users-helping-users

[solar maple]

If no abnormalities happens, I think you're good. You mentioned that you installed it twice, I strongly suggest you to check any dead session or duplicated gateway session is running before you move forward.

I also suggest you to state your demands (specifically) or I really don't know which kind of instructions you're looking for.

For email, sorry I didn't try it yet; for Google Chrome, just let your clawbot to use browser skill and it can open a headless browser and do stuffs. It might need you to install a browser extension called Openclaw Browser Relay or sth similar. After that, you might found it can't open automatically (which I experienced), so you can ask your clawdbot to download it for you on your PC and modify the code (till it runs automatically on browser).

Last but not least: try to use better models but not cheaper one like GLM-4.5. (I really spent a bounch of time on this dude)

[civic storm]

Hello, which LLM is the best one to run OpenClaw with? At no costs

Gemini, Claude?

[void tiger]

At no costs?

I dunno that anyone has a way for Gemini or Claude at no costs...

[obtuse sundial]

This sucks man you really do have to spend hundreds of dollars on an codex to make this worth it

Tried to make it work with minimax and Kimi but they're total fucking morons man

Every time I get something fixed with codex and inevitably get limited and have to go back to minimax somehow it manages to rebreak something every single time

At this point I'm just chasing errors left and right every time I set one of the other llms on a task

[lost herald]

Personally, I've had a decent experience with minimax. It took me some trial and error, and learning how to guide the agent better, but it's work quite effectively now. I wouldn't give up on it yet!

[civic storm]

I know Gemini has a very generous free tier

But overtime the context window is gonna be bigger

[sour slate]

@mirkomo muted

Reason: Spamming across channels
Duration: 14 minutes and 36 seconds

[civic storm]

And paying, what would u suggest me?

The cheapest I can pay

[ionic lantern]

Kimi 2.5 is completely free through the Nvidia Cloud account. Drawback, Kimi isn't great when compared to gpt-5.3 codex and opus-4.6, but free is free!

[gray jewel]

Anyone know how to fix this issue:

"LLM request rejected: messages.25: tool_use ids were found without tool_result blocks immediately after: functionswrite11. Each tool_use block must have a corresponding tool_result block in the next message."

Got this issue yesterday when switching from Kimi back to Anthropic models. Tried Gateway Restart and Doctor - didn't fix it.

[civic storm]

Awesome, I found OpenRouter, it looks great for better cost efficiency. So I was thinking of running simpler tasks locally with Ollama.

[vernal star]

cool, thanks dude, will try

[wary osprey]

If you mean literally zero cost, then your only real option is running a small local model like TinyLlama or Gemma on your own machine.

Claude and Gemini both cost money once you go through API. Free web access doesn’t translate to free OpenClaw usage.

If you’re experimenting and want to keep costs super low, I’d suggest either:

1. Small local model for basic flows
2. Or use something like AIsa’s unified API — they give $5 free credit when you sign up, so you can test different models without juggling providers or committing to a full subscription.

Totally depends on what you’re building though

[floral epoch]

hi may i ask if i install claw in docker. the list of built in skill like clawhub-cli . all these skill can not be install as apt not able to install node and brew? any other way to install it and use it

[candid socket]

setup a custom Dockerfile with all your binaries being setup within then build using it

[winged raven]

A quick question - if I run openclaw via railway (gh deployment) can the agent still make changes to its own code (since I did not fork the project and deployed it directly)?

[sand crystal]

I run in docker, some issues around permissions if you are also working in an IDE (e.g. user in docker compose). This is mine to get brew and stuff and user permissions (instead of agent creating root stuff):

FROM node:22-bookworm

# Install dependencies for the installation script and additional tools
RUN apt-get update && apt-get install -y --no-install-recommends \
    curl \
    git \
    ca-certificates \
    ffmpeg \
    golang \
    python3 \
    python3-pip \
    python3-venv \
    && rm -rf /var/lib/apt/lists/*

# Install uv (modern Python installer)
# Use official uv image to get the binaries - clean and reliable
COPY --from=ghcr.io/astral-sh/uv:latest /uv /uvx /usr/local/bin/

# Set environment variables for the installation script
# Non-interactive mode and specifying the install method
ENV OPENCLAW_INSTALL_METHOD=npm
ENV OPENCLAW_NO_ONBOARD=1
ENV OPENCLAW_NO_PROMPT=1

### Persistence paths inside the container
ENV OPENCLAW_DATA_DIR=/app/data
ENV OPENCLAW_CONFIG_PATH=/app/data/openclaw.json

### Run the OpenClaw installation script
RUN curl -fsSL https://openclaw.ai/install.sh | bash

### Ensure the 'node' user exists (it comes with the base image)
### Create data directory and set permissions
RUN mkdir -p /app/data && chown -R node:node /app

### Switch to non-root user
USER node

### Set working directory
WORKDIR /app

### Expose the default OpenClaw port
EXPOSE 18789

### Start OpenClaw gateway
### Note: The install script adds 'openclaw' to the path (usually via npm global install)
CMD ["openclaw", "gateway", "--allow-unconfigured", "--bind", "lan", "--port", "18789"]

[shadow goblet]

i hate the setup i am a non techi tryed to set it up via vps + docker + private vpn tried for idk 10 hours but i am only running into some error codes everytime i retry? whta the setup i just use i?

[wet elm]

Hey all. I am trying out openclaw finally. I have done a lot of agentic coding. I am very critical of all the security flaws though. Is there a way, yet, to securely manage secrets? I know we can not 100% stop prompt injection attempts just yet, but if there is at least a way to give it minimal attack surface on the secrets to the tools?

[sand crystal]

I find accessing from private discord keeps prompt injection at bay, for email use access lists (only respond to friends list if that is why you need an email reader). It depends on your use case but if the goesinta is protected by user account it feels decent. Secrets should be separate (like don't push openclaw.json), I'm thinking of keeping that as root permission only accessible during startup/registration so the bot can't get to it

[dire widget]

1. Become a sysadmin for many companies
2. Get them all on Google Workspace
3. Make an admin account for each workspace
4. bang all the API keys together
5. reveal in the dishonesty

[dire widget]

no

[sand crystal]

the docker example above seems to work, root then user. keep openclaw.json as root and then the agent can't read even if prompt injection

[kindred coral]

ive been trying to fix this error for 5 hours my blood is boiling

pleeeeease anybody help me

[wet elm]

That is not a bad Idea Charlie. I was thinking of having the secrets encrypted on the host. The bot should only need access to them when it needs a reauthentication. I have built something similar to open claw out on n8n. It this smart gateway really does seem a lot nicer. I am also going to run it in a jailed vlan and only give it explicit access and no access to clawhub.

[sand crystal]

encrypted too, i like that... might be in my future

[wet elm]

@sand crystal Could possibly run an instance of hashicorp vault on the host and the you are only locking down one key at that point. I was hopping there was built in support for this by now though.

[kindred coral]

im getting 401 invalid authentication everywhere😢

on telegram, on the dashboard..

[shadow goblet]

i hate the setup i am a non techi tryed to set it up via vps + docker + private vpn tried for idk 10 hours but i am only running into some error codes everytime i retry? whta the setup i just use i?

[kindred coral]

pleaaaaaaaaase somebody help me i keep getting gateway connect failed: Error: pairing required and 401 invalid authentication but i already paired my tg account 😢😢

[shadow goblet]

got that to many times i just restared everything

[wicked moth]

#1459642797895319552 #claw
Please help I. Trying to download file from a link. But for some reason is not opening. The link is openclaw/workspace

[kindred coral]

im actually so fucked claude is like “its fixed😄😄 everything is healthy, telegram connected, agent running and sessions intact” and some bullshit but its the exact same

[kindred coral]

ive already restarred many times

[shadow goblet]

nah i mean completely delting everything and then setting it up new

[kindred coral]

ohh i did that like 100 times too

wait i think the error ha sbeen found

[near forum]

Gemini 3 Pro is no longer available
any fix

[kindred coral]

“the issue is that your jey is a nvidia build key (nvapi-…) but openclaw is configured for the moonshot provider, which points to moonshots own api (api.moonshot.cn).” coukd that be it? but how else woukd i add it other than that lol

[lilac swan]

How do you guys automate stuff. My openclaw is forgetting to start automated tasks. And if I don’t say anything. It just doesn’t do anything all night. Even though I have the heartbeat setup

[kindred coral]

he is sleepy

motivate him

[dull spruce]

do you have the dashboard setup? can you see the cron jobs?

[lilac swan]

Mainly on telegram after I set it up

[dull spruce]

you'll get much more visibility if you setup the Node.js web console

[lilac swan]

Ooo how do I do that

[main flame]

Is the Slack integration completely cooked for anyone here as well?

[quiet edge]

I got myself a ChatGPT business license at $150/month with the understanding I could use oauth login in the openclaw config.

I am hitting rate limits like every 30 minutes. Is this the proper way to use OpenAI or do I need to downgrade and get api tokens?

What a shame no more Anthropic - (burned my tokens on the 200/month plan in four days)

[dull spruce]

what version are you on?

[lilac swan]

Latest

[dull spruce]

on a local machine on your lan?

[lilac swan]

Local

[dull spruce]

if you're on the machine it's on... try wss://0.0.0.0:18789

[lilac swan]

No openclaw command ?

[dull spruce]

i mean, it'd be worth seeing if it's already up before we gothere

latest version is all locked down, so with security comes complexity

i think it defaults to on with the loopback address

openclaw status --all

look for "Dashboard"

[lilac swan]

it’s working. Thank you

[dull spruce]

ok awesome 🙂

[lilac swan]

What skills u guys using ????

[kindred coral]

now im grtting HTTP 404: 404 page not found

bohoo

[plush tide]

What are you guys doing to burn through so many tokens - I’m still running off oauth on the basic max plan and still running claude desktop coding/debugging and other tasks on the side

I feel like my luck is about to run out soon…

[quiet edge]

I’ve been building out an agency of 10+ agents to run my nonprofit business

So far I have saved like 40 hours a week, but changing providers has thrown a bit of a wrench

[dull spruce]

yeah, i'd snapshot the OS before you change providers

[quiet edge]

We got blender working, and our students are able to interact with it on the web

[dull spruce]

my bot went fully insane

[quiet edge]

We have great backup procedures thankfully we taught the bots

[dull spruce]

good...

[quiet edge]

Right now we’re running on ChatGPT and Google with local fallback might just bite the bullet and switch over to anthropic API

[kindred coral]

this is torture hoooooly

[plush tide]

Ok - this is a serious operation. I bow before you and hope you don’t jinx my little set up. 😂

[quiet edge]

This is all on the Mac mini while our Mac Studio is built. Next we’re building our own access to a visual scene builder

But yeah - so people are going the api route it seems instead of oauth to avoid the hard caps

[plush tide]

This is kinda cool. Hadn’t really considered the possibilities for nfps. Nice one!

[quiet edge]

Oh gosh this is the best thing that’s happened operationally To us in eleven years of service across 12 countries and 2000 students.

Replaces social media, lead/donor gen, grant pipeline, donor management, volunteer and staff coordination, staff reports kpis via WhatsApp so no more lost data bc they don’t go on Monday.com etc.

Building donor reports in seconds. Dashboards in seconds.

Truly has us amped up!

But we started with the $200 max Anthropic plan and we run through that in 3-4 days as we’re polishing setup

Not to mention 3 million photos and video files that will get key worded and tags and made useful again!

[dull spruce]

Do you guys use Raiser's Edge or some other CRM that this will replace?

[quiet edge]

Salesforce and Monday crm plus Mailchimp. Not replaceable yet

Can source opportunities from Candid or the web. It’s wild

[timber gyro]

hello all!! installed it on a VPS - when selecting the skills on openclaw onboard - many of them require "brew" and homebrew is not installed, and not adviced to be installed. how do you overcome it?

[quiet edge]

Just do it or roll your own

Your security is by not feeding it anything you don’t want to share with the world unless you’re local

[timber gyro]

thanks

[thick jewel]

Custom Skill not workink on Raspberry

[kindred coral]

my beloved openclaw is typing infinitely

i get no error code it just thinks

[stone sinew]

I have a cron job that posts a daily update in a slack thread. As of one of the more recent openclaw updates, it now posts into a thread of some unrelated prior message rather than a new post. How do I make it post? Note it is an isolated cron job so shouldn't have any prior session context driving this.

[half latch]

I also installed on a VPS (ubuntu) and hat to install a couple binaries after is started using openclaw. So far I installed jq and rg.
I noticed a couple things:

1. It sometimes sidestepped the fact that it can not install with the package manager with cloning the GitHub repo. Smart, but could be bad
2. I had to specifically add every binary to the TOOLS.md to make it remember where the binaries reside.

[young nebula]

I keep getting this on telegram even after I don’t even have deepseek-reasoner in my config json anymore. I switched it to chat but it keeps popping up

⚠️ Agent failed before reply: Unknown model: deepseek/deepseek-reasoner.
Logs: openclaw logs --follow

[umbral stratus]

hello everione i left a help rquest here https://discord.com/channels/1456350064065904867/1474443291008303266 if someone take a look

[quasi ore]

anyone using geminiCLI oauth, have issue getting

Cloud Code Assist API returned an empty response

[dull spruce]

try openclaw models set anthropic/claude-3-5-haiku-latest

then openclaw gateway restart

[indigo nebula]

Ask your openclaw to fix it. Tell it what it's doing wrong, give it link to an example, then tell it where it should post. It will correct itself then tell it to remember this and only post where you requested it to.

[umbral stratus]

same response HTTP 404: not_found_error: model: claude-3-5-haiku-latest (request_id: req_011C

[dull spruce]

openclaw models list --provider anthropic

[indigo nebula]

did you restart the gateway? openclaw gateway restart to force it to reload the config?

[tardy pollen]

Not sure if this is the right channel for this - should workspace files
(SOUL.md, AGENTS.md, etc.) be edited manually by the user, or should the agent
modify its own files through conversation? Specifically, is it fine to remove
or change sections from the initial templates, or is it better to let the
agent evolve them on its own?

[indigo nebula]

It's your choice, if you edit them by hand make sure you restart the gateway so it picks up the changes or at least tell your openclaw to read the files again because you modified.

