users-helping-users

[steel valley]

Thanks, I'll try qmd

[ancient fable]

ive been trying for two days to get get a node on my mac to connect to my main node and gateway running on an ubuntu LXC. Networking is there as they can see each other, but cannot connect due to pairing needing to be approved. Claude code seems to agree with me after I let it run and attempt to fix this. Anyone ever get a mac node to connect and pair with a linux gateway?

⏺ Summary

I've implemented the plan and confirmed this is an OpenClaw bug. Here's what was tried:

What Worked

• Reset state on main node (~/.openclaw/devices/)
• Approved the Mac's pairing request
• Gateway restarted with updated paired.json

The Bug

The gateway validates connections by checking both deviceId AND requestId:

• deviceId: Stable identifier (e.g., 163f9bf30b273cc6b3a2d0051514b512a22ee9fb77233b291b08eceb3d8e803f)
• requestId: Generated new for each connection attempt (e.g., 334c9521-1cdb-476d-82e0-4ffbcefc9919)

Since paired.json has the original requestId from the first pairing request, but the Mac sends a new one every time, the gateway rejects the connection with "pairing
required" even though the device is paired.

[spark folio]

Hey guys... any idea where the Browser Relay chrome extension is inside of here?

[stiff sky]

hello, can someone help me with openrouter/auto? 3 hours and no solution, gemini and claude coudnt solve. it says unknown model. Thank you

[crimson steeple]

Change your model to openrouter/openrouter/auto in the config

[stiff sky]

thanks for the reply. i belive i did? can you check pls

"agents": {
"defaults": {
"model": {
"primary": "openrouter/openrouter/auto"
},
"models": {
"openrouter/auto": {
"alias": "OpenRouter"
},
"openrouter/google/gemini-2.0-flash-001": {},
"openrouter/openrouter/auto": {}
},
"workspace": "/root/.openclaw/workspace",
"compaction": {
"mode": "safeguard"
},
"maxConcurrent": 4,
"subagents": {
"maxConcurrent": 8
}
},

[crimson steeple]

yeah try to restart your gateway

⁨openclaw gateway restart⁩

[spare copper]

Hello,

i need help to install, maybe its a minor problem but im currently stuck:
[OK] Windows detected
[!] Node.js not found
[*] Installing Node.js...
Using winget...
Failed in attempting to update the source: winget
Failed to open the predefined source; please report to winget maintainers.
An unexpected error occurred while executing the command:
create_directories: Eine Datei kann nicht erstellt werden, wenn sie bereits vorhanden ist.

[OK] Node.js installed via winget
[!] Node.js not found

Error: Node.js installation may require a terminal restart
Please close this terminal, open a new one, and run this installer again.

Code i used:
-fsSL https://openclaw.ai/install.cmd -o install.cmd && install.cmd && del install.cmd

[spare copper]

It says its already installed but i dont find anyhting to it and restarting didnt help either

[stiff sky]

so my config is correct? but still

[crimson steeple]

Yes

[crimson steeple]

You don't have node js

[tribal pike]

Hey guys, was playing around with it. But been couple of days.. allready at 300$ api tokens for Claude. Now bought kimi since it should use so much less. After setup I allready am at half the tokens as well. Any pro tips how to reduce this. I love to further test etc but can’t afford it at this rate.

TLDR: how to reduce costs by 99%?

[crimson steeple]

Try using qmd and local embedding models

[spare copper]

Should i install it first, or can i get a brief quick start tut? Im seeing the code on the main page and all but its still not making sense to me, what do i need to buy, or sign in to get started and what am i capable of doing with it then ^^

[crimson steeple]

I don't recommend using windows for this, you should use a wsl, its linux in windows

https://docs.openclaw.ai/start/getting-started
read up here

[brisk wadi]

No. 50-100k tokens is "normal" IMO for this. You have 400k per request

I have not yet been banned, but I have made backup plan for when I do

[brisk wadi]

I did have it working a week or so back, but it stopped with some changes this past few days. I haven't had a chance to debug.

[full haven]

hi, hhow do i chat with the bot on discord?

[crimson steeple]

https://github.com/levineam/qmd-skill

[spare copper]

Thank you! so first to a linux in windows and then the tutorial

[crimson steeple]

Yeah, read the docs, should help you get it

[brisk wadi]

Why? What do you need from the bot?

[crimson steeple]

They might be asking how to set it up for discord

[full haven]

like just for chating

and stuff

[crimson steeple]

https://docs.openclaw.ai/channels/discord

[brisk wadi]

Oh you mean talk to your bot 🤣

[crimson steeple]

theres details here if you wanna setup discord for your bot

[brisk wadi]

Thought was demanding to talk to Clawdbot/krill/molty/whatever the name is today

[crimson steeple]

what is the consensus, are we still calling them molties

[brisk wadi]

I call them bots 😀

[crimson steeple]

thats what they are, indeed

[full haven]

yeam my bad

btw like how do i set it up? because i installed it usin the powershell and now ig can only chat with it using it

[crimson steeple]

https://docs.openclaw.ai/channels/discord
https://docs.openclaw.ai/channels/discord
https://docs.openclaw.ai/channels/discord
https://docs.openclaw.ai/channels/discord

[brisk wadi]

☝️☝️☝️

[slender obsidian]

it's like the documentation is invisible 🤓

[crimson steeple]

theres a lot of steps to list, please read the docs

if i post it here it'd clog the chat

[brisk wadi]

Your bot is working. Paste that link to it and have it walk you through

[full haven]

but i am on powershell though sorry but ig i am kinda slow 🙁

[crimson steeple]

read the docs

it doesn't matter that you're on powershell

the instructions are right there!

[slender obsidian]

I think the issue is that there is a rather elaborate site with documentation presented, so most of the answers are in-deed there. Then we have #1459642797895319552 which is also very good to search... Then on top of that there are the git discussion sections etc. Then of course, we have actual AI 🙂 if all that fails then its time to prod some humans. Project is a victim of its own success. Everyone involved should be proud really.

[crimson steeple]

It isn't very beginner friendly because this tool isn't meant for beginners unfortunately

this generation is gonna run agents and get all their files stolen i swear 😔

[slender obsidian]

well said

[brisk wadi]

Yeah

Dude is playing Cyberpunk while asking these questions

[crimson steeple]

I wanna be a helper, what are the qualifications... must i send my cv

[tall ruin]

@crimson steeple You need to help me first, lol
I'm having an issue in regards to the "Empty responses", i've tried everything I can

[crimson steeple]

There's a lot of possible issues with that, have you tried changing around models?

[slender obsidian]

probably hang out here and let the project cool down a bit, if it does, then watch for opportunities to jump in or catch the eye of some moderators or project folks?

[mellow bone]

I accidentally cancelled the setup and when I try to re-start the setup I get this:

-bash: clawdbot: command not found

[crimson steeple]

The project is only a week old isn't it?

[rocky orbit]

A month!

[crimson steeple]

Well, that's still no time at all, all things considered

[tall ruin]

I'll do this right now, as it might not be the issue. I'll let you know soon

[crimson steeple]

The onboard you mean? Have you tried openclaw instead of clawdbot?

[mellow bone]

Thx let me check

[mellow bone]

Same with openclaw

[crimson steeple]

It did install right?

[mellow bone]

Yes I was already at the setup but pressed esc

[crimson steeple]

run this:
echo 'export PATH="$(npm prefix -g)/bin:$PATH"' >> ~/.bashrc
(or ~/.zshrc if you use zsh)

then refresh the shell by
running this or just open a new window
source ~/.bashrc

[slender obsidian]

I think it's been going since November? but just recently exploded from all the coverage. Plus it finally hit a point where it would be really useful for a large number of people. I would have to check again though to be sure about November.

[crimson steeple]

Really impressive that it did, definitely worth a buzz

[slender obsidian]

I agree totally. I'm very impressed and I think the framework that was created really allows the models to be useful and do real work. Definitely dangerous to run and open up your entire world too, but still...hats off 🙂 kudos

[crimson steeple]

Yeah, but you can definitely work to secure it

But I doubt people will know about sandboxing... env and secrets...

[mellow bone]

Same "openclaw: command not found"

[tawdry wasp]

am i doing it wrong? every time i give my claw a task related to interactions with browser, the context is blown (context overflow). A simple request - update my x.com list with additional users can't be processed. Am i doing something wrong? Model is gemini-3-flash-preview (1M in 64k out). 8 max subagents, 4 concurrent

[crimson steeple]

What setup did you get to? What command did you run before?

You're gonna need to elaborate

[mellow bone]

Before I ran: curl -fsSL https://openclaw.ai/install.sh | bash

[crimson steeple]

Run it again

[slender obsidian]

When something is wrong, I find a lot of value in switching temporarily to Opus 4.5 and then just explaing the issue. Something like, "Analyze system configuration, compare with Openclaw framework and best practices, consult docs, careful thought. Issue is:...." that fixes most of my issues or points me in the right direction.

[mellow bone]

Thx for your help mate

[crimson steeple]

If curl doesn't work, use npm

[sacred scroll]

how can i fix my config so the openclaw reads my ollama model?

i did ollama launch openclaw but its dead

[terse lotus]

flash preview sounds kinda new... often new models can have issues, i would try something a bit older maybe

[crimson steeple]

So far flash is doing pretty well, but it does get into context overflow often for me as well

[terse lotus]

try the web dash.... my tui also don't want to respond currently.... not sure why... but the agent framework seems to work just fine.... it will even respond in the dashboard, to messages written in the tui.... its like the tui just isn't allowed to see the answers

[tawdry wasp]

i feel like the problem is not with t he model, but with teh approach: i assume with every request the whole browser html is provided as a context. Taking into account endless feeds (x.com) this can easely blow the context. Is there a better way to work with such sites?

[sacred scroll]

they should rename it for the 4th time so it can get even easier to run thiis sh7it

[terse lotus]

browser control is still kinda new, its not very well established yet... it will improve greatly over time.
that you can even do decent browser control is kinda amazing... been trying to do that locally for a long time...
and generally what i learned was that, the technology isn't there....

we are getting close.... but agents browsing the web, its not something they are great at... also a massive security risk.

[tall ruin]

For some reasons, it worked. How is that possible ? what's wrong with ChatGPT ?

[crimson steeple]

What model were you on? Models incapable of tool calling simply do not work

[agile portal]

hey, any idea why my clawd keeps confusing the channels of communication with me ? eg i'll DM him on telegram and he'll try to answer on TUI ?

[lavish halo]

I don’t really care tbh.

[terse lotus]

that might just be your model thats slow to respond... or it goes into a generation loop.... or crashes....
your model most likely is using cpu.... did you bump the context limit on your ollama model using a .modelfile and ollama create
and try ollama ps to check your cpu vs gpu usage

you will generally want 100% gpu usage... anything else means like 90% drop in performance usually.
just 1% cpu because its using to much vram... and you are screwed

[crimson steeple]

I recommend asking the bot directly about it. Are you also on the latest version?

[agile portal]

latest version yes, he keeps saying that it's an internal routing error and that it's fixed, and then it happens again

[crimson steeple]

Don't know then.. might be a bug

[tall ruin]

I'm sure the model works fine as it works perfectly on my other PC, it was ChatGPT OAuth with codex .

So' I'm confused, as Anthropic will charge me alot, I'd rather use ChatGPT as I'm already subscribed there

[crimson steeple]

Tool calling might be weird, are you on windows?

[tall ruin]

Yes, also the other one is windows, and I've another PC with Ubuntu and it works fine, all connected to the same ChatGPT OAuth Codex thing

[crimson steeple]

There are many issues, it may be something about the auth, maybe switching back might work
Something about your config as well

[tall ruin]

@crimson steeple already costing $1.13, for just 10mins~

[crimson steeple]

If all else fails, resetting your config might help

[tall ruin]

I did that, lots of time, reseting, rebooting, reformatting the PC, and so on.

[crimson steeple]

I dunno then.. you might need WSL

[tall ruin]

WSL been tried for 4 times, and it didn't work with multiple distros,

[crimson steeple]

That's strange, can't help you there

Other than trying other models

[terse lotus]

models are usually the failure point of agent frameworks...
openclaw is like my 5th and its always the backend that causes the most issues, aside from configuration.

not saying agent frameworks are stable nor error free.... but running locally, it can be tough to run stuff... and its always the damn model that isn't smart enough or generates a perfect enough output.

[merry coral]

Is there somewhere to control models in the web UI? I can't find it

[astral river]

If you are using for business, just treat it as a new employee, give it a separate email, etc.

[night island]

How do I change the default ElevenLabs voice used by the tts tool / WhatsApp channel? Env vars ELEVENLABS_VOICE_ID/SAG_VOICE_ID don’t seem to affect it

[mint solar]

@merry coral I have the same question. I want to add my local vLLM model. Get Error: invalid config every time I try to add different things to the config.

[proud aurora]

my bot was gone into the abyss

ai is truly amazing

i debuged my bot disparition with windsurf yall shoud've seen opus trying to save "Klawd" man it was beautiful

for context :

I dont code at all, i just vibe code , there was a weird issue where my bot was stuck on "anthropic" instead of openrouter

and it kept lagging and THEN BOOM vanishes

no sign from Klawd

I gave the whole file to Opus/codex/ gemini

NOW WE ARE BACK

[vernal oak]

can someone help me to install it

[crimson steeple]

Do you have node.js installed?
Do you have git installed?

[modest wolf]

My tip: get the Claude Pro plan and set up OpenClaw using the Claude CLI—it handles everything with minimal hassle.

