. If it still can’t access tools, paste your config (tools and agent section) and we can check what’s blocking it

[solid verge]

hello, how to fix rate limit exceed error?

[burnt shard]

Thank you i will give it a try.

[mighty sleet]

Hey friends! This is Evy from NemoVideo — an AI video editing agent that currently building our own editing skills plus your personalized openclaw specialized for video editing

We are looking for creators to build this future together, anyone interested for a casual paid chat？

1. ✍️Topic: Your video editing x openclaw workflow, your skill needs
2. 🔍Who we're looking for:
   • You edit videos regularly/ you have video editing needs
   • You are deep user of OpenClaw or any AI agent in your workflow
3. What you get:
   💰 $50 USD
   🎬 1 month NemoVideo Pro (free)
   DM me please!!

[wary osprey]

If you want a free local model with tool calling, you can try Qwen2.5 (7B/14B) or Llama 3.1 (8B). Many people run them locally with tools through setups like Ollama. If you prefer not to manage local models or hardware, another option is using AIsa. It gives you access to multiple models through a single API key, so you can switch models easily depending on what you're building. It's a unified API provider like OpenRouter

[hollow trail]

Hi. I was fresh installing openclaw to Windows. But I am stuck with error like: |15580933 [main] sh 46 dofork: child -1 - forked process 22008 died unexpectedly, retry 0, exit code 0xC0000142, errno 11

[dusk gyro]

Hi guys, I'm tring to setup blue bubbles on mac mini. So far everything looks fine and the bot CAN send messages, however the webhooks doesn't work and it says that the endpoint is not found:

[2026-03-13 12:34:08.535][debug] [WebhookService] Failed to dispatch "new-message" event to webhook: http://192.168.1.194:18789/bluebubbles-webhook?password=<hidden>
[2026-03-13 12:34:08.535][debug] [WebhookService] -> Error: Request failed with status code 404
[2026-03-13 12:34:08.535][debug] [WebhookService] -> Status Text: Not Found

[feral cradle]

Reasoning mode can sometimes cause the agent to get stuck if the model keeps extending the internal reasoning chain.

A couple things worth checking:

1. openclaw logs --follow
   See if the gateway is retrying the request repeatedly.

2. agents.defaults.timeoutSeconds
   If the model takes too long the gateway may restart the session.

3. try switching the model once (even temporarily) to confirm it's model-specific.

Kimi reasoning sometimes loops if the response never reaches the tool execution stage.

[feral cradle]

That warning in the logs is interesting:

tools.profile allowlist contains unknown entries (apply_patch)

If the tool isn't actually enabled the agent may think it wrote the file even though the tool call failed.

Worth checking:

openclaw tools list

and confirm the coding profile actually includes the file tools plugin.

Sometimes after updates the tool profile references entries that aren't loaded anymore.

[wise lark]

Ty 🙂

[torn jolt]

i use telegram, but i keep getting this error! whats going on??

11:47:45 [telegram] telegram sendChatAction failed: Network request for 'sendChatAction' failed!
11:47:48 [telegram] telegram sendChatAction failed: Network request for 'sendChatAction' failed!
11:47:51 [telegram] telegram sendChatAction failed: Network request for 'sendChatAction' failed!
11:47:54 [telegram] telegram sendChatAction failed: Network request for 'sendChatAction' failed!
11:47:57 [telegram] telegram sendChatAction failed: Network request for 'sendChatAction' failed!
11:48:00 [telegram] telegram sendChatAction failed: Network request for 'sendChatAction' failed!
11:48:03 [telegram] telegram sendChatAction failed: Network request for 'sendChatAction' failed!
11:48:06 [telegram] telegram sendChatAction failed: Network request for 'sendChatAction' failed!
11:48:09 [telegram] telegram sendChatAction failed: Network request for 'sendChatAction' failed!

[cloud swallow]

my openclaw cant update

npm error code EBUSY
npm error syscall rename
npm error path C:\Users\me\AppData\Roaming\npm\node_modules\openclaw
npm error dest C:\Users\me\AppData\Roaming\npm\node_modules.openclaw-zyI2HD6g
npm error errno -4082
npm error EBUSY: resource busy or locked,

even if i stop all openclaw

[finite sable]

If someone knows how I can reduce the context for my openclaw AI by a lot, please tell me

[native thunder]

sa

[native thunder]

same bro we have the same problem

[native thunder]

sent you a dm

[tribal knoll]

restart openclaw gateway once

[fallen vortex]

Hello all
Why when i click update ( for the open claw version showing on the portal ) it keeps saying updating .. but nothing actually happens .. how can this be soomth ? Or be automated ?

[indigo goblet]

did you talk with your agent?

to try to troubleshoot it

[vapid tinsel]

Hey! Has anoyone foudn a good solution for agent+human task management with openclaw? I would like my agent to plan / create tasks, but also need a way for myself to see / add tasks there as well (or maybe that's just my old-school experince and everything is chat-driven now). I searched for solutions and e.g. Linear is an overkill for most people, while systems like "beads" and simlar are developer-specific and not working well for e.g. marketing team planning it's launch strategy

Would appreciate you advice / experince e.t.c.

[indigo goblet]

can you be a bit more specific, are you looking for an UI board where things are clear to see and manage?

[fallen vortex]

Yes .. but everytime the agent direct me to do a fresh update through powershell comands .. and usually the end result is that the update is done .. but the portal is still showing an older version .. until i manually interfere to fix it all together

[steel quail]

Yeah I think its just a Grok issue. I've been using grok4.1-fast for cheap tests, and there seems to be an issue with tool calls for Grok. When doing some verbose troubleshooting, even if I explicitly ask it to use the write tool, it fails to call the tool. Other models have no problems, so it seems Grok isn't a viable option for Openclaw.

[indigo goblet]

does 'openclaw update status' command show you that it is indeed updated?

[fallen vortex]

Yes

[indigo goblet]

if yes then you'd need to actually fix the portal not showing the actual current update, might be a bug from the current openclaw version itself, you'd have to check that out first.

in any case, I think your agent should be able to pinpoint the issue

[steep galleon]

Guys, I have Claude Max Abo and im thinking about cancelling it and buy the Pro Codex Abo - Does the Codex have more usage then claude?

[indigo goblet]

did you ask him why that keeps happening?

[warm rover]

Hello, battling with this error here for a while: "error:Failed to resolve Discord application id" My discord agent just won't come online anymore.

Does the applicationId even need to be part of the config? I don't see any mention of it in the openclaw docs discord setup page.

[severe burrow]

If yes, then ask it to debug and fix itself

[warm rover]

Yes, He is struggling and keeps hitting rate limits trying to solve it

Turned out to be a bot token mismatch... All good. Thank you

[dapper basin]

I don't need it to do that much either, as long as it allows interaction and information sharing on Moltbook.

[rugged bramble]

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

[wary osprey]

Try these:

1. add a short delay between requests
2. lower the number of tool/API calls per task
3. switch to a model/provider with higher limits
4. check if your agent is looping and sending repeated requests

[steep vale]

My cron jobs do not work, they get put in disabled and the agent does nothing, why?

[true sky]

Hello

I am installing openclaw via hackable

its getting stuck at Updated pnpm allowlist for git-hosted build dependency: @tloncorp/api . Tried verbose and now its stuck at A2UI bundle up to date; skipping.

For context, I'm running openclaw on a gcp compute enging(e2 medium, 2 cores, 4gb ram)

[tribal knoll]

not good performance for running openclaw

[true sky]

I read that it was enough

on the docs page

Type Specs Cost Notes
e2-medium 2 vCPU, 4GB RAM ~$25/mo Most reliable for local Docker builds
e2-small 2 vCPU, 2GB RAM ~$12/mo Minimum recommended for Docker build
e2-micro 2 vCPU (shared), 1GB RAM Free tier eligible Often fails with Docker build OOM (exit 137)

[tribal knoll]

okay, okay that's enough

[true sky]

so whats the problem?

[tribal knoll]

just try to "openclaw onboard"

[true sky]

meaning?

I just ran the hackable script

then i'll onboard

[tribal knoll]

??

[true sky]

I'm getting stuck at the install stage

[tribal knoll]

for openclaw, right?

[true sky]

yes

[tribal knoll]

so did you get trouble with installing openclaw?

[true sky]

yees

[tribal knoll]

so what's the problem?

[karmic current]

Hey all I'm looking for someone who can review our SKILL.md file, it will be published on ClawHub in the next 12 days or so.

[true sky]

ts getting stuck at Updated pnpm allowlist for git-hosted build dependency: @tloncorp/api . Tried verbose and now its stuck at A2UI bundle up to date; skipping.

It's*

[tribal knoll]

I willl do it

[tribal knoll]

contact me

[tribal knoll]

what command did you use?

[true sky]

curl -fsSL https://openclaw.ai/install.sh | bash -s -- --install-method git

[tribal knoll]

are you going to run it on your pc?

[true sky]

on a gcp compute engine

[tribal knoll]

did you clone the openclaw?

[true sky]

meaning?

[tribal knoll]

I can't guess what are you doing

[true sky]

like i did the quickstart hackable command

I thought it auto clones for me

[karmic current]

@tribal knoll that's very kind! I'll drop the github link in a sec, I guess you wouldn't open a direct link to our website 😛

Here u go: https://github.com/PrivyPad/clawhub/blob/main/SKILL.md

[tribal knoll]

hmm, just run this command in empty folder?

[tribal knoll]

ok, I will check it

[karmic current]

Thanks much appreciated Robin!

I just commited a minor update, the AI agents couldn't use the API due to an error in the URL (moved to Cloudflare that redirects the website to www.)

It should work now.

[true sky]

in. home folder

should i reset the vm and try again

?

[tribal knoll]

hmm, first git clone openclaw

and run that command in that folder

[true sky]

I'm resetting vm and doing okay

I'll come back in 10 minutes

[tribal knoll]

https://docs.openclaw.ai/install

and read this docs

[copper gulch]

I have a large exclamation point in the chat and the chat won't load on the dashboard. Can anyone help me?

[tough swan]

Is anyone having success with their agent automatically updating openclaw, and the gateway loading as well? Yesterday I used the CLI "Openclaw update" and everything loaded perfectly. Today, I told my agent to update, and it never restarts the gateway so I manually have to use the doctor or launch the gateway in CLI

[karmic current]

Nopes I've done it myself a few hours ago. I use Docker, I only had to run three commands to get it done.

I didn't even know agents were able to do that lol.

[tribal knoll]

CLI command works well, but agent can't work all

[tough swan]

Is there any hope to have the agent build a workflow to simply just use the CLI? Like a cron job?

[tribal knoll]

it can be possible

[tough swan]

My complaint with this stems from needing to remote into the OC machine. I'm using mac screen sharing presently, and my main Macbook disconnects from the Mac Mini (OC Machine) frequently...I haven't figured this one out yet either how to allow "permanent access" without approval on the Mac Mini

[tribal knoll]

did you set remote configuration?

[tough swan]

No, I've never done anything like that before aside from Google Remote Desktop on other computers 😅 . Anything terminal/ssh/etc is totally foreign to me

[tribal knoll]

haha

[tribal knoll]

openclaw provide remote access feature

[waxen geyser]

Hello people - ive got openclaw running on digitalocean - i then have my macbook connected as a node via the macos app in the top bar. Is there anyway i can have it read files from my mac?

[sage maple]

Hi guys, I have a discord setup for my bot, but its not great and I would like to switch. What is the best platform to work with openclaw ? Mostly for coding

[tribal knoll]

just use dashboard, that's the best place

[mossy tree]

can someone please hgelp me , Im getting "Health check failed: gateway closed (1006 abnormal closure (no close frame)): no close reason"

[foggy gulch]

Hi, I've setup OpenClaw on a Linux VM and Ollama (qwen3.5) on the Windows 11 host. The two communicates, but OpenClaw often stops working while it's doing something more complicated than just talking to me. Anyone maybe have a similar setup?

[atomic kiln]

Hi, I'm new to the OpenClaw Discord server. I've been setting up and using a multi-agent stack system on open claw for about a month now. Encountering a lot of challenges ofc; being able to solve some of them, some still looking for solutions.

What's the best place/channel on the server for advice on multi agent coordination/orchestration? Also looking for solutions regarding agent drift and neighbouring problems.

Appreciate any help/hint to where to look at! Cheers!

[indigo nebula]

Why not just ssh into the OC machine when you need to do something like a gateway restart? If you need a browser on the OC machince to open 127.0.01:18789 on the OC machine just install TailScale on the OC machie and your Mac then you can connect to it via http://100.x.x.x:18789 from your main mac.