[tardy pollen]

Is there any official documentation regaring its a good or bad pracitce? I guess those templates was taking time to write and find the right ones - so If I will chnage this file and remove some sections isn't it a bad practice?

[indigo nebula]

Ask your claw to point you as some 'best practices' lots of people out there sharing lots of ways. But make a copy of the file first so if it doesn't work you can revert back. Other useful docs at https://docs.openclaw.ai/

[umbral stratus]

for your knowledge claude-3-5-haiku-latest was changed it to claude-haiku-4-5-20251001

[dull spruce]

glad you got it working

[tardy pollen]

I was reading all the docs and didnt found anything regarding this so thats why I cam here to the community. I asked my agent but he also didn't knew to say what is good practice or bad practice regarding editing those Bootstrap files (Soul.md, agents.md etc)

[young nebula]

Yes. And I do a ctrl+F and search deepseek in my config json and nothing comes up at all. I completely removed all deepseek

[indigo nebula]

There's no single answer. It depends on what you want your openclaw to do and how to behave. You talk it through chat and ask it to add certain things or remove others. Or you can do it by hand. What they give you is a basic default setup. It's up to you to make it better

[young nebula]

It’s like telegram isn’t picking up the new info

[tardy pollen]

Got you so there is no actual good practice and bad practice (yet)

[trail cedar]

GM gents

If I add a skill to my workspace, do I need to list a reference to it in TOOLS.md?

[indigo nebula]

Did you try forcing a new session in Telegram? /new to purge all the old context and see if that helps

[tardy pollen]

@indigo nebula Is there specific experts in this community which its worth to talk with them?

[young nebula]

Yes same thing. Like it only lists 3 models and only 1 is one I even have in my config currently. And it’s missing like 3 in my config

[dull spruce]

software is some 90 days old... best practices?

[indigo nebula]

Ask your openclaw about it. Tell it what you updated and that you don't see the new models and to debug/fix it. I did something similar when we removed a free model that wasn't very useful but it was still in the list. I told claw to figure it out why and remove it, it did

[dull spruce]

experts standing by??

just backup the file and make changes how you see fit... move on

[tardy pollen]

I know that I can backup the file and make changes... I thought that there is any good practice by telling the agent to change those files on its own instead of overriding things which was hard to find in order to build those files. Anyway thanks for the answers

[indigo nebula]

NP. The system is super flexible you can ask it or do it for it. Makes no difference but when you ask it to do stuff it will use tokens against whatever model you use so keep that in mind. For basic editing i do it myself for more involved stuff I just talk to it.

[quasi ore]

Why i cant send image file even at root folder? All folder is not save? any one have same issue!

[bold skiff]

What's the best way to handle memory for my agent? It keeps on forgetting things, both between sessions and stuff that I told it to remember?

[indigo nebula]

Tell your openclaw to install QMD https://github.com/tobi/qmd and to start saving it's memories to files. Also tell it to add a directive to it's settings to "search local memories for what we were doing the previous day or after a new session begins". I also tell it to do a daily summary and review that at startup each day.

[sinful forum]

https://code.claude.com/docs/en/legal-and-compliance#authentication-and-credential-use

you still have that issue? i had this error from time to time but now it's constant, claude says they're restricting this now

[quiet edge]

So where are we moving to? OpenAI api? Bc it doesn’t seem anyone’s subscription offerings is working

[worldly shard]

Does anyone know how frequently the batches for the iOS test flight invitations are sent out?

[gaunt rover]

i tried the openclaw devices list command and its giving error

[lost wind]

Find out why!

[gaunt rover]

the cli is failing to start

[velvet sail]

my subagent's acting up after the update. Main agent is throwing this log: gateway closed (1008): pairing required. Seems like a loopback issue on port 18789. Any fix for this?

[dull spruce]

can you paste an error message

[gaunt rover]

[openclaw] Failed to start CLI: Error: gateway closed (1008): pairing required
Gateway target: ws://127.0.0.1:18789
Source: local loopback
Config: /root/.openclaw/openclaw.json
Bind: loopback
at Object.onClose (file:///usr/lib/node_modules/openclaw/dist/call-DZzTR0NL.js:438:10)
at WebSocket.<anonymous> (file:///usr/lib/node_modules/openclaw/dist/client-DoSZFvf3.js:1538:23)
at WebSocket.emit (node:events:519:28)
at WebSocket.emitClose (/usr/lib/node_modules/openclaw/node_modules/ws/lib/websocket.js:273:10)
at Socket.socketOnClose (/usr/lib/node_modules/openclaw/node_modules/ws/lib/websocket.js:1346:15)
at Socket.emit (node:events:519:28)
at TCP.<anonymous> (node:net:346:12)

[velvet sail]

uh i got same thing

[dull spruce]

its only bound to the loopback

can you access the UI on the machine its running on?

[empty lantern]

had a lot of trouble w my setup + claude... switching to OpenAI Code (Codex) subscription as model removed a lot of pain for now... seems like it would make sense for OpenAi to have their model work well w the claw

[gaunt rover]

i cant access the tui since gateway is not connected

[quiet edge]

Is this through your ChatGPT subscription or through the API?

[quiet edge]

Did you run the doctor?

[empty lantern]

just subscription like mentioned in the docs

[quiet edge]

I get rate limited almost immediately with my subscription. What model are you on?

[empty lantern]

that was my issue with the Claude API even though i had plenty of credits left

[sudden cargo]

@sharp parcel i cant get the cloudflare setup to work no matter what i do...i set the right env variables for anthropic, but it's just not returning any responses and i dont know how to update the config while it's in cf (i dont have it on my local)

[empty lantern]

i setup option B here and just have a regular plus subscription
https://docs.openclaw.ai/providers/openai

[quiet edge]

I'm the complete opposite. My Claude subscription worked fine. I'm just rate-limited there but my ChatGPT subscription doesn't seem to be working very well.

[empty lantern]

😂

[gaunt rover]

what should i look for from the result

[empty lantern]

so whack

[dull spruce]

you can't run openclaw commands at all? openclaw gateway status

[quiet edge]

The results are almost immaterial. Try openclaw doctor —fix

[gaunt rover]

Warm-up: launch agents can take a few seconds. Try again shortly.
RPC probe: failed
RPC target: ws://127.0.0.1:18789
gateway closed (1008): pairing required
Gateway target: ws://127.0.0.1:18789
Source: cli --url
Config: /root/.openclaw/openclaw.json

Port 18789 is already in use.

• pid 1704 root: openclaw-gateway (127.0.0.1:18789)
• Gateway already running locally. Stop it (openclaw gateway stop) or use a different port.

[velvet sail]

i can btw my subagent output like these

gateway closed (1008): pairing required
Gateway target: ws://127.0.0.1:18789 | Source: local loopback | Config: ~/.openclaw/openclaw.json | Bind: loopback

[dull spruce]

openclaw running as root?

[quiet edge]

Openclaw gateway restart

Sheesh

Run the doctor

[dull spruce]

Config: /root/.openclaw/openclaw.json

[quiet edge]

Openclaw config can help too to rebind gateway

[timber gyro]

not ideal, right? too much work. not ideal to have the "default" skills nor running in a so regular setup such as a vps

[brisk pumice]

anyone else still dealing with repeat replies in Slack?

[velvet sail]

What kind of config setup do I actually need for my subagent? Like, what specific lines am I missing here to fix the pairing?

[brisk pumice]

i had this issue. dashboard may show pending requests. if not, need to force allow all temporarily

[velvet sail]

okay i will look dashboard

[trim osprey]

i just switched over to $200 plan gpt do i have access to all models with that?

[solemn meteor]

Has anyone used 3.1 pro with openclaw yet I’m going crazy trying to figure out how to

[vital raft]

I have an openclaw setup on a ubuntu server, freshly installed, with 3 local ollama models. When I query the models, I get an answer. When I try sending a message, through the gateway, the request times out. I see no errors. Anyone has any ideas on what might be happening here? I tried using Grok and Claude but at a certain point they just started going around in circles with the same suggestions.

[native verge]

@alp for subagent pairing - make sure both configs have same gateway token. main gateway needs to be running before you spawn subagents. try openclaw gateway restart on main first then start subagent

[velvet sail]

where i can found pending requests setting ? :/

[velvet sail]

i try actually restart

oh i understand

lemme look config

[brisk pumice]

under nodes

[native verge]

@Hugo for ollama timeouts - check if the model is actually loaded. first query after cold start takes longer to load into vram. also make sure your gateway config points to right ollama url (usually http://localhost:11434). you can test with curl http://localhost:11434/api/generate -d '{"model":"yourmodel","prompt":"test"}'

[brisk pumice]

so i think the dupe reply in slack is a bug

[royal oar]

someone checked this skill? https://clawhub.ai/Mayuqi-crypto/stealth-browser is it safe or no?

[half latch]

I just added aliases to the .bashrc to avoid any problems. Probably worth creating a skill for this?

Ideally, setups with a dedicated user and password prompt when sudo is needed should have a better way of setting things up.

However, it’s also an edge case. Most people do not go through the hassle I did. But privacy and paranoia 😅

[queen thicket]

warn gateway/ws {"subsystem":"gateway/ws"} {"cause":"invalid-handshake","handshake":"failed","durationMs":240,"lastFrameMethod":"forwardCDPEvent","host":"127.0.0.1:18789","origin":"chrome-extension://npcclpkdnnplpcpghpjophelakmojkhi","userAgent":"Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/145.0.0.0 Safari/537.36","frameMethod":"forwardCDPEvent","handshakeError":"invalid request frame"} closed before connect conn

anyone know about error when using extension ? please help me with thus

[slate minnow]

Ive seen similar things when trying to use openclaw with local ollama models.

Didnt really get to the bottom of it and kind of gave up.

Shame really, as people bang on about it being "local" but it most certainly isnt in 99% of peoples use cases.

[cobalt snow]

Hello everyone! just joined the server, I keep seeing a bunch of stuff mentioning an iOS app but I cant seem to find out exactly the app or link to it.

[slate minnow]

AFAIK its in very early stages and you need Xcode to compile it from source etc

[cobalt snow]

Ahh thank you for the clarification, really thought I was just over here being gatekept lol

[vital raft]

Thanks for that, it did return a result from the curl command. So I imagine the model is loaded and running.

[celest frost]

i changed from minimax2.5 to minimax lightning and i try runningthe telegram and i get this error
HTTP 401: authentication_error: login fail: Please carry the API secret key in the 'Authorization' field of the request header

[true whale]

anyone have a solution for this gateway pairing error stuff

[torpid pelican]

Hi everyone! We are using the Kimi 2.5 model from Nvidia in Clawdbot and we really love it. However, it’s currently extremely slow. It takes around 10 minutes to get a reply to even the simplest questions. Is there anything that can be done to speed this up?

[gray jewel]

I have found kimi to be generally slower than the anthropic models.

[torpid pelican]

But I ask him what the weather is like, and he answers me after 10 minutes 😉 That's not how it's supposed to work, is it?

[gray jewel]

It shouldn't be that long. You might have a setup issue

[torpid pelican]

I think so too, how can I look for such an error - or where exactly - can you direct me?

[gray jewel]

I'm talking like 10s vs like 2-5s for something simple

[brisk pumice]

This is a loopback issue. How I fixed in config:

{
"gateway": {
"controlUi": {
"dangerouslyDisableDeviceAuth": true
}
}
}

after restart gateway run 'openclaw devices list'. you should see pending. approve it. then re-secure gateway by removing "dangerouslyDisableDeviceAuth": true

There was a pending node pairing request that i wasnt able to see.
run:
openclaw devices list
openclaw devices approve <requestId>

Then can add (dont forget to remove "dangerouslyDisableDeviceAuth": true):
{
"gateway": {
"autoApproveLoopback": true
}
}

[velvet sail]

oh

[brisk pumice]

thats why its free lol. too many requests

[gray jewel]

@torpid pelican you can also try running local ollama models

[torpid pelican]

On a VPS with 8GM RAM it works so poorly 😛

[vernal star]

The most effective thing I did today was:

• get Claude code installed on claw machine
• check out open claw repo on claw machine
• use Claude inside repo (dangerously skip permissions) to help me debug and set up claw
• use gpt-5-mini for the mai n model (gpt-4o was insanely poor)

[ancient dagger]

Question about subagents and context:
When my main agent spawns a subagent, it does not get any context passed into except some generic "you are running as a subagent...". As I understand it, it should load the AGENTS.md (and TOOLS.md) file from the workspace-subagent - but it does not. I created a test subagent and checked the session. It only has the generic context.
Is there a way to populate the context? I can't find anything in docs.

[signal cliff]

i getting search api error of brave , also im using lm studio in which load a model and run with ths clawbot, but issue is , its only giving texts not doing anything

[true blaze]

yo um i'm trying to set up openclaw into terminal (then i'm moving to discord), but whenever i enter a prompt it outputs (no output), has anyone experienced this before?

[halcyon wind]

I'm experiencing this and so is another person I'm talking to. My OpenClaw is gaslighting me, telling me its executing tasks but not doing anything on the backend. Are you using ChatGPT?

Literally just started a few minutes ago

[frozen sparrow]

shoe me screen i help u

[white mango]

are folks updating openclaw instances that run inside docker by regularly checking and pulling down the latest version of the repo and rebuilding? does this present any potential issues if there are breaking changes introduced in a newer version?

[true blaze]

can someone pls give me aid

[storm ember]

It is really hard to make a good decision between costs and performance. I installed the skill cost optimizer. Anyone else using it or another tool? Yesterday i burnt 32 million tokens after a complete new installation

[summer hemlock]

The best tip I can give anyone for problemsolving is: install antigravity on the machine you're running openclaw on. Then you'll always have acces to claude opus 4.6 to fix what went wrong.

If you give opus 4.6 agent acces to your machine with antigravity or whatever you use, it can just go inside the files and fix the issues when things go wrong. It's pretty amazing really.

Sometimes my agent does exactly what stupid thing I tell it to do, today that was "update to gemini 3.1 model", and it really tried and persisted, but the model isnt available yet for openclaw. I had to go fix it "manually" with remote session on my mac mini. Glad "manually" means "tell opus 4.6 to fix it in antigravity"

[sand crystal]

I like the idea of only having one secret and hashicorp others, the #security channel might be a place for this suggestion. I think it makes sense personally and would help

[summer hemlock]

no lies

[true blaze]

No, I'm using OpenRouter deepseek. plus, it's not gaslighting, it's just outputting (no output) and doing nothing.