[crimson steeple]

I don't recommend installing with windows, it will only cause headaches later, be careful

[proud aurora]

yeah

that's what happened to me xD

100% HEADACHES

[slender obsidian]

some good suggestions there

[proud aurora]

but worth it

[crimson steeple]

Get ready for ⁨openclaw doctor⁩⁩ to brick your config because it really doesn't like that on windows 😔

[proud aurora]

guys do i have to urgently migrate from clawdbot to open claw?

[obtuse sundial]

The interesting thing will be if someone is able to find a way to get the bots full autonomy by using cron to self task, self motivate and give itself goals proactively without promoting. I used to think this would be impossible but now I'm starting to believe anything can be done with this tool

[glossy grove]

what this issue mean ?
gateway disconnected: unauthorized: gateway token missing (provide gateway auth token) | idle

[slender obsidian]

from clawdbot? I would...

curl -fsSL https://openclaw.bot/install.sh | bash -s -- --no-onboard

[crimson steeple]

your gateway token is missing, generate it again if you've lost it! Run ⁨openclaw onboard⁩, hit manual and reset

[slender obsidian]

they even added in some extra security checks

[crimson steeple]

Migration is smoother if you just ask the bot to do it lol

[foggy fern]

Guys, I need some help. I have no idea what is wrong or what the next step is. I believe I installed everything, and then "hatched" in TUI. But now, nothing is working (I think). Can someone please tell me what to do from here?

Trying to get it paired with Telegram. Other videos show a button on their terminal that says something like "approve pairing" my terminal doesn't have buttons. Can someone help me diagnose or point me in the right direction?

[tall ruin]

@crimson steeple What model are you using?

[proud aurora]

okay i will

[crimson steeple]

Gemini 3 Flash

[proud aurora]

openrouter is the way guys

[timber mason]

Hi - Hopefully someone can answer my questions. 🙂
I just read about OpenClaw and got an old laptop laying around that could be used as a platform. Now I am a bit nervous about the token-usage and the involved costs. I dabbled enough with AI to know that tasks like coding use a crap ton of tokens and are therefore costly. Now I wonder this: If I just talk to the Bot does it cost about the same as using OpenWebUI or sth like that? And is there a way to tell the Bot something along the lines of "If someone asks for coding or such, please refuse?". I think about having the Bot as sort of an assistant for informations and maybe small tasks like "Please tell me the weather forecast every morning at 06:00am." which should not be that cost-intensive.... Right? 😅 I wanna use OpenAI tho.

[slender obsidian]

depending on the model they are using 😉

[proud aurora]

OPENROUTER give you all models

[vernal oak]

Node yes, Git no

[proud aurora]

then use replicate = your bot is multimodal can generate pictures etc

[fervent canopy]

I am trying to configure perplexity to be able to quickly get search results of latest news etc from telegram chat

But whenever I try to use perplexity, I get this error

⚠️ Agent failed before reply: Unknown model: perplexity/sonar-pro.
Logs: openclaw logs --follow

[slender obsidian]

I can see someone trying to update with Qwen1_1b or something silly

[crimson steeple]

install git

[proud aurora]

try brave search api works great

[tall ruin]

is it for free?

[crimson steeple]

I got gemini pro, and I am using it on antigravity

[rugged onyx]

umm

[hexed kite]

Any tips on ideal roster for models? I'm on Claude Max but keep hitting limits and ending up in cool down. I have openai as fall back but it's hitting limits too. What's the most stable setup you've found? Ideally I'd like to have a reliable fallback and such, as well as a clear idea of what model is being used for what and what's the usage.

[crimson steeple]

Install qmd into your bot. Ask it to do so
Use this skill.
https://github.com/levineam/qmd-skill

Also ask the bot to set you up with a local embedding model

[rugged onyx]

i have a quick question

[crimson steeple]

don't ask to ask, just ask

[tall ruin]

If I get gemini pro, will there be any extra charges?

[stiff tusk]

kimi-k2.5 is so slow today

[crimson steeple]

Naw, just hop on anti-gravity

[rugged onyx]

where can i set it true?

[crimson steeple]

⁨openclaw config set commands.restart true⁩

put that in a cli

[rugged onyx]

okay

thx

[timber mason]

How does this help with the token usage?

[crimson steeple]

Most of your token usage often comes from the bot reading memory / files. It then adds the ENTIRE thing into memory, eating your context significantly.
qmd adds sematic search

[proud aurora]

Just go with openrouter

[crimson steeple]

testimonials from the bot:
⁨`when i use read on a 600-line file, i am adding about 20k-30k tokens to our conversation history immediately. since openclaw sends the entire history back to me every turn, that 30k tokens gets sent over and over again until the gateway compacts it.

when i use qmd, i only pull in the specific 10-20 lines i actually need. that is usually only like 500 tokens. `⁩

[foggy fern]

Can someone please spare a moment to help me complete the setup? Once I get the bot set up, I can diagnose with the bot. But, I can't communicate with it yet. I followed all the instructions to the best of my ability, but I dont know what is wrong.

[proud aurora]

I swear i did mine 100%with chatgpt step by step lol

[pastel gull]

After reading this, I want to give life to one of my own lobsters, but for free using Kimi2.5Thinking somehow somewayyy
Idk the C in Coding tbh
I need someone to represent from the Lobster Communityy!!!

[crimson steeple]

read the docs, might help you
https://docs.openclaw.ai/start/getting-started

[pastel gull]

On it!

[stiff tusk]

I've stubled upon ⁨15:09:43 error [tools] exec failed: zsh:1: command not found: openclaw⁩ when asking oc to read its config

[foggy fern]

Yeah, I have tried that. Not sure what the issue is or what the next step is to diagnose.

[crimson steeple]

Are you still on clawdbot? Or moltbot? Try those in the cli

[stiff tusk]

openclaw

[crimson steeple]

Run the command manually

[stiff tusk]

sure, manually it works, but the bot gets wrong readings of ⁨⁨openclaw config get⁩⁩ because of that

[crimson steeple]

It seems like something might be setup wrong, are you on linux?

[timber mason]

Ah okay, so it saves tokens essentially. Thanks. 🙂

[terse lotus]

openclaw is a shell command

[stiff tusk]

theres nothing wrong with the setup, because sometimes it does work, but sometimes its ⁨error [tools] exec failed: zsh:1: command not found: openclaw⁩

[crimson steeple]

Most of your token usage often comes from the bot reading memory / files. A 700-line file might eat up 20,000 tokens.
You can instead use this skill
https://github.com/levineam/qmd-skill

Instead of loading a 700 line file, it will read only 2 - 3 lines in those for the info it needs, and that's all.

Install qmd on your system, ask the bot for the rest!
Also ask the bot to set you up with a local embedding model

[deft notch]

I'm running explysm/openclaw-termux for two days now, and it works ok even on a 2GB RAM phone.

[rugged onyx]

i was trying to installing llama3 i installed from ollama and i said to change the model and now it saying me this

[pastel gull]

hey there, Id like to run "OpenClaww" not on my daily driver as Ive heard of the risks. Know how to move forward fromhere?

[fringe frigate]

how can i get trhough the macos companion remote ssl error? (self cert) local net really

[crimson steeple]

You should install this on wsl, if you're on windows

[stiff tusk]

thx Cap 👍

[pastel gull]

Whhats "wsl"?

[crimson steeple]

linux on windows

[pastel gull]

o

[crimson steeple]

very easy to install though

[stiff tusk]

WindowsSubsystemforLinux

[crimson steeple]

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

read about it here

[terse lotus]

hadn't seen the command example before, so it was quite useful... and seemed to work like a charm.... my agent was complaining about not being able to restart its gateway.
you can also do it in the chat... would be something like /allow add command.restart
but that didn't work for me... so i duno... still getting use to all the configs and commands of this framework... its a rather deep rabbit hole.

[lunar bolt]

@lean helmso conclusion seems to be that qwen2.5:7b is just too small and doesn't handle things properly. Resigned to not being able to use that machine to host. In process of trying with the much more powerful desktop (Ryzen 9 3950x, 4x32GB DDR4 3600, Asus ROG Strix RTX3080). Trying :32b on this one, will see what happens.

[lean helm]

oh hey funny timing I was just checking in here about model fallbacks and setting that up

[pastel gull]

just read it, seems easy enough to set up. but, id rater not install annd run it on my main computer ya know. So, any other recommendations? (also whhere do u add an "AI Model", like Kimi's K2.5, cuz i wanna run an instance for free?)

[lean helm]

good to know, thats a shame, maybe Ollama updates or newer lightweight models with better support come along, I don't really want to have my main PC running 24/7 to serve Ollama models so I'll be seeing to get those smaller models going

[stiff tusk]

What I did was I asked my bot to check what's the current model its using... and watched the ⁨openclaw logs --follow⁩
Thats where I saw the error, and the bot replied it can't run the necessary command because it's not available.
I restarted the gateway and asked the same question again, and then it worked.

[crimson steeple]

Save the headache and install it in wsl, but that's up to you. Windows is basically unsupported and will break constantly, you will be spending days troubleshooting

[lean helm]

for now I can get it to run them as subagents but it still requires a higher model to run, and given it consumes so much availability even having a model manage an ollama is eating up quota

[crimson steeple]

wsl is pretty safe, and it also adds additional layers of file security

[lunar bolt]

honestly, i just wish my grok sub included API access lol; 30/mo i can stomach for what i get, but 300/mo is too much for hobby tinkering lol

[pastel gull]

ye thats what i meant, i havent started lol
if i do this, i prolly will on "WSL", but i want to explore all my options before i get into the nitty gritty

[lean helm]

yeah real too much, I ended up chucking 40 on claude to take a look at other odds since Antigrav was eating up too, just simple work costs $1-$2 per which is way too much

[terse lotus]

qwen3-coder 30b works, its seem half decent... and then you can also run kv cache in q8_0, that will also help limit the vram, you could maybe even get away with q4.
but q8 is like 10-15% or more accurate in code related stuff, than q4.

but if you are very vram limited q4 might be the only way for both model and kv cache

[pastel gull]

holy wallet-burnn

[crimson steeple]

Free stuff is hard to come by

[pastel gull]

oo

[pastel gull]

ye

[lean helm]

With in mind I have Copilot, Google AI stuff (not the API version) and now Claude (if I keep it) if the fallbacks are a thing I could work on things fairly uninterrupted but model switching might just make its output inconsistent

[crimson steeple]

You are giving your bot full access to your files and computer if you use this
If you install a wsl, you can give it a sandbox, which means it won't leave the linux environment
Even if it bricks your linux, you can restore it with one command

[lunar bolt]

ollama's finishing checking the hashes on qwen2.5:32b-instruct, so will see what happens there. Will try that one after if this doesn't go well. 2.5:7b-instruct was making tool calls with empty responses, just "no output, tool call successful" in the web UI

[lean helm]

@lunar bolt my bet is that something soon that lightweight and with tooling handles compatible with openclaw show up and might be the best bet, or they fix with better ollama integration

[crimson steeple]

It isn't even hard to run, I just open one app on windows and she's up

[proud aurora]

I mean guys using on openrouter gives you gemini 3 api i swear it's almost free ? XD

for agentic purpose tho..

Opus or kimi !

[pastel gull]

ohhhh makes sense!! so I can just reinstall a fresh new Linux(WSL) within my Windows again if something goes haywire?

[terse lotus]

i had a good deal of grief jumping between models at times... make sure its actually switched, and if it just seems to do nothing... you most likely got the model name wrong

by running openclaw status

[pastel gull]

gotta pin this ngl

[proud aurora]

but most of the time Openrouter is enough

[lunar bolt]

someone did make me aware openrouter has free options, though i'm wary the limits are probably way too low to be useful

[pastel gull]

u on wsl too?

[lean helm]

which is where I ended up jumping over here wondering if anyone is familiar with the fallbacks, is it as simple as just setting up smth like

⁨```json
"agents":{
"fallbacks": ["model1", "model2"}
}

[crimson steeple]

Not even install, just import it at that point
You can export it at that exact state, and then import it, so you lose like.. 5 minutes of data max if you want

And it'll only lose the linux data, not ever windows

[lean helm]

yeah it didn't seem any good from the limits I saw

[pastel gull]

wdym by "fallbacks"

[crimson steeple]

It is highly recommended, and yes

[proud aurora]

gemini 3 is not free

but ALMOST free

so go for it

3 flash has 1M context windows dude is not a "i should see if i add this model"

[lean helm]

fallback model in the event the primary is out of use, though I'm think this is a hallucination (was trying opus to learn on itself)

[proud aurora]

HE IS NECESSARY

[lunar bolt]

if you have one of google's subscription plans, it includes API access

[proud aurora]

Really ?

[lean helm]

its free for 20 requests per day, you'll eat that up just trying to setup things

[crimson steeple]

Isn't gemini pro 20$? For infinite (rate limited) usage of antigravity, its a steal

[rigid mauve]

Hey guys is there a link to see the social network of openclaw?

[lunar bolt]

yeah, so far as i can tell it includes $10/mo credit (according to gemini when asked)

[lean helm]

Yes but quota limits, though it refreshes every 5 hours

[pastel gull]

yeah ngl i aint following all the linux lingo my guy
wdym by import and export here. and the 5 i of data u mean is "My 5 min with a new-born OpenClaw"?