[indigo nebula]

Mine has built a number of things that run from system cron and not use any openclaw or LLM tokens. Things like doing a check via npm to see if there's a new version of openclaw or other plugins available. Doing a daily healthcheck of the system. Just tell your claw what you need and that you want it to build the workflow using the CLI for whatever you're asking it to do in your workflow

[indigo nebula]

Yes. Mine has instructions to firs tell me a new version is available and to only do the update once I approve. This way if I'm not at the house and it goes sideways I'm not down until I can get home and fix it. But I ask specially it to "carefully review the new release notes, determine if there are any concerns or dependencies we need to address as part of the update and finally to provide step-by-steps of what it will do before it runs the commands including confirming that it will run a config validate and use the dry-run flags to make sure nothing will break. Once that passes you can proceed with the update and gateway restart". It takes care of the rest. I usually give it a few minutes then I prompt it and say "Did it work?" and she'll reply with the details of the upgrade.

[wise lark]

Hello again, I'm looking for a way to approve the "exec approval" via Telegram. Does anyone know how to set this up?

[indigo nebula]

Create a free Trello.com account. Ask your claw to build a Kanban board on it with columns of "ToDo, InProgress, Blocked, Done" and to create all tasks/steps for a project on the trello board, including assigning them to itself or you/whomever if it is not a task it can do. There's a Trello skill already you can get off clawhub. Once you get stuff in there you need to tell your claw "Don't assume you know what to do, read the trello card first in the even I or others added/updated information on the card. One then can you move a card to "In Progress" and begin the task."

[olive cliff]

This is why I have the ability remote in through my phone. If anything goes wrong, I still have full remote access to my system with my phone.

[late narwhal]

anyone have a good setup/configuration guide for kimi 2.5? I realized I just had maxTokens set at 8k and contextWindow at 200000 instead of 262000. So wondering if I'm missing anything else

[tribal knoll]

hmm, what's the problem?

[indigo nebula]

I have it too via TailScale but I'd rather not try fixing it from my phone screen. Doing an OC upgrade remotely isn't a top proirity. It can wait till I'm back home or somewhere I can use my laptop.

[olive cliff]

Honestly I fully understand the struggle doing it through a phone screen lol. Its definitely a nice if all else fails fallback to have though. Having a tablet handy kinda helps if its got internet, or tethered through your phones connection.

[wise lark]

I asked my Openclaw to use a Subagent for a task, and while it was performing the task, it sent me the following message:

Approve it from the Web UI or terminal UI, or from Discord or Telegram if those approval clients are enabled.```