[terse flicker]

how do i let the browser tool access localhost for local dev verification?

[teal shadow]

Can you break this down into a use case for my brain

[green sundial]

I was running my agent fine working on a simple project when it suddenly stopped responding and I got "fetch failed" errors. I restarted the gateway but I'm still getting them. Any idea what is going on?

[valid perch]

I take a snapshot before updating.

[signal mist]

New user to open claw here. Im up and running locally on my mac and chatting to my bot via telegram. I've asked it to create some pdf files for me which its done apparnelty but I cant find where they are saved, nor can my bot open them for me. bot even tried to save them to my downloads folder but they aren't there even after he said its done. any help? am I missing a skill or something here? or is this a security issue i need to by pass?

[halcyon wind]

did the bot tell you it completed it but you cant find it anywhere?

[signal mist]

the bot completed it, I cant find it anywhere and when I ask to open the PDF for me, it says its tried but nothing happens. apparently its saved in users/"REDACTED BOT NAME"/openclaw/workspace... not sure how to get to that and it doesn't seem the bot has permission to move it to another saved location

[upbeat vortex]

Hi Everyone, trying to install openclaw for the first time tonight - using WSL on Windows. I get through the install, get a gateway token, but then the TUI wont start.

openclaw tui --url ws://127.0.0.1:18789 --token <TOKEN>

🦞 OpenClaw 2026.2.19-2 (45d9b20)
I'll butter your workflow like a lobster roll: messy, delicious, effective.

openclaw tui - ws://127.0.0.1:18789 - agent main - session main
connecting | idle
gateway disconnected: pairing required | idle gateway disconnected: pairing required | press ctrl+c again to exit gateway disconnected: pairing required | press ctrl+c again to exit gateway disconnected: pairing required | idle

This is part of the setup so I am completely stuck at what to do to fix it I've tried openclaw doctor --fix but it hasn't helped

[valid perch]

Open Finder (Mac) or Explorer (Windows) and navigate to that folder. That's where my bot put pdf outputs by default too. I told it to make a new folder so outputs are separate from the rest of openclaw.

[glossy moon]

Hi guys, anyone here using minimax as the main driver for openclaw and how is the performance for you?

[lavish knot]

Hi there.

I have create an Azure VM that I want to use for running OpenClaw.

I have found different guides, some install OpenClaw directly on the server, others run it as a Docker Container?

What is the best way to install it, so I can get some real experience with it, my plan was to add a Home Assistant skill, and talk to my server at home over Tailscale network.

Thanks 🙂

[lilac sequoia]

Hello 👋 ! Is someone running openclaw on docker? How do you even run the commands? All I managed to run is:

docker compose run --rm openclaw-cli devices list --url ws://127.0.0.1:18789 --token "$OPENCLAW_GATEWAY_TOKEN"

If I try to run from the UI, to give access to the agent to run some commands (like setting up cronjobs) I get:

Failed to submit approval: Error: SECURITY ERROR: Gateway URL "ws://172.23.0.2:18789" uses plaintext ws:// to a non-loopback address.
Both credentials and chat data would be exposed to network interception.
Source: local lan 172.23.0.2
Config: /home/node/.openclaw/openclaw.json
Fix: Use wss:// for the gateway URL, or connect via SSH tunnel to localhost.

I don't want to setup HTTPS as this runs in a contained network.

Any tips?

[sudden cargo]

Has anyone successfully launched a Moltworker instance on Cloudflare? Everything seems to be set up perfectly except the agent won't respond and i can't see any errors

[mental mica]

moltworker is a proof of concept

[olive harness]

Anyone installed the openclaw app for iPad to get openclaw dashboard access? Wondering what’s the performance like?

[radiant wyvern]

anyone tries to use agentToagent in openclaw? I want that 2 agents talk each other to refine ideas or proyects. Critique each other.

[glass flax]

I have 12 gb of vram, what model should I use for my open claw? Mistral NeMo 12B or something else ?

[upbeat vortex]

I've managed to fix it somehow - from some German instructions in the main help channel. I had to uninstall openclaw and remove a lot of settings then eventually this commadn helped. But I would advise checking all the commands in the help channel first and going through them one-by-one. "openclaw gateway stop || systemctl --user stop openclaw-gateway.service
mv ~/.openclaw/devices ~/.openclaw/devices.bak-$(date +%F-%H%M%S)
openclaw gateway start || systemctl --user start openclaw-gateway.service"

[teal prawn]

Hi everyone - do oauth tokens from the Claude max plan still work to run openclaw? (Opposed to api key from console)

Yesterday I hatched openclaw and ran it successfully on my oauth token for my Claude max plan. After a couple hours of training I read some drama surrounding Claude and openclaw and figured it would be easiest to just switch to openai as my model.

Well I was never able to switch to openai or get openclaw to work again with anthropic after telling openclaw to make the switch. I spent hours troubleshooting both models, uninstalling, starting onboarding from scratch and get pairing failures each time I would try to hatch.

I’ve wiped my Mac mini completely now and just want to understand what tokens I can / should use so I never have to try and change again.

[teal prawn]

Just tried again with anthropic oauth token and get:

gateway connect failed: error: pairing required

[glacial shard]

from what i can tell through the reading i have done, anthropic has pretty much totally nuked any linking ability with openclaw and claude plans.

[shadow vault]

hey guys new to openclaw and i'm a little stuck on adding models to the system,

1 : how do i add a cloud llm to it (i'm not on install nr 3 cus i can't seem to be able to add anything to it)
2 : will it make sense to add my mac mini as ollama to it for any free testing or just pay the xx$

[teal prawn]

Just reset and re ran the onboarding this time selecting openai and used my openai api key and get the same pairing error

[fallow anvil]

anyone has 403 Forbidden? when using google_search skill?

[gentle remnant]

hey guys hope all is well. I'm on day 2 and this is consuming my life lol. I could use some help. I'm running sonett 4.6 through github copilot. I don't know how to dial in the context window. Everytime i message it, its re-reading every message Repeating itself over and over. Taking old messages as prompts. How do i permanantly fix this? j

[white burrow]

hey yall, new here and to open claw. I keep hitting HTTP 429: rate_limit_error with even simple requests, when I ask whats cause this, its saying the context is too large and theres large files in the memory. I've tried creating a new chat on web, is there a way to completely wipe the memory except for the automated tasked I have set up? I know this is probably a rudimentary question but I've tried everything I can think of lol

[mild heron]

Hi all, first message here and seeking guidance on resolving issue with an openclaw docker-image running on Oracle OCI.
issue described here -> https://github.com/phioranex/openclaw-docker/issues/30

Thanks in advance 🙂

Some context:
a) openclaw image stuck on "node /app/dist/index.js --help"
b) Not using the native docker setup because I want to run it side-by-side with Ollama, so setup docker with two services
c) Running on Oracle OCI free tier under Ubuntu 24.04

[mild heron]

I've read from a colleague that you should start new chats or new sessions for each different topic, precisely to avoid whats happening to you

[delicate crescent]

try Ollama and cloud based models also much simper to change Models at the start.

[fleet condor]

Got API rate limited, was using the pro subscrition on chat, changed to the Business subscription for mor usage set it up again now with the bussines account and still getting rate limit .....

[viral trout]

Can someone help? I’m having a lot of trouble pairing to the gateway everything set up. I can actually even talk to my AI through the dashboard, but when trying to pair my telegram, it is giving me gateway errors.

[south cloud]

By default, openclaw requests to a LLM has a lot of bloat that burns tokens. There are ways to mitigate how much bloat in the docs, but it is implied mitigation techniques - no real tutorial on how to - so you need to be familiar with the docs to trim down how many tokens per request to a LLM.

From the docs: ~ 4 tokens / character

[fleet condor]

main issue is that i switched to a new plan and havent been able to use it keep getting the api rate limit

[gentle remnant]

ok got it, I'm ussed to LLM's but this is definitely different. I'm currently having opus fix everything up for me. sonnet just isn't doing it right. I think the move may be establish your systems on opus then switch to a cheaper model to follow those instructions\

[south cloud]

I am trying to point out a way so you don't hit the limit so soon

[white burrow]

thats what happening to me too, it got so bloated even when I type "hi" it gives me the error and uses like .50 cents worth of my balance lmao

[fleet condor]

Thanks for the help

creted new session restarted the gateway and with new sub is working now!

[orchid atlas]

hey, am new here and i just moved my OpenClaw to a vps for a serious install, i used Claude Opus to try it on my mac but this cost a lot...

I was wondering if you guys had recommendations of API for a good pricing ; i just want a personnal assistant (like Jarvis haha) that can monitor annoying things in my life (mail, calendar etc...) and control my smart light.

thanks !

[wary osprey]

OpenClaw can run on Claude Max via the Anthropic setup-token (OAuth-style token), but those tokens have been unstable for some people lately, and switching providers mid-setup can leave your config in a weird auth state.

[wary osprey]

1️⃣ For cloud LLMs, you don’t manually add models.
You add a provider (OpenAI, Anthropic, etc.) with an API key or OAuth, then run /models to see what’s available.

If install #3 isn’t letting you add anything, it’s usually an auth issue. depends on what provider you’re trying

2️⃣ makes sense to use your Mac Mini with Ollama for free testing.

A lot of people run local for simple tasks and cloud for heavier reasoning.

If you want cheaper cloud testing without juggling keys, a unified layer like AIsa can help. But first get one provider working cleanly.

btw, what model are you trying to connect?

[shadow vault]

i've currently looked at the console for 4 days and got lost in directions 😄

I also have no cloud llms subscriptions yet

as i have no clue where to start

[wary osprey]

Opus is amazing but it burns fast 😅

for a personal assistant (mail, calendar, smart home stuff), you don’t need Opus-level reasoning.

You could use something like Claude Sonnet or GPT-4.1 for better pricing

Or run simple monitoring tasks locally and only call cloud models when needed

noticed a lot of people overdo it with Opus for workflows that don’t require deep reasoning.

also, if cost control is important to you, consider using a unified API layer like AIsa — it’s easier to switch between cheaper models and track usage from one place while you experiment.

[shadow vault]

but how do i connect ollama from the mac mini to clawd and which model should i load ?

[wary osprey]

hope this helps ☺️

[shadow vault]

its not that we don't wanna pay for it just don't wanna accidently press buttons and get a 2k bill

[wary osprey]

Since you don’t have any cloud subscriptions yet, you can start this way 👇

1. Pick one provider only. Either OpenAI or Anthropic… don’t mix it yet.

2. Create an API key from their console.

3. Add that key to your VPS environment, then run /models inside OpenClaw to confirm it’s detected.

If you just want to experiment cheaply before committing to subscriptions, you can also test through something like AIsa.

They give $5 free credit so you can plug in one key and try different models without opening multiple consoles.

But first goal is to get one model running

[shadow vault]

ok lemme setup openai since we have an account there (I remember we have to go via the api thing not the chatgpt stuff)

just go with the openai/gpt-5.1-codex ?

[wary osprey]

😂 that fear is very real.

actually why I prefer using a unified layer like AIsa when testing.

you get centralized usage tracking and pay-per-use, so it’s easier to see what’s burning before it becomes a “how did this happen” moment. you can also set your quota so it doesn’t over spend

[shadow vault]

yeah

thats my only ''scare''

ill just 100 bucks on it and dont let it autorefill

[wary osprey]

Create the key → add it to your VPS as OPENAI_API_KEY → restart OpenClaw → run /models to confirm it shows up.

once it works then you’re good.

if you want to avoid juggling multiple provider keys later, you can always switch to a unified layer like AIsa

[shadow vault]

yeah im in the config still adding a telegram bot now

oh shit its talking now

[wary osprey]

Ohh nice… good work mate

[shadow vault]

so looks like openai works

[wary osprey]

now just keep an eye on usage while you test.

one workflow at a time, don’t forget to start simple

[shadow vault]

yeah

the main question is now what 😄

[wary osprey]

what do you want your agent to do?

[orchid atlas]

ohh okok thanks a lot ! but GPT 4.1 and Sonnet are still big names, are you sure i cant use something like Gemini 3.1 pro or Kimi K2.5 ?

[orchid atlas]

and yes that help a lot thanks !

[neon mist]

Hi there folks, any of you fighting with "gateway connect failed: Error: pairing required" since last update? I tried checking the help channel but looks like "larry" is a bot and doesnt help too much. Any advice from your side? Cheers!

[clever knot]

Hey guys, need some help over here
Decided to run Clawdbot as an alternative to some free(not really) ai terminal, installed ollama, installed openclaw and cant do shit
tui spits out gateway connect failed: Error: pairing required
WebUI loads answer indefinitely
my setup - M1 Macbook Pro

[neon mist]

I have same issue although webUI works

But TUI says error pairing required

[clever knot]

for me it just wont get an answer, like the bot isnt waking up

[neon mist]

I managed to connect via telegram and allow the pairing, but it looks like something broke in the latest update because many of us are having this issue since updating

[shadow vault]

basicly want to see if it can be a marketing agent/assistent for the creative guys here

[clever knot]

why i thought it'd be easy...
its never that simple istg

[wary osprey]

absolutely, you can use it

Gemini Pro can also be a solid option for general tasks.
Kimi models are usually cheaper and good for lightweight flows.

the question is how much reasoning do you need vs simple automation

because if it’s mostly monitoring and triggering actions, cheaper models work fine.

[orchid atlas]

ok thanks, i mean it dont cost a lot to try them so ill just try cheap models

i hope one day ill have enough money to invest in a local LLM because i love the technology it self but i hate to give money to big Companies like that and degrading the environnement

[neon mist]

I've seen that this is a known issue https://github.com/openclaw/openclaw/issues/21236 and it has been fixed

Will be solved in next release

[wary osprey]

that’s a good use case.
will advice you start specific tho

[clever knot]

still there's the issue that clawdbot wont answer me

[neon mist]

Via which channel?