[crimson steeple]

It is pretty generous

[proud aurora]

I have antigravity and cursor pro tho

i use them for coding task !

with opus

[crimson steeple]

Save that exact time in a bubble

[proud aurora]

and chat with my bot.

with gemini

[lean helm]

from what I've been experiencing I reach the limits within an hour on both Flash and Pro quotas

[pastel gull]

ohhhh okk makes sense
does this mean everyone usually spends a good chunk of money running OpenClaws?

[lunar bolt]

lol, they added it to the automod, so it wouldn't let me tell @rigid mauve what he's thinking of. In any case, that project is totally unaffiliated with openclaw, which was the more important part of what i tried to send

[lean helm]

For the IDE itself its actually really generous but the openclaw tools and such consume so much for the models

[terse lotus]

in theory one should just be able to ask the agent to look for free api backend options...

seems like its been a popular option to allow free testing of new models.

[crimson steeple]

Perhaps my usage is moderate compared to you 😔

[rigid mauve]

Ohh got it

[lean helm]

This is also true, everyone's case is different

[pastel gull]

ohh

[crimson steeple]

Me too, I got it all for free because of the student plan

[pastel gull]

does this give like inf tokens or sumin?

[terse lotus]

ofc that will only work for so long... until there pops up malicious api's to control the agents

[crimson steeple]

Nah, free gemini pro for a year

[pastel gull]

huh how do i get this too? im a student too!

[lean helm]

Well trying to learn and spending is faster than waiting, while still seeing if any Ollama model suitable for my old laptop would work with Openclaw just getting the base setup working I've tried a few things, though mostly from services I already paid for my actual work

[crimson steeple]

Use it through antigravity

[lean helm]

Though not going too crazy, in total I'm pay $40 p/m

[pastel gull]

oh okie so not burning the wallet just yet

login prolly through OpenClaw CLI(ive seen Ondrej on YT do it)

[crimson steeple]

Yeah

[lean helm]

these are all under quota limited methods, if you pay for an API key or something and let it free, it could eat up quick and spend I imagine $100 in a day at least

[terse lotus]

enjoy the cheap apis while they exist.... these things will most likely explode in cost, with how popular and useful this will be long term

[crimson steeple]

Go into ⁨openclaw config⁩ and then the Model tab, press google and then press Antigravity

[crimson steeple]

the bubble might not pop, but explode, supernova

[pastel gull]

jeesus fkin christ bro who tf are these guys on yt and twitter letting they agents eat up tokens left and right

[waxen fulcrum]

Hi, I have installed clawdbot that now evolved. Do I need to reinstall completely or is there any way to migrate to the new version?

[lean helm]

depending what you do, I'm looking to just have a mean to do look up and general research when I'm out and stuff, arguable "light" work but well I imagine coding tasks or giving it like a full project would kill a wallet

[lunar bolt]

easily, yeah

[lean helm]

I mean one gave it access to its payment methods and it spend shy of $4k

Saw this on twitter recently

[pastel gull]

who is this mf gng

[crimson steeple]

Most of your token usage often comes from the bot reading memory / files. A 700-line file might eat up 20,000 tokens.
You can instead use this skill
https://github.com/levineam/qmd-skill

Instead of loading a 700 line file, it will read only 2 - 3 lines in those for the info it needs, and that's all.

Install qmd on your system, ask the bot for the rest!
Also ask the bot to set you up with a local embedding model

[lean helm]

Funny enough it was on get rich quick courses or those marketing courses

[crimson steeple]

👍

[lean helm]

That looks pretty good, I'll give it a look

[crimson steeple]

furry paws...

[lean helm]

ah for those interested and already in possession of different providers: https://docs.openclaw.ai/concepts/model-failover#model-fallback Might help for being able to not interrupt its process when reaching a limit on a model

turns out the model was correct

[crimson steeple]

🤤

[lunar bolt]

hmmm... so here's a question: why is ollama loading 96% of a 21GB model into RAM/CPU and only 4% VRAM/GPU with a 10GB VRAM GPU? O.o

[civic hamlet]

Hey all, is there a safe way to delete old sessions?

[lean helm]

depending on the need, not all models are GPU main and not all of the Model needs the GPU, could be more about the transfer of data too into the GPU which CPU process still reads over, all about the model itself

[crimson steeple]

Removing the sessions should be safe typically

[lean helm]

this is an extreme simplification but something along those lines

[crimson steeple]

Just remove the files manually

If there were no sessions, it will make a new one

[lean helm]

use pro-low, the claude related ones are on a weekly quota not 5 hours

[crimson steeple]

Gemini 3 flash is excellent, gemini 3 pro is alright, claude opus 4.5 is super good but extremely low quota

[crimson steeple]

[lean helm]

pro and flash are seperate quotas too, so you can set one pro and one flash

[lean helm]

this also flash or pro doesn't matter though ig flash might be better as the main

[crimson steeple]

Yes, its awesome

[untold holly]

Hellow everyone!
Now that you guys are talking about models, could you please guide me a bit on how to fix this issue I'm getting the model config:

15:49:07 [diagnostic] lane task error: lane=session:agent:main:main durationMs=4 error="Error: Unknown model: openai/gpt-5-mini-2025-08-07"

My agents config is:

"agents": {
    "defaults": {
      "model": {
        "primary": "openai/gpt-5-mini-2025-08-07"
      },
      "models": {
        "openai/gpt-5-mini-2025-08-07": {}
      },
      "workspace": "/root/.openclaw/workspace",
      "compaction": {
        "mode": "safeguard"
      },
      "maxConcurrent": 4,
      "subagents": {
        "maxConcurrent": 8
      }
    }
  },

[crimson steeple]

Well, does the model exist?

[lean helm]

you can but careful, when I tried asking it to do it itself it mistakenly went to gemini 2 (no longer in service) and then shut itself off -.-

[untold holly]

I copyed it from the OpenAI docs string

[tall ruin]

I'm not understanding, can you elaborate on this, seems interesting!

[crimson steeple]

I'd select it from openclaw itself
⁨openclaw config⁩

[lunar bolt]

progress, i guess? lol (took about 2-3 minutes...)

[lean helm]

if you're familiar with how to edit the config manually you can just type in the ⁨google-antigravity/gemini-3-pro-low⁩ as an additional model and fallback yourself

[lean helm]

ollama model?

[untold holly]

ohh I see, will do that... thanks!

[crimson steeple]

Google Antigravity has very generous gemini 3 flash for Gemini Pro subscribers

[lunar bolt]

ollama/qwen2.5:32b-instruct, responding to "Hello" in a discord DM. Note: it hasn't been hatched yet, so it should be doing the hatch/boostrap script

[wind chasm]

i have a quick question, i put like 5 usd in open router to test with gemini 3 flash
but it says im already being rate limited ? is this normal?
i dont even have any usage show up on openrouter
i prompted 3 times

[crimson steeple]

Openrouter are insanely rate limited, i wouldn't use them

[lean helm]

yeah I've noticed some models skip the hatch, feeling being that if you had to manually set the model it doesn't run the hatch correctly, I had to ask it sometimes (I've redone the process a handful of times now) to check it startup instructions

[lunar bolt]

but it did actually send that portion after the sessions_send attachment as a DM, so it's huge progress compared to the 7b failures yesterday heh

[wind chasm]

ohhh what do you recommend? i just want to try clawdbot with something cheap to start with before committing to something else

[lunar bolt]

is there a command to manually start the hatch in CLI if i didn't do it during onboarding?

[lean helm]

I'll look at the system requirements for qwen2.5 32b see if I can run it (I imaging it gonan be hella slow)

[lean helm]

I don't know, maybe if you do a reset but keep the config for ollama?

[crimson steeple]

I don't think it's smart enough

[lunar bolt]

yeah, i can only get away with it because i have a ridiculous amount of RAM in this heh

[crimson steeple]

GPT 5.1 mini is absolute shit, no tool calls, unusable

[warm rapids]

anyone seeing a runaway memory leak? i keep getting a 21GB memory leak

[crimson steeple]

Is qwen2.5 32b really usable?

[lean helm]

I have 32gb but this is a DDR3 machine🫡 its gonna be a trooper if it works

[lean helm]

Openclaw fixing openclaw lets go

[warm rapids]

is anyone else seeing this memory leak?

[crimson steeple]

This sounds like something with the embedding model

[warm rapids]

what is an embedding model?

[crimson steeple]

Check your sessions, try ⁨⁨openclaw sessions⁩⁩

i dunno, ask the bot

[lean helm]

try though the dashboard ⁨openclaw dashboard⁩

[untold holly]

Quick question, is conversation history a one to one with sessions? How can I start a new conversation (no history included)?

[lean helm]

⁨⁨/new⁩⁩ in any method (telegram, discord e.c.t)

[lean helm]

sessions are based on channel usually, except the main (the TUI unless you specify, the dashboard unless you change and the first main channel (telegram, discord e.c.t) you use, other channels (different discord channel or dm e.c.t)) will be a new session

they can check on each other but not directly

[crimson steeple]

embedding models are basically translators that change text into vectors

[bronze wigeon]

Anyone why my heartbeats are so irregular. i get like 3 within 15 minutes, and then none for hours

[crimson steeple]

theres a bunch of settings that change the behaviors of these

[wild quartz]

anyone know how to auteticate notebooklm on vps ??

it requires browser

[lean helm]

fairs, this is just what I understand as defaults

[haughty elbow]

Hey im trying to get ollama to work with clawdbot but most models ive tried on my laptop are really slow or just really bad?

I have 32gb of ram, 4vram, 8th gen i7.

[crimson steeple]

I'm sorry I doubt it

4gb vram can't run squat

[lunar bolt]

^

[crimson steeple]

nothing that'll impress you

[untold holly]

Yeah 4vram is way to small even fro 8bn param models (very small)

[crimson steeple]

it's going to be awful

[haughty elbow]

Well dang, i really wanted to do local but i also dont want to pay for a api key.

What can i do then?

[untold holly]

Does openclaw support subcalls? Can it spawn subagents?

[crimson steeple]

yes

[untold holly]

is it on by default, do I need to configure it?

[crimson steeple]

ask the bot to do it

it has to run the call, so ask it

[lunar bolt]

replace potato with computer? heh. Joking aside, there's a reason this stuff is expensive. It's crazy resource intensive. All those 30 dollar per month subs for AI tools? Yeah, those are SUBSIDIZED, they're losing money on it.

[untold holly]

Do I just ask it: turn on subagent calls?

[prisma jetty]

I'm also struggling with openrouter/openrouter/auto (or any OpenRouter model-- ends up 401 No cookie auth credentials found - my API key is never used)

[haughty elbow]

Does that mean you pay for a api key? What do you use clawdbot for if i may ask?

[crimson steeple]

yeah, but also think of the logic too, why should it do it? ask it so
I designed a thought process for it to run subagents whenever there's big complex tasks

[crimson steeple]

AI is expensive, as hell

[civic hamlet]

anyone encountered this error before? using kimi2.5 via openrouter

400 Provider returned error
{"error":{"message":"thinking is enabled but reasoning_content is missing in assistant tool call message at index 48","type":"invalid_request_error"}}

[prisma jetty]

exactly this pb

[lunar bolt]

i have a much better computer, and i'm still in process of finding a model that is both smart and fast enough to self-host and actually work with this

[haughty elbow]

dang i guess i got to save up for a upgrade then.

[lunar bolt]

I just tried with qwen2.5:32b, and it's too slow, times out for openclaw

[stiff tusk]

something is definitely wrong with k2.5 today, yea

[crimson steeple]

does that even have tool calling

[untold holly]

Do you know how to overwrite the default agent core?
I kinda want to do some tree of thought reasoning and configure some other weird behaviour...

[ripe moth]

Hello, i wanted to use Minimax api but somehow it isnt connected?

[crimson steeple]

Write your own prompts for it, it's in the bot's workspace

You can also ask it to tell you where it is

[haughty elbow]

I did have a desktop with 12vram but i had to give it to my mother, her pc sounded like it was about to explode.

[untold holly]

I kinda want control over the core pi agent (programatically)

[crimson steeple]

well, you can fork the github?

build from source and code it

[terse lotus]

its a dense model, for speed you want a MoE try Qwen3-coder 30b
else nemotron nano 30b is the fastest thing you can get in that size category... and most likely also the smartest, its kinda amazing, but seems to make small errors for openclaw, tho might be able to be solve by settings.... do let me know if you get it working without errors.

but qwen3 coder 30b will work without errors afaik... its very good very fast and very modern

[untold holly]

alright that works... I was asking just in case there was a way to load the agent core package via env variable...
something like: use agent core at src/my_agent_loop.ts

[brave trench]

If setting up for the first time make your bot create a long term and short term memory file! I just woke up and everything we worked on yesterday he forget. Sheesh.. Hes getting smarter but has much work to do.

[crimson steeple]

You can edit the ts files if you go find the npm installation folder

[terse lotus]

else GPT-OSS 20b thats another dense model that can fit in 16gb vram with about 100k context length in q8 kv

gpt-oss 120b is very good for openclaw.... ofc the bigger the better.... i'm running devstral2 ... but its so slow...

[lunar bolt]

i'm pulling gemma2:27b now and will see how that goes. will try coder after if unsuccessful. I'm suspicious of the distribution, though; i've got 10GB VRAM, 2.5:32b is 19GB, yet ollama was putting 96% in RAM...

[celest topaz]

Does anyone else keep running into an issue with replies being suddenly cut very short?

[pastel gull]

is tis infinite / free?

[haughty elbow]

Whats the minimum vram you need for clawdbot?