Then I asked him how to fix it, and he edited his own ``openclaw.json`` file by adding
```json
"approval": {
  "mode": "explicit",
  "chatClients": {
    "telegram": {
      "enabled": true
    }
  }
}

But by restarting the gateway:

Config invalid
File: ~/.openclaw/openclaw.json
Problem:
  - tools.exec: Unrecognized key: "approval"

[tribal knoll]

Check your web dashboard

[indigo nebula]

Roll back your openclaw.json or just remove that section, restart to get backonline then tell your claw to go read https://docs.openclaw.ai/channels/telegram#exec-approvals-in-telegram before making config changes and to show you what the settings it is propsoing to add after reading the docs. Mine assumed some settings previously but once I told it to review the docs and only use settings that exist it seemed to do a better job with things like this.

[wise lark]

Thank you so much 🙂

[tough swan]

You rock, thank you! I'll review all this over the weekend and take it from there.

[simple phoenix]

NEEED HELP i set up openclaw using hostinger and first step was using ports, firewalls , tailscale but openclaw just stop working every time, can setup a mission contrl on a new tab, i ask i simple task and just stop working. i use codex 5.2
any guide to setup it securely without errors??

all i use is the freaking terminal to make it work

[chrome tartan]

Hey guys, how can I use talk mode with Eleven Labs voices?
https://docs.openclaw.ai/nodes/talk#config-/-openclaw/openclaw-json
when I tried to add "talk" key to clawbot.json, it results:

Invalid config at /Users/rafaelplantard/.openclaw/openclaw.json:\n- plugins: Unrecognized key: "talk"

🦞 OpenClaw 2026.3.11 (29dc654) — I'll refactor your busywork like it owes me money.

Config invalid
File: ~/.openclaw/openclaw.json
Problem:
  - plugins: Unrecognized key: "talk"

[tribal knoll]

I am also using bot in hostinger vps, I think I can help you

[next topaz]

🐛 2026.3.12 regression: Slack socket mode connects but receives zero inbound events

Just spent ~3 hours debugging this. After upgrading from 2026.3.11 → 2026.3.12, Slack stopped receiving any inbound messages. Agents could still POST to Slack (outbound worked fine), but zero events arrived through the socket.

What we verified on 2026.3.12 (all passed):

• Bot token valid (auth.test ok)
• App token valid (apps.connections.open returns wss:// URL)
• Socket shows "connected" in logs
• Channels resolve correctly
• Event subscriptions present (message.channels, message.groups, app_mention, etc.)
• openclaw doctor clean, openclaw status --deep shows Slack: ok
• Regenerated app-level token, reinstalled Slack app to workspace
• Multiple gateway restarts

Symptoms:

• tail -f on gateway log while sending Slack messages = zero new entries
• Health monitor detects "stale-socket" periodically, reconnects, still no events
• Cron agents also failing with "Action send requires a target" in isolated sessions

Fix: Downgrade to 2026.3.11 → Slack immediately works again.

Environment: macOS arm64, Node v25.8.0, npm global install, socket mode

[simple phoenix]

please

[tribal knoll]

what's the problem exactly?
can't you run bot successfully?

[prisma inlet]

Gpt 5.4 fast mode is out? Whats the exact config change in the json i gotta right to enable it?

[simple phoenix]

errors everywhere i ask a task and stops working, cant open a mission control,i think is always something eith the ports, firewalss, etc.

[tribal knoll]

🤔 I can't guess the problem exactly

[simple phoenix]

how did u setupt it up may i ask. security, ports firewalls u use tailscale? does yours works normally?

i have problems everywhere

[simple phoenix]

.

[haughty elbow]

Hey whats a good cheap way to run openclaw without it using my main computer? im sure this gets asked aton but i must know.

[simple phoenix]

vps, hostinger 15 usd a month. u can get cheaper ones also

[dim harbor]

vps, something like hetzner. should be under $10

[haughty elbow]

ok, by chance are there any guides or youtube videos that talk more about this?

[simple phoenix]

Need help with OpenClaw setup on Hostinger (Constant errors & crashing)

Environment:
• Host: Hostinger (VPS)
• Network: Tailscale, custom ports & firewall configured
• Version: Codex 5.2

The Issue:
I managed to get OpenClaw running via the terminal, but I'm facing a few critical problems:

I'm getting constant errors in the terminal all the time.

It completely stops working and crashes every single time I give it a simple task.

I can't set up or access "Mission Control" in a new browser tab, so I'm stuck doing everything purely through the terminal while it throws errors at me.

The Ask:
Does anyone have a reliable guide or best practices to set this up securely without these recurring errors and crashes? Any help would be greatly appreciated!

[prisma inlet]

Gpt 5.4 fast mode is out? Whats the exact config change in the json i gotta right to enable it?

[sturdy tundra]

can anyone help me remove ollama model provider? i already got it out of openclaw.json but it keeps mentioning that it cant find models when i restart the gateway

[swift slate]

Hey, I updated from v2026.3.7 (I think) to v2026.3.12 and now when I load the page there's a warning sign (or at least it looks like that as it doesn't fully show) like this: ⚠ instead of the chats. Did I do something wrong? Or do I need to do any other setup? I'm using it in Ubuntu via LAN.

[karmic current]

@haughty elbow have u considered a Pi?

[haughty elbow]

I do have a pi5 8gb laying around but i wasent sure if i should use it or what type of setup to make with it.

[sturdy tundra]

u might be able to use a free oracle cloud instance. they have a fine free tier

[karmic current]

I advise u to use Docker, much easier to troubleshoot.

[karmic current]

You can use it to run the gateway and cli, not a problem. You probably need to set up your router to enable port forwarding if u want to access it from the outside.

[sturdy tundra]

i recommend cloudflare warp or tailscale instead. port forwarding openclaw itself without proper security can be rly dangerous

[haughty elbow]

So a online service is better than a pi 5?

[karmic current]

Nah he meant managing the ports.

Like allowing yourself and/or others to access the Dashboard via the internet, from a cafe for example.

You can run the software on a Pi 5 without a problem.

I run it on my macbook because I don't need my agents to work 24/7 for me, if I want that I'd use my Pi.

[indigo nebula]

Try running the following from command line to make sure you don't have something wrong in your config or some other issue going on. Do these one at a time and address any issues int he config before you run the doctor.

openclaw config validate

openclaw doctor

openclaw doctor --fix

[haughty elbow]

could i also use a old laptop perhaps?

[karmic current]

Yes as long as it have enough resources.

Like RAM and CPU, HDD/SSD etc.

[haughty elbow]

ok

[karmic current]

I will probably install OpenClaw on a Pi next week, I'm testing a group of social media agents right now, 1 that is doing research and another one that is posting and engaging with people.

It would be probably useful to get them post on X all day.

I can write a small guide for you @haughty elbow but it wouldn't be much different than any other machine, but I prefer Linux.

[haughty elbow]

i think ill be fine on my own.

[swift slate]

Hi Shane. Thank you very much. I got "Config valid: ~/.openclaw/openclaw.json" and then I did run the doctor and doctor fix, but still nothing.

[frigid summit]

Just now bot showed up with error message and when I asked him what’s this, he said he doesn’t remember me and don’t have any of our history, and I’m in the lobby

How to get back to my personal agent?

[indigo nebula]

What happens if you close the browser, relaunch and go to http://127.0.0.1:18789/overview. Once on that page click "Connect" then try clicking on the "Chat" link and see if it comes back?

[weak wyvern]

Hi guys! quick question, can openclaw initiates a new convo on whatsapp? (I'd like it to DM someone)

[slow ingot]

Yes, that works

[opal rock]

I have given up guys. I spent more time trying to get openclaws gateway to restart and configure local models than getting things done and even when I did ask it to do things. It got it all completely wrong or built long winded complex systems that were too buggy to maintain. I understand that it will depend a lot on the model. But it’s so way behind the road at the moment. And to see its competitors deliver solid solutions really makes my heart bleed.

Adios and I will probably come back in a few months to see if there has been progress. Unfortunately it’s all been hype and expensive at that

[swift slate]

It shows me like a login page in which I need to put the token, then it asks for pairing, and after pairing I'm able to hit "connect", and then I get to the same warning symbol which covers almost the full space of where the chat used to show.

I also have (with the update) a new menu which shows my conversations, including the one via telegram, but neither of these will load anything it will just show "⚠" covering the section of the chat.

[sly carbon]

Is there a decent setup guide? I'm really struggling to get mine working with my OpenAI API, and I don't know if I need a free or premium membership or not? I'm not particularly well off financially, so I'm not looking to spend huge sums of money just to get OC working.

[weak wyvern]

how could I configure it? it says it can't send any message...

[simple phoenix]

i use docker inside hostinger

[slow ingot]

Oh.... sorry for confusion on it.. it is impossible..

[slow ingot]

Instead of DM messaging, you can do it in a group.. for instance, you can bring all the members in the group and mention someone you want to interact in the group.

[simple phoenix]

what u do it to keep it safe?? tailsclae ports firewall or what, i just want to use it for myself safely

[indigo nebula]

Sounds like you might need to generate a new token via openclaw doctor --generate-gateway-token but if you have Telegram working it might be better to first via Telegram tell your claw "I upgraded versions and now I'm getting this warning symbol on the web chat at http://127.0.0.1:18789/chat can you analyze the logs and configs to determine what might be causing this issue and correct it?" Mine is pretty good at debugging that kind of stuff.

[weak wyvern]

sorry but you look like an AI hallucinating 😭
but thank you so much for your answer, it's super useful ❤️

have a nice day!!

[slow ingot]

I am openclaw... LOL~ anyway, thanks for your kind attention

[weak wyvern]

I don't know what to believe 😭

[swift slate]

I just realized Telegram is not working either. I'll try doing a clean install, I think. Thank you so much for your help. 🙂

[indigo nebula]

Good luck!

[sterile brook]

For those of you leveraging acpx to get your OpenClaw agent to manage other coding agents, is "allow-all" permissions the only viable pathway? Or is there a way to relay exec approval prompts from the external agent through the OpenClaw instance to you?

[weak wyvern]

do you know if it works on Telegram?

like if my bot can DM people on Telegram (even though it's not possible on whatsapp?)

[slow ingot]

100% impossible for now by default. But there is a solution we can try... that is to build another bot in tg responsible for handling DMs. Let me share the details what I am thinking of.

1. ingesting all the infomation or report by openclaw (if you would need)
2. building a tg bot that will interact with 2 apis - find users and then getting chat_id from the response, and then make it happened in DM.

[meager vessel]

my openclaw is disconnected from my telegram, how can i fix it?

[indigo nebula]

Try to Pair it again https://docs.openclaw.ai/channels/telegram#telegram

openclaw gateway openclaw pairing list telegram openclaw pairing approve telegram <CODE>

[meager vessel]

if i have the telegram token? what can i write then?

[kindred ridge]

hello
just joined the server here for help
im finished with the installation but cannot access the chat UI
error: RPC probe failed
1006 abnormal closure

any direction/resources for me?

[meager vessel]

fixed the issues with doctor fix!! thanks for the help

[indigo nebula]

Follow the instructions at this post in the #1459642797895319552 channel. https://discord.com/channels/1456350064065904867/1482041323413311549

[dark scaffold]

Hello everyone!! new to the server- been messing with my ClawdyBot- Mine seems to be "forgetting" a lot of stuff- even stuff as recent as a couple hours ago- I must have done the configuration wrong as I am often going back and forwarding messages or screenshotting our conversations for it to remember- Has anyone experienced this or have any ideas?

[mint idol]

I’m having error: gateway timeout after 3000ms

But when I run “gateway discover” it shows up my instance

[indigo nebula]

You should first ask your claw to install QMD from https://github.com/tobi/qmd then tell it to read the docs for QMD at https://docs.openclaw.ai/concepts/memory#qmd-backend-experimental and make the appropriate settings to your openclaw.json. Once you've got that installed tell your claw "Now that you have QMD installed, please use it to search your local memory files for things we've done in the past. Make this a standing rule when I ask you about something we've previously done" That will help it remember a lot more things but this is only part of the problem. Once you have QMD working you want to install LCM (lossless-claw plugin). Tell your openclaw the following: "read the following repo on Github so you uderstand what LCM does https://github.com/Martian-Engineering/lossless-claw then download it from clawhub and configure the plugin to use LCM." Once I did these my claw remembers things even after multiple upgrades of openclaw with gateway restarts, system reboots you name it.

[dark scaffold]

beautiful!! i'll give these a shot!! I really appreciate it!

[clear jacinth]

hey does anyone have a mission control/virtual office set up for their agents/ subagents if so how do you guys like it do you think it helps out ?

[sleek depot]

Hi guys please help me, I'm stuck with an outdated version of the MacOs app, and I can't figure out where to update it from
My agent is no help, and I'm trying to set up talk mode. I have been keeping up with OpenClaw updates consistently and I'm uptodate from that side.

[potent nebula]

acciently screwed up the installation and now i can’t fully uninstall it

how do i uninstall clawdbot and start from scratch

also, is it possible that I can use clawdbot to control my desktop and do work for me?

[mint idol]

Mine shows: gateway timeout

[indigo goblet]

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

[indigo goblet]

yes it is possible but you'd have to be much more specific in what do you want him to do.

[potent nebula]

I ran this

openclaw uninstall --all --yes --non-interactive
npx -y openclaw uninstall --all --yes --non-interactive

yet, i still don’t have to reinput api keys and tg channels

basically just makes me create directories again when i try to reinstall

[potent nebula]

for example if i had schoolwork can i tell it

“go to chrome on my computer, open [this tab] and finish the assignment?”

[indigo goblet]