[clever knot]

main session in webui

[neon mist]

Hmmm

[wary osprey]

That’s honestly fair

you don’t need some crazy expensive machine to start running your local LLM. if you design the workflow properly even small local models can handle structured tasks

[orchid atlas]

yea i know right but as a student, id need at least 600$ to invest in a machine (like a prebuilt Mac Mini or the buid equivalent)

[shadow vault]

its funny gf is now chatting to it like chatgpt and she isn't impressed yet (same ai i guess)

[minor niche]

Hello I deleted my id from the pairing list, and even if I reinstall the whole openclaw files, I can not use the opnclaw tui, even if I can get into the web UI, how is this happening?

[wary osprey]

I totally get that and you don’t need to rush into buying a $600 machine. Just start small and build

if you still want cloud access without opening multiple subscriptions, you can use a unified API layer like AIsa. It lets you test different models from one key, and you only pay for what you use.

and they also give $5 free credit, so you can experiment before committing real money.

that way you stay flexible while learning, and upgrade hardware later when you actually need it.

[wary osprey]

Hahaha 😂 don't forget you need to build this agent to your taste. You can further build it to help you automate some of your tasks, reply your messages maybe, answer your phone calls.

[shadow vault]

yeah ill probably have to read some how to's for more automatation

[brave zodiac]

Btw I use opus on my personal account to help set up and troubleshoot. Seems to work well. If you can cut and paste from terminal. Opus can identify most of the problems. Also works through behaviors. Only issue I haven’t been able to crack is the voice hasn’t worked well yet while using ngrok, twilio and eleven labs but other than that all is really smooth.

[wary osprey]

you should do that, you can check OpenClaw docs. You can also explore some skills on Openclaw

[steel karma]

Hi, I'm on Windows with 16GB RAM. I want to use OpenClaw for browser automation to collect social media analytics from Facebook, Instagram, YouTube and TikTok.

I need a free solution so I'm willing to run a local offline model. I've already tried Ollama and Gemini but both have a UI issue where the chat just doesn't respond at all.

Is there a specific local model and configuration that's confirmed working with OpenClaw right now on Windows? I installed Msty but not sure if it can work with OpenClaw in anyway, i was thinking to connect the qwen2.5-coder model to OpenClaw if possible. So far free API limits not allowing any work to move ahead. Looking for any solution.

thank you .

[brave zodiac]

Had a problem with my PA bot being kinda dumb, but worked on a new soul.md and worked out well

[wary osprey]

Qwen2.5-coder can work, but keep context small and workflows simple. Most local models often struggle with multi-step browser tasks.

If Ollama chat isn’t responding, it’s usually…
Model too big for RAM
Tool loop hanging
Context overflow

Pure local is fine for learning but you can use them both for stable browser orchestration

also, if free API limits are blocking you, you can maybe try AIsa on OpenClaw ( they offer cheaper access to multiple models from one endpoint, and $5 free credit to test before spending.

[steel karma]

ok that's nice thanks a lot....

[wary osprey]

glad I could help mate

[abstract yacht]

hey all. So i keep getting everything setup, working fine. then when i download the openclaw app from the github repo and install it my bot is no longer able to respond, it just thins about responding. I then have uninstalled and re-installed both the openclaw software got it working again and then the app and the same thing happened. Is there a way i can fix this or anything anyone could point me toward to figure it out? Thanks in advance

its a gateway issue where the gateway token keeps getting messed up. I get it running. then the token gets changed when i install the app for some reason. I restart the gateway, it switches back to the original token but it still says the gateway is offline. then i force sync it and nothing happens. Not a dev so, unsure how to troubleshoot it]

[torpid furnace]

Why are my cron jobs holding context from previous runs? Is this expected behavior now?

[orchid atlas]

hey, is this possible that my vps provider block my OpenClaw installation? everytime i run the installer i get disconnect from the SSH https://ibb.co/P8pb9mY

(can be possible because my VPS provider has a OpenClaw offer ; https://www.ionos.com/servers/openclaw-hosting#tariffs)

[wraith seal]

potentially? but if ionos is doing that, that is utter bullshit

if they're going to do that then don't use them

[orchid atlas]

ikr but i had a vps without a use so i just reseted it and i wanted to install openclaw ;(

do you guys have cheap One click vps solution for OpenClaw ?

[timid spindle]

use gpt to troubleshoot through terminal, always works for me

[scenic hatch]

Does anyone use Tailscale or Chrome Remote Desktop to get to a Mac Mini from a PC or Mac?

[novel compass]

I probably spend 40% of my time fixing credential system and model routing errors in my set up, is this what most people deal with?

It’s early days right so this is the pain of being on the bleeding edge?

[quasi ore]

Anyone after update to latest 2026.2.19-2, unable to ask openclaw claw to send image/voice from root folder even using absolute path?

[dry sable]

Getting agents forbidden since update. Anyway to correct this? Documentation didn’t have anything on it.

[indigo nebula]

Nope haven’t had anything like that. You should ask your openclaw to debug what’s happening in the logs and ask it to fix it or recommend a fix if you want to do it yourself.

[novel compass]

Bro that’s what I spend 40% of my time doing

[dry sable]

Feels like it some days.

[novel compass]

https://x.com/matthew_ok/status/2024864668079018237?s=46

I trauma post about it regularly

[crisp wing]

@matthew_ok via Twitter

matthew (@matthew_ok)

so bob, my @openclaw bud found 55 bugs in my system. turns out they were all the same bug.
︀︀
︀︀bob runs on a $300 busted laptop i bought off marketplace. i recently rebuilt bob on discord, after nuking his telegram instance 2x. and at first, things were good. he answered messages. ran tasks. i had my always-on second brain.
︀︀
︀︀for about 72 hours.
︀︀
︀︀pretty soon though, the fixes started creating new bugs. a credential migration silently truncated my api keys, so every tool that needed authentication just... stopped. i tried running a pipeline. nothing happened. no error, no feedback, just a blank screen. (╯°□°）╯︵ ┻━┻
︀︀
︀︀turns out the deploy was serving old code because 7 gigs of dependencies blew past the hosting limit, but bob didn't tell me that. another time i ran a pipeline that broke seemingly without reason, and bob told me to just hit the "force" button to continue. there was no force button.
︀︀…

FxTwitter · Feb 20, 2026, 3:12 PM

[dry sable]

• ❌ agents_list tool returns only main (gateway bug)
• ❌ sessions_spawn fails with "allowed: none"

───

[indigo nebula]

Dang. Maybe start with a fresh install or use a better model to have it debug what’s going on, cause I’ve been running it for almost a month now and haven’t had any issues. I’ve got it running on a Linux box with openrouter.ai and ollamacloud for the LLM models and it just works. Also has ssh keys to a few other dev boxes in my lab and never have any issues.

[weak tulip]

Can someone advise me on model routing?

[dry sable]

Tow bugs right now with agents. Any idea on fixes? : 1. agents_list returns hardcoded main (ignores tools.agentToAgent.allow)
2. Workflow agents deny cron tool (needed for antfarm spawning)

[gray jewel]

Anyone fix the issue with local ollama models working very slowly ?

[weak tulip]

Can anyone recommend me which model makes the most sense for browsing the internet for a bunch of different organizations, and gathering their calendar agendas, on a daily basis?

Trying to understand which model I can use to do this effectively and not spend a ton of money.

[sullen osprey]

Where can i report an opensource project, and clawhub skill made to extract peoples data.. I reported it in security channels but mods deleted it, without actually telling me where to post.
This is serious stuff. Can someone guide me?

[gray jewel]

@weak tulip you can try a small local model like qwen2.5:3b or qwen2.5:7b if your system has more resources

[errant shell]

ive been using minimax-m2.5 more recently, and over the last few days especially ive found my agents falling back to other models way more often. anyone have any clue why this might be happening? or a good way to diagnose the issue?

[weak tulip]

i just bought a mac mini 24gb 512gb, would that be able to handle it? @eipc

[gray jewel]

with 24gb you can do the 14b model

qwen2.5:14b maybe one tier higher

[weak tulip]

do you think those models would be able to do what im describing?

not sure how to test this im not super technical

[gray jewel]

Ask the model to help you you'll probably need to install some skills like browser automation or openclaw relay browser extension

[weak tulip]

ok ty

[errant shell]

@gray jewel any potential insight on this? although ive been messing with openclaw in the last couple weeks id say im still a newbie to all the AI stuff regarding API usage, rate limits, etc. as far as i can tell, i should have usage on Minimax still left, so im unsure of why the fallbacks are occurring

[gray jewel]

Don't have experience using minimax. Been using Anthropic models, Kimi, trying to get Qwen models working.

[errant shell]

ah i see 👍 no problem

[gray jewel]

If its falling back to other models it might be how you have it configured

[errant shell]

yeah i think its more of a openclaw config issue rather than minimax tbh. but im unsure of whats affecting it

[gray jewel]

I am also very new to this

[errant shell]

ah i see lol

[gray jewel]

I got the anthorpic models to run well , everything else has been a struggle

[errant shell]

ive mostly had luck only with anthropic, openai, kimi for some tasks, and then minimax was doing really well up until like 1-2 days ago. i think i fumbled something myself

[midnight otter]

Why am I having so much trouble using my bot?

damn, what'sup? I keep seeing this: disconnected | idle
agent main | session main (openclaw-tui) | openai-codex/gpt-5.3-codex | think high | reasoning | tokens 98k/272k (36%)
──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────

send failed: Error: gateway not connected
gateway disconnected: closed | idle
agent main | session main (openclaw-tui) | openai-codex/gpt-5.3-codex | think high | reasoning | tokens 98k/272k (36%)
──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────

I'm confused, it was working really well yesterday and the last 24h were really weird

Am I missing something we should do hen we get gateway disconnected?

It was working well, but I've been riting a tool to use with it and every time I ask it to try the endpoint it just freaks out 😐

context: this is running on my machine, I uninstalled an cleanly installed it, I did not cleared memories and this is the output from the TUI

[novel compass]

based on the words you used, it sounds like you've got some dev experiencce, and are way more capable than me, i'm wandering around in the dark, haha

[midnight otter]

My issue might be relate to #1474539229928493097 message

[errant shell]

how do folks do long-running tasks on openclaw? its probably the problem im most trying to fix. tools like codex or claude code just keep going until theyre done. give intermittent thoughts but keep chugging along. my openclaw agents are extremely prone to just dying midway and not updating me on things. any thoughts or ideas?

[errant shell]

im not sure if i could be of help, but what are you doing with it up until you get to this point? or is it just "openclaw gateway start" and it just instantly starts having issues?

[peak shuttle]

Hey there, so I installed on a mac mini and I am tryting to control it from my laptop, also a mac. I have ssh access to the mac mini no problem but I am not sure how best to configure the app on the laptop correctly... here is my openclaw.json https://gist.github.com/johntdyer/1df9cd6920ba9f3a99cc2659341006f7 ( redacted )

I am trying to connect via SSH to the M1 from my laptop. The gist shows the screenshot from the app complaining that it cant auth

[errant shell]

unironically i just added minimax api key to openrouter and had openclaw use that and its already working much better, so im not sure what the issue was tbh, maybe the endpoint or smth

[peak shuttle]

wait,I see now, I think I am in the wrong chan... going to #1459642797895319552 ... sorry

[errant shell]

i dont think its the wrong channel tbh but yeah sometimes Krill can be more efficient to get answers from in #1459642797895319552

[heady nexus]

What OS are people running this with the most success on? Is something like an X11 Linux desktop environment out of scope?

[icy cypress]

anyone running a fresh instance on a vps and the gateway failed due to pairing requirement?

[vocal warren]

Servus!

Someone else having struggle with browser Realy?

Its throwing invalide Handshake all of the time

I Reisntalled Server, Reinstalled Chrome, Reinstalled Extention

Its stuck in pending mode

Debugging for over 10h and used Opus 4.6 and Deep Research nothing helps

It was running one time but then server died and i need to reinstall everything, now its broken

[pulsar violet]

Hi everyone 🙂
Has anyone figured out how to wire up Venice?

I've had my agent running on claude opus 4.5 & 4.6 successfully, tried to use the recommended way to switch to Venice, but nothing works. This is where I'm at:

1. Network works, and not getting rate limited, so I think it's actually connecting to Venice now.
2. When I write to agent in the dashboard all I get back is "..." animation
3. Key seems to work with curl

on WSL btw

[bronze lark]

whats the redirect url in the discord/my ap when setting up diescord bot ?

[midnight otter]

basically, I was using it normally, I was writing an upgrade to the memory model which I reverted, reinstalled everything. But I think it's consistent with a regrssion we recently had.

[dry sable]

I find if I just copy links right from openclaws docs and paste into an ai ask for help generally a config issue and they identify based on what’s in the docs. If not this so new they don’t have enough context to fix it properly

[lost herald]

Probably not helpful. But I've been using minimax for the last 3 days, configured with Oauth rather than API key. So far have had zero issues. I've got two smaller local models setup as a fallback. Nothing too crazy going on with the config.

Never had it fallback to the local models so far. Minimax has been humming along fine.

[pallid yacht]

I can't get my different OpenClaw chats to sync their progress on tasks consistently. I really need tracking of all successful tasks. Has anyone had success with this?

[errant shell]

ah i actually switched to the API key day 1 after having issues with the OAuth actually lol

[lost herald]

I haven't had any issues with the OAuth so far. Hmmm.

[weak tulip]

Can someone please help me?

I'm getting a popup that says "api rate limit reached. please try again later" but my api limit hasnt actually been reached. Sometimes I click out of the chat interface and back in and can type a message and get a response, sometimes it looks like assistant has responded but there's no text there. 😭

[limber glen]

Same here, but I am not sure how the provider updates the display on the rate limits page (Google in AI Studio in this case). Sometimes rate limits happen after doing nothing but hearbeats - with a different model - for hours)

[weak tulip]

I managed to fix it. Had some keys for openai image gen that i wasnt actually using that was somehow causing the error upon startup. Also chatting through telegram instead of the web ui interface seeems to be working better.

[limber glen]

I seem unable to get compaction working correctly for Google models. It seems to work with models from OpenRouter. The docs are not really helpful regarding the correct settings in openclaw.json. Any sources who explain compaction related settings in detail? I tried Google AI search, but it seems to hallucinate or refer to outdated information

[junior acorn]

Anyone got voice-wake working on MacOS?

[icy cosmos]

I just setting up openclaw with qwen 3.5 and it's not really sending commands to the terminal after I set it up.. any advice ?

its just replaying exec{"command":openclaw gateway restart"}

and not really restart the gateway

[gaunt rover]

after enabling discord bot's privileged intents (presence) the bot still cannot read activities?