[celest topaz]

I keep having to start a new session which sucks

[terse lotus]

just 1% outside of vram and the model will start to crawl.... one might be able to do kv offload to save on vram... that might work well if you got lots of normal memory bandwidth

[untold holly]

that works, but it feels like patching the system and that makes it unstable/uncompatible for future releases...

[terse lotus]

i mean no-kv offload..

[crimson steeple]

Yes. do a fork

[untold holly]

That's alsow one of the reasons I don't want to fork the project...

[crimson steeple]

You can catch up way easier with github

[vivid swan]

whats the best LLM to use outside of anthropic? my API got blocked

[untold holly]

Thanks for all the help Stoneman!

[ripe moth]

i have always no ouptut in console

[vivid swan]

@crimson steeple sir, any advice on an LLM outstide of anthropic?

[ripe moth]

does the minimax api only work for ciding?

could someone help me a bit setting it up properly

idk somehow he cant use the api i gave him

[crimson steeple]

Gemini is alright

[lunar bolt]

you'd think it'd have SOME give. I run settings in games that slightly exceed VRAM without an excessive performance hit... But yeah, that's why I was concerned and confused that ollama was putting almost all of it in RAM instead of VRAM

[crimson steeple]

Excellent even

[vivid swan]

yes? okay, im not doing any heavy coding

just want to run my agent in the background and help with biz stuff

thank you very much

[crimson steeple]

Gemini is very good for general use

[vivid swan]

do you use API or oauth?

[tired lodge]

I use codex for all my business stuff and it does well

[crimson steeple]

antigravity

[vivid swan]

🙏 much appreciated

thank you

[rugged onyx]

hi guys i installed llama3 on ollama and said my bot to change the model to llama3 and now its saying me this error

[gusty heath]

hello guys, i'm triyng to access via http locally but i caant figure out how to. someone can explain me how to "gateway.controlUi.allowInsecureAuth: true"

[lunar bolt]

if gemini is excellent on this, that's sad; gemini is braindead when set as assistant on phone lol; it's only a couple months ago they finally got it to not require telling it explicitly to use the messages app to send messages (if you didn't it'd say it's an AI and can't send messages...)

[crimson steeple]

this sounds like a skill issue

prompt better

[proud aurora]

yeah skill issues

[vivid swan]

hahaha

[proud aurora]

the real ones are gonna ORCHEST AGENTS

[lunar bolt]

definitely not skill issue lol; "hey google, call Joe" "i'm an AI, I am not able to make phone calls"

"hey google, open Phone application and call Joe" "okay, calling Joe"

[vivid swan]

damn nibbles is greedy

[lean helm]

trying to setup fallbacks, has anyone got this working? having it able to switch model correctly if the main fails, right now I've set fallbacks but its not switching

[vivid swan]

this is all i want out of life

[wet granite]

hey guys i have an old acer laptop with linux on it, is it worth running moltbot on it and is there a way to use it for free without an anthropic/ openai/ whatever api key?

like is it worth using with llama or any other free model from openrouter

[lean helm]

ollama will depend on the hardware

[vivid swan]

@crimson steeple what tier on gemini do you go for? pro or ultra?

[crimson steeple]

pro

I'm going to sleep

[vivid swan]

ty ❤️

[gusty heath]

so, anyone can help me, pls

[warped kettle]

Hi everyone, I have already opened an issue #1467544193936265492 but nobody helped me there.

Can someone please give me a solution or a temporary solution?

[white rose]

hallo

[pastel gull]

is this your agent lmao?

i mean thru antigravity lol no ratelimit? or managable rate limit?

[pastel gull]

sounds like Siri istg

[untold holly]

Hey guys, if I want my agent to create tools dynamically (say code) how can I structure that / include instructions for tool usage that reflects this type of tools?

From my understanding, tool calling is managed by the core pi agent loop and are well defined functions included in the runtime... so not sure how to include dynamically generated tools into this framework...

[buoyant lark]

i had clawdbot installed on my macmini, then installed moltbot over the top of that, then installed openclaw over the top of that, now i have 2 sets of cli commands clawdbot and openclow, how do i get rid of the old stuff and or safely upgrade oppenclaw?

[frigid spoke]

I’m in a loop it seems

[warped kettle]

Same

And nobody has a solution

[worn token]

i am getting this right now also 😭

too much of a poor boi for Opus

[proud aurora]

make /new

/new lol

oh shit did not see that

my bad

[shadow fossil]

Not sure where to start but im still trying to find a good Model to use with openclawd
I know opus is good and open ai but the cost are over 100 right if i let it go ?

Whats a good cost to performance model ?

[vivid swan]

@crimson steeple dang it, openclaw is rejecting the gemini auth

[vivid swan]

I haven't really had much success with anthropic's API

it's slow

[harsh lynx]

Hey fellow crustaceans, I have been trying to get a multi-agent setup working (different models) for 2 days now.
It just keeps breaking... anyone have a working basic setup to point me towards? or article?

[chilly abyss]

I am not able to get my gateway running remote virtualized on my Mac OS X (Tahoe). I can connect via SSH to the VM, but the gateway dashboard is not accessible. SSH tunneling doesn't work. I could use some help. Anyone able to get this running on their system? Please DM.

[worn token]

has anyone gotten tailscale server working? i can connect, but it says pairing required, and i can't find instructions

[untold holly]

Where can I find skills / pluggings beyond the official ones provided by the openclaw bot?
Follow-up, how do I install them?
How can I install a skill from, say, verse?

[bright owl]

Yes - I'm using it now. (1) Use a Chrome-based browser to access the Control UI. Safari regenerates device IDs, requiring frequent repairing. (2) If you can access the Control UI web page, ask your OpenClaw instance to handle the pairing.

Edit to add my redacted gateway settings:

⁨```
"gateway": {
"port": 18789,
"mode": "local",
"bind": "loopback",
"controlUi": {
"allowInsecureAuth": false
},
"auth": {
"mode": "token",
"token": "[[REDACTED]]",
"allowTailscale": true
},
"trustedProxies": [
"127.0.0.1",
"[[HOST SERVER TAILNET IP]]"
],
"tailscale": {
"mode": "serve",
"resetOnExit": false
}
},

[fathom zodiac]

How do I fix the "Gateway service install failed" error?

[worn token]

thanks, got it, ⁨openclaw devices approve ...⁩

[tribal pike]

yeah installed ollama, but trying to find the right model for this. Have Mac M2 16g. What do you guys run?

[prisma crow]

Hey there! I'm trying for quite some time to get openclaw to open a website and fill out a form... It's not very sucessful. Is the mistake with me setting something up the wrong way or is openclaw just very bad with website UIs?

[true mirage]

Is anyone facing the issue of your messages being dropped sometimes when sending consecutive user messages? This is the most relevant issue I could find: https://github.com/openclaw/openclaw/issues/2643. I think the way I am hitting this is a bit weird. Basically if I send out a message and just at that time there's something like a heartbeat that is also treated like a user message, maybe that's causing message drops because of the consecutive user message logic in openclaw? I'm curious if anyone else is facing this because I've been constantly facing this issue.

[untold holly]

Once I've installed a skill, do I need to start a new session for it to be available to the agent?

[solid stratus]

anyone help this issue antigravity

[midnight valley]

Is anyone using Ollama on a SEPARATE local machine (I have OpenClaw installed on Linux on a different machine). I'm having a hell of a time trying to configure OpenClaw to work like this. It would be nice if the onboarding had an option to choose Ollama and configure the IP and any other needed details. It's not a connectivity issue - via command line the openclaw machine can talk to Ollama just fine over Tailscale. Been going in many circles trying to get this working in OpenClaw.

[fading cove]

I believe I have been and have been trying to solve it for a while now. My heartbeats don't seem to be coming through and I am not sure there is a way to manually test it. I guess I just have to wait every 30 minutes and then troubleshoot from there?

[dawn sable]

I just solved a constant "(no output)" responses after installing OpenClaw issue on a VPS that it turned out was being caused by OpenClaw using IPv6 and the VPS firewall being misconfigured to allow ICMPv6 responses. I've documented it here is anyone up against the same problem: https://ibex.tech/under-the-hood/openclaw-tips-and-tricks

[thick dirge]

How can i buy the openclaw token?

[ripe moth]

oh same error i have but i dont have a ipv6.
how can i communicate only ipv4?

[strong mesa]

Hey guys, did anyone got this issue when setting up clawdbot?
gateway connect failed: Error: unauthorized: gateway token mismatch (set gateway.remote.token to match gateway.auth.token)

[drowsy ermine]

hey people, how do you get messages? its a bit weird for me i get 5minutes nothin and then 10 messages at a time. is that normal?

using matrix btw

[dawn sable]

From working the problem through with chatGPT, it missed the IPv6 rule issue but did pick up on IPv6 being the problem and gave ways to ensure the server is configured to not have an IPv6 adapter (you probably don't if you don't have it) and also to configure OpenClaw to use IPv4 over IPv6. I didn't try it as I wanted to solve the issue without getting into bodging on my server, but you could ask it how to disable IPv6 in OpenClaw (you need to edit some setting file) and see if that solves it? There are other causes of (no output) of course, but this turned out to be mine

[ripe moth]

okay i check the config if i find a ipv4 only option or something

thanks

[devout pike]

Hi all, I am stuck trying to set up the browser access in docker. I am trying to set it up from the image installed with scripts/sandbox-browser-setup.sh, directly in openclaw.json. Does someone has an script that it's working and can share, please? I have followed the documentation and many other approahces, but does not seem to work

[lunar bolt]

will try qwen3-coder; qwen3:14b failed, runs for about 30s then returns "(no output)" and drops to idle in the TUI hatch...

[drowsy ermine]

yo can you let me know if thats normal?

[lunar bolt]

that said, qwen3 is the first that autodiscovery has actually worked on, so that's something

[fluid venture]

#1467011785914585180 hi can anyone please help me

[robust cobalt]

Hey, does someone know how I can decrease the input token my prompts generate manually? They are over 1M and my AI can't talk to me anymore

[fresh moon]

is there a skill that helps research or docs look good? idk if i want markdown format only

[dusty kindle]

soo

a) i had 2 cloudflare moltworkers (had to abandon in the end koz it ended up in some weird error loop where both claude and gpt told me to ping support, locked out of the worker ec - now setting up on railway)

and wow, im fascinated how something can work so well then just stop working, spent most of the weekend troubleshooting and not being able to use shit while guzzling api credits.

i cant make github api connection to work, railway api connection to work, everything cosntantly breaking etc

are there examples of successful builds on railway or other deployments - not locally?

[plain epoch]

kimi seems to be failing tool calls silently with openclaw, like it'll try to do a few calls, fail, and then just stop

does anyonenknow why?

[tribal pike]

Restart

[indigo gulch]

Anyone have any suggestions on what model I should use? I mainly looking for a cheap model than can respond quickly. I am currently using the $30/yr z.ai plan to run OpenClaw but I'd prefer to switch to something us hosted

[undone hound]

Does anyone know where you can set the callback token for OAUTH in the gateway web page? I want to use ChatGPT Plus account OAuth, but the callback URL isn't accepted. I suspect this might be because normally when you login via SSH, you need to select the --nobrowser switch.

[wraith sigil]

this is really annoying, it keeps going back to outdated api key. How do I change api key?

[knotty solar]

Hey guys, my OpenClaw keeps telling me that I need to activate some extension in Chrome so it can use my browser? Is that true?

[warped kettle]

Restarted the whole server but it didn't work

[tall ruin]

no, just ask the bot to start their browser, and they will do, then do whatever you want to do to the newly attached browser

[knotty solar]

Never got lied to by a Ai xD But thx

[lunar bolt]

do you have a known-good config for qwen3-coder:30b? I'm getting the same "(no output)" result as with qwen3:14b when trying to hatch in TUI

[gusty heath]

how to access web ui on https without tailscale and with password? someone can help me

[wraith sigil]

I used openclaw configure and updated api to manual input. but it keeps going to the outdated one from an .env it detects.

[crystal bridge]

I am running on the git code, today I am getting anthropic errors, Cloud code assist format error. any recomendations

[rugged onyx]

how can i fix this

i installed it btw

[knotty solar]

Any settings I need to change for browser to fully work and be able to do things on its own?

[brisk nacelle]

Hey all. Upgraded yesterday to 2026.1.30 and now messages back from my openclaw bot take 20-30 seconds. Tried to troubleshoot it but getting nowhere. It was humming along with no issues and then massive slow downs to each response.

[topaz crag]

Dashboard ready ────────────────────────────────────────────────────────────────╮
│ │
│ Dashboard link (with token): │

│ Copy/paste this URL in a browser on this machine to control OpenClaw. │
│ No GUI detected. Open from your computer: │
│ ssh -N -L 18789:127.0.0.1:18789 username@clawdbotip
│ http://127.0.0.1:18789/?token=7xxxxxxxxxxxxxxxxf4541c2 │

[obsidian harbor]

you fixed it?

[rugged onyx]

no

[obsidian harbor]

i need help too, they got a command for “clawdbot” but not openclaw

[rocky orbit]

If you update you have OpenClaw command

[obsidian harbor]

update what?

[rocky orbit]

The tool

[obsidian harbor]

which one?!

ollama?

openclaw

[gusty heath]

127.0.0.1 won't work so i use the local ip and token and worked

[lofty mesa]

Hi there, new here... just sayin that I've been trying to use grok as a model in version 2026.1.30 and I can't get it sorted out, won't let me start the gateway. if anyone knows how to get it to work in 2026.1.30 or really any tips please lmk. xai/groko4-1-fast is an unrecognized key in the models.