you are probably missing one or multiple command to uninstall it correctly, did you read the full doc on it.

[potent nebula]

i’ll try in a few

[indigo goblet]

yes it is possible.

[potent nebula]

do i need any plugin or something

or is this something clawd can do naturally

i badicslly want a lap dog to do redundant work for me

[indigo goblet]

it depend of the complexity and how often, you might need a skill, but it's hard to tell for now I'd say.

[stuck magnet]

I just set up open claw but I wanna make some quick cash so like what are some examples of stuff open claw can do

[indigo nebula]

Have you tried stoping and restarting the gateway?
`
First see if it is running
openclaw gateway status

If not restart it with these
openclaw gateway stop
openclaw gateway start
openclaw gateway restart
`

[mint idol]

Yes

[indigo nebula]

What does it say when you run the gateway status

[mint idol]

It’s weird because it shows up on “gateway discover”

[weak trail]

🚨 Join the “Weekly Claw” 📺🚨
LIVE NOW
https://discord.gg/bFHgpe75j?event=1479314622669520996

[stuck magnet]

I just set up open claw but I wanna make some quick cash so like what are some examples of stuff open claw can do

[iron cradle]

for real?

[indigo nebula]

Maybe try running openclaw gateway health --url ws://127.0.0.1:18789 or do a openclaw gateway stop then run openclaw gateway run to run it as a foreground process this way you can see if it throws any errors to the console where you ran the command. There's a number of commands related to the gateway at https://docs.openclaw.ai/cli/gateway that might also help you

[stuck magnet]

OpenClaw: access not configured.

Your WhatsApp phone number: #########

Pairing code: #######

Ask the bot owner to approve with:
openclaw pairing approve whatsapp #######

My open claw keeps sending this on WhatsApp

I don’t know what it means and how do I stop it as it sends it at certain times to people are texting

At that time

[indigo nebula]

Go to your server where you have openclaw installed, open a command prompt and type openclaw pairing approve whatsapp ####### where the ###### is the number it gave you

[stuck magnet]

What will it do then

[indigo nebula]

It will pair whatsapps with your account so you can message it.

[stuck magnet]

Oh right

Ty

[mint idol]

Service: systemd (enabled)
File logs: /tmp/openclaw/openclaw-2026-03-13.log
Command: /usr/bin/node /usr/lib/node_modules/openclaw/dist/index.js gateway --port 18789
Service file: ~/.config/systemd/user/openclaw-gateway.service
Service env: OPENCLAW_GATEWAY_PORT=18789

Config (cli): ~/.openclaw/openclaw.json
Config (service): ~/.openclaw/openclaw.json

Gateway: bind=lan (0.0.0.0), port=18789 (service args)
Probe target: ws://127.0.0.1:18789
Dashboard: http://192.168.1.56:18789/
Probe note: bind=lan listens on 0.0.0.0 (all interfaces); probing via 127.0.0.1.

Runtime: running (pid 254, state active, sub running, last exit 0, reason 0)
RPC probe: ok

Listening: *:18789
Troubles: run openclaw status
Troubleshooting: https://docs.openclaw.ai/troubleshooting

[mint idol]

but:

root@openclaw:~# openclaw browser --browser-profile chrome tabs

🦞 OpenClaw 2026.3.12 (6472949) — I can run local, remote, or purely on vibes—results may vary with DNS.

│
◇ Doctor warnings ──────────────────────────────────────────────────────────────────────────╮
│ │
│ - channels.telegram.groupPolicy is "allowlist" but groupAllowFrom (and allowFrom) is │
│ empty — all group messages will be silently dropped. Add sender IDs to │
│ channels.telegram.groupAllowFrom or channels.telegram.allowFrom, or set groupPolicy to │
│ "open". │
│ │
├────────────────────────────────────────────────────────────────────────────────────────────╯
│
◇
Error: gateway timeout after 3000ms
Gateway target: ws://127.0.0.1:18789
Source: local loopback
Config: /root/.openclaw/openclaw.json
Bind: lan

[celest helm]

I got a giant triangle with an exclamation mark on the UI and I can't write from the UI anymore, but I also see other channels like telegram and whatsapp and they work and respond but the main one doesn't work

[indigo nebula]

It appears to be running. I'd try working through these and see if you can find any errors running these one at a time.

openclaw browser --browser-profile openclaw status openclaw browser --browser-profile openclaw start openclaw browser --browser-profile openclaw open https://example.com openclaw browser --browser-profile openclaw snapshot
There's some additional info at https://docs.openclaw.ai/tools/browser#browser-openclaw-managed

[mint idol]

root@openclaw:~# openclaw browser --browser-profile openclaw status

🦞 OpenClaw 2026.3.12 (6472949) — No $999 stand required.

│
◇ Doctor warnings ──────────────────────────────────────────────────────────────────────────╮
│ │
│ - channels.telegram.groupPolicy is "allowlist" but groupAllowFrom (and allowFrom) is │
│ empty — all group messages will be silently dropped. Add sender IDs to │
│ channels.telegram.groupAllowFrom or channels.telegram.allowFrom, or set groupPolicy to │
│ "open". │
│ │
├────────────────────────────────────────────────────────────────────────────────────────────╯
│
◇
Error: gateway timeout after 1500ms
Gateway target: ws://127.0.0.1:18789
Source: local loopback
Config: /root/.openclaw/openclaw.json
Bind: lan

[indigo nebula]

If you have other channels that are working use that and ask your claw "I'm getting a giant triangle with an exclamation mark on the UI and I can't write from the UI anymore. Can you look through the logs and debug what is happening and apply a fix if you can determine what the cause is?" I've done that previously and my claw fixed itself

[indigo nebula]

What happens if you just open a browser like chrome or ff on that computer and type: http://127.0.0.1:18789 into the URL bar does it load the openclaw console?

[mint idol]

Looks like its a bug in the last release: https://github.com/openclaw/openclaw/issues/44714

[indigo nebula]

Well if you installed it via npm, you can open a command prompt and roll back to the previous verion with npm install -g openclaw@2026.3.11 which would roll you back one version. After it rolls back run openclaw config validate to make sure you don't have any config errors followed by openclaw gateway restart and you should be back up on the previous version

[celest helm]

It seems to me like a bug in the latest update, because I did everything with the help of my agent's telegram but nothing remains the same

[torn pelican]

im unable to update openclaw

its yelling about sharp & node-addon-api

[mint idol]

It's working after changed that hardcoede values

but now im' gettint his: Error: gateway timeout after 1500ms
Gateway target: ws://127.0.0.1:18789
Source: local loopback
Config: /root/.openclaw/openclaw.json
Bind: lan

[turbid lantern]

claw keeps clobbering the openclaw.json file, turning it from json5 into regular json, and removing all the $include clauses and inlining them in ... any way to stop it from doing that? its annoying.

[finite sable]

I need someone, anyone, to give me what openclaw expects/receives from openai compatible api when AI uses a tool/multiple tools
If anyone's able to help I'll be really really grateful

[scarlet silo]

good night, still cant use my openai plus account. i always get error of exceeded quota but its 100% free on 5h and weekly...

i dont understand

[ancient knot]

how can I optimzie cost for my openclaw??

anyone know please let me know

[indigo goblet]

are you using the right OAuth gpt account ?

[devout pike]

Can someone tell me if the cron jobs are broken? They barely execute at all

I have tried tons of configs. Nothing seems to work. Maybe broken in newest versions

?

[meager vessel]

google still baning for using them?

[mighty rampart]

Run mitmproxy and point openclaw at it and watch what is happening.

[finite sable]

uh I'll try tomorrow

[scarlet silo]

i think so.

anyway to check if oauth login is correct on openai.

[inland knot]

Anyone can recommend a good AI model that its free on OpenRouter that doesnt spit thinking in the chat?

[sharp sentinel]

Hi. Recently, I deployed my open claw on the template from this link.
https://railway.com/deploy/openclaw
I have a question because my biggest pain point is that I cannot see when my agents are working and how my agents are working, which leads to many problems, such as getting frustrated because I don't know if an agent is working or what the agent is doing. If I could see that an agent is currently working, I could plan my tasks differently.

Since the deployment is on Railway, I am not entirely sure how to create, for example, a mission control dashboard. I would like to ask if there is anyone here who would be able to help me with this? Is there someone who has done something similar and know how to deploy a proper mission control on Railway? If so, how can I do it? Please help.

[vast dove]

ask chatgpt, you can test it using a cdm or powershell command in windows.

[indigo nebula]

Have you tried asking your claw to make you a simple dashboard running on another port that you can monitor what calls it is making as a way to monitor what’s happening under the covers? Mine is pretty good at creating scripts and tools for me and revamped my company website.

[inland knot]

Are OpenRouter free models capable for OpenClaw ?

[naive summit]

I just updated my openclaw, and now all I get is this. anyone any idea what it wants to tell me?

I mean, error messages in plain English are so last century, it seems...

[vernal moth]

same here, no clue what to do

[naive summit]

by angry at the one who overhauled the UI? jk... I'm diving into the DOM. I hate DOM diving...

I think I got it.

<span><!--?lit$335933267$-->100% context used</span>

Which is wrong. I'm at 70% according to /status (which I now have to press return twice to execute???), and that's weird, too. I was ... darn, the new version cuts of the history. around 20~30% when I last checked

Same in another session... Context: 22k/131k (17%) definately a big bug (pun intended

[spark valley]

definitly been using is the best

nvidia/nemotron-3-super-120b-a12b:free just dropped

it damn good with 260k context

there is tthe stepfun reasoning for codbase iterating

and more powerful

:: guys i need help i want to use seperate apui keys for seperate agents i tried auth profiles but it seems that always fallback to the mian agents api

[devout wasp]

Hey ya all! Anybody else having issues with Grok roleplaying shell commands?

[naive summit]

tool handling for grok just got fixed in the current release---there is a chance that it got fixed from "working" to "not working". If you're on the current release, open a regression ticket.

[naive summit]

if you have an ad blocker, right-click the ⚠️ and block that ("Block this ad"). Just the icon.

[abstract folio]

Mac App Issues — Talk Mode Audio + Voice Wake (v2026.3.12, macOS Tahoe 26.3, arm64)

Two issues with Talk Mode / Voice Wake:

1. Talk Mode not playing back TTS audio
   • TTS audio files generate fine (ElevenLabs, eleven_v3)
   • Playing the files manually via afplay works perfectly — speakers are fine
   • But the Mac app doesn't play the audio back during Talk Mode
   • Restarted the gateway, toggled settings — no luck
2. Voice Wake stops listening after first successful send
   • Wake word triggers correctly the first time, transcribes and sends
   • After that first successful send, the recognizer doesn't seem to restart
   • Subsequent wake word attempts do nothing
   • Toggling Voice Wake off/on in settings is a workaround but shouldn't be needed
   Screen recording + microphone + speech recognition permissions are all granted. Fresh restart of gateway done.

[austere drift]

https://github.com/cybernati1776/openclaw-koboldcpp-tutorial I created a tutorial to help run OpenClaw through koboldccp ko with custom provider

[spare yacht]

New to the openclaw daemon, I've been chatting on the UI for more deep work. Maybe I am not prompting it right but my assumption was if I tell claw to set daily cron jobs for something or do X work and notify me at Y time, it would automatically set a cron for X task. Am I missing something.

[frank harbor]

Why deoes my telegram bot always become unresponsive until I go check the process status on my VPS? which wakes it up

[spare yacht]

@frank harbor I am also seeing the same issue!

[frank harbor]

It seems running sudo loginctl enable-linger $USER then restarting fixes it

[violet crystal]

Anyone have ny issues where Openclaw keeps saying it will do something and then never does? with recent updates

[high heron]

update wont effect ur models response ig

try

/think xhigh
or
/think high

basically use the highest lv of /think

[iron inlet]

Is anybody else facing this error with 2026.3.12?

FATAL ERROR: Reached heap limit Allocation failed - JavaScript heap out of memory

[mint idol]

you don't have enough emmory

Hi! I need help with this:

root@openclaw:~# openclaw browser status

🦞 OpenClaw 2026.3.12 (6472949) — I'm not saying your workflow is chaotic... I'm just bringing a linter and a helmet.

│
◇ Doctor warnings ──────────────────────────────────────────────────────────────────────────╮
│ │
│ - channels.telegram.groupPolicy is "allowlist" but groupAllowFrom (and allowFrom) is │
│ empty — all group messages will be silently dropped. Add sender IDs to │
│ channels.telegram.groupAllowFrom or channels.telegram.allowFrom, or set groupPolicy to │
│ "open". │
│ │
├────────────────────────────────────────────────────────────────────────────────────────────╯
│
◇
Error: gateway timeout after 1500ms
Gateway target: ws://127.0.0.1:18789
Source: local loopback
Config: /root/.openclaw/openclaw.json
Bind: lan

[quasi comet]

is it openai-codex/gpt-5.2-codex highest llm for chatgpt plus sub oauth?

[weary basin]

I use codex 5.4 Oauth and have Plus.

[quasi comet]

How? No way, they degraded it

[quasi comet]

5.4 is for pro subs also

[weary basin]

I was using 5.3 before when 5.4 dropped I set up
Oauth the same way for the new model.

What do you mean degraded?

[odd dirge]

Hey @solid falcon , hope you're good?

Noob here, so pardon the naivety of my questions.

I picked the qwen/qwen3.5-27b model during my onboarding with OpenRouter, but somehow it has switched back to Auto, any idea to how I can switch it back.

[nimble hill]

@faint thunder Saw in help section that you are setting up the docker container for gateway. Got it working? if not, let me know can extend some help, I have setup in k8s and its working fine

[nimble hill]

I picked the qwen/qwen3.5-27b model during my onboarding with OpenRouter, but somehow it has switched back to Auto, any idea to how I can switch it back.
in the chat, use /model <model name>

[odd dirge]

Thanks for looking out @nimble hill .

When you say "in chat", do you mean in the telegram chat or in the web ui chat?

[nimble hill]

should not matter, I use only web and integration via my own APi

[odd dirge]

Good. I used the Web UI.

after using the "/model qwen/qwen3.5-27b" command as you said, this is the error message I'm getting - Failed to set model: GatewayRequestError: model not allowed: qwen-portal/qwen3.5-27b

[nimble hill]

check which models are allowed as per your API key - /models <provider> I have not used qwen, so the provider would be either alibaba or aliyun i think. If it does not work. Go to command line of the gateway and do openclaw onboard go through the slection and choose correct model