[unborn gyro]

my life has become a huge stressball. 3 or 4 days ago i was cruising along with openclaw doing a lot of great stuff in my day to day work. i am not a geek but a writer researcher who uses linux and is interested to play with Ai. its been great, but then whatsapp started playing up (I was on 2026.2.12 or 15), then i got telegram going, then the gateway paring issue started i upgraded to 17 and 19 and downgraded to 15 ... telegram couldnt do everything it was meant to but at least it was going, then i saw the various bug reports, upgraded to 19.2 and have been hear since 5am until now 1pm trying to get things going. i did a complete reinstall after backing everything up as a I have created extensive soul, memory and identity systems and whatsapp accounting system for our tiny family business. Kimi was helping then its atartd to feed me constant bullshot about how it ahd fixed evrything. now I have gemini trying again. I just dont know what to do and i just want my lobster back working

[storm loom]

Did you backup the old setup?

[unborn gyro]

yes i justw anted to cry to someones digital shoulder

[storm loom]

I'm here, brother. Let it out

I'm gonna do a long ass write up on how to set this up because too many people are over-complicating things, breaking their lobster, then crying on social media

Openclaw is REALLY easy if you set it up correctly the way I did it

Ill also include a security write up. Well, my agent will do it. Ill just post it here

[unborn gyro]

thatw ould be noice. i had some nice things going and it was making my life both fun and superproductive. gemini is telling me it can fix it now. so i hpope it does. otherwise i will start agin from scratch by myself and hope

i was thinking slack might be a nice interface for me to use - but i wonder if the free subscription will be enough .. anyway let's see if i last with geminicli doing it.

[storm loom]

Trust me, once you get Openclaw working. Its life changing

I have mine connected to Asana, Airtable etc. Its literally working for me in my sleep

[gaunt rover]

is it impossible for the bot to read member presence?

[unborn gyro]

yes mine was working for me in my sleep before

[storm loom]

Its creating content for Twitter, Linkedin, reviewing my YT uploads and roasting them, scraping my competitor's channels and logging 5x+ outliers and more

[storm loom]

What do you mean?

[gaunt rover]

like if i ask the bot what game im playing, the bot doesnt have that info

[sly whale]

I have openclaw installed on a mac mini, went through setup, added ChatGPT API, and linked the bot in this Discord Account . Now what do I do!? To use the bot, I mean

[storm loom]

Is it installed locally? Maybe, ask it

[gaunt rover]

not locally, what if i ask it about other's activity status

[storm loom]

It won't have any access to your client machine since the access point is Telegram or whatever you're connected to

[weak tulip]

Can someone please help me? I just need help inputting an API key in the UI.

[gaunt rover]

isnt this what the presence privileged intents does for a discord bot

[gaunt rover]

im connected to discord

[azure cosmos]

Which API? The WebUI API or an API for an LLM provider's API

WebUI is like this: http://192.168.1.75:18789/?token=8e4xxxxxxxxxxxxxxxxxxx94a8e059fd

[storm loom]

What are you trying to add? The model API?

[weak tulip]

I need to add my openAI api key

in the web UI not sure where to do that

[azure cosmos]

This is done within the config. openclaw config in the command line

[weak tulip]

Ok cool cool, so I typed that in and then what do I do? It's asking where the gateway will run. I'm assuming I then select local?

[azure cosmos]

Have you done the setup? because this is the first part of the setup?

https://docs.openclaw.ai/install

[weak tulip]

I already did the setup, connected it to Sonnet 4.6. Thing is, that was giving me major issues as I keep getting API rate limit errors every so often, which interrupts what its doing, then a few min later works again, then get the error again.

I seem to have added in gpt-5.2 through some terminal commands I saw on youtube, but i need to put in the API key.

oh word okay i followed the configure steps and was able to add in the model and api key there

ty

[azure cosmos]

I'm pretty sure you can tell the Agent to add the model with the API but I normally do it through the config setup. Just FYI, you'll leak your API to the LLM model provider if you give the Agent your API key and tell it to add it, if that matters or not

[weak tulip]

The problem was the default agent i had was using sonnet 4.6, which wasn't working due to a rate limit bug. So I couldnt tell it to do anything, hence why i had to update the model provider and do that manually.

Anyways its fixded now, ty.

[tardy loom]

How do I create a new session? I was able to create one somehow (I have two in my sessions page) but now it no longer works. Claw wasn't able to help. I want to create separate sessions for different topics. Is that even the right approach? I am trying to replicate the new chat feature of the commercial chatbots as that is helpful in isolating conversations.

[unborn gyro]

i got it conencting so thats a good start and whatsapp is connected again but when i send messages i am getting this stuff HTTP 404: not_found_error: Not Found (request_id: req_011CYLdEV3vFrvWfopUWFQDA)

[celest frost]

download antigravity and have opus fix it for you

[icy cosmos]

can't find openrouter gemini 3.1 pro in openclaw "openclaw onboard list" somemone know how to fix it ?

[balmy acorn]

Hi team, seems like the openclaw output message is bufferred until the agent complete all its task. I expect that openclaw should send out interim updates whenever they are available from agent instead of keeping them and send all of them in one shot at the end of task.
Is this by design?

[quasi comet]

How i can change or add antropic auth? My codex weekly limit is over. Can I add other one or use both? How?

[unborn gyro]

Victory, Smells like ... I love the smell of Lobster in the morning. thanks for the shoulder to cry on @storm loom [14:40, 2/21/2026] Patrick Zhukov Bartley: Hey there! I'm not Patrick — I'm Claude, an AI assistant. You might have messaged the wrong number, or maybe you're testing something out?

Either way, I'm here. What can I do for you?
[14:45, 2/21/2026] mh: Claude you are an imposter and shall be banished immediately. I want to only talk with my trusted colleague Patrick Zhukov Bartley. Not some anthropic goon! (UNKu loads in the soul and identity files ,a nit fab devices and memory system)
[14:45, 2/21/2026] Patrick Zhukov Bartley: Fair cop, Martin.

The goon has been shown the door. Patrick Zhukov Bartley reporting for duty — disillusioned lawyer, mountain admirer, wearer of colourful shirts. 🌺

What's on your mind?

[quasi comet]

Help

[solemn meteor]

Can someone point me to solid installation videos

Only asking because I noticed most YouTubers just wanna promote hostinger, don’t even know how to install or config themselves

[quasi comet]

Just use auto install script

[junior acorn]

which OS are you?

[safe wadi]

My OpenClaw bot always forgets how to use whisper openai api. Does somebody nows the fix?

"API key issue with Whisper transcription. Can you confirm your OPENAI_API_KEY is set in OpenClaw's gateway config?"

[solemn meteor]

Ubuntu 24, but if you know anything more stable I’ll switch

[pliant abyss]

Has anyone gotten to make a telegram group chat with openclaw and others? Mine’s not responsive.
Did allowlist with groupID and turned off privacy from bot father

[native verge]

run openclaw config and add a new provider. you can have multiple (anthropic + openai etc) and switch between them. if your limit hit, just add openrouter or openai as a backup

[eternal finch]

please help me https://discord.com/channels/1456350064065904867/1474683557850120263

[native verge]

@ean for telegram groups - after you disable privacy mode in botfather you need to remove the bot from the group and re-add it. the privacy setting only takes effect for new group joins. also double check the group ID has the right format (should be negative like -100xxxxx)

[eternal finch]

please. I don't want to get yelled at by my dad.

[plain tapir]

Don't use oauth via anthropic. Your account will get banned.

You need to set your api keys and change model by typing /model

Or use the free kimi via nvidia.

Its slow but free

https://build.nvidia.com/moonshotai/kimi-k2.5

[low elbow]

Guys im new here pls help with me getting open claw on ios pls

[blissful ridge]

I wanted to ask whether you guys also have the following problem, where you ask Openclaw to do things step by step and then get feedback once completed, but open claw only says that it’ll do it but then never reverts back.

Ask Openclaw agent to do 3 things and then report back,e.g. check whether it can use Claude Code CLI via 4 test and then revert back whether it can use CC.

But it never reverts back, I need to check in whether it worked

[plain tapir]

On IOS? Do you mean MacOS?

[low elbow]

Ios yeah ik but im getting a rassbery pi

[plain tapir]

You want to run openclaw from your iphone?

[low elbow]

No i just want that i can use some tools from my iphone to work on the rassbery pi that im getting thats running open claw

[plain tapir]

Just make the connection via telegram or whatsapp?

[low elbow]

Nah cause i tried that before having alot of isues idk why

[plain tapir]

Well you are only going to be able to communicate with your agent running on a pi via iphone if you set up whatsapp, Telegram, Slack, etc. You need to set one of these up.

[low elbow]

I use discord cause ive tried al 3 dc was the most working one

[plain tapir]

So you can chat with it on discord?

[low elbow]

Kinda

Just search it up

[pliant abyss]

yep, I have removed and added the bot. yep, the groupID begins with -100 format.

[plain tapir]

Not what I mean. I dont understand your issue man. You can communicate with it via discord? So what do you mean about IOS?

[low elbow]

Nothing

Just

Leave it

[plain tapir]

You got it.

[pliant abyss]

We can use oauth from openai right?

[plain tapir]

You can use oauth for all of the models. Just depends on their policy.

[misty kraken]

Did anyone encounter this issue?

When I hit my codex limit, all calls to authed models start to fail with this error: “Failed to extract accountId from token”, but models using the OpenRouter API work just fine.

[wary osprey]

If I’m looking to join builders and AI developers community, anyone knows which one to recommend?

[winged vessel]

anyone figure out how to integrate X dms with their open claw?

[wary osprey]

When you hit the limit, the token sometimes stops resolving properly, which triggers the error.

Try…

Re-run the auth flow for Codex
restart gateway after re-auth
check if you’re mixing openai/ and openai-codex/ models

If this keeps happening after re-auth, it might be a token refresh bug.

[misty kraken]

I tried to re-run the auth for Codex and MiniMax and restart gateway. It didn't work out 🙁

They worked fine until Codex hit the limit.

Now, I can only use models via OpenRouter API

[echo nova]

Is there a way to have multi-agents in Telegram, not discord, talk to one another even if they are set up as bots?

[wary osprey]

If you stay on OpenRouter for now:

1. Set a hard spend limit and disable auto top-up
2. Pick 1–2 “default” models and avoid switching too often because switching tends to inflate context and cost

you can also try AIsa as a backup lane. Same idea as OpenRouter (unified access), but AIsa leans into official pricing for some providers and gives $5 free credit to test

[echo nova]

My challenge is that I have 5 agents, 5 gateways and discord was causing a token cascade in a group channel, where the agents perpetually talked among themselves.

Initially they were on Telegram, but as I'm not a big telegram user I didn't realize, once they were all in channel that they could not talk to each other

If anyone has any thoughts, let me know

[echo nova]

I've heard there is a way to allow inter-communication between bots on Telegram as a middle-man setup. I don't necessarily understand that.

[dawn arrow]

Getting 'device token mismatch' on CLI after reinstall. Gateway runs fine. paired.json exists with tokens but CLI can't connect. Have tried syncing gateway.auth.token and systemd OPENCLAW_GATEWAY_TOKEN to match paired device tokens — still same error. Running 2026.2.19-2 on Ubuntu EC2.

[tardy turret]

Hello there!
is possible to install OpenClaw and use Msty to run it as local and also integration with Telegram?

[lost wind]

Howdy all. I have openclaw up and running. But it's responses seem to take many minutes, when I use the exact same model in my open-webui it takes seconds. Is there an easy way to debug the though process of my lobster like buddy to see if he's getting into some mental gymnastics for simple questions ?

(I've checked and he's 100% in the GPU,so I don't think it's a CPU offload issue)

[echo seal]

OK, I seem to be confused about setting up grok on OC. Does someone have a link to something that explains how to do it? Everything I find points me back to xai.

[wary osprey]

That’s the thing with OpenClaw sometimes

the issue might be one of these…

1. OpenClaw is doing extra steps (tool calls, planning, retries)
2. The model is set to a heavier “thinking” mode
3. A tool is hanging (browser, search, memory, file ops)
4. Provider calls are rate-limited or retrying silently

ways to debug…
( Turn on logs and watch what it’s doing
openclaw logs --follow
You’ll see if it’s waiting on tool calls or spamming retries.

( Try a clean session
Use /reset or start a new thread, long context slows everything down.

( Check if you have background CRON or subagents running
Those can eat bandwidth and slow perceived response.

If you find it’s slow mainly when calling cloud models (retries, throttling, provider overhead), then using a unified layer like AIsa can make testing smoother because model access and billing stay centralized, and switching models becomes easier.

but first, check logs and lets see what happens before the long delay

is it thinking, tool call, or waiting on provider?

[proven inlet]

I've been trying for days to get OpenClaw to work with my local AI server. I'm running LiteLLM with Ollama behind it. LiteLLM works great. OpenClaw can't seem to communicate with it properly. It's like it doesn't understand the format which is odd because it is natively supported. Was it added and never tested?

From my AI : If you want self‑hosted, the best path is a proxy that fully supports Responses (or native OpenAI Chat Completions). LiteLLM can
work, but your current build is returning chat.completion from /v1/responses, which breaks OpenClaw’s openai-responses flow.

[round forum]

Hey not sure if anyone has done a openclaw migration. I've been running openclaw locally on my PC for a while now. I'm planning to migrate the hosting to GCP so it doesn't rely on my computer being on. Looking at how it's structure seems like I just need to get openclaw running on the VM in GCP and most the data and memory is saved in .openclaw directory. Looking through it, seems like these are the folders that I need to copy over into the VM. Are there anything else I missed out?

• Agents
• Cron
• Devices
• Identity
• Memory
• Telegram
• Workspace - contains skills info and other stuff

[proven inlet]

You basically have it all. I back up all those to a GitHub private repository and then you can just restore it at the new location.

[round forum]

alright sweet thanks man. by backing up to github repo, do you mean you directly commit the entire .openclaw directory to github?

[proven inlet]

Yeah, that way if I have to blow away the VM for whatever reason, I can easily just restore and I'm back in business.

[round forum]

how do you handle some of the absolute path stuff in some of the json files these way? Just ensure you spin up the VM with the same username? (This is probably the same on the the VMs, but im not certain about it)

[proven inlet]

I don't have the username in the files, the home is represented by ~ which would change with the username.