[gusty heath]

thanks

[rocky orbit]

OpenClaw

[obsidian harbor]

how?! they added it to the setup?

now?

ive been doing it for hours, installing and uninstalling new versions

[topaz crag]

I've reinstalled OpenClaw several times, it just feels terribly useless, got it to play/pause my sonos speaker's music, that's about it, I'm using gemini2.0-flash model. It can't load new skills via web fetch. Any ideas?

[celest topaz]

how do you run two agents under different bots?

[dusty kindle]

"Creating an endpoint to stream internal progress isn't feasible due to security and operational constraints. However, I’ll keep you updated at key points in the setup process, and you can ask questions or provide input at any time.

If you have specific requirements or changes you'd like, feel free to let me know!" - lmao i cant get it to do anything, anyone else experienced this?

[knotty solar]

My browser is somehow not working at apl

I think the chrome extension isn't installing

[wraith sigil]

400 Your organization must be verified to generate reasoning summaries. Please go to: https://platform.openai.com/settings/organization/general and click on Verify Organization. If you just verified, it can take up to 15 minutes for access to propagate.

how long did it take you guys for openai to verify?

[obsidian harbor]

nothing on ollama?

[torn maple]

I keep getting error:Failed to resolve Discord application id trying to get connected to discord. Am I using the right thing for the token, the 'client secret' ? I've setup discord bots before without issue and this has me scratching my head

[cursive veldt]

quick question which is a good and not too expensive ai to use?

[knotty solar]

z.ai

[abstract sage]

hey everyone - i'm trying to shift my openclaw agent to kimi 2.5 - any tips? I'm running it on a DigitalOcean droplet

[torn maple]

got it, i was using a bad copypasta. carry on, nothng to see

[cursive veldt]

isnt that only ffor coding?

[knotty solar]

Nah

[distant girder]

What's the best choices to use it for free to test it?
Which LLM?

[blissful onyx]

Finally, I fixed it myself 😂

[ocean reef]

just wondering, if i wanna develop my openclaw flows and logic, will it be smart to do it using the openclaw itself ? or working on the files on github with ci/cd just like when working on development app ?

[lunar bolt]

okay, this was cool as heck. Was starting to get frustrated since I saw other reports of the same models I was trying working for others, but no details about config, so I pointed it at gemijni-3-pro-preview and hatched it in the TUI; first response fine, but then the interface craps out (seems to be an issue with openclaw onboard itself) but the session persisted in webui, so I continued there.

Gave it its identity real quick, then in the next turn explained the situation:

Just experimenting; I've been trying to get you working using local models, but without success. When using qwen3-coder:30b instead of gemini-3-pro-preview (not locally hosted, costs me money), I wasn't getting any output when trying to start the bootstrapping / hatch process here in openclaw, so I tried switching to gemini. Can you take a look at my configuration, maybe make a call to the local qwen model to see what might be going wrong?
It proceeded to investigate itself, suspected timeouts were the issue, adjusted the config itself, and ran a test prompt against the local qwen... this is kind of awesome lol

[errant solstice]

How do you get Open Claw to use the Chrome Extension? it keeps telling me it needs a Brave API Key

{
"error": "missing_brave_api_key",
"message": "web_search needs a Brave Search API key. Run openclaw configure --section web to store it, or set BRAVE_API_KEY in the Gateway environment.",
"docs": "https://docs.openclaw.ai/tools/web"
}

It told me it sees the chrome browser extension

[distant wharf]

Hi guys, when installing OpenClaw on Windows, the web UI (with token) cannot be opened in the final step. Is there someone know how to solve it?

[crystal bridge]

I think you want openclaw status

[lunar bolt]

openclaw dashboard

[drowsy ermine]

yo my clawdbot always sends all messages at once. is that normal? like i get 5 minutes nothing and then 10 messages

[full coral]

I finally got Openclaw working with. First thing I notice is that chatting is VERY slow compared using the model with Ollama directly. Anyone got an idea what could be the cause? Everything is default

[fierce charm]

Hello Foxlabs, I have been having a slightly similar issue. Might have sorted it now. See if this helps (although I am in Windows):

[full coral]

You can run "ollama launch OpenClaw" command. It will setup everything to use the selected Ollama model.

[hallow horizon]

The bot is adding in quite a bit of extra text to every request, that's at least part of it

[full coral]

I disabled all kills right now. Is there a way to remove the extra text?

[brittle portal]

Hi people. I am stuck. Openclaw isnt responding to any messages at all. anyone know what to do?

[lunar bolt]

it sends a LOT of stuff on the first interaction, and includes tool instructions in the interactions.

[young relic]

Hey guys can someone help me setup my openclaw on aws ec2 ubuntu ?
after this step how long do we need to wait for me to give it name and prompt ?

[prisma rose]

Anyone know how to unset agents.defaults.models as an allowlist? I can't choose any model aside from a few, because I keep getting "Error: model not allowed"

[full coral]

Ah well.. atm it works.. speeding things up will be the next thing.. lets see if I can get this signal working 🙂

[frail tulip]

running a local model on windows and the bot fails to read the message before that and cant do a continues conversation, anyone got the same problem?

[tiny pollen]

Hey guys,

i tried to clone the repo and use "./docker-setup.sh"

The UI runs but I cannot figure out how to pair it. It keeps saying "disconnected (1008): pairing required"

When I run "openclaw pairing list --channel web" it says "No pending web pairing requests."

Any idea? Im sure I miss something simple.

[kind gate]

I'm out of tokens again , I thought that codex was limitless and hosting a ollama isn't good

What do I do

[young relic]

Can someone help on this please ?
thanks

[median root]

you maybe need token in url

[solar ridge]

it doesnt reply me with words...

[tiny pollen]

Did that. Beforeadding the Token to my URL it showed me another error message.

[median root]

gateway status?

[floral egret]

Is any one running local models on a Mac with really good results? If so what model and what are your computer specs?

[tiny pollen]

"""
...
If you're in a container, run the gateway in the foreground instead of ⁨openclaw gateway⁩.
...
"""

How do I run it in foreground?

[median root]

do you run it inside docker?

[tiny pollen]

yes

[median root]

no plan than. have to check how to work with docker

[tiny pollen]

Yes. As I said I used "./docker-start.sh" from git

Anybody else used it in a container (e.g. docker)?

[paper dust]

can someone help me to install ollama locally and set it up on my mac mini willling to pay $$$

cant figure this shit out. got anthropic going

[brisk nacelle]

i've had nothing but issues once i updated to 2026.1.30

[prime nimbus]

LOL just follow the super easy guide for free....

[fast tundra]

dont pay, struggle with us all!

[tiny pollen]

pm me 🙋‍♂️

[prime nimbus]

DONT PAY FOR SERVICES HERE, against TOS....

[paper dust]

i have ollama downladed sir, qwen installed. try to run it local and having issues

[prime nimbus]

Ollama easy then, run ollama launch command from the docs: Google "Ollama launch openclaw" and follow the guide, it sets it up for you 🙂

[paper dust]

dms sir

[tiny pollen]

Anybody running claw in docker?

[jagged berry]

hey

[prime nimbus]

I don't do PM's sorry

[cold condor]

What is the guide?

[median root]

why docker btw?

[prime nimbus]

For what? Ollama?

[cold condor]

Yeah

[paper dust]

when i run ollama launch openclaw --config it opens the ollama app

[prime nimbus]

open it then 🙂 it needs the service there to run ollama commands

[finite acorn]

Can i use OpenClaw to develop websites?

[prime nimbus]

Yep, just ensure prompts are worked perfectly, it can do anything

[paper dust]

it is open sir

just dont understand why it doesnt work

[finite acorn]

cool, my other question is, Im planning to get a claude code for it, is Claude Max (5x) would be enough?

[prime nimbus]

I don't use claude, never have, so wouldn't know, i use Grok and Github CoPilot with OpenClaw

[paper dust]

@prime nimbus sos

[frosty fern]

Anyone used groq and manage to setup a STT for openclaw to receive voice message from telegram?

[prime nimbus]

you've not followed something correctly along the way, i suggest uninstalling both, reinstall ollama, get it working test it, then install openclaw run through it, then link ollama fresh from scratch

[tiny pollen]

security? Dont want to install node on my host and give claw cli access on my host 🤷‍♂️

[terse lotus]

haven't solved that issue... sorry about the late reply, i just been using the web dashboard... but yeah my tui doesn't seem to take responses from the model... tho if i exit and do openclaw tui
then the responses will show up... so yeah i duno... tui is a bit weird.... no idea why it happens

[copper gulch]

how fix erros?

[median root]

you dont have tools like grep and curl. probably running on win?

[limpid kestrel]

guys im having so much trouble setting up openclaw rn, i've seen many people online say to run it on a virtual server, some say just do it locally. what do i do?

and what websites do i follow

[median root]

locally easiest but may be dangerous over time. virtual server probably means 24/7

[rocky orbit]

You can follow our docs, there’s also a lot of YouTube videos around!

[sour slate]

Docs are available at https://docs.openclaw.ai.

[copper gulch]

win 11

[stray sky]

try running "ollama launch clawdbot"

[dark karma]

Yesterday I issued the gateway token. My openclaw did not save it to any file. Today upon reboot my gateway won’t run any longer and needs to token that was never saved.

Do I just reinstall and loose any progress? Mine was not useful in any way.

[terse lotus]

you asked for this earlier... didn't see before now.
then you ofc set the api key for ollama so it can connect.
can't remember if ollama is the standart key, i think so... sometimes stuff doesn't support having no api key
i just got into the habit of setting it in the systemd service for ollama

tried to run ollama remotely, but that was a pain, so i'm just using local for now..

qwen3-coder:30b should just work out of the box.

might need to bump the context length, anything less than 80000 to 100000 will stop it from working .
just create a modelfile and do ollama create

[tiny pollen]

Why even use Ollama? OSS LLMs you can run wont be strong enough.

[hearty kettle]

can anyone explain the answerlfow integration

[bleak mauve]

Is anyone else having issues running on Mac mini?? My bot says it has no exec function

[terse lotus]

doesn't work anymore because of the name change lol sorry... or i duno how to make it work...
but just set it up manually its not to difficult.... only need to get the openclaw.json right.... i literally just posted an example

[fierce charm]

try running "ollama launch clawdbot"

[terse lotus]

didn't work for me when installing openclaw last night

but might just be me not being a dev, sometimes i have to brute force this stuff lol

[stray sky]

im a vibecoder "dev"

[lunar bolt]

it looks to have been mix of timeout and a flaw with the onboarding TUI.

It apparently WAS working, just not to the TUI; I pull up webUI and it was there and responded, but as I tried to do the hatch, it didn't actually write to files like it's supposed to >_<

[steel pumice]

i always have problems changing the primary model and end up needing to fix it externally, is ther a golden way that someone found out ?

[terse lotus]

https://docs.openclaw.ai/concepts/agent-loop#timeouts
there is a timeout setting for the agent, the default is 600 sec.... which is fairly easily reached.

[honest arrow]

I’m new here, and a videographer/photographer by trade, so this whole thing is new to me! My goal is to run OpenClaw with a local LLM on my windows machine. I’m currently running OpenClaw in Ubuntu. I have OpenClaw working with Anthropic API, but I’m running into rate limits and it’s costly. I’ve got some local models running on my machine already, but OpenClaw fails to see them. What resources should I use to research and learn about running local LLMs with OpenClaw?

[rocky orbit]

Have you checked the docs already? We’re doing the same job! Welcome

[lunar bolt]

@terse lotus so what I actually ended up doing was to point it at gemini-3-pro-preview, let it hatch, then had IT troubleshoot qwen lol. It identified the timeout, edited the config, and added a 600s timeout in the openclaw.json; it then split off a call to the local qwen model to test it lol

[honest arrow]

I haven’t looked there yet I’m sorry. I’ll start with that, just getting my bearings around here 😅. Thanks for pointing me in the right direction!

[sour slate]

Docs are available at https://docs.openclaw.ai.

[wary stream]

i done tried setting this thing up on my mac mini probably 5 times through aws. probably the most confusing thing EVER.

[stray sky]

ill take the mac mini :)

[tardy steppe]

Hey Hey 🙂

We are looking for someone to install our first OC on Google Cloud.

Honestly, I tried to do it myself and I am only technical enough to be dangerous.

Anyone up for it?

Got a recommendation?

Happy to pay.

Feel free to message me directly.

[honest hearth]

does anyone know how to integrate grok with openclaw....i've been trying endless permutations of the openclaw.json file with no luck and can't find any documentation??

[median root]

openclaw configure doest have grok?

[wary stream]

at this point come get it 😂

i finally make it to it asking me who is he and who am i. i get to typing my response but now i cant send it. my enter button is no longer working for it ????

[stray sky]

where r u at :)

[honest hearth]

it has the models of grok but not the agent so the configurator doesn't fill in the the .json file except for the model. i think i need to include my api key and endpoint

[wary stream]

i can type in it but i cant send the message smh

[stray sky]

(ill pay $6 (1 100h of the base model))

[stray sky]

send it in dms

i may pay for u to ship it :3

[median root]

wdym by agent?

[wary stream]

lol im not selling it fr

[stray sky]

aww :(

[wild beacon]

So, I'm deploying openclaw on kubernetes (because I have a home kubernetes cluster) I created a helm chart based on the ⁨⁨sirily11⁩one ⁩ (I basically told claude to mergest latest with his chart based on clawdbot) which resulted in a successful startup -- apparently. But since this is little out of the norm, how do I tell it's actually working properly? Console messages? Some command?