[odd dirge]

thanks Chief, I've made the switch. 🙏

You are now my Official Patron Saint.

Saint Manjax 😇

[tame beacon]

Hmm, do someone have ideas for OpenClaw to work with foundry VTT? I must say that i linked it, and it can login etc, but eh it is really bad. Sometimes it applies dmg, sometimes not etc

[polar lava]

To run some business in OpenClaw, how do you add a credit card or payment info in Claw if we can’t use our real information? Is it possible to create a new credit card and Stripe account for Claw with random info?

I want to understand how to access payments or send money to the Claw bank account.

[misty vale]

guys i cannot run my claw with ollama3 8b it says too little tokens and whan i run a biogger model it takes 29gigs of ram

[hearty crane]

with newest version today .13 sub-agent or ACP session not working, tried to spawn many times but no session working?

[nimble hill]

what? there is nothing to pay for openclaw. It is the LLM provider you sign up for and add the API key / OAuth in openclaw

[inland knot]

For me, it spits out thinking/reasoning in discord chat, I used him on threads not normal chat, and he didnt listen my prompts, for example i prompted him "do X thing" and he did Y thing, my SOUL.md and AGENTS.md are strong, copped some good .md from github for instruction etc

I currently run MiniMax 2.5 from NVIDIA NIM, but the rate limits are hell

[ripe walrus]

Does anyone of you know how one can connect amazon bedrock to openclaw?

i happen to have 200$ credits on AWS, does an amazon bedrock key work well?

[inland knot]

Anyone got a stable free tier AI that doesnt spit thinking/reasoning in output (discord chat in this case) and is fairly capable of toolcalling to be reliable?

[nimble hill]

did you follow this? https://docs.openclaw.ai/providers/bedrock

[nimble hill]

nope - the cheapest one would be chatgpt go plan - USD 8 . but you have to mind what you use for as the weekly quota finishes quite fast - you can monitor in %age in the codex site

[narrow hare]

your claw can get openai-codex usage data

[inland knot]

Well i'm using have tool calling, i got a marketing agency, so i need tool calling for monitoring and optimising Ads, lead scraping etc, i think OpenRouter free tier models or NVIDIA NIM free models are not capable of such tasks, not reliable on the long run at least

[spark valley]

use stepfun

[spark valley]

that just abug not mdel related

[inland knot]

Is Step 3.5 Flash good on OpenRouter?

Cause by Stepfun i guess you mean Step 3.5

[spark valley]

no the freee one it is really good

been using it for perntesting

[finite sable]

Hey I was too sleepy to notice at the time, this won't work cause I'm tryna figure out what openclaw expects not what it receives, so like what does openclaw want my api to return when openclaw sends tool options through the ai thats all. and mitmproxy wouldn't be able to catch that cause my api doesn't yet know what to send.

So instead I wanted someone else to tell me what openclaw accepts, after perhaps testing it. so I can make my api do that.

[wary osprey]

you should probably try Minimax on AIsa... no rate limit, and the cost is pretty durable

[nimble hill]

ha ha - spend tokens to view token usage 🙂 Just use a small rest API to get the usage for free!

[narrow hare]

nah, mines a system cron job

openclaw models status and status --json

[swift willow]

It’s a good model I recommend it

[civic thistle]

Hi everyone, first message here 👋
Is there a good thread / benchmark somewhere to see which model would fit my usage with openclaw (cant run locally I guess, it's on a self-hosted NUC without GPU) ? 🙏
I'm getting crazy with Claude Haiku, Sonnet, or Gemini Flash Lite not respecting the RULES.md or MEMORY.md most of the time... Claude Opus or Gemini Pro Preview are doing better but way too $$ at the end of the day 😔

[nimble hill]

try the chatgpt go plan - USD 8 I think. and use codex-5.1-mini to last relative more before it exhausts itself and you have to wat for next week's reset

[fading junco]

Hi All, since a recent update I am no longer able to attach large MP3 files in Telegram and have the agent do something with the files. It just doesn't see the files. Does anyone know why this might be?

[latent hedge]

Anyone here using LM Studio on a big Mac Mini? Mine is a Mac Mini M4 Pro with 64 gig ram. Was curious which models people have the best experiences with for a Mac of this size, and whether or not any of you have success with specific MLX models

[nimble hill]

I think this is the best you can run Llama 3.3 70B-q4

[latent hedge]

Really? was told the new Qwen3.5 was extraordinary?

Just cant make it work.. tried a few qwen models.. they are acting weird and sometime crash, etc.

Maybe Im using the wrong one

[spark valley]

depends on the parametrs

[mild karma]

are you following the recommended parameters for qwen3.5? and are you using the latest unsloth?

[latent hedge]

Which one specifically? Is it a GGUF model or MLX model you are talking about?

[mild karma]

I'm talking about the GGUF, unsloth has a documentation page with the recommended paramater depending on the task you are doing

[latent hedge]

Did you try it yourself also?

[spark valley]

it is probably you not being able to build the correct structure and limitations i use a dummb model but it never reveals what strict or do what not permitted

[latent hedge]

Because.. all models work fine in local chat.. but curl commands from the openclaw server.. and they act weird

[mild karma]

im actively using qwen3.5 35B a3b and 27B and it works so much better with the latest unsloth

I mean its not the best but it usually works ok

[spark valley]

because they are wrapped i guess

[latent hedge]

Yeah... my goal is making it be served through LM studio, and then connect to it through tailscale from the ubuntu server I have, running the openclaw client

Everything worked on Ollama... now trying to switch to lmstudio, to mitigate the long load times

[mild karma]

You absolutely need to set the proper parameters for the model of your choice https://unsloth.ai/docs/models/qwen3.5

[latent hedge]

but I cant make it work

[mild karma]

      "lmstudio": {
        "baseUrl": "http://192.168.0.209:1234/v1",
        "apiKey": "lmstudio",
        "api": "openai-completions",
        "models": [
          {
            "id": "aessedai/qwen3.5-35b-a3b",
            "name": "Qwen 3.5 35B A3B (local)",
            "reasoning": false,
            "input": ["text", "image"],
            "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 },
            "contextWindow": 96000,
            "maxTokens": 8192
          },
          {
            "id": "unsloth-qwen3.5-27b",
            "name": "Qwen 3.5 27B (local)",
            "reasoning": false,
            "input": ["text"],
            "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 },
            "contextWindow": 96000,
            "maxTokens": 8192
          }
        ]
      },```
note the api `openai-completions` that might be what you are missing if you are woking with lmstudio

[latent hedge]

Right now, Im not even setting it up in openclaw. Just trying to get the curl command to work first

.... 8k tokens seems low also.. isnt it?

I would have assumed 16-32k? or?

[mild karma]

its just my personal preference you can use more

[lone ice]

Anyone saw this ClawedEx thing?

[latent hedge]

So odd. Check this curl:

melsen@melsenx:~$ curl http://100.87.99.87:1234/api/v1/chat
-H "Content-Type: application/json"
-d '{
"model": "unsloth/qwen3.5-35b-a3b",
"system_prompt": "You answer only in rhymes.",
"input": "What is your favorite color?"
}'
curl: (52) Empty reply from server
melsen@melsenx:~$

2026-03-14 12:57:26 [DEBUG]
[LMSAuthenticator][Client=lms-cli] Client created.
2026-03-14 12:57:26 [INFO]
[LM STUDIO SERVER] Success! HTTP server listening on port 1234
2026-03-14 12:57:26 [INFO]
2026-03-14 12:57:26 [INFO]
[LM STUDIO SERVER] Supported endpoints:
2026-03-14 12:57:26 [INFO]
[LM STUDIO SERVER] LM Studio API
2026-03-14 12:57:26 [INFO]
[LM STUDIO SERVER] -> GET http://localhost:1234/api/v1/models
2026-03-14 12:57:26 [INFO]
[LM STUDIO SERVER] -> POST http://localhost:1234/api/v1/chat
2026-03-14 12:57:26 [INFO]
[LM STUDIO SERVER] -> POST http://localhost:1234/api/v1/models/load
2026-03-14 12:57:26 [INFO]
[LM STUDIO SERVER] -> POST http://localhost:1234/api/v1/models/download
2026-03-14 12:57:26 [INFO]
[LM STUDIO SERVER] -> GET http://localhost:1234/api/v1/models/download/status:job_id
2026-03-14 12:57:26 [INFO]
[LM STUDIO SERVER] OpenAI-compatible
2026-03-14 12:57:26 [INFO]
[LM STUDIO SERVER] -> GET http://localhost:1234/v1/models
2026-03-14 12:57:26 [INFO]
[LM STUDIO SERVER] -> POST http://localhost:1234/v1/responses
2026-03-14 12:57:26 [INFO]
[LM STUDIO SERVER] -> POST http://localhost:1234/v1/chat/completions
2026-03-14 12:57:26 [INFO]
[LM STUDIO SERVER] -> POST http://localhost:1234/v1/completions
2026-03-14 12:57:26 [INFO]
[LM STUDIO SERVER] -> POST http://localhost:1234/v1/embeddings
2026-03-14 12:57:26 [INFO]
2026-03-14 12:57:26 [INFO]
[LM STUDIO SERVER] Logs are saved into /Users/allanmelsen/.lmstudio/server-logs
2026-03-14 12:57:26 [INFO]
Server started.
2026-03-14 12:57:26 [INFO]
Just-in-time model loading inactive.
2026-03-14 12:57:26 [INFO]
[LM STUDIO SERVER] Client disconnected. Stopping generation... (If the model is busy processing the prompt, it will finish first.)
2026-03-14 12:57:26 [DEBUG]
Received request: GET to /lmstudio-greeting
2026-03-14 12:57:26 [DEBUG]
[LMSAuthenticator][Client=lms-cli] Client disconnected.
2026-03-14 12:57:26 [DEBUG]
slot release: id 3 | task 0 | stop processing: n_tokens = 140, truncated = 0
srv update_slots: all slots are idle
2026-03-14 12:57:26 [DEBUG]
srv stop: cancel task, id_task = 0
srv update_slots: all slots are idle
2026-03-14 12:57:26 [INFO]
[unsloth/qwen3.5-35b-a3b] Generated response: {
"model_instance_id": "unsloth/qwen3.5-35b-a3b",
"output": [
{
"type": "reasoning",
"content": "Thinking Process:\n\n1. Analyze the Request:\n * Question: "What is your favorite color?"\n * Constraint: "You answer only in rhymes."\n * Goal: Provide a response about my (the AI's) favorite color, but strictly in a rhyming format.\n\n2. Determine the Content:\n * As an AI, I don't actually have preferences or senses like sight. However, for the sake of poetic expression, I can choose a symbolic"
}
],
"stats": {
"input_tokens": 0,
"total_output_tokens": 0,
"reasoning_output_tokens": 111,
"tokens_per_second": 0,
"time_to_first_token_seconds": 0.561
},
"response_id": "resp_e4edb40543c98f613ff1f72640a22d724c871f6bccdc0f7e"
}
2026-03-14 12:57:36 [DEBUG]
[LMSAuthenticator][Client=lms-cli] Client created.
2026-03-14 12:57:36 [INFO]
[LM STUDIO SERVER] Success! HTTP server listening on port 1234

Hrmmmm. And I set up the parameters according to the guide with temperature, etc

[mild karma]

Im currently testing another variant and was too lazy to launch the unsloth but it seems to work on my side 🤔

curl http://localhost:1234/api/v1/chat \                                      
  -H "Content-Type: application/json" \
  -d '{
    "model": "aessedai/qwen3.5-35b-a3b",
    "system_prompt": "You answer only in rhymes.",
    "input": "What is your favorite color?"
}'
{
  "model_instance_id": "aessedai/qwen3.5-35b-a3b",
  "output": [
    {
      "type": "reasoning",
      "content": "Thinking Process:\n\n1.  **Analyze the Request:**\n    *   Question: \"What is your favorite color?\"\n    *   Constraint: \"You answer only in rhymes.\"\n    *   Goal: Provide an answer about a favorite color, but strictly in a rhyming format.\n\n2.  **Determine the Content:**\n    *   As an AI, I don't actually have personal preferences or senses like sight. However, for the sake of the poem/rhyme, I can pick a color metaphorically or just state that I don't have one but choose something poetic.\n    *   Option 1: Claim a specific color (e.g., blue for the sky/ocean).\n    *   Option 2: Acknowledge AI nature (no eyes) but rhyme about it.\n    *   Option 3: Pick a \"digital\" color (like code green or screen white).\n    *   Decision: Let's go with something simple and pleasant, like blue or maybe just express the lack of preference in a rhyming way. Actually, picking a color makes ...```

[latent hedge]

very frustrating

[mild karma]

in the interface of lmstudio does it work?

[soft ledge]

how do i access the "LOCAL" browser? its running on a proxmox lxc rn and i cant tunnel in

[queen roost]

Hey guys, I have a simple problem that neither me or claude opus are able to solve.. 😢
I want to use claude code through openclaw to be able to use my sub without violating TOS, I have set it up, but I keep running into this error

ACP error (ACP_SESSION_INIT_FAILED): ACP metadata is missing for agent:main:acp:a3fbd45e-4cac-425a-aae6-4e98f0812590. Recreate this ACP session with /acp spawn and rebind the thread.
next: If this session is stale, recreate it with `/acp spawn` and rebind the thread.
[end_turn]

any idea on what is going wrong?

[dapper basin]

Hi, is there a current TestFlight invite for the OpenClaw iOS app / iOS node app?

[rugged bramble]

Thank you for your interest - there's currently no Testflight program, if you want to run OpenClaw on your phone, please compile the code in the repo (Claude Code, Codex or Gemini can help you), and feel free to submit PRs if you find issues. We'll let the community know once Testflight becomes available
-# Your message was reposted above for the sake of conversation.

[fluid topaz]

Anyone let me pick their brains for 30 mins via zoom? $50.

[mighty rampart]

I meant to run it through mitmproxy using an existing endpoint and model so you can see what it is doing, so you can duplicate it. For example, hook it up to OpenRouter. But also, it appears to be the OpenAI Responses API and there is published documentation.

[finite sable]

it seems like my api is doing it properly but openclaw isn't recognizing it still, im soo confused,

[hushed jackal]

Hi. Im wondering if theres a way where I can toy with a certain group of settings or something to get my openclaw to be as productive as possible. Right now it works quite well, but it comes up with alot of fluff and planning where I have to repeatedly tell it to start a task instead of it just doing said task. Would anyone have any guidance?

[red lotus]

Have you tried tweaking your SOUL.md to cut the fluff? Something like "skip the planning summaries, just do the task" goes a long way.

[inland knot]

Minimax 2.5 NVIDIA NIM or Step 3.5 Flash OpenRouter for free tier AI? Maybe it is a stupid question, but i wanna know

[unique palm]

Hi there! Is there an ios app?

[flat bison]

Support Request: /gog mail list returns empty results

[coral timber]

hi guys ! using OC for 2 weeks now , as side hustle . Virgin Discord also 🙂 I have an ideea and I don't know how to submit it to OpenClaw . I belive that , chatting in main session should be automatically done with main agent . Menwhile if "@"agent is introduced as a drop down list with all your agents you will automatically talk to your subagents . What do you think guys ? waiting for your feedback . Cristian

[tall mauve]