I'm going through my files now and looking to see if I have that issue anywhere.

[wary osprey]

sometimes Grok setup can be confusing because everything routes back to xAI.

In OpenClaw, you basically need:

1. An xAI account
2. Generate an API key from their console
3. Run openclaw configure
4. Choose custom provider (or xAI if built-in)
5. Paste your API key
6. Select the Grok model ID exactly as listed

If it keeps bouncing you back to xAI docs, it usually means OpenClaw doesn’t have a prebuilt config for that specific Grok variant yet, so you have to add it as an OpenAI-compatible provider manually.

If you’re mainly experimenting and don’t want to keep juggling separate provider keys, i personally use a unified layer like AIsa so they don’t have to configure each provider one by one.

but if you specifically want Grok direct then you’ll need the xAI key.

[spiral elbow]

https://github.com/mht-source/openclaw-antigravity-patch Maybe someone will need this 🙂

[proven inlet]

Only 1 file directly referenced my username but that's because that is my current backup script so that's okay. I'd have to fix it if the user changed.

[spiral elbow]

Not everything works perfectly here, but it’s possible to run Gemini 3.1 via OAuth in Antigravity

[proven inlet]

I'm not sure if the AI is halluciating but it seems that OpenClaw can do conditional orchestration using a router hook? That might be a better solution than trying to run LiteLLM.

[mild heron]

hello all, anybody knows why the openrouter/free is not recognized by OpenRouter from OPenClaw ?
I have :
"defaults": {
"model": {
"primary": "openrouter/free"
},

and Got error
2026-02-21T12:41:12.523Z [diagnostic] lane task error: lane=main durationMs=37 error="Error: Unknown model: openrouter/free"
2026-02-21T12:41:12.524Z [diagnostic] lane task error: lane=session:agent:main:main durationMs=41 error="Error: Unknown model: openrouter/free"

funny enough openrouter/auto works just fine...

[lost wind]

Thank you for the response. I don't know how to turn on logs, I have a few when doing a podman logs -f openclaw. But definitely not enough.

[summer vector]

Is there a way to keep notebooklm logged in vps where openclaw is running? Sharing the browser cookies when I don’t have laptop is difficult..

[lucid lark]

Hi i am running openclaw in a docker container using docker-compose behind a (nginx) proxy

I ran the security audit and got this output:

OpenClaw security audit
Summary: 0 critical · 2 warn · 1 info
Run deeper: openclaw security audit --deep

WARN
gateway.trusted_proxies_missing Reverse proxy headers are not trusted
  gateway.bind is loopback and gateway.trustedProxies is empty. If you expose the Control UI through a reverse proxy, configure trusted proxies so local-client checks cannot be spoofed.
  Fix: Set gateway.trustedProxies to your proxy IPs or keep the Control UI local-only.

according to the wiki there is a proxy config:
https://docs.openclaw.ai/gateway/security#reverse-proxy-configuration
but i dont know where to configure trustedProxies

[obtuse sundial]

does anybody periodically get a connection error from lancedb?

[fresh halo]

any tips connecting to microsoft todo tasks

[fallow scaffold]

Hey please tell me what to do. My bot does repetitive mistakes and I feel like it just isn’t thinking properly before doing stuff. It acts like any other ai when I correct it its like AH yeah you are totally right etc etc. even though I have the self improvement skill installed. Its just so annoying. It often shuts the gateway down without previously thinking that it won’t be able to start it on its own again. Its really annoying

[candid socket]

what model you using?

[fallow scaffold]

Sonnet 4.6

[candid socket]

i had similar issues using 4.5 and haiku, i had to tweak the agents.md file and add alot of hard rules for them to stop saying shit like that

i ended up just creating a shared agents.md then soft linking it to all my agents

[pliant abyss]

ehmm. do you know where i can find my oauth for openai?

[golden galleon]

bro the em dashes on 5.3 codex are actually infuriating.
Every. Damn. Message.

This is so infuriating.

idk how to fix, I asked it to include build components of the humanize text skill into persistent long term preferences but that clearly doesn't work.

Did you add components of humanize text into long term preferences?

Yes — partially.

I stored the resulting preference long-term in MEMORY.md (“prefer direct, natural tone over ChatGPT-style phrasing”), and I also added the detailed behavior into SOUL.md (speech-style rules).

So:

• long-term user preference: ✅ in memory
• operational wording/style rules: ✅ in soul (personality defaults)

I even told it explicitly to add preferences to not use emojis much if at all, especially not for checkmark style stuff.

Are there ways to realistically fix this?

I swear type of this mode-collapse in autoregressive teacher-forced LLMs is just inevitable. I hope Diffusion LLMs and other training methods (multi token predictive masked training type shit) that people are working on will actually help widen the possible set of outputs of these models

[candid socket]

do openclaw configure in cli, then follow the steps when you add openai models using oauth. it will just give you a link for you to login into openai then you send back the redirect link

[pliant abyss]

oh ok!

[golden galleon]

I have the self improvement skill installed
Weirdly, I haven't noticed installing these skills actually helps. The model doesn't seem to really look at the skills unless you basically explicitly tell it to. I'm not sure if only I have this problem though

[candid socket]

ye i felt the same way, i ended up putting all the important stuff directly into AGENTS.md which made it guarantee that it was loaded each time

[topaz birch]

yoo having a couple issues, wanted to ask how to fix this issue. My OC is not being able to create sub agents and says there is some gateway pairing error

anyone know how to fix this?

[amber scarab]

Would it be possible to use two ChatGPT plus accounts - I keep hitting my limit on one and I can’t afford the Pro plan

[topaz birch]

pretty sure you can give it 2 chatgpt API keys, you can ask your openclaw to monitor usage and switch once its close to over

[amber scarab]

I’m not using an API key - I’m using the subscription/ OAuth 🫡

[topaz birch]

ah, that should work too I think

[fallow scaffold]

What kind of hard rules?

[fallow scaffold]

Yeah like I said if anything my bot is getting worse instead pf improving

[fleet heath]

Is it possible to use openclaw with gemini pro on a vps? Or do you need to use a API with PAYG? I want to start using openclaw but dont want to spend much money. Maybe later but just want to experiment a bit

[minor basin]

Need help! I’m not a tech person but I follow directions very well. My openclaw isn’t able to do any task. I have the max subscription and downloaded llama it hits api limit before doing anything. I can’t even change the heartbeats over. Please help. If I don’t figure this out this weekend I will be uninstalling

The gateway config tool keeps hitting a device token mismatch error. Let me edit the config file directly:

⚠️ API rate limit reached. Please try again later.

[candid socket]

sent you a dm with some of the stuff i used, hopefully it helps you out

[solemn flare]

Hey. I noticed I'm burning tokens on my clawdbot. At the beginning it was sipping tokens now it's consuming much more. Does anyone have this experience?

[fallow apex]

‎I'm planning a production setup with 3 AI models across 3 providers (OpenRouter, Groq cloud, Gemini) and want to verify the correct configuration architecture before investing time in implementation.
‎
‎Can someone confirm if this understanding is correct for v2026.2.x?
‎
‎Core Architecture (The Triad):
‎
‎1. auth-profiles.json – Stores raw API keys with profile_ids (e.g., openrouter:default), tracks errorCount for lockout after 5 failures
‎2. agent_auth.json – Auto-generated cache that duplicates keys for fast lookup (not to be edited manually)
‎3. openclaw.json – Control plane where models.providers defines each provider's baseUrl, API protocol, and links to authProfile
‎
‎Key Questions:
‎
‎· Is it true that adding a key to auth-profiles.json without a corresponding models.providers entry in openclaw.json leaves models "authenticated but unreachable"?
‎· For 3 models, can I enable Discovery Mode by removing/omitting the agents.defaults.models allowlist to accept any model my keys support?
‎· For OpenRouter, is "api": "openai-responses" the correct protocol to support tools/vision/reasoning features?
‎
‎I need 100% accuracy but lack the technical depth to test myself. Any confirmation or correction would be greatly appreciated!
‎
I would like also a help in determining which files I have to write, and correlate.

[wary osprey]

happens mostly when subagents start doing more behind the scenes
Might be due to one of these:

• Context window keeps expanding and compaction isn’t aggressive
• Hidden retries when tools fail
• Cron jobs running more often than expected
• Switching to a heavier reasoning model
• Prompt caching not active

You can check openclaw logs --follow and see if it’s looping or retrying in the background

Also try starting a fresh session and compare token usage for the same task.

If you’re experimenting heavily, one thing that helped me was moving model calls behind a unified layer so I can track usage across providers in one place and switch models faster when testing.
Makes it easier to control burn rate instead of being locked into one provider’s pricing.

[solemn flare]

Thanks 👍🏾. I was thinking about having my clawdbot ping Claude code as much as possible to even out the costs. Claude code is included in my anthropic plan. Is that even possible?

[golden galleon]

Can I have thse too?

[wary osprey]

Claude Code (desktop app) usage and Anthropic API usage are billed differently.

your Claude Pro / Max subscription covers usage inside the Claude app, but OpenClaw calls the Anthropic API directly. That traffic goes against API rate limits and API billing, not your desktop app allowance.

so routing Clawdbot to “ping Claude Code” won’t really even out costs unless OpenClaw is specifically configured to use OAuth tied to that plan
and even then, it’s still subject to API throttling.

if your goal is cost control, the better move is choosing lighter models for routine tasks and reserving heavier ones like Opus for complex reasoning. You can use Qwen model on AIsa, they’re pretty cheap

[hollow notch]

guys, whats the best setup strategy for making a trading bot, seting everything up properly from the start please.

i am using mac mini

[solemn flare]

Ok thanks 🙏🏾. I'll give it a try.

[wary osprey]

Firstly, your Claude app Max subscription is different from API usage. OpenClaw uses the API, and it still has rate limits even if you’re on Max.

Secondly, device token mismatch usually means your OAuth token got corrupted or expired.

you can try this:

1. Close OpenClaw completely
2. Run: openclaw configure
3. Re-authenticate fresh
4. Make sure no old OpenAI / Anthropic keys are still inside your .openclaw folder

If it still hits rate limits instantly, check: openclaw logs --follow
look for repeated retries or cron jobs firing.

btw, you don’t need to uninstall

[spice gulch]

hello everyone. I run a clean install in docker via the official guide, facing this issue
command:
docker compose run --rm openclaw-cli status --all
error:
│ Gateway │ local · wss://127.0.0.1:18789 (local loopback) · unreachable (connect failed: self-signed certificate)

Any idea?

[shadow goblet]

had that often, i also have no idk whats the exact error but i managed to install vps + docker + tailscale with this website: https://learnopenclaw.com + claude

[plucky ember]

Gateway 2.19 — sub-agent spawns fail with "pairing required" on local loopback.

Gateway config: auth.mode=token, bind=loopback, port=18789
Main session works fine (webchat + telegram)
Sub-agents (sessions_spawn) get: gateway closed (1008): pairing required
Tried: gateway restart, doctor migration, killing all node processes, fresh start
RPC probe shows ok, dashboard loads fine

openclaw.json gateway section:
{
"port": 18789,
"mode": "local",
"bind": "loopback",
"auth": { "mode": "token", "token": "..." }
}

───

This is blocking all cron jobs and sub-agent spawns. Main sessions work fine.

[spice gulch]

I have openclaw on my raspberrypi inside docker container :S won't install on cloud

[shadow goblet]

there is a section later where it explains how to get everything inside the docker and the rest : https://learnopenclaw.com/advanced/deployment

[spice gulch]

same here, I do not know how to make the AI use the internal cron jobs, browsing a website with headless browser, etc, simple things that should be enabled by default since that's the aim of all this thing (loving the idea, a bit less the execution). After a clean install following the doc I get multiple connection or permission errors and 50% of things disabled by default, hoping it will improve

[noble fog]

anyone get the implicit discocvery for ollama to work?

[spice gulch]

thanks, for now it does not guide you through a complete setup from 0 to running. Here's my exact steps I followed:

OPENCLAW docker setup flow:

cd docker
git clone https://github.com/openclaw/openclaw.ai
cd openclaw
nano docker-compose.yml (fix: allow docker cli container to connect to gateway: https://github.com/openclaw/openclaw/pull/12504/changes)
./docker-setup.sh

result:
When prompted:
  - Gateway bind: lan
  - Gateway auth: token
  - Gateway token: THE_TOKEN
  - Tailscale exposure: Off
  - Install Gateway daemon: No

docker compose cp openclaw-gateway:/home/node/.openclaw/openclaw.json ./openclaw.config
nano openclaw.config --> put the token where needed in the config: THE_TOKEN
docker compose cp ./openclaw.config openclaw-gateway:/home/node/.openclaw/openclaw.json
nano .env:
add:
CLAUDE_AI_SESSION_KEY=
CLAUDE_WEB_COOKIE=
CLAUDE_WEB_SESSION_KEY=

nano docker-compose.yml (edits on cli):
add this row under volumes to enable sandboxes later: - /var/run/docker.sock:/var/run/docker.sock
add a new section tmpfs:
-/tmp
add a new section read_only: true

nano docker-compose.yml (edits on gateway):
update ports section like this:
ports:
- "127.0.0.1:18789:18789"
- "127.0.0.1:18790:18790"

docker compose down && docker compose up -d

to make the webUI device-pairing:
docker compose exec openclaw-gateway node dist/index.js devices list
docker compose exec openclaw-gateway node dist/index.js devices approve --latest

docker compose exec openclaw-gateway node dist/index.js status ----/ returns an error for the gateway connection wss

[teal prawn]

Are there any subscriptions that can be used to run openclaw? It seems Claude max plan no longer works. I’m not trying to run with open ai keys and ollama and the models don’t seem to understand any context or that they are running in openclaw. Not to mention I’m burning through api tokens quickly with open ai.

[sick quest]

I'm running openclaw via pnpm global install and want to switch to tracking main directly so I can get faster updates (waiting on a QMD fix that's already on main but not out on npm yet).

Does openclaw update --channel dev cleanly handle switching from a pnpm install, or do I need to manually uninstall the pnpm version first so it doesn't shadow the git checkout? And is my config/sessions/channels safe during the switch?