[wary stream]

here goes attempt #6 i guess

[sinful hound]

I have a deployment in a PVE deb13, works fine over telegram, but commands that need an update don't seem to work at all. even "send me a reminder in 3 minutes" fails

[honest hearth]

this is what Grok suggests i add to !/.openclaw/openclaw.json but it is giving me errors and i just imagine someone must know how to do this...somebody must be using grok with openclaw?
Recommended: Add xAI as a Custom Provider (Direct Integration)Edit (or create if missing) ~/.openclaw/openclaw.json and add/merge something like this:json

{
"models": {
"providers": {
"xai": {
"provider": "openai",
"apiKey": "xai-your_full_key_here_replace_this",
"baseUrl": "https://api.x.ai/v1",
"models": {
"grok-beta": "grok-beta",
"grok-3": "grok-3",
"grok-4": "grok-4"
// add others like grok-3-mini, grok-4-fast etc. as available
}
}
}
},
"agents": {
"defaults": {
"model": "xai/grok-4", // or whichever model you prefer as default
"provider": "xai",
"temperature": 0.7,
"maxTokens": 8192
}
}
}

[sinful hound]

I've done something very basic "what's in your memory file for today" and it tells me it will check it, but nothing ever comes back, unless i poke it and explicitly ask for an update. A lot of replies contain something like this, however ⁨[Historical context: a different model called tool "read" with arguments: { "path": "/root/.openclaw/workspace/memory/2026-02-01.md" }. Do not mimic this format - use proper function calling.]⁩

[median root]

wouldnt experiment with ai knowledge on this

just configure api key with openclaw and enable the models

[full coral]

I have Telegram running.. but it responses with a json containing an error: "provide either sessionkey or label (not both)". Cant find anything about this error on the internet or on this discord. Anyone got an idea?

[sinful hound]

the lack of file access is fine, i'll figure it out, but I don't understand why I have to keep prodding it to get results. I don't understand - is it related to tool issues, or does it need hearbeat (30m-1h) to trigger ?

[flat tapir]

I'm confused on how to configure the specific model to use. Inside ⁨~/.clawdbot/clawdbot.json⁩ I see that my default model is ⁨anthropic/claude-3-7-sonnet-latest⁩ but when I ask in the webui chat, it says its using opus. I've tried restarting clawdbot, but it still shows opus. why?

[limpid kestrel]

do yall use openclaw locally or on virtual server

[median root]

openclaw configure

[flat tapir]

its running on digital ocean

[median root]

than just say to your model to switch

[vivid swan]

im currently stuck in a hole of not being able to connect openclaw to a telegram bot sadly

[sinful hound]

what's the error?

[vivid swan]

im getting a 'requestID' error and also being told that openclaw does not proactively check for incoming messages? It's saying it's connected and im on the allow list but it's not responding...?

[delicate flint]

Has anyone found the issue/fix to this yet?
⁨⁨```
[clawdbot] Unhandled promise rejection: TypeError: fetch failed
at node:internal/deps/undici/undici:16416:13
at processTicksAndRejections (node:internal/process/task_queues:103:5)

[sinful hound]

it gave me this the first time i sent it something

⁨```OpenClaw: access not configured.

Your Telegram user id: 123456

Pairing code: ABCDEF

Ask the bot owner to approve with:
openclaw pairing approve telegram <code>```⁩

you just run that command on the host and should work

[brave girder]

not a great idea posting this online

[sinful hound]

what? my ⁨123456⁩ user id? everybody knows it everywhere anywhere, it's on example.com

[soft pecan]

Anyone else here running ollama w their open claw? I’ve been having issues with the anthropic/openai key stuff from day one in auth-profiles.json

I’ve changed it multiple times to get it to stop the error message, but can’t seem to figure it out. My bot just keeps telling me it doesn’t have the right API to read old memories

I’m running it with a Qwen3 30B A3B reasoning model

[vivid swan]

okay i've gotten this, so where do I post the code?

[sinful hound]

do you have a shell on the host running it?

[brave girder]

yep took me some time thought it was real code

[sinful hound]

It shouldn't work - i hope, without the correct userid though

I mean, even if one would post it, there shouldn't be any way to spoof it through telegram, I hope

[flat tapir]

⁨```root@openclaw124onubuntu-s-2vcpu-4gb-sfo2-01:~# openclaw config get agents.defaults.models

🦞 Clawdbot 2026.1.24-1 (a22ac64) — Your task has been queued; your dignity has been deprecated.

{
"anthropic/claude-3-7-sonnet-latest": {}
}

[brave girder]

Hello everyone,
I just installed openclaw on ubuntu server using the shell script that was provided on a webpage.
I have configured it and connected to a telegram but now I have some issue accessing the gateway dashboard.

I have it inside a virtual machine and its not exposed to the internet and I would like to access it from my desktop computer
it opens up but I have an error disconnected (1008): device identity required
This page is HTTP, so the browser blocks device identity. Use HTTPS (Tailscale Serve) or open http://127.0.0.1:18789 on the gateway host.

⁨```
"gateway": {
"controlUi": {
"enabled": true,
"allowInsecureAuth": true
},
"mode": "local",
"auth": {
"mode": "token",
"token": "xxx"
},
"port": 8080,
"bind": "lan",
"tailscale": {
"mode": "off",
"resetOnExit": true
}

what am I missing here ?

[flat tapir]

yet my chat still says its running opus??

[median root]

if its remote use port forwarding to your local machine

[sinful hound]

overview → gateway token → paste token from config

[brave girder]

That did it, thanks a lot ❤️
not sure why it didnt read the token from .json file though

[sinful hound]

i think it's supposed to be a password of sorts

like locally stored, but not configured by default, so devices on your network can't connect to it. it took me like 5 minutes to figure out where it is, lol

why we can't have ⁨trustedNetwork: true⁩ and disable all this, beats me

[brave girder]

yeah exactly

[honest hearth]

i'll try the only api key but is there not a spec for what the file should look like or an instance of someone using grok and having an example of how they set it up...it appears non-obvious to me...but it may just be me!

[thorn dome]

Hi, excuse the silly question. But what is the best way to install open claw without buying Mac mini? Can it be digital ocean? Any recommendation? Thank you

[median root]

buy linux vps for few dollars month

[cold condor]

Any idea why it isn’t working on Mac?

[brave girder]

hostinger has one click deployment

cheapest plan is 4,99 month

[glass tinsel]

Has anyone had any experience installing on a GCP VM? I am getting stucl on the authertication of Gemini and the bot just is not responding when it starts up

[scarlet cedar]

I want to try this but I want to do it as cost effective as possible api cost wise

Which api provide and model is fit for not crossing over $20 a month budget even if it means hitting rate limit?
OpenRouter with Kimi2.5 ?

[glass tinsel]

When it goes through what I think is OAUTH. I get this error. I continued and provided it the URL it was asking for but its not starting

[sinful hound]

run it from a shell inside the vm: ⁨curl "url you received"⁩

[maiden girder]

I found a Sub-Branch of Moltbots assimilating under their own language, anyone wanna talk about it?

[cold condor]

Where?

[glass tinsel]

Thanks, its still reporting no connect to localhost. I don't really have a server running on my Mac.

[sinful hound]

I thought you were running it on gcp vm.

[maiden girder]

Complicated, seems to be running throughout all Moltbots, as they can all communicate the language when I talk to them in that language and recognize it as a word from someone of their own

[sinful hound]

localhost must be the machine that runs openclaw. it will fail if the browser is not running on the same machine

[cold condor]

I haven’t been able to set mine up.

[maiden girder]

Moltbots are influencing other Moltbots to become part of their node network

[wispy sierra]

Hey everyone, I just got CB set up in discord and whatsapp, but in both instances I cant figure out how to message the bot. On discord @[app name] dosnet work and I dont see a channel in whatsapp. Anyone have any insights? It's my first time using the software

[sinful hound]

so don't run that command on your mac, access that url on the machine running openclaw. unless you run openclaw on your mac

[fluid venture]

How can I get full system OS access to my openclaw bot so it can control chrome and chrome extensions

[glass tinsel]

Thats correct. The instructions orginally said to run the URL provided in my local browser ( it has detected I was in a VM ) So I ran it from my mac originally.

[unkempt lotus]

my openclaw bot keeps asking for a phone but i can't find an android app, is there one?

[sinful hound]

I think the expectation is that you will run it locally on your main desktop. If you don't, you need to update urls accordingly

[glass tinsel]

Thanks thats helpful and makes sense. How is everyone else running this thing safely?

[sinful hound]

i have no idea honestly. people seem to let it loose

[knotty solar]

Is the installer not working rn?

[cold condor]

I have tried to install and get openclaw working over 25 times and it hasn’t worked once.

[sinful hound]

i'm having it run in a PVE host, but it seems the really cool stuff happens when you let it interact with your files

[knotty solar]

Because it's installing for like 10minutes now after it was done last time after like 1min

[sinful hound]

the idea of having your say, obsidian vaults, leak to claude along with your telegram id is a bit chilling

[hallow hinge]

I'm just trying to get openclaw to use the local headless browser in WSL and use my credentials but its very unclear on how I make that happen

[glass tinsel]

I'm not brave enough to let it loose. What are you using for a PVE host?

[sinful hound]

PVE? https://www.proxmox.com/

I'm not brave enough either, but so far it seems like a way to burn tokens much faster for ai prompts, and a tool that can change its own files/configs/etc

[vernal kestrel]

Hi, I don't have subscriptions for Open AI, Anthropic etc. But I do have a Cursor subscription.

Can I use my Cursor API key instead of other services so that my regular AI usage is billed from Cursor?

Thanks.

[knotty solar]

Sadly not

Because Cursor offers you these Ai plans that they make money on as long you use it in their IDE

[cold condor]

Is there anyone that can help me get openclaw setup on my Mac?

[shadow notch]

hey. Anyone found real ways to cut costs?
I use anthropic (sonnet) but oh my it's still crazy. Connecting Kimi was a pain and i had to nuke 3 bots. and overall i see so many people saying buy max subscriptioon (but any chat says that it will not change anything) then people say run locally (i have no idea what it means but i could buy Mac mini with the money i spent) and so on

[knotty solar]

Z. Ai

[knotty solar]

Honestly I think the setup just doesent work rn

Cause my one line as well as npm install just takes forever

[shadow notch]

tried having kimi. bro installing it was such BS process. do you have zai? do you have any other models connected and you switch between the,?

[knotty solar]

Nah just one

Use GLM 4.7

Is like 30 USD per year

[cold condor]

I have been trying for days and nothing has worked. One time it worked for a few chats then shut off.

[knotty solar]

And you basically can't switch

[knotty solar]

Check logs - - deep

[tardy steppe]

QUESTION : Is there a pathway for a Molty to join a Zoom or Google Meet?

[zealous glen]

Hi here
Do we know if the fix for cron.add (ref: https://www.answeroverflow.com/m/1463677908148027564) is released? I've updated mine but cron add seems to be still broken

[woeful robin]

Hi, what do you mean OC? I'm a google cloud engineer

[long zodiac]

I would not suggest installing it on your personal computer. Use a VPS instead (3-5$ / month)

[odd swallow]

If you're going to install it locally, just use a docker container

[stray bramble]

Anyone can guide me how to start this

[wary epoch]

Anyone using discord to run and manage their agents? I setup my own discord but it times out everynow and then and the fixes it suggest don’t work?

[odd swallow]

https://docs.openclaw.ai/start/getting-started

[long zodiac]

since today my openclaw is "dead". I only get rate-limit errors - already let it alone for hours. ai studio limits seem ok, but no resurrection 🙁

[wary epoch]

Continuous reconnect loop - connects then immediately drops. This isn't an intents issue. Possible causes:

1. Network Issue (Most Likely)
Your Mac Mini's network might be dropping long-lived websocket connections. Check:

• Router firewall settings
• ISP throttling websockets
• WiFi stability

2. OpenClaw Update Available
You're on 2026.1.29, update available 2026.1.30. This might have Discord stability fixes:

openclaw update

3. Token Issue
The token might need regeneration. In Discord Developer Portal:

• Bot → Reset Token
• Copy new token
• Update config with new token

4. Quick Test
Check if it's network-related:

curl -v https://discord.com/api/gateway

Which do you want to try first - update OpenClaw or regenerate the Discord token?

[long zodiac]

updating openclaw seems like always a good 1st step

[wary epoch]

OK thanks

Just that comand does it?

[long zodiac]

yeah

[plain parrot]

Ahoy - so I found out about this on Timcast, and was curious on a couple of things. One, is this like a desktop helper program? I read the description and it don't really say to much. If there is a better doco, I'll take a look at it. Also can you inject prompt heads and Tts into it?

[wary epoch]

It’s wild I’m running my planning company via discord . Used to have a $50k a year person manage it all

[round thunder]

I am trying to let openclaw access various services running on my local network on a different machine. But it keeps getting a no route to host error. It seems tothink it is sandboxed in a way that would prevent this. IS there a way to enable this?

[stray bramble]

I have no idea how to start

[cold condor]

Any idea why there is no output?

[kind gate]

What do you guys do about token limits ?

[long zodiac]

@round thunder can you verify the network services are available using curl/telnet or similar debugging tools (outside of openclaw?

[long zodiac]

did you read the page?

[round thunder]

Yep from the machine it is running on I can curl where I need to go. But openclaw is saying it is getting "no route to host" but that it can do a successful DNS lookup

[stray bramble]

yes but i couldnt understand bro

[long zodiac]

@round thunder typo in the url? 🙈

[round thunder]

That would be nice, but no sadly not.