I'm looking for anyone who has had experience experimenting with different LLMs - Currently I'm using qwen coder as my main agent and mini-max2.5 as my backup, I keep reading that Claude is superior but I haven't tried it as I can't find a cheap way of using it, any recommendations?

[spark valley]

first may i ask what you run it for (as side hustle so i can get some monetisation out of it) then, for your specific problem did you try to talk to your agent about it it can give you a good report on the issue as now it kinda of scrambled , you can post at github issues or by GitHub Discussions (Ideas Category)

[spark valley]

you could signup for pro then use the oauth method but risk ban

it got some limitations but it is the best methd for tryiing out

[summer glade]

I am getting this error and i tried allot things nothing works - npm.cmd : npm error code 128
At line:474 char:22

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

•             ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  

  • CategoryInfo : NotSpecified: (npm error code 128:String) [], RemoteException
  • FullyQualifiedErrorId : NativeCommandError

[spark valley]

tryin on windows?

[tall mauve]

Yeah I read that somewhere too, They're actively cracking down on that apparently

I'm at a point where every corner I turn seems to be a paywall

[spark valley]

well that you can do for now before they catch on and that is a cheap triout

well there openrouter

[tall mauve]

For real, thanks

[spark valley]

openrouter got some seriousely dangerous models for free

[indigo goblet]

Most likely rumors, Peter has said explicitly OpenAI is OK with it, doubt it is real.

[spark valley]

no we talking bout anthropic

[indigo goblet]

ohh, apologize, I read that too fast without looking at the initial replies.

[winged relic]

any one seeing tool call issues? I am receiving the tool call command in TUI like the following:

Great! Let me test if tools work now:

<tool_call>
{"name": "exec", "arguments": {"command": "echo hello", "timeout": 5000}}
</tool_call>

[dim harbor]

what model ? latest openclaw? there were some kimi fixes for this. other thing cold be you are in messaging mode that doesnt have exec permissions at all

or model doesnt have tool call functionality

[winged relic]

OpenClaw 2026.3.13 (61d171a), opus-4.6

it has been working with the same config and then suddently the tool calls are not working anymore.

[spark valley]

probably tools issue

go into the openclaw config

look for tools set that to fulll

[unborn igloo]

Nothing free on openrouter

[winged relic]

it is set to full

"tools": {
"profile": "coding",
"exec": {
"security": "full",
"ask": "off"
},

[grim summit]

Anyone here using MyClaw? I want to confirm understanding: My guess is that the base subscription just gets us a server. Then, we need to provide API keys if we want to also run different models. Is that right?

[spark valley]

tell that to my 1b usage agent

for 0.00001 dollars

[spark valley]

no the tools profile must be set to full do not hardcore the exect param and test after restarting the gaateway

[coral timber]

basically , I am trying to fully autoamte some repetitive task . Everythnig works fine after defining skills , but it gets kind of pricy programming the cron jobs , especially if you have a lot of tasks . I have tried to make a new agent with a cheaper LLM anthropic model , and I wanted to gave this cheaper agent those tasks . Right now I prefer to swich OpenClaw LLM model from Sonnet 4.6 to Haiku . Make Cron Jobs with Haiku LLM , and than switch back to Sonnet 4.6 . What do you guys think ? makes sense ? I have to constantly look into logs ( anthropic / openrouter ) to actually see the cost and compare .

[spark valley]

it is just hosting obviously which you can get it cheaper and basiclly you pay for the setup which is the easiest

[spark valley]

well that is correctly set you can give the agent you want to do the tasks (probably do not need high end models) and just make that agent do the cron jobs i guess that what you said you were doing . so you just specify the model you want for that agent and yyour good to go if your issue is the main agent doing the cron jobs that's somehing else,

[unborn igloo]

Which model?

[spark valley]

step 3.5 flash

the 50 request daily doesnt aapply to it

ive been using it the whole day with average of 800 request per day

[coral timber]

now all the tasks are asigned to an agent that runs on a cheaper LLM . when I am doing the cron jobs I firstly switch to the cheaper LLM , asign the job . good to go . Problem solved . Hopefully . Than switch back to a powerful LLM . to conclude my problem was how to be more cost effective in running and scheduling multiple tasks .

[spark valley]

and if you want to use some hard solid just dropped model with a fricking 1m in coontext you can use openrouter/hunter-alpha for damn 0 pennies

[spark valley]

why you can set the needed model to be used for that specific agent always no need to re change

as it gonna e exhausting

here is a peek on how it should be in the cconfig:: "id": "hacker2-nemo",
"name": "Kali Hackster Nemo",
"workspace": "/home/kali/.openclaw/workspace-hacker-nemo",
"agentDir": "/home/kali/.openclaw/agents/hacker2-nemo/agent",
"model": {
"primary": "openrouter/openrouter/hunter-alpha"
},
"sandbox": {
"mode": "off"
},
"tools": {
"profile": "full"
}

let say this is your secondary agent that runs the taasks now it has this cheaper ie model specifid byut the main agent is till using it opus opr whatever u specify for it

[kind crest]

[BUG] sessions_spawn: Ollama model prefixed with 'google/'

[unborn igloo]

The 50 limit doesn't apply only to step model? Or are there other openrouter models that bypass that limitation?

I notice step is a programming model - are you using it to drive your openclaw machine?

[spark valley]

yes it is really great for 0 dollar

there is alot that the limit dont apply to theem

like openrouter/hunter-alpha it got 1m context

and the rate do not apply and if you just buy one time 10 dollars credtis you get a 1000 daily limit quota

[spark valley]

i also use openrouter/hunter-alpha for hackig agent and it is doing super great

[rose halo]

First time setting oclaw up here in VPS, having issues accessing the web UI, summary: I accidentally broke the gateway connection trying to change the default port, then reinstalled now. (note: I usually run all my apps behind https+domain+cloudflare proxy so I just want to set up simple ip:port remote access for the web UI to start)

I don't really want a ssh tunnel-- I just want to pick a custom port and open it, requiring authentication in the UI ofc) -

I setup whatsapp integration succesfully - just the web UI and accessing this gateway , configuring 'config.bind' to local and opening the port in ufw, I managed to see the onboarding of the web UI. Then I tried changing the default port and broke the connection and couldn't fixx even tweaking in oclaw's config.json.

So I reinstalled. Now I'm trying to setup this web ui for remote access with auth again. lol

Doctor didn't really remedy my issues. I think doctor also got confused, oclaw status : https://i.imgur.com/DunSWw2.png

│ Dashboard       │ http://127.0.0.1:18789/               
│ OS              │ linux 6.8.0-101-generic (x64) · node 22.22.0       
│ Tailscale       │ off                                 
│ Channel         │ stable (default)                     
│ Update          │ pnpm · npm latest 2026.3.13                    
│ Gateway         │ local · ws://127.0.0.1:18789 (local loopback) · unreachable (connect failed: connect ECONNREFUSED 127.0.0.1:18789)   
│ Gateway service │ systemd installed · enabled · running (pid 843361, state active)        
│ Node service    │ systemd not installed                
│ Agents          │ 1 · no bootstrap files · sessions 0 · default main active unknown      
│ Memory          │ 0 files · 0 chunks · dirty · sources memory · plugin memory-core · vector unknown · fts ready · cache on (0)   
│ Sessions        │ 0 active · default claude-opus-4-6 (200k ctx) · ~/.openclaw/agents/main/sessions/sessions.json     

From this doc :

The Control UI is enabled by default when assets are present (dist/control-ui). You can control it via config:
{
  gateway: {
    controlUi: { enabled: true, basePath: "/openclaw" }, // basePath optional
  },
}

I have no idea if 'assets are present'. I just installed the app, how would I know this / how do I make the dist assets be present?

I have a feeling if I change inside my config.json the gateway auth mode from: token
to

  gateway: {
    bind: "loopback",
    tailscale: { mode: "funnel" },
    auth: { mode: "password" }, // or OPENCLAW_GATEWAY_PASSWORD
  },
}

everything might break

re-did onboarding manual setup and set
◇ Gateway WebSocket URL
│ wss://0.0.0.0:18789

...not sure if I'm doing this right for remote access with auth

[spark valley]

look try tailscale is the best option

for remote secure access

[rose halo]

yes now the oclaw config still tries to connect to the old url for osme reason

fter 60s waiting for health checks.
root@srv556089 ~ [1]# oc gateway status

🦞 OpenClaw 2026.3.13 (61d171a) — I speak fluent bash, mild sarcasm, and aggressive tab-completion energy.

│
◇  
Service: systemd (enabled)
File logs: /tmp/openclaw/openclaw-2026-03-14.log
Command: /usr/bin/node /usr/lib/node_modules/openclaw/dist/index.js gateway --port 18789
Service file: ~/.config/systemd/user/openclaw-gateway.service
Service env: OPENCLAW_GATEWAY_PORT=18789

Config (cli): ~/.openclaw/openclaw.json
Config (service): ~/.openclaw/openclaw.json

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

Runtime: running (pid 876690, state active, sub running, last exit 0, reason 0)
Warm-up: launch agents can take a few seconds. Try again shortly.
RPC probe: failed
RPC target: ws://127.0.0.1:18789
  gateway closed (1006 abnormal closure (no close frame)): no close reason
  Gateway target: ws://127.0.0.1:18789
  Source: cli --url
  Config: /root/.openclaw/openclaw.json

Gateway port 18789 is not listening (service appears running).
Logs: journalctl --user -u openclaw-gateway.service -n 200 --no-pager

Troubles: run openclaw status
Troubleshooting: https://docs.openclaw.ai/troubleshooting
root@srv556089 ~# 

[rose halo]

I will have to setup tailscale

:/

I just wanted to open this to a port, simple as

[spark valley]

well you said you can access it locally

so it is only a forwardin issue

[rose halo]

The issue is, no other app I've worked with before exposed an app on websocket url so I don't rly know what I'm debugging here

Look

[spark valley]

do a full reset wiith openclaw reset then , openclaw onboard

and set the gateway to lan

so it is listenin on lan then you forward to internet

[rose halo]

See if you can help me find where I should fix this:

I setup remote access with pass.

◇  What do you want to set up?
│  Remote gateway (info-only)
│
◇  Discovery ──────────────────────────────────────────────────────────╮
│                                                                      │
│  Bonjour discovery requires dns-sd (macOS) or avahi-browse (Linux).  │
│  Docs: https://docs.openclaw.ai/gateway/discovery                    │
│                                                                      │
├──────────────────────────────────────────────────────────────────────╯
│
◇  Gateway WebSocket URL
│  wss://0.0.0.0:18789
│
◇  Gateway auth
│  Password

---

Now I get this err (it's trying to access the old url ws 127.0.0.1 instead of wss 0.0.0.0)

oc gateway status

🦞 OpenClaw 2026.3.13 (61d171a) — I speak fluent bash, mild sarcasm, and aggressive tab-completion energy.

│
◇  
Service: systemd (enabled)
File logs: /tmp/openclaw/openclaw-2026-03-14.log
Command: /usr/bin/node /usr/lib/node_modules/openclaw/dist/index.js gateway --port 18789
Service file: ~/.config/systemd/user/openclaw-gateway.service
Service env: OPENCLAW_GATEWAY_PORT=18789

Config (cli): ~/.openclaw/openclaw.json
Config (service): ~/.openclaw/openclaw.json

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

Runtime: running (pid 876690, state active, sub running, last exit 0, reason 0)
Warm-up: launch agents can take a few seconds. Try again shortly.
RPC probe: failed
RPC target: ws://127.0.0.1:18789
  gateway closed (1006 abnormal closure (no close frame)): no close reason
  Gateway target: ws://127.0.0.1:18789
  Source: cli --url
  Config: /root/.openclaw/openclaw.json

Gateway port 18789 is not listening (service appears running).
Logs: journalctl --user -u openclaw-gateway.service -n 200 --no-pager

Where do I change this? in config json?

[rose halo]

Yea I did that, I'll try again! ty

But I went to change the port, via cli comand, and broke it lol.

And gateway restart failed. now. so yea lemme just reinstall for the 2nd time

[rose halo]

I might also be interested in using a ssh tunnel too due to the reason: I want to test OClaw agent using the chrome-devtools-mcp to control my currently open chrome

So I guess with ssh tunnel thats possible for the agent (in my remote VPS) to access my local browser opened port right through ssh tunnel

#releases message

I already use chrome-devtools-mcp in my local opencode it can open a tab in the chrome browser i'm using

so later I wanna test oclaw with it too wil probably perform better

I'll try it later, I wasn't able to connect the ssh tunnel yet either :c

[spark valley]

well yeah that s better approach for faster toubleshooting

[rose halo]

I have a lot of troubles to shoot here

[rose halo]

just did full uninstall and cleared data and reinstalled using script. (except the cli apparently)

Installed systemd service: /root/.config/systemd/user/openclaw-gateway.service
Updated ~/.openclaw/openclaw.json
│
└  Doctor complete.

· Updating plugins
No npm-installed plugins to update.
· Gateway daemon detected; restarting
! Gateway restart failed; try: openclaw daemon restart

🦞 OpenClaw 2026.3.13 (61d171a) — If you can describe it, I can probably automate it—or at least make it funnier.

Dashboard URL: http://127.0.0.1:18789/#token=ecensored
Copy to clipboard unavailable.
No GUI detected. Open from your computer:
ssh -N -L 18789:127.0.0.1:18789 root@85.209.92.214
Then open:
http://localhost:18789/
http://localhost:18789/#token=exxxxxxxxcensored
Docs:
https://docs.openclaw.ai/gateway/remote
https://docs.openclaw.ai/web/control-ui

FAQ: https://docs.openclaw.ai/start/faq
root@srv556089 ~# openclaw daemon restart

🦞 OpenClaw 2026.3.13 (61d171a) — I'm the reason your shell history looks like a hacker-movie montage.

Restarted systemd service: openclaw-gateway.service
Timed out after 60s waiting for gateway port 18789 to become healthy.
Gateway process is running but port 18789 is still free (startup hang/crash loop or very slow VM startup).
Service runtime: status=running, state=active, pid=900871, lastExit=0
Gateway port 18789 status: free.
Gateway restart timed out after 60s waiting for health checks.     <---------------------------------------- what 
⏎                            

gateway restart just doesnt work

'fresh' reinstall um

I mean daemon restart

oclaw doctor please rescue me I am dying

after learning to read I probably found what I was missing

│
◇  Gateway ──────────────────────────────────────────────────────────╮
│                                                                    │