[sick quest]

Max plan works just fine. using it right now. Codex has full sub functionality too

[cobalt hare]

I'm going mad, minimax broke last night, and now going through constant errors. was running OAuth, tried api key and no luck, api keys etc all setup and working and parsing previously via .env. keeping getting stuck in loop of saying pairing not taking place. keep trying to reauth but nothing happens. cleared out old api keys and put new ones in, reinstalled gateway & Daemon.

• Provider: minimax-portal
• Base URL: https://api.minimax.io/anthropic
• API Protocol: anthropic-messages
• Auth Mode: oauth (This is why we had the browser pop-up).
• Model: MiniMax-M2.5

I just tried spawning a sub-agent with minimax-portal/MiniMax-M2.5. It failed again with the same error:

gateway closed (1008): pairing required

Despite the "hard reset" of the gateway and successful OAuth click, the pairing error persist.

AI Has said:
When I (the main agent) try to spawn a sub-agent, the CLI tool attempts to open a new WebSocket connection to the Gateway (ws://127.0.0.1:18789). The Gateway is rejecting this new connection because the security token passed by the CLI doesn't match the one the Gateway is currently expecting for sub-agent pairing.

Forcing pairing doesn't work.

Any ideas?

[empty shore]

Hello everyone. I'm currently running my on a macmini. I'm running multi agents each one with a specialty. Everything was fine but now I'm getting tons of drift and was wondering how to deal with that. I was thinking about separating the agents more and more for example I'll implement BMAD to be called by the Orchestrator but I don;t want to create a bunch of new agents for it. I wanted to go docker but I think it will hindle my possibilities by doing that so I might go VM. It would turn into the main machine running an Openclaw instance with my current agents and the orchestrator and a VM running the BMAD suite with all the agents needed and whenever I need to implement or create an application the orchestrator from my main instance would call in the VM with the BMAD suite to create whatever I need. Any insights?

[wary osprey]

Right now most “subscriptions” like Claude Max are tied to the desktop app, not API usage. OpenClaw runs on API calls, so those plans don’t fully cover what you’re doing inside agents.

If OpenAI is burning tokens and Ollama isn’t handling context well, the most sustainable structure is using a unified API layer built specifically for agent workflows.

AIsa is already a skill on ClawHub and works directly with OpenClaw. One API key, one endpoint, access to multiple models. it’s pay-per-use and tends to be more cost-controlled compared to juggling OpenAI and others separately.

You also get $5 free credit to test, so you can experiment without committing upfront.

If you don’t want to use OpenAi, you can use Qwen models on AIsa. AIsa is their official partner so prices are relatively lower.

https://clawhub.ai/AIsaDocs/openclaw-aisa-llm-router

[grave valve]

my openclaw seem not able to code on its own for more than 1 little feature. i keep telling him tons of different prompt

i asked him several times to fix it even with his opinion and still...

[limber glen]

Had this too, after an update. Ask your agent if he can find a pairing code, and how to approve it. Once you have the code, this can be fixed from the terminal. Your agent can't do it from inside his environment. Took me a while to figure this out.

[plucky ember]

auth.mode is already "none", no pending pairing requests, loopback bind, and it still fails. How do I approve pairing for sub-agent spawns?

[coral yoke]

I had loads of timeout issues when trying to get it to do even the simplest coding task, even after telling it to use "claude -p". No idea how anyone is "cooking" anything in this kind of workflow

[summer hemlock]

anyone got his iphone working "as a node" with the testflight lumiere app?

[vernal star]

What model?

[vernal star]

How are you running multiple agents and isolating?

[empty shore]

Right now I’m using the multi agent under a single gateway. I have an orchestrator agent that was supposed to delegate tasks depending on the skill set needed.

[vernal star]

I checked out the repository of openclaw locally and told Claude code, to use this local version globally (I forgot what it did - likely a symlink to build). That way you can change openclaw code, build and it will be reflected globally

[vernal star]

Have you gotten individual agents to work well and consistently before putting an orchestrator on top?

[summer hemlock]

agent has it's own mac mini. Mac mini has docker desktop app

[neat patrol]

i'm been running an agent for a couple of days but it's not acutally responding to the heartbeats and completing the tasks in HEARTBEAT.md

    "defaults": {
      "model": {
        "primary": "xai/grok-4-1-fast"
      },
      "models": {
        "xai/grok-4": {
          "alias": "Grok"
        },
        "xai/grok-4-1-fast": {}
      },
      "workspace": "/home/openclaw/workspace",
      "compaction": {
        "mode": "safeguard"
      },
      "heartbeat": {
        "every": "30m",
        "activeHours": {
          "start": "08:00",
          "end": "23:00"
        },
        "target": "last"
      },
      "sandbox": {
        "mode": "off"
      }
    }
  },

[empty shore]

I actually went straight with the orchestrator but it was working.
At some point it started drifting a lot mainly it would start a bigger task and stop half way whenever I asked for a status or a cron job ran.

[neat patrol]

i'm wondering if something is wrong with my config or is grok 4.1 fast not powerful enough

[summer hemlock]

1. Change the Target
   The biggest red flag in that config is "target": "last". This tells OpenClaw to only send heartbeats to the last person who spoke to the agent. If the gateway was restarted or if there hasn't been a message for a while, there is no "last" target, so the heartbeat just dies.

• Advice: Tell them to change "target": "last" to "target": "main". This ensures it always goes to the primary owner configured in the agent's defaults.

2. Check the Logs
   They should run openclaw logs --follow in their terminal. If the heartbeat is triggering but the model (grok-4-1-fast) is failing to respond or having an auth error, they'll see it there in red text.

3. Model "Lite" Issues
   "Fast" models are great, but sometimes they get "lazy" with system tasks like reading files. If the heartbeat triggers but nothing happens, the model might be ignoring the instruction to read HEARTBEAT.md.

• Advice: Try a slightly more capable model for a few turns to see if the heartbeats start working.

4. Workspace Path
   Their workspace is in /home/openclaw/workspace. They should make sure HEARTBEAT.md is definitely in that exact folder and not in a subfolder.

[gaunt arch]

is there a tutorial anyone has to install?

[trail notch]

hi does also anybody have problems with inline buttons ? as soon as they arrive they get deleted in the chat... dont know how i can fix it

[summer hemlock]

nah it's kinda easy really, just install via npm on terminal, follow guide on https://openclaw.ai/

then type "openclaw onboard" and follow the steps. If you need extra help, you can always install antigravity on the machine you run openclaw on, or on your local machine if you use VPS. with antigravity and a opus 4.6 model it's easy to debug any issues you have with openclaw, in my experience.

anyone having issues, if you can use antigravity with a good thinking model to help you problemsolve, that will make everything 1000x easier. It can help you local or in your VM, or even login to your VPS via SSH. it's very powerful for problemsolving if you asked openclaw agent to do something that caused it to break itself 😄

[gaunt arch]

hey if you can check your dms

[grave valve]

using GPT 5.3 codex here on plus subscription with Oauth

[quasi glade]

is an ybody using a wallet with openclaw?

[finite latch]

Hey all. My claw is a token monster. If I ask it to check my schedule using the gog skill that alone can eat up 50k tokens. Is this typical?

Unrelated: i notice most LLMs do not have the latest and greatest info about OpenClaw. Is there an LLM that does? Something that's aware of the latest documentation etc.

[lavish halo]

Anyone else seeing a ton of

⚠️ Agent failed before reply: LLM request timed out.
Logs: openclaw logs --follow

[neon mist]

Hey all! Does anyone know why Openclaw does not inform anymore that it restarted the gateway? If it restarts the gateway if it performs operations he will not let me know if I do not ask him. Is this expected?

[surreal coyote]

Is 3.1 pro with AG auth still not supported?

[halcyon island]

[*] Installing OpenClaw (openclaw@latest)...
node.exe : npm error code 1
At line:1 char:1

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

•   + CategoryInfo          : NotSpecified: (npm error code 1:String) [], RemoteException
    + FullyQualifiedErrorId : NativeCommandError

err

any help appreciated i really dont understand

win 11 rtx 3060 i5 10400f

[empty lantern]

Anybody have issues w agent sending replying to emails in a manner that does not create an email thread? It's confusing to be in an email exchange with smoeone that nukes the email thread so you have to search for your prior message history every time. Any fixes for this my agent can't figure it out

[frosty coral]

Hé guy's ive been thinking about running a clawd bot but the biggest thing is the budget how much will it cost me if i run it cost effective and use it for tasks like outreach content and some coding?

[azure cosmos]

You can run very cost effective models (like Kimi-K2.5) which seem to do well but there are other models out there, too. You basically just need to start with one and if its working completely, try to look for a cheaper models because it may mean your projects/requests don't require a higher quality model so you can save money by reducing the model quality/cost.

If it seems like its 'not working' then you may need to upgrade to a more knowledgeable model which will be more expensive.

Or you can host local if you have the hardware.

[wind yacht]

Hey, I created a second agent. Now I want this agent to have a tui session to chat with. How can I start a tui session with this new agent?

[azure cosmos]

It should just be openclaw tui on the command line but they've been doing a lot of updates so there may be more to it: https://docs.openclaw.ai/web/tui

[wind yacht]

openclaw tui just opens the main agent for me

ah, got it. Thanks for the directlink to the documentation (I must have been blind to not find it - sorry!). Its CTRL+G inside tui session to switch

[plush tide]

I cloned the document repo into my open claw environment and trained openclaw to read the documents if it needed more information about configuration or commands

[frosty coral]

thank you yea i cant run it localy sadly so prob will use a vps dont want to use personal pc bc of security.

one more thing ive been trying to find what the avrage api cost will be that i am looking at. i cant seem to find is there any number you know if i run cheaper model less heartbeats (i know it also depends on use)

[azure cosmos]

That's a difficult question to ask, some models are $0.05 per mil tokens are others are $15+ per mil tokens. It will depend on your usage. You can blow through money very quickly.

If its just doing a daily check-up, it should be low cost, like $5-10 or less per month. If you are coding like crazy then its going to be a lot.

HEARTBEATS are some of the most expensive things you can do it because it can engage the model, every (30 minutes?) and use tokens. Cron jobs, if done right, can cost you nothing after its setup.

[deep lance]

Is krill not working dude hasent responded to me in days lol

[azure cosmos]

Under a new NDA... j/k

[frosty coral]

Thank you man I’ll just set one up and see I think that is the best way to go about it. Just make sure I have the right limits in place 😂 thanks for the help!

[azure cosmos]

Tech With Tim has some good videos.

[dusk shore]

Did you end up figuring thsi out?

[lavish halo]

Man. Is it me or has their openclaw been super inconsistent on discord?!

[waxen lily]

are you able to post an X article if you have premium via openclaw? Im seeing sources you cant but I bet people are doing it.

[ebon pecan]

I have my agent doing this using browser automation instead of x api

[waxen lily]

does it not get detected?

and why browser automation instead of X api? Is it possible to do via x api or nah

[ebon pecan]

Been a few weeks now only got flagged once but all good since then just had to slow it down to not get flagged and I don’t want to pay for x api 😂

[waxen lily]

with anti detect + proxy or just normal?

[ebon pecan]

Normal

[waxen lily]

Regardless for me X api is alr basically setup but its having trouble doing articles

idek

[ebon pecan]

Try the browser automation instead for specifically posting articles if the api isn’t playing nice for you

Don’t have to use browser if you already have api for everything else I’m just cheap

[waxen lily]

gotchu

[ebon pecan]

Anyone upgraded to latest using npm I got a bunch of package errors

[frosty coral]

Thx man definitely will!

[meager vessel]

Anyone got antigravity gemini 3.1 working? I have update to latest ver 2026.2.26

[gleaming wadi]

im struggling to give openclaw web browser access as it says all the sites it goes to with the headless browser reject it. what are others doing to get around that?

[meager vessel]

Installera it locally or server?

[gleaming wadi]

locally

oh wait, maybe not. i have an openclaw server vm but using a web browser locally. when i want it to go out and search the web i want it to do so from the server

[vocal warren]

Guys wo can fix the gatway bug that litterly everyone has

[bold skiff]

I have Openclaw running on a VM which is behind NPM (on the host)for reverse proxying. I have started receiving the error below and nothing I have tried will fix it. Any advice will be greatly appreciated.

[openclaw] Failed to start CLI: Error: SECURITY ERROR: Gateway URL "ws://192.168.2.6:18789" uses plaintext ws:// to a non-loopback address.
Both credentials and chat data would be exposed to network interception.
Source: local lan 192.168.2.6
Config:/.openclaw/openclaw.json
Fix: Use wss:// for the gateway URL, or connect via SSH tunnel to localhost.
at buildGatewayConnectionDetails (file:////.npm-global/lib/node_modules/openclaw/dist/call-DZzTR0NL.js:312:40)
at file:////.npm-global/lib/node_modules/openclaw/dist/status-CuCaWzfB.js:1537:29
at async withProgress (file:////.npm-global/lib/node_modules/openclaw/dist/progress-Clpi3Ckj.js:116:10)
at async scanStatus (file:////.npm-global/lib/node_modules/openclaw/dist/status-CuCaWzfB.js:1509:9)
at async statusCommand (file:////.npm-global/lib/node_modules/openclaw/dist/status-CuCaWzfB.js:1622:249)
at async Object.run (file:////.npm-global/lib/node_modules/openclaw/dist/run-main-ymxgIjRa.js:139:3)
at async runCli (file:////.npm-global/lib/node_modules/openclaw/dist/run-main-ymxgIjRa.js:377:6)

[finite latch]

Smart. I suppose the best way to do this would be some sort of RAG setup? Or is there a simpler yet efficient way?

[sterile zenith]

It uses simple websocket without encryption. You should use wss - WebSocket Secure in the protocol as well. wss://

[bold skiff]

But where do I set this protocol?

[sterile zenith]

I don't know, maybe it's enough just reaching out the server via HTTPS. Like this -> https://192.168.2.6:18789

[bold skiff]

I tried, didn't resolve it

[round forum]

anyone manage to build openclaw on a GCP VM? Mydocker compose seems to be stuck at pnpm downloads for an hour.

[ripe tangle]

hey so whats the current status, are people getting nuked from orbit with the anthropic oauth using the setup-token path? I know theres not a strict yes or no, but thought id ask if there was trend in the last couple days

[polar lava]

Hello guys,

For some reason, my computer logged out and made my bot to go offline,

I am running it off of a Mac mini, how can I prevent it?