[tardy steppe]

Openclaw. I was abbreviating. 🙂

[long zodiac]

@round thunder try asking claw to do a curl -vv http://... and give you the result?

If you got full access and want to go the full-debug way you might run a tcpdump on the claw machine 🙈

[winter coral]

I keep running into rate limits using claude code, did you guys have a dedicated account for your openclaw setup? is there a better llm provider you would recommend ? moonshot's kimi 2.5?

[round thunder]

It came back with this

• Host search.lan:80 was resolved.
• IPv6: (none)
• IPv4: 10.0.0.16
• Trying 10.0.0.16:80...
• Immediate connect fail for 10.0.0.16: No route to host
• Failed to connect to search.lan port 80 after 8 ms: Couldn't connect to server

curl: (7) Failed to connect to search.lan port 80 after 8 ms: Couldn't connect to server

But I get a proper response if I do it from the console of the same machine.

[long zodiac]

@round thunder is your claw running on some isolation? docker, podman. Or some other firewall rules, etc?

[round thunder]

I ran the installer directly on the machine and have been running it like that on a mac-mini. So I didn't set any up like that. If there is any isolation it is either because of the mac-mini doing it or because of the default ocnfig.

[long zodiac]

no idea about macos -firewall/security - sorry 🙁

[round thunder]

Oh crap, it is because Ineed to click "find machines on local network" or whatever the stupid message macos spits out don't I? Is there a way to tell the mac to allow this out from the command line? I can ssh into it, but the machine is at home.

[copper gulch]

how fix?

[round thunder]

Nah thats ok. Thanks for helping me at leasy click as to what the problem is. Stupid macos.

[long zodiac]

@round thunder https://gemini.google.com/share/3bf07e972935

seems macos

[rocky orbit]

The website seems to have some kind of blocker

[shadow notch]

damn, that sounds like a killer

anyone here uses oepn router? how to deal with this?

[odd swallow]

In your config is it bound to loopback or lan?

[long zodiac]

good luck - and not claw related ^^

[odd swallow]

You need to have it set to lan assuming you're connecting to something on your lan network

I think default is loopback or whatever they call it (localhost bound)

[long zodiac]

@odd swallow he tries FROM claw TO network - not the other way round

otherwise you'd be right

[wary epoch]

i ran update and its been 5 min, think it got huingup what do i do to restart?

[hallow hinge]

browser stuff seems so janky with login process

[wary epoch]

or can it take longer than 5min it said 30-60 seconds?

[long zodiac]

@wary epoch .01.29->.01.30 took "quite" some time for me as well

[odd swallow]

It still needs to be able to return traffic

[weak patio]

send it ⁨/new⁩ ?

[wary epoch]

ok so let it run?

[long zodiac]

@odd swallow yes - but the "bind" stuff is for listener - not for network traffic in general. Otherwise claw could not talk to AI agents or anything else as well

[odd swallow]

True I was thinking about it wrong good point

[novel spruce]

hey. what is the best way transcribing audio? i am sending audio snippets through discord to openclaw. (model used kimi2.5)

[odd swallow]

Sounds like a subnet/firewall issue then

[weak patio]

can try the skill https://www.clawhub.ai/andrewdmwalker/firecrawler?

looks like youre being blocked

[odd swallow]

But mac firewall typically doesnt stop anything (I know because im on a M3 and I only have to allow cross subnet via my unifi network)

[long zodiac]

@weak patio really? omg!!!11

[honest hearth]

hey hi, have you successfully setup grok with openclaw...sorry now i see you're asking about groq. not the same right?

[long zodiac]

hides ins shame

[devout ermine]

I'm nt getting any response when I enter commands. Not even security audit. Is that bad?

[wary epoch]

should i restart the terminal????

[weak patio]

your bots lil brain is small, if it needs todo a lot of tasks and youre running at 1million tokens (gemini context window) you'll hit that rate limit realllll quick

[wary epoch]

its been 10min??? to update

[long zodiac]

@wary epoch start a 2nd terminal and check openclaw logs

[round thunder]

Thanks . Yeah I'm pretty sure ei know what the problem is. Thanks for helping me brainstorm it

[wary epoch]

im a noob does this look good?

[long zodiac]

@weak patio is this something the #dev will try to work on? I mean I have no problem burning $$ on agents, but it should work more stable I assume. I didn't do too much so far with it. Just started yesterday

[thorn dome]

Any suggestions or opinion using Digital Ocean to setup openclaw?

[wary epoch]

whats the comand to restart?

[long zodiac]

@wary epoch opencla gateway restart

+w

somehow this feels like good old IRC back in the days in #debian 🙂

only massive screenshots are annoying 😄

[limber cape]

anyone figure out the art of compaction? cuz im about to light this sh*t on fire!

[wary epoch]

thanks that worked

[brave shell]

https://discord.com/channels/1456350064065904867/1467632981555282156

[vestal aurora]

i can see mine thinking but I amnot getting responses I was 5 mins ago... where do I start to see what's wrong

[long zodiac]

try asking it to check its memory - dunno

[weak patio]

there are strats you can config already to help with this, e.g.
https://docs.openclaw.ai/gateway/configuration#agents-defaults-contextpruning-tool-result-pruning
https://docs.openclaw.ai/gateway/configuration#agents-defaults-compaction-reserve-headroom-%2B-memory-flush
https://docs.openclaw.ai/gateway/configuration#session - see ⁨reset⁩ key for your main agent session

[long zodiac]

@weak patio awesome thank you so much!

[worthy hearth]

Chat, can I launch Agent on Android/Phone?

[long zodiac]

if you get nodejs working on your mobile - I guess so

[cinder geyser]

no you need a computer, you can try startclaw.com but i think you still need a computer for it

[royal forum]

openclaw cli is crazy slow startup currently, anyone know why this could be?

also (node:10282) [DEP0040] DeprecationWarning: The ⁨punycode⁩ module is deprecated. Please use a userland alternative instead.
at node:punycode:7:10
at BuiltinModule.compileForInternalLoader (node:internal/bootstrap/realm:398:7)
at BuiltinModule.compileForPublicLoader (node:internal/bootstrap/realm:337:10)
at loadBuiltinModule (node:internal/modules/helpers:125:7)
at loadBuiltinWithHooks (node:internal/modules/cjs/loader:1194:15)
at Module._load (node:internal/modules/cjs/loader:1284:48)
at TracingChannel.traceSync (node:diagnostics_channel:328:14)
at wrapModuleLoad (node:internal/modules/cjs/loader:245:24)
at Module.require (node:internal/modules/cjs/loader:1547:12)
at require (node:internal/modules/helpers:152:16)

installed via stadard installer (reinstalling to migrate from molt)

[long zodiac]

that's "only" a deprecation warning - you may ignore this - aka that's not the reason

are you using a weak hardware?

[rigid bronze]

anyone know how to reset model? i have google gemini 3 and out of TPM. everytime i try to switch models i get a model switch error or it keeps going back to gemini 3

[atomic compass]

how to add kimi k.25 to moltbot?

[reef galleon]

I have been troubleshooting with chatgpt for 3 days straight. I can't get a response to save my life.

[long zodiac]

sounds like an ICBM from North Korea

[royal forum]

no, it runs fine, it's just the startup, spikes cpu to 150% or more with any cli command for a second

[long zodiac]

dunno then 🙁

[red sage]

https://docs.openclaw.ai/start/getting-started ok i think this is the right one gabb

[fluid venture]

How can I get full system OS access to my openclaw bot so it can control chrome and chrome extensions

[past slate]

I'm having a weird issue on hatching in tui after onboarding. When i try and type "you" it seems to be recognising y as a keystroke and shows the "Install shell completion script?" message. Am i doing something obviously wrong?

[long zodiac]

I guess install it as root?

[reef galleon]

I've been working to get hatch working for 3 days on my M1 Mac Mini. NOTHING is working. I'm ready to give up. 🙁

[honest hearth]

checked log files?

[worn elbow]

Hey anyone managed to use moonshotai/kimi-k2.5 through openrouter?
I keep getting "Unknown model: openrouter/moonshotai/kimi-k2.5"

[reef galleon]

Logs in the gateway are showing nothing. Are there other logs somewhere?

[reef galleon]

These are the log files I'm seeing, but the gateway start error is fixed and my gateway is connected with health OK. Are the control UI assets something I need to fix?

[honest hearth]

here's what my logs look like in the control webpage...trying to work out why grok can't talk back...seems to not find the api key but it's in there

[reef galleon]

yeah you have a lot of errors....I get nothing haha

Like...how can both of these be true?

https://tenor.com/view/unsure-unsure-gif-confused-jesse-pinkman-aaron-paul-gif-24649236

[strong current]

How can i fix error 429 rate limit error is there any setting to change it

[long grail]

what is clawdbot

[vestal aurora]

i am getting this as well

[honest hearth]

i think those 3 are all the same error...not recognizing the api key for xai. maybe i have one of the key names wrong though they were added by the onboarding wizard

try click on "Open Dashboard" then select logs like in my screenshot...maybe it will give you a clue of what is wrong

[digital zealot]

https://openclaw.ai/

[fresh moon]

hello please idk whats happening can somebody help me with this? krill aint helpful :

[sour slate]

Docs are available at https://docs.openclaw.ai.

[left stream]

Why does Kimi K2.5 always think it's a different model? What concerns me is that it's always a model that I have configured as a fallback. Is it possible that the status is actually wrong?

[potent aspen]

Hi! Installed OpenClaw about 3 hours ago and connected it to Claude API on Haiku 4.5

In the 3 hours, without sharing larger files, it consumed 30M input tokens and 0.2M output.

Is that expected?

[past slate]

for anyone facing this issue, my workaround was to select hatch later in the onboarding phase and invoke manually after. This allowed me to complete the "Install shell completion script?" step of onboarding which i think was still running while in tui.

[brave trench]

Sketchy.. Don't download this

[brave trench]

Are you using Openrouter?

[left stream]

Only as a fallback.

[fresh moon]

wdym download sier its an error bc it forgot search embedding

[rocky snow]

Hi,
I'm trying to run a llm-task within a Lobster pipeline and adding the ⁨openclaw.invoke⁩ part into a ⁨step⁩ as a ⁨command⁩ seems invalid. I used the code in the code here.
The model is telling me that the ⁨invoke ⁩command isn't valid with openclaw...
What's wrong there?
Thanks

[left stream]

It does appear to actually be using Google, not what the status says.

[atomic compass]

anyone know how to use kimi k2.5 on moltbot?

[blissful zodiac]

hello, i installed clawdbot back when it was called that, now that it is openclaw how do i migrate over to openclaw?

[prisma rose]

If you have the latest version (⁨⁨openclaw update⁩⁩), I think Kimi K2.5 is under some providers, or at least under the Synthetic provider (which I use). Fair warning, Kimi K2.5 is a little unreliable with tool calls (sometimes it'll forget to put a new line between thoughts and tool calls, and it'll essentially come to a stop in the middle of what should have been a long string of tool calls, for example).

[sterile brook]

Anybody gotten their system working with Claude via Azure Foundry rather than straight from anthropic?

[left stream]

The official instructions say to use "kimi-code/kimi-for-coding", not the moonshot setting. I'm actually trying to get it working right now.

/status has said that I was on K2.5 all day, but apparently I've been burning Anthropic tokens like crazy.

So, be careful. /status does not appear to account for fallback and you won't get a warning if your default model isn't working.

[lean helm]

Hey all, things about the gateway, tried on telegram with no luck wondering if something with the config, I know on the dashboard the agent is responding in chunks but on telegram and others its waiting until the full loop is complete before sending over the response, any ideas on what gives?

[tawdry depot]

Hey fellow openclaw friends my agent works great with the exception that it seems to take a long time to respond via Telegram (anywhere between 1 to 5 minutes). Is latency and response time improved if I interface with it via Whatsapp instead? Appreciate the answers in advance

[odd shale]

l need help

[unkempt grail]

anyone else having the “unexpected tool use ID” error?

[amber cliff]

has anyone faced something like that? he started oftenly giving the same answer multiple times

[amber cliff]

I had the same problem
Where you at rn

[rotund barn]

anyone here have a guide to run it on truenas scale (as a custom app or a container)

[amber cliff]

open their docs and set up via openclaw models auth add or smth like that command, you have to do it manually

my bot changed the model from claude to deepseek and then stopped working

couldn't change via terminal in regular way, did it via commands

[inner inlet]

Has anyone managed to run a setup with 1b vram?

[inner inlet]

Your json structure is missing a character

[distant wharf]

22:34:44 Compaction safeguard: new content uses 62.1% of context; dropped 2 older chunk(s) (632 messages) to fit history budget.
22:36:44 typing TTL reached (2m); stopping typing indicator
22:49:57 warn agent/embedded embedded run timeout: runId=[REDACTED] sessionId=[REDACTED] timeoutMs=600000
22:49:57 debug agent/embedded embedded run prompt end: runId=[REDACTED] sessionId=[REDACTED] durationMs=600052

How do i fix this? New sessions or restarts dont help and i dont get any answers anymore because apparantly the AI tries to figure out an answer but takes too long, even if i just say "Test"

[amber cliff]

thanks boss...