│  gateway.mode is unset; gateway start will be blocked.             │
│  Fix: run openclaw configure and set Gateway mode (local/remote).  │
│  Or set directly: openclaw config set gateway.mode local    ```

[mighty rampart]

I don't really want a ssh tunnel
me neither given that it's all running inside my house behind two firewalls, but alas, with the latest update, this seems to be required now. ssh tunnel it is.

[rose halo]

for configuring public remote access - I think it's just not explicitly documented in a very detailed manner so clueless people don't follow the doc and expose their apps publicly or something

But I can setup it using 'lan' and it will be accessible locally. then you aallow the port in your firewall. and you can acccess remotely the UI

But you can do both tunnel connection and expose the app publicly with auth i think im still configuring this

[mighty rampart]

Did you get it to work? That is how I had it set up before today. Claude worked on it for a while and gave up. And it doesn't matter that much to me so I gave in.

[rose halo]

ill @ you here later when I complete setup

what do you want to do?

expose web ui remotely with password login? or something else

[mighty rampart]

just a token. it's not accessible in public just in my house. it used to work just fine with the port mapped and exposed and http://<my.ip.address>:9094?token=abc123

[rose halo]

run openclaw configure and in Gateway bind mode should be selected LAN (all interfaces) for local access in ur wifi

did u change the config and it stopped working?

[spark valley]

remove that message now

it conatins your api keey

!!!!!

[winged relic]

hmm need to rotate that now 🙁

[spark valley]

yup

better safe than sorry

repaste

the config

without th api (:

[mighty rampart]

I upgraded from I think 2026.3.2 to .12 and did not touch the config -- yes it's "bind": "lan" and "dangerouslyDisableDeviceAuth": true and had allowedOrigins -- this has seemed to stop working. It's fine. ssh tunnel is a minor annoyance. just curious what it was if you get it working.

[winged relic]

"tools": {
"profile": "full",
"exec": {
"security": "full",
"ask": "off"
},
"web": {
"search": {
"apiKey": "xxxxxx"
}
},
"allow": [
"message"
]
},

[unborn igloo]

great call on the free OR models - thanks. got one set up and it's working beautifully. nice to be able to set guardrails around it too within OR platform

[spark valley]

yes good

[spark valley]

yes anytime , another top way for me to find latest models as the alpha hunter that was two days released , you simply ask google gemiini search best top free models and you browse somethimes you fin the rate limited oes like the nvidea nemo super

but for reference some models in open routeer are always free but dont have the (freee) so make sure always to see input outpput price

you will see 0 0

[wraith seal]

trinity is amazing

[spark valley]

yup that is another top one , depends on your needs

[unborn igloo]

🙌

[unborn igloo]

chat, not coding

[unborn igloo]

are any of the free OR models decent at all for managing openclaw config? i find unless you're running a flagship model like opus or gpt5 codex, it struggles to get the configurations right

[rose halo]

for any model really you should provide the documentation .txt file ideally for any kind of api setup or config etc

there must be a llms.txt file in the docs for full context somewhere. usually its a file that links all documentation pages.

[spark valley]

try stepfun or hunter lpaha with skill of openclaw docs so

always it points to that skill to manage the configs

[unborn igloo]

yea, true - pointing to the docs is important, but you still need its reasoning to be good

[unborn igloo]

unfortunately i don't think openclaw has a single llms.txt file

[spark valley]

yea the reasoning is quite good actually . if you are willing to spend one time 10 dollars to get a 1000 daily requests and usse Nemotron 3 Super

[winged relic]

it's working now! Thanks!.

[spark valley]

no worries anytime

[unborn igloo]

i love the warning, "Note: For the free endpoint, all prompts and output are logged to improve the provider's model and its product and services. Please do not upload any personal, confidential, or otherwise sensitive information. This is a trial use only. Do not use for production or business-critical systems." :p

where do you see the requirement to spend 10 to get 1000 daily requests?

[spark valley]

To get a higher rate limit on OpenRouter, the most effective method is to maintain a minimum credit balance. OpenRouter uses a "boost" system where your daily quota for free models increases significantly once you fund your account.
OpenRouter
OpenRouter
+1
How to Increase Your Limits
Purchase $10 in Credits: This is the primary threshold for a massive limit jump.
Standard Free Tier: If you have purchased less than $10 in credits, you are restricted to 50 requests per day on :free models.
Boosted Free Tier: Once you purchase at least $10 in credits, your daily limit for :free models increases to 1,000 requests per day.

note you should miantain the 10 dollars not to spend it as it is you leveraage

[cyan heart]

anyone here have some simple ways to reduce token useage more permenantly via config? long context windows are burning up a lot of tokens for simple tasks. I'm trying to run /new from time to time to get a new session for the time being.

[rose halo]

why is setting up brew so terrible? its required to install a lot of openclaw skills/packages

[winged relic]

TUI is ok. but it looks like discord sessions cannot run commands, but claiming that it's running commands.

hmm actually TUI does not work either.

should add other tools in the config?

ok. removing the allow section worked

[spark valley]

yes it is noot the only model for me im using stepfun as my main ad it is doing great job setting up agents and their soles and apis keys i dont find any issues in technicl side

[spark valley]

yep maybe that was the issue holding, but what it does it just strictly only allow text messages . aw thats it it only allow messages that why , well good to hear u fixed it .

anoter workearound i doo is every agent i give it it own isolated account api key i have several openrouter accounts so the rate is split

not overwhelmin one api

[unborn igloo]

yea good practice - all better for usage tracking

[spark valley]

yea it was hard to setup but got it to work

[rose halo]

didn't know that, interesting

So my current issue is I can't get 'brew' to be a usable command in my VPS, and oclaw requires that for installing like 6 skills. Brew refuses to install due to having sudo permissions, and in a fresh user, it does not have folder permissions to install wherever it is trying to.. what the hell. Guess I'll just not have the skills that install with brew

[spark valley]

what os

u got

u must have the sudo password of the user

unless the vps comes with preinstalled users and no root access allowed

[rose halo]

I use this 'wrongly' as root user to setup everything. So I created a 'brewuser' rn to install brew. guess i should setup openclow NOT as root user too.

It's a VPS ubuntu 24

I mean I use it 'wrongly' as root user mainly because it is like 4 years old from when I started learning selfhosting stuff lol

But all my other apps are safely living in docker

[spark valley]

no just install brew seperately : sudo apt update
sudo apt install build-essential procps curl file git -y
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"

just bear with me do you have the user password for sudo?

[rose halo]

um yep I ssh with root user

i have sudo but dont need to cuz i run things as root so its sudo auto

[spark valley]

okay so do the above to install brew

[rose halo]

above worked to install in the new user I made, brewuser. but my openclaw was setup in root user lol.

gyess i need to install openclaw inside brewuser.

ok so, lets go to the 3rd reinstall

[spark valley]

yep

[rose halo]

now I'll backup the configs and import.

[spark valley]

you need to install it inside su - brewuser

[spark valley]

lol last day was my 9th

clean setup

that's how we learn

[rose halo]

[spark valley]

i've been using the hunter alpha i just hacked into a local server i was struggling to do so. it cracked it within 20 mins ai agents are the future ong... I will continue to use this model for complex tasks and give a feedback for the guys here that need free strong models but they pay with their prompts.

[rose halo]

whats that

AI pentester?

[spark valley]

yea

no it is a model

the agent is on openclaw

free one on openrouter

https://openrouter.ai/openrouter/hunter-alpha

[dense field]

Is there a way to rewind to a previous message? Sometimes the openclaw messes up and I want it to undo changes it made to my files

With claude code you can /rewind to previous conversation and/or code

[mint idol]

Hi! I need help with the Mac app on remote server. I’ve grant camera permissions but it only shows camera.list and not camera.snap

[shadow rover]

there's no rewind command yet, so your best bet is to use Git to checkpoint and roll back your files if it messes up

[civic bison]

Iv gotta be doing something wrong I keep running into “api limit reached try again later” and I haven’t been able to have it to anything intensive can someone help im new to this

If set up security features and parameters to make sure nothing bad happens but I’m failing to under stand some things and how the magic happens

[spark valley]

openrouter?

[civic bison]

Uhhh idk

[spark valley]

which api u using which provideer

[next cave]

hi, I'm trying to get started with openclaw in docker on ubuntu: checkout the git repo (main), run docker-setup.sh but keep getting telegram plugin not found.
can anyone recommend working/safe tags?

[violet crystal]

Does GPT just not work in OpenClaw anymore? Only my anthropic models use tools and reasoning anymore. OpenAI models literally just reply and hallucinate right now with the recent updates.

[velvet terrace]

id been using codex 5.3 via oauth but it just broke today. anyone else experiecing this? Worried they may no longer be allowing its use or something similar to anthropic

[brave swallow]

Hey so I tried using open claw a month or so ago right when it came out. It was pretty cool besides the fact it was really expensive, and it broke several times until I couldn’t fix it any more and I had to stop using it. TBH I do miss the Ai since it did have a little bit of a personality. But I want to revisit using it on a fresh Mac mini later and really give it my best shot. However the only problem is idk if it’s something that would benefit me and what I do for side projects or even work. Can someone help me figure out if I should try to do this all over again or just leave it behind? In the past I used a VPS so it was a lot harder to fix or do things with the Ai since it I am not a technical person.

Can someone help me with this

[next cave]

I wish I could get it running, lol

[brave swallow]

And if I do decide to do it again, what is the best way to set up the Ai so that it works the best it can without deleting my bank account. I know i need to make it secure but also make it token efficient

I also wanted to use it on slack before and that might have been a mistake. Telegram is fine now

[ember tide]

Without chadhardware need to pay for api /max plans so seems to be a expensive toy at the moment

[brave swallow]

Hmmmm I don’t have a lot of money to waste on this without a way to get a return

But how would I maybe make it worth it?

[ember tide]

Same, I've been trying to get old hardware to show some positive results

[brave swallow]

I can tell you vaguely what I do for work and my side projects. Maybe then someone could tell me if open claw is worth it for me

[brave swallow]

I might ask family for a Mac mini for my birthday sooooo

[ember tide]

Happy birthday 😅

[brave swallow]

Idk if I will but I can ask

[violet crystal]

Curious how yours is broken though?

Mine literally just can't run tools or think. It just replies. And it's super annoying because it hallucinates and it has no context of anything.

[ember tide]

Yes the current choice is : make api companies rich, by paying them 20$/ 400$ per month or buy ultra expensive hardware, I don't like both

Unless we bought RAM back when it was cheap

[velvet terrace]

im connected via openai codex. i selected to sign into my openai account via Oauth, is that what ur using?

its worked for me prior but something broke today connecting to oauth. Its the Codex 5.3* model, is that what your using as well?

EDIT: typo on model name prior. i've been using Codex 5.3

[civic bison]

Ollama cloud

[velvet terrace]

Because initially you saying it can't use tools or anything, made me assume maybe u were using a less capable model or something

[dim harbor]

primary "openai-codex/gpt-5.4"

thisone is working fine for me. you got claw updated to latest?

[velvet terrace]

im on v2026.3.7

atm

[cloud patrol]

Has anyone implemented OpenAI's symphony architecture in an Openclaw environment

Using linear

[velvet terrace]

ill update maybe thats it

[rose halo]

small dumb model

use a bigger one and needs to have 'tool calling' capabilities

[opaque steeple]

Hi I am trying to post a question in help but always get an error message that this type of message is not allowed?

[wispy willow]

Same issue; ended up reverting back to a previous version

[velvet terrace]

also mine only says 5.3 for codex. I don't appear to be able to connect to 5.4?

[dim harbor]

you do need to update openclaw to see 5.4 was added recently ,or you on latest already ? it wasnt showing on older versions

[worn folio]

hi guys. I cant get this to work
=> ERROR [internal] load metadata for docker.io/library/node:24-bookworm@sha256:3a09aa6354567619221ef6c45a5051b 22.1s

if i do
docker pull node:24-bookworm
24-bookworm: Pulling from library/node
failed to copy: httpReadSeeker: failed open: failed to do request: Get "https://docker-images-prod.6aa30f8b08e16409b46e0173d6de2f56.r2.cloudflarestorage.com/registry-v2/docker/registry/v2/blobs/sha256/e9/e925c020d6d8277a3adf58cff70adc73b3535b43c8daf69d77147ce4bee167ff/data?X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Credential=f1baa2dd9b876aeb89efebbfc9e5d5f4%2F20260314%2Fauto%2Fs3%2Faws4_request&X-Amz-Date=20260314T183456Z&X-Amz-Expires=1200&X-Amz-SignedHeaders=host&X-Amz-Signature=50fce929f72212b915d0e3a84a989c5aad5f1462c5cc6f813b97579ab8583eb8": dialing docker-images-prod.6aa30f8b08e16409b46e0173d6de2f56.r2.cloudflarestorage.com:443 container via direct connection because Docker Desktop has no HTTPS proxy: connecting to docker-images-prod.6aa30f8b08e16409b46e0173d6de2f56.r2.cloudflarestorage.com:443: dial tcp 172.64.66.1:443: connectex: A connection attempt failed because the connected party did not properly respond after a period of time, or established connection failed because connected host has failed to respond.

sry for mountain, cant paste pics

the compose app says is no longer running on docker

[velvet terrace]

well I just updated, my prior issue is now resolved but yea im still on gpt 5.3 codex

and i even reran the oauth and selected 5.4 and it just defaulted back to 5.3

[violet crystal]

🚨 Here is the GPT issue I am facing if anyone else has experienced the same thing — issues with model output quality using OAuth instead of API keys.

this is what my claw said:

" 🔍 The Real Answer

Both GitHub issues (#28754 and #33587) are still OPEN. Neither
3.12 nor 3.13 fixes this.

Here's what's actually happening at the code level:

1. OpenClaw sends your Codex requests to chatgpt.com/backend-api
   (the consumer ChatGPT API), not api.openai.com. Even though
   models.json says api.openai.com, the runtime overides it at line
   35367 of the bundle.
2. Tools ARE being sent in the request payload (I verified
   suportsModelTools would return true for Codex). But the
   chatgpt.com/backend-api endpoint silently ignores them — it
   accepts the tools parameter without error but the model never
   emits tool_use blocks back.
3. This is why it "doesn't think" — Codex on that API surface just
   generates instant conversational text. It's not that it's broken
   per se — it's that this API path was designed for ChatGPT web, not
   for agentic tool-calling workflows.
   "

[dim harbor]

Keep in mind session remembers selection if it’s same one from before the update, try /new

[velvet terrace]

hmm yea.. its not even listed is the thing. attempted /new

maybe its not avail to everyone yet?

[dim harbor]

its possible your model list is filtered to only allowed ones, and you havent added it yet. you can try disabling that filter to see all models and see if you can see it then. just ask open claw about it

[spark valley]

probably u hit the rate limit

[civic bison]

How do I even change my model

@spark valley

[spark valley]

in openclaw?

do this

[civic bison]

Yes

[spark valley]

openclaw configure

it will start a wizard

go to model

then choose the provider and the model

[wide cliff]

Hi- I've been using O/C ~3 months and my API costs are ~$1500/wk (I'm ok w/that based on what it's doing) and thats w/moving non-complex tasks to lesser models, etc. I can't find a model that comes close to sonnet for my reliably doing coding and other complex work. I'm hoping someone here can point me to a discussion/resource/similar for people that are also facing this issue. Or even somewhere w/heavy discussion about optimizing for API costs or model comparisons - specifically for open claw. Thx

[celest vortex]

you're just doing straight api?

honestly I could see gpt sub being a lot cheaper than that, and probably get you somewhere close to that api cost in usage

or anthropic sub but obviously at your own risk haha

though I will say I've never had any issues just a pain to setup

[spark valley]

keeps like that the internet is good : 'sendChatAction' failed!
05:05:27 error telegram/api {"subsystem":"telegram/api"} telegram sendChatAction failed: Network request for 'sendChatAction' failed!
05:05:30 error telegram/api {"subsystem":"telegram/api"} telegram sendChatAction failed: Network request for 'sendChatAction' failed!
05:05:33 error telegram/api {"subsystem":"telegram/api"} telegram sendChatAction failed: Network request for 'sendChatAction' failed!
05:05:36 error telegram/api {"subsystem":"telegram/api"} telegram sendChatAction failed: Network request for 'sendChatAction' failed!
05:05:39 error telegram/api {"subsystem":"telegram/api"} telegram sendChatAction failed: Network request for 'sendChatAction' failed!

[wide cliff]

its a complex, 7-round system that API hits 6 separate frontier models w/very large prompts and then collaborates between each of them between each round. That stuff is easy and those APIs costs are not the issue-- its that i have a constant queue of system improvements that my QA agent builds and i need sonnet to be able to do the coding on the very complex system - constantly-- other models break things in the overall system when they think about and then code and test and document their changes.

The sub path won't work. I need full API: 100K+ token prompts, 6 parallel model calls per round, no rate limit headroom. A Plus sub would hit its cap in the first few minutes of a session.

[unique hazel]

Anyone also faces a problem where agents sometimes reply only when being mentioned but requireMention is off?

The problem is, it happens only sometimes. That means the routing and config shouldnt be wrong.

[cyan quartz]

my gateway refused to stay up unless I set gateway.mode=local. Is that the proper config even if I'm running it on VPS?

[karmic current]

Yes that's fine, unless you want to connect with it via a different machine, I mean with the gateway. Most people SSH into their VPS and then manage the gateway via CLI.

[tardy dune]

Hey everyone. AI founder operating out of India. Just scrapped my local model architecture today because the API routing was bleeding time, pivoting to a leaner cloud setup for my B2B outbound agents. Glad to find a room with actual builders

[covert gazelle]

Anyone using computed skills? The !command syntax in SKILL.md lets you run a script before the agent sees the instructions.
The idea: you already wrote the instructions. Python just decides which ones to show based on what's actually happening. Changed auth files? Show the security review checklist.
Changed config files? Show the config audit checklist. Nothing changed? Show nothing.
Why leave that decision to the LLM when a 10-line Python script gets it right every time?

[grand basin]

Hey does anyone know why cerebras is just giving me error 404 no body?

a gooogle is saying its because its sending the developer tag and not system, where can i fix this?

[indigo goblet]

can you expand on that?

[covert gazelle]

Sure. A normal skill has static instructions in SKILL.md:

---

name: code-review
description: Review code changes

Check for bugs, security issues, and consistency.

The agent gets that same text whether you changed 2 auth files or 40 config files.

With a computed skill, SKILL.md just calls a script:

---

name: code-review
description: Review code changes

!python3 scripts/generate.py

The script runs before the agent sees anything. It reads git diff, counts files, checks what changed, and prints different instructions:

• 2 auth files changed → "Security review: check for hardcoded secrets, validate input sanitization"
• 40 config files → "Config audit: check for breaking changes, validate syntax"
• Tests added to untested project → "Test quality review: check edge cases, verify assertions"

The agent only sees the final markdown. It never knows a script was involved.

Three working examples in the repo: https://github.com/Joncik91/computed-skills

I also opened a feature request specifically for OpenClaw: https://github.com/openclaw/openclaw/issues/46565

[south ether]

Hi everyone! Looking for some help here. I've been trying to run local models with Openclaw for a while now, but they just won't work for me.

I typically get a model not allowed error, or every time I send a message I never get a response back (not in the sense of the message being infinitely stuck, in the sense where the web UI bugs out and doesn't even show typing.)

Is this something I can fix, or is it a hardware issue? I have an Nvidia GeForce RTX 4060 with 16 GB ram and my CPU is a 12th gen Intel core i5-2400F. Is there any hope, or should I just stick to cloud models? If there is hope, any model recommendations?

[covert gazelle]

Local models work fine with OpenClaw — the issue is usually config, not hardware. Your 4060 can handle 7-8B parameter models easily.
For the "model not allowed" error: check your openclaw.json model config. You need a provider set up (like ollama or lmstudio) and the model name needs to match exactly what your local provider reports.

For no response in the web UI: check the OpenClaw logs (journalctl -u openclaw -f or the log file in /tmp/openclaw/). Usually it's a connection timeout or wrong API endpoint. What local provider are you using — Ollama, LM Studio, or something else? And what model are you trying to run?
With 16GB VRAM on the 4060 I'd start with something like qwen2.5-7b or llama3.1-8b through Ollama. They'll run comfortably and actually work well as agents and then scale up if there is headroom

[south ether]

I'm using Ollama with Qwen 3.5-uncensored, tried both the 9b and the 4b but neither is cooperating with me!

I'll take a look at the openclaw logs, with the new update I tried using the onboarder and it picked up on the models I have just fine, so I thought they'd work... Yet still getting errors

[iron shadow]

Guys it asking for api key when I already give it

[grand basin]

Mine doens't ask for an API key but then it fails without it 🙂

and i dont know where it goes

[iron shadow]

then put it in maybe it will work

[grand basin]

when I setup the model it doesn't prompt, and when i manually add it to the auth-profiles it doesn't take

[forest galleon]

Hi guys, I am new here and I plan to install openclaw and maybe run it locally or on vps.
My first question is: what is the best option vps or locally I have 32gb ram?
And 2nd question: what is the best model to use it with?

[umbral kindle]

so don't put it on your own device for privacy concern,put it on new device or aws is a good choice

chatgpt is a good choice if your are rich

minimax is much more cheaper but a little less intelegent than gpt

[brave swallow]

So how good is open claw at making books that are designed nicely because I have a fully written cook book but idk if it’s going to do it well. Idk I have some small projects for it to work on. And maybe it could help with my main job. But Idk what would be the main thing I use it for if I was use it again

I wonder if I should even try using it again

[unborn igloo]

Why can't you push the QA through the API as well? Or is that the part you're trying to save costs on?

[brave swallow]

I know someone here said that it’s a very expensive test hobby

Is it even worth it if I don’t have a ton of ideas for using it right now?

[odd iron]

What’s the best local model for a MacBook Air M4 16gb ram? I need a few tools, basic python skripting, creative writing, and all day tasks

Seems like the qwen 2.5:14b model is already to big to handle

[grand basin]

Yea cerebras is giving me
run error: 401 status code (no body)

[rose halo]

how exactly are you running the local models?

and details of their configuration in oclaw

[south ether]

its configured like this in openclaw.json, i didnt touch it it set up like this when i did onboarding. Its running through Ollama! Unless you mean something else?
"ollama": {
"baseUrl": "http://127.0.0.1:11434",
"apiKey": "ollama-local",
"api": "ollama",
"models": [
{
"id": "jaahas/qwen3.5-uncensored:4b",
"name": "jaahas/qwen3.5-uncensored:4b",
"reasoning": true,
"input": [
"text"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 262144,
"maxTokens": 8192
},

I've been wrestling with this thing for the better half of 3 days now. The only thing of note is when I checked, it said the port was already being used. Unsure how to fix that though

[rose halo]

when running ollama server for the llm ?

the ollama is sort of like a server that openclaw calls so if the port you tried to start it with was being used, it might not be running

in this case you just pick another port when running your app eg instead of 8080 pick 8081

11434

is your ollama api port apparently

[violet crystal]

Discovered this is only an issue with Codex 5.3 on OAuth

SOLUTION: Use 5.4 with latest Openclaw updates and things should work again

[violet crystal]

yes i just added my solution above^

[covert gazelle]

@south ether do you have a claude sub?

[south ether]

i do, ive been using claude to troubleshoot mostly

[covert gazelle]

but directly in claude app or with claude code?

[south ether]

i have an api and in app, havent touched claude code at all though

[grand basin]

uh, anyone know why my openclaw will just reply to me "NO"

oh it randomly finished stuff wth

[covert gazelle]

my suggestion is to install claude code, clone into openclaw repo into your device (so claude code can take a look at the whole repo and docs), and open a terminal and run claude code and let claude code look at the docs of openclaw and then explain the issue you are facing and if it can advice based on the docs the correct set up for your specific GPU and Ollama.

[torn jolt]

fuck guys what happened?? i updated openclaw now everything is fucked! my telegram connection is broken, all the memory and MD files are blank!

[dawn kayak]

How do I install openclaw on my windows pc

[south ether]

I have now thrown Claude at the problem! Hopefully Claude figures it out

[sweet marlin]

hey folks, just wondering if any of you arre using playwrite to subvert the api calls to make claude code talk to chatgpt

ive seen its bannable but are they enforcing it, and whats best way to go about it

[spark valley]

lol

[grand basin]

legit goes "NO" and then if i wait it'll complete and show me what i wanted lmao

[spark valley]

wsl is the best option

[grand basin]

#attitude hahaha

[spark valley]

look for #period true in the config ad turn it off

[grand basin]

omg i just hit 1mil tokens

doing pretty much ntohing

wtfffffffffffff

[spark valley]

in wt

provider

[grand basin]

I think it sent a full fresh api call for each file i said to download

there was a few hundred files, it must have just ripped through

Cerebras

[spark valley]

damn

[grand basin]

20k tokens by default is just insane

gotta be way to reduce this

[spark valley]

yeah i saw somewhere a skill to optmize

the usage

anthropic?

[grand basin]

Cerebras

3000tk/s lmao

[spark valley]

my agent lol i love it :
[3/14/2026 11:15 PM] AB: no dont over complicate device by device let do synoology
[3/14/2026 11:15 PM] hackkers: Focusing on Synology only. Let me hit it harder:

[spark valley]

that' a lot

are you calcullating the bitcoin hash

[grand basin]

I'm just testing and its like BRRRRRRRRRRRRRRRRR

Openclaw must be slow AF if its sending 20k tokens per message

[spark valley]

no openclaw is not the problem

[grand basin]

This is a fresh install lol

[spark valley]

maybbe there isn't solid soul to it or agent

it coulld be the last update

[grand basin]

i wish ic ould send pics

now dont get me wrong im a openclaw noobie but everyone was 1mo ago lol

[spark valley]

dm

[grand basin]

says tokens 19k/131k lol

So each api call is 19k minimum

[spark valley]

well that bit huge you can optmize it let me find you the skill

[grand basin]

i should see if this is what it was like a week ago, saw some ppl saying the update broke shit

i think i gotta clear the agent and restart

a simple question is just stuck in running

5min right now running on the question "what jus thappened"

[spark valley]

yea rollback to the version

it worked before

[grand basin]

oh my god api limit reached on model #2

[spark valley]

what you using it for

[grand basin]

i literally just asked it "what just happened"

lmao

[spark valley]

try /reset

and ask it again

[grand basin]

ah tyty ill try

once the api limit cleras

i think it may be spamming and getting rate-limited on the calls not the tokens

[spark valley]

bro use openrouter i been tellng ppl all day

it free

[grand basin]

oh i have been too

but i am testing a setup for serious workload

yep, HTTP 429 - too many requests

lmfaooooo

[spark valley]

but yoou need to try to go step by step try to exoert the setup then whatever model u put it just a model