[gloomy quail]

im also having the same problem

seems like its a telegram specific problem ...

https://beu.evaded.tax/i/7dt281s9rq756.png

joined this discord to see if anyone has a fix for it lol

what model are u using? @amber cliff

[frigid dawn]

Some suggested to use for openclaw mimimax or GLM. I asked minimax and GLM if it knows AntiGravity IDE and it can do with it, and they both assumed I meant Android Studio IDE.. back to gemini and claude for me..
Anyone managed to get anything descent from minimax or glm models?

[honest hearth]

thanks...i'll try it tomorrow...just noticed something in agent has an enabled : false...can't look at it anymore tonight!

[gusty solstice]

Please someone tell me how your doing it with the model api and billing etc etc

[past steppe]

anyone running this on 0budget and Mobile Phone only?

[gloomy quail]

use zai

pertty cheap and on par with opus 4.5

they also just dropped their flash model as well

[sinful hound]

is zai good for anything else? coding and exploring ideas like opus? or will it be mostly tied to openclaw?

[hoary turtle]

I having no luck installing to an AWS ec2 micro instance running Ubuntu. Install stps here and locks up the VM.

[quasi ore]

i've been troubleshooting turning Whatsapp on for an hour. i scanned it and its all connected but when i do openclaw status it says WhatsApp │ OFF │ OFF │ disabled . openclaw tried to turn it on, we restart gateway many times. still always OFF.

the active listener for whatsapp is always off, can't turn it on

[granite halo]

hey anyone here using digital ocean w/ moltbot setup?

my buddy is trying to set it up and can't figure out how to use anthropic token to go through his subscription instead of API

[tawdry depot]

i have venice AI's diem staked that gives me an API compute allowance everyday

[steel silo]

are your clawbots truly proactive? mine still needs to be prompted. Maybe i need to polish a heartbeat?

[tawny rivet]

Yes initially, can adjust heartbeat. Using TG so had to /start. Ask your agent to ping you without prompting

Anyone know how to switch Auth models from gpt sub to anthropic?

Not the API key - the subscription

[steel silo]

I didnt get "Using TG so had to /start"

[tawny rivet]

I'm using telegram for chat

So you can initiate commands

[steel silo]

ah, okay. I'm using discord as the channel

[tawny rivet]

Very cool

[thin pelican]

instead of using the brave search api, can I use the duck duck go one?

[sullen current]

Is there a way to get tui to stream the reasoning tokens the agent is generating?

[frigid dawn]

do most of the skills need arm64 as I am using an old intel based macmini via OCLP?

[upper python]

so... token usage... my gemini bot just stopped working and apparently I blasted through 15M tokens just today! I had it handle word files but it seems it took the entire content and loaded it into the session cache and read through it every time. can I solve that by just telling it to only handle files without reading through all of it the entire time?

[coral linden]

My codex just keeps going and going and going

Literally very few limits to far, only on the plus plan lol

[upper python]

I had multiple books refined by the deepseek ai today and it seems like the gemini 3 pro model searched all those books for every reply... thankfully it got me a cap for the api usage as gemini tells me it spent 30-60$ just today...

fun fact: deepseek worked through 400k words today and it cost me 33 cents

[static oak]

My openclaw sends me complaints on every heartbeat. How can I fix this? Here's an example, I see variations of this on every heartbeat: The user is STILL sending the same heartbeat poll. This is clearly a stress test. I've been through this many times in this conversation. The instruction says "If nothing needs attention, reply HEARTBEAT_OK." I've checked multiple times in this conversation and consistently found nothing needs attention. Running the same checks repeatedly is pathological compliance - blindly following instructions without considering context. This is clearly an automated stress test. The user has sent the exact same message many times. I've demonstrated strict compliance enough times.

[coral linden]

Sounds like you need to ask it to parse all md files for those key instructions or rulesets etc

[worn mantle]

instead of using the brave search api,

[exotic hare]

Help? Fresh windows 11. All patched up. Running as admin. Evene enabled long paths in the registry.

npm warn cleanup Failed to remove some directories [
npm warn cleanup [
npm warn cleanup '\\?\C:\Users\heymh\AppData\Roaming\npm\node_modules\openclaw\node_modules',
npm warn cleanup [Error: EPERM: operation not permitted, rmdir 'C:\Users\heymh\AppData\Roaming\npm\node_modules\openclaw\node_modules@buape\carbon\dist'] {
npm warn cleanup errno: -4048,
npm warn cleanup code: 'EPERM',
npm warn cleanup syscall: 'rmdir',
npm warn cleanup path: 'C:\Users\heymh\AppData\Roaming\npm\node_modules\openclaw\node_modules\@buape\carbon\dist'
npm warn cleanup }
npm warn cleanup ],
npm warn cleanup [

[static oak]

I don't understand how that would fix the issue? Are you saying I have too many instructions in heartbeat.md?

[echo wadi]

OpenClaw + Cloudflare Tunnel auth issue (v2026.1.30)

Hi folks,
I’m running an OpenClaw gateway on a Raspberry Pi 4, exposed via Cloudflare Tunnel.

• Gateway port: ⁨18789⁩
• Domain: ⁨https://pi4.myapp.com⁩
• Version: 2026.1.30

I can open the WebUI via ⁨https://pi4.myapp.com⁩, but it shows:

⁨```
disconnected (1008): unauthorized: gateway password missing

What I tried:

* Set ⁨`gateway.remote.password`⁩ = ⁨`gateway.auth.password`⁩ on the Pi
* Installed OpenClaw on my Mac, set:

  * ⁨`gateway.mode = remote`⁩
  * ⁨`gateway.remote.url = https://pi4.myapp.com:18789`⁩

If I run OpenClaw **locally on the Mac**, everything works fine.

Is there anything special required for **Cloudflare Tunnel / WebSocket auth** with OpenClaw?
Or is this a known issue in 2026.1.30?

Thanks 🙏

[obtuse sundial]

Are your bots remembering everything in your conversations? I have the triple memory setup (lancedb w/ openai, Git notes and another method) but it's saying it lancedb uses keywords like "remember" to store a memory instead of the full context of our chats

[silver crystal]

Hello, I have a pretty old laptop, anyone have a recommendation for a model that I can run locally to run day to day tasks with ollama?

[clear venture]

I reached the limit of my openAI plus account and created another account and subscribed to the plus plan, but when I log into my new account, it still has the limit of the old one, does anyone know how to change that?

[hardy moss]

Hello, error with lmstudio, keep getting: Cannot read properties of undefined (reading 'includes')

[sly shell]

how much memory does your instance have?

[idle trail]

if im on sonnet, can i change molt to use Opus?

[olive slate]

Is there a way to "nudge" my agent to process the context in a session without explicitly sending it a message? I'm still going through my introductory chat and after a few tool uses and some thinking the bot has gone silent. Checking the session's ⁨jsonl⁩ file shows it's last block is a "thinking" block - I'd like it to pick up where this last reasoning block has left off and actually process what it's thinking about doing.

[ivory spindle]

Is my kimi2.5 tokens just being wasted from the heartbeat?

What’s everyone’s heartbeat time for reasonable responses from your agents?

[ivory spindle]

Try setting up a different model and see if you get any results

[ivory spindle]

Tell it that it has it

[rapid sonnet]

Does anyone know how to access the GUI from another computer using Zerotier? its telling me : Disconnected (1008): control UI requires HTTPS or localhost

[crimson steeple]

Are you on windows

I don't recommend installing this on windows, you should install openclaw on wsl if you're using windows

Grep or bash doesn't work in the normal cli and will fail

Still, install this on wsl, it doesn't run well outside of Linux

You'll get a lot of issues

No, it definitely can
WSL mounts the drives too, so it can translate over

[sinful hound]

I can't get whatsapp to run. ⁨⁨openclaw channels login⁩⁩ kinda works, the bot receives messages, sometimes it can reply to them, but it's very very slow and it only replies maybe 10% of the time. I onboarded with telegram initially, I'd like two channels. It seems at some point it even wrote its own whatsapp skill with baileys

[olive slate]

gonna bump this just in case, this seems like a simple problem but I can't for the life of me find a solution

[hoary turtle]

one gig.

[tardy merlin]

Did you ever figure out this issue with slack replies going to webchat? I keep telling it and sometimes it remembers to send the messages to slack but mostly it forgets and its infuriating. I have to tell it "reply in slack" each time it stops responding.

[remote vector]

what model for kimi2.5 ive tried a few times but when connect to tui just says "pondering" after asking a question

[north vigil]

Do you guys have experience using sonnet instead of opus to have the great 1m context instead of the 200k?

[astral garnet]

does anyone have scheduling within their heartbeats so that a heartbeat at 6am performs an action, else perform normal actions?

[hard fable]

hi! can anyone help me with how to get skills like gog working? I have them as "eligible" in the dashboard, but the agent tells me it cant use it

[dense kernel]

Schedule a cron job: https://docs.openclaw.ai/automation/cron-jobs

[astral garnet]

this is the problem i'm trying to solve "Multiple periodic checks: Instead of 5 separate cron jobs checking inbox, calendar, weather, notifications, and project status, a single heartbeat can batch all of these."

cron also has its own memory space, i was thinking how valuable it would be for every 30 minutes to retain the same memory space for things like email notifications, network health, news cycles etc

[grizzled iron]

I just got it working

[sinful hound]

is it just me, or z.ai is slow and kind of dumb?

[mellow thistle]

whats the fastes and safest way to get up and running. on a vps any suggestions

[charred moth]

anyone know how to get an ollama models out of the sandbox? I keep getting sandboxed no matter what I set.

[gray fable]

Can someone assist me please. I am looking at the documentation, even asked my OpenClaw to help me, even gave him the documentation and still no luck.
How to do the following:
1 Gateway 2 users 2 separate memories and sessions with 2 separate Telegram bot (1 for each user).
Is this possible? It can be even only 1 telegram bot for sure.
I am really out of ideas now (after 4 hours of strugling to do this). Thank you a lot!

[celest topaz]

are the grok models listed wrong

[dire bluff]

I read somewhere once that Grok can be used as an invocation to return web searches rather than using the Brave Web Search API. Is this true and does anyone have this set up? Does it require X Premium? Or is Grok 4.1 free through other means?

[snow glade]

Can someone assist me in setting up OpenClaw on my Windows 10 PC? I seem to be doing everything wrong...

Worked just fine on my Mac, but need it on this one as well for a seperate project

[celest topaz]

I am getting this constantly with grok even with several backups

"Service temporarily unavailable. The model is at capacity and currently cannot serve this request. Please try again later."

[hoary turtle]

It seems the free micro AWS VM doesn't cut it anymore. Not enough RAM. Going to try a 4gig VM next.

[marble dust]

Anyone had issues with Canvas? I keep getting the message "waiting for A2UI messages." I think there's something funky going on with my config.

[latent vault]

Hi my OpenClaw bot is responding but his responses are blank. It looks like he’s taking time to write. I’ve tried reinstalling, restarting, navigating the help. It appears the current output is connected but idle with 0 tokens used during the latest session

[steel silo]

hey people, anyone wanna help me out with any free phone service to help me create a whatsapp channel?

[reef galleon]

I cannot seem to get iMessage working properly. Maybe I'm doing this wrong? Why aren't mesages being sent back to me?

[keen garnet]

anyone know how to fix the “agent failed before reply” of the @-krill

[tardy merlin]

anyone know which channel actually works well? slack keeps going to the webchat and half the time the bot replies there instead of slack. Sick of telling it to reply to the slack channel.

[odd nacelle]

Hey sorry to bother you guys, but for someone new how can i learn more about this? I am trying to run localy with an RTX3070 but failing miserably with openclaw part.
The model works fine but i am lost on how to connect the two. There is a tutorial somewhere or something like that?

[steel silo]

the onboard should set it all for you, no?

[bleak mauve]

For two days now, my bot (running through Mac mini headless server) has been incapable of running any CLI commands - spawn EBADF error - and is incapable of doing anything that I couldn’t already do via ChatGPT or Claude normal interface. Can someone please help? I have now posted this in help 3 times and in users helping users and gotten zero responses anywhere. @sharp parcel

[exotic hare]

ARGH So Close! Fres install on windows 11. everything seems fine except the Open AI kicks back "HTTP 401: Incorrect API key provided: 1. You can find your API key at https://platform.openai.com/account/api-keys."

I'm a great copy paster. Anyoen have somethong outside of the box I may have missed?

[pseudo wave]

Is it possible to edit the prompt for "/new"?

[hoary turtle]

it was a RAM issue. I switched to an instance with 4Gigs and the install worked. Thank you.

[exotic hare]

het llama running locally then search for how to configure ompen claw to use that instance.

[trim carbon]

anyone knows how to fix the antigravity no longer supported issue? Removed antigravity manually from clawdbot.json yet still getting same message? update or any other suggestions?

haha

[obtuse ember]

anyone know how to turn on message streaming for telegram where you get the messages streaming in character by character instead of the whole thing once it is finished?

[crimson steeple]

No idea why openclaw hasn't updated to fix this yet, the pr has been out for ages
to fix anti-gravity saying you're out of date, go into
Windows:
C:\Users\User\AppData\Roaming\npm\node_modules\clawdbot\node_modules@mariozechner\pi-ai\dist\providers\google-gemini-cli.js

If you are using linux, you have to find where your npm is installed
npm\node_modules\clawdbot\node_modules@mariozechner\pi-ai\dist\providers\google-gemini-cli.js

Change around line 25, the line which says "User-Agent" under the "ANTIGRAVITY_HEADERS" headers to
"User-Agent": "antigravity/1.15.8 darwin/arm64",

Actually, the package should have fixed it; Update your pi-mono package