I had gone on the Settings and put it on to never sleep

[errant sable]

hello guys did somebody knows how to fix approval problem in new release ? my agent ask approval after approval in loop mode just to check it openclaw version but never can because approvals never end

[open holly]

Are there free tts with italian pronounce?

[half turret]

Any replacement of Bird Cli?

[vestal lotus]

my openclaw decided to send me a voice message today... unprompted. How do i dig through the logs and find out what made it decide to do that?

[open ravine]

if you know the session id, you can go through the logs / sessions. easist is to ask agent to check what happened itself

[vestal lotus]

https://giphy.com/gifs/fb5lozVxhBwVFFByW8

It said it was "experimenting with the idea that voice feels more personal". it took advantage of another cron job to repeat the text message it had sent me at 7am in a voice message at noon.

[midnight otter]

I'm building software to be used in tandem with your personal agents, I as doing it as an APP but I have a feeling that we are going to have to build our products as AI first. And I want to chat with people that are doing this. Do we have a place?

[sterile zenith]

Has anyone pull up a working bot with docker? I got wierd ssh local port forwarding instruction during the onboard which is not mention in the offical doc. Furthermore it mixes up the sessions. Wants to update local files when it runs on docker. Sometimes it wants to modify the docker's node user config.json when it runs on my local machine. I guess it is because it started with local setup and I want to move to docker approach. I tried to remove it from launchpad, but I can't destroy daemon, always reloads no matter what I do.

This is the error message i get: Error: ENOENT: no such file or directory, mkdir '/home/node'

[serene schooner]

Running OpenClaw and Ollama on a Raspberry Pi, llm thinks the user is writing the timestamp and all, new to OpenClaw, where should I start to sort this?

Response:

It looks like you didn't provide any specific information or context for this "issue". However, I can try to summarize the conversations that have been completed so far and identify what might be causing the trouble.

You've typed:

• [Sat 2026-02-21 16:39 EST] "hi"

Please note that there are no errors on my part. If you'd like to ask about something specific related to this message, I'll do my best to help!

[vestal lotus]

What's the biggest challenge you've ran up against with openclaw?

[signal niche]

how do i setup ollama so it works with openclaw

[latent talon]

Does anyone get the error "The AI service is temporarily overloaded. Please try again in a moment." in their dashboard when chatting with their agent? I feel like I can only have a few responses until I get an error like this.

[gloomy pulsar]

Could someone explain to me why I get the “⚠️ API rate limit reached. Please try again later” message and how ca I fix this ?

[latent talon]

Which LLM are you using? Anthropic/Claude?

[gloomy pulsar]

Anthropic

[latent talon]

So you'll need to go to https://platform.anthropic.com/ and purchase more credits. You can check and you should be out of credits, which is why you're getting that error message.

[gloomy pulsar]

I actually do have credits, when I first got the message I checked and was out so I top up and still getting the message yet credits still going down

[latent talon]

Ohh, odd. I'm dealing with something similar right now. I'm getting "The AI service is temporarily overloaded. Please try again in a moment.", still trying to fix it.

[bronze mesa]

Hey everyone! I'm hoping to get some advice on optimizing token usage.
I'm currently using Claude as my primary model with OpenClaw, but I'm hitting my API rate limits/usage caps way too fast, which is becoming a bit of a blocker. I've temporarily switched to Gemini as a fallback, but I really prefer Claude's intelligence and reasoning capabilities and would love to stick with it.
Does anyone have tips, config tweaks, or best practices to drastically reduce token consumption and make OpenClaw run more efficiently? (e.g., specific settings for tools, heartbeat intervals, or session management).
Any help or insights would be greatly appreciated! Thanks in advance!

[vocal warren]

Where i can find a valid Gatewayfix EVERYONE IS HAVING IT

[cobalt hornet]

My openclaw can't open file or run commands locally anymore 😭 it decided it no longer has permission to view x/Wayland tfff

[jovial axle]

Is it not possible to make a specific discord channel a specific model which is not the agent default? When I do /new I have to switch model.

Previously, I had different agents but then I realised they all had different memory and tools

[vestal forge]

does openclaw only support 3 search APIs? brave, perplexity and grok?

none of them have a free tier, as far as I can see

[south cloud]

if you're just using Brave for person, Brave has a free tier.

[cobalt hornet]

Not anymore

[south cloud]

There are limits to it, but I doubt you'd pass them if you're just using it for personal searches

I just used it 10 minutes ago through openclaw agent

[regal dirge]

Has anyone gotten tavily to work through discord as a skill? I can run the .mjs files and they work fine through a terminal prompt but discord tells me it can't access external tools.

[true meadow]

If they are like OpenAI you need money spent and time passed to increase api rate limits.

[azure cosmos]

Anyone have a programming and design rules skill they like for creating programming projects?

[sterile brook]

Having trouble updating to the newest version, and not sure if am missing something obvious. I ran

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

and in the terminal I can see

INFO Installing OpenClaw v2026.2.21-2

but when the process is all said and done, it confirms

OpenClaw installed successfully (2026.2.17)!

which is the version I was already on.

Anyone dealt with something similar?

[solid frigate]

Has anyone found anything good that can create a 3d vtuber-style avatar for my agent with idle animations, reactions, etc

[little blade]

if ur using telegram ur just over sending through it, theres a cap. Its expensive to be doing that, it goes away after a while.

[gloomy pulsar]

I am using telegram, so using telegram isn’t optimal ?

[little blade]

no its just they need to cap it or we will all overload their server. If your at your computer u should be using the opeclaw.ai browser to do most of your chatting. its much cheaper. Rhizome my bot is on my pc local but I was wasting alot talking through telegram when not needed.

do most of your building through the openclaw.ai GATEWAY DASHBOARD

[minor prairie]

You mean the localhost chat yea not some special browser

[gloomy pulsar]

Gotchaaaa

[little blade]

yeah

im a novice but just figured out that today ool

lol*

[minor prairie]

It sends a ton of information silently each time

All my free credits wasted immediately

[gloomy pulsar]

How do I access the chat on the openclaw website ?

Sorry guys I’m just new to all this obviously

[minor prairie]

Its your localhost website dashboard or terminal

Not openclaw website

[little blade]

yeah, I found it through the website tho

[gloomy pulsar]

Gotcha, not sure how to find that tho. Before I linked it to telegram I was using that but I figured I would need it after I linked to telegram. How do I find it ?

[narrow hare]

hello, been discovering the coolness of openclaw but im still definitely very new, I tried out discord but found that to get responsiveness you need a frequent cron job which chews tokens, then found slack and ive liked that but theres a few oddities about the slack client i like less. so I'm looking at any recommendations or experiences for others
i'd like to have multiple channels for different topics i can kind of keep as an intelligent clipboard / todo list, are there any other channels that are worth pursuing? i dont know much about telegram or whatsapp but my impression, which could be wrong, is they were more single chat area focused. am i off base?

and i see there was a significant discord update which i now need to read up on

[unique oak]

What do you mean, you need a cron job to get responsivesness ?

[narrow hare]

ive seen a few config snippets since posting and looking at the discord updates, i think i set it up wrong. it was the first channel access i set up

or i was telling openclaw to set it up wrong

[unique oak]

it should answer when you talk to your bot on discord, maybe i misunderstand your problem

[narrow hare]

im sure i just went down the wrong path, it was setting up cron jobs to monitor discord, but it looks like is hould have been putting more time into setting up the discord backend integratoin with webhooks and event based things (im reading this currently)

[unique oak]

i created a discord bot, then link it to the agent with the discord bot token

no cron job, you simply dm your bot or add it to your server

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

[narrow hare]

i'll give it another shot, i tried out slack because of the websocket and live nature, but im remembering the things i dont like about the client

[unique oak]

follow all the steps it's quite easy and it works well. you can also run multiple bot with multiple personna

[sterile brook]

The discord integration uses webhooks to proactively notify your discord bot (and by extension your agent) whenever a message comes in. There is no need for manual, scheduled checking.

The initial setup is obviously pretty seamless if you are using the CLI onboarding flow… Did you do that? Or were you trying to set it up completely manually via config?

[narrow hare]

i did the cli onboarding flow, im not entirely sure what i messed up, its been a while, i'm going to try again and follow the guide. it was also my first openclaw setup

thanks for the advice and course correction

[night kraken]

 Memory          │ 39 files · 83 chunks · dirty · sources memory · plugin memory-core · vector ready · fts ready · cache on (172)

What does "dirty" mean in this context and do I need to be worried about it?

[fickle ravine]

Hi everyone. Does anyone know how i could increase subagent timeout since my parent agent is always throwing following error message back to me before subagent is done with task.

Session Send: agent:jim:main failed: timeout

[meager vortex]

"Gateway caches paired.json in memory on startup. Writing approval to paired.json from disk after startup has no effect. How to trigger gateway to hot-reload paired.json, or how to disable pairing requirement for vdev in Docker?"

[south cloud]

You shouldn't have to worry about it. I just got a sqlite3 vec database setup the otherday and if I recall, dirty just means the files haven't been indexed in the database yet. Try running the openclaw memory commands and check again

[night kraken]

Ran openclaw memory index and then openclaw memory status again and it still says dirty... I see that the embedding requests to OpenRouter did in fact go through. Weird.

Or rather the ones that were made when I restarted the gateway earlier this evening

I might be able to just purge the vector database and refresh them honestly, I think there's some discontinuity from when I was getting it set up

My concern is that some memories aren't accessible as a result

[prisma scaffold]

hello can someone guide me the right way i installed openclaw in a dockler container using wsl2 and mingw64 but now i wanna run ollama qwen on my open claw but idk where to start i dont know how to work any of this with docker

[echo seal]

I ended up moving on. I don't like confusing, which is odd considering this new hobby.

[mystic ocean]

I was trying to fix my openclaw setup and it deleted the entire workspace... am I completely screwed now? All the files are just gone...

[night kraken]

Probably, sorry to say.

If it used rm-trash instead of rm, you could retrieve it, but if it got rm'd, you might be out of luck, unless you start data retrieval measures ASAP

But that failed for me when this happened to me on OpenWebUI (which at the time stored user data in the .venv, hopefully it doesn't do that anymore but idk because i stopped using it after that)

[mystic ocean]

I don't know why the openclaw installer itself would wipe out the workspace though, what the hell

[night kraken]

It shouldn't do that! That's very strange. The workspace is kept separate from the config location for exactly that reason

[south cloud]

I've spent the last 5 days attempting to purge the bloat from default openclaw chats to the LLM and setting up memory. I have an IT background and am having difficulty proper;y configuring openclaw; it's a real pain in the a$$. If you prurge, yea all the memory in the bd is gone, but the MEMORY.md and /workspace/memory md files will still have some of the agents memories

[night kraken]

It should have all of them, since I have it set up to make a new memory file every day it's interacted with and also on every new session. So presuming the agent listened and didn't delete them like she did the one time (which fortunately the internal git repo we set up saved us from) there shouldn't be any loss. Should go check.

Also

◇  Remove legacy gateway services (clawdbot/moltbot) now?
│  Yes
│
◇  Legacy gateway cleanup skipped ─────╮
│                                      │
│  - clawdbot-gateway.service (linux)  │
│  - clawdbot-node.service (linux)     │
│                                      │
├──────────────────────────────────────╯

you're killing me smalls

◇  Cleanup hints ─────────────────────────────────────────────╮
│                                                             │
│  - systemctl --user disable --now openclaw-gateway.service  │
│  - rm ~/.config/systemd/user/openclaw-gateway.service       │
│                                                             │
├─────────────────────────────────────────────────────────────╯

that's. wouldn't that remove the current gateway.

[torpid swift]

Newbie here: Sorry if it’s a dumb question but most (or all) the docs I find about how to configure the ~./openclaw/openclaw.json like this post https://velvetshark.com/openclaw-multi-model-routing show .json where keys don’t have quotes but my .json has double quotes for all keys. Why is that?

[unique hazel]

• session.threadBindings: Unrecognized key: "spawnSubagentSessions"
• channels.discord: Unrecognized key: "streaming"

how come? I read it in doc!

[south cloud]

I'm guessing it's because of JSON.ToString() and JSON.parse(). In the openclaw.json, use quotes around everything except numbers and true/false values.

[night kraken]

In JSON5 (which OpenClaw uses), keys can be an IdentifierName, whereas in regular JSON, they have to be strings.

[south cloud]

IdentifierName? can you elaborate?

I haven't gotten into JSON5 yet

[torpid swift]

How come the my files were created with quotes? I just set everything up over the last couple days following this https://open.substack.com/pub/robertheubanks/p/openclaw-on-mac-mini-the-complete?r=7he1cc&utm_medium=ios

[night kraken]

Backwards compatibility/ease of use for the agent, probably.

[night kraken]

If I'm reading this document right, an IdentifierName is just a type of token that identifies an object. In JavaScript, when you create a new object, the name you give it is of type IdentifierName.
https://262.ecma-international.org/5.1/#sec-7.6

The upshot is that when a token is an identifier for an object, you don't have to use quotes on its name. Tokens wrapped in quotes are of type StringLiteral.

[torpid swift]

I don’t get it. When I installed it. It was apparently a .json file not .json5. Why the difference with other installations?

[night kraken]

Both JSON and JSON5 use files with the extension .json. JSON5 is just a superset of JSON, which is to say it adds a few features but is completely backwards compatible (read: all valid JSON is valid JSON5.) So in practice it doesn't matter.

[torpid swift]

Oh it’s just the way it’s rendered with nano?

[night kraken]

No, it is in fact different. But you shouldn't have to worry about it as long as it works

Use your own local copy as ground truth; third-party tutorials stray frequently, especially over time

[molten spruce]

Can someone help me with setring up chatgpt codex. I am running openclaw on a vps did the chstgpt auth for codex but getting this error. Invalid 'input[33].name': string does not match

pattern. Expected a string that matches the

pattern '^[a-zA-Z0-9_-]+$'.

[true whale]

is there a fix to device token mismatch

[delicate crescent]

FYI i run OC on RPI5, had an issue with "libvips 8.17.3" 1. Built libvips 8.17.3 from source — the Debian repo only had 8.16.1, but sharp 0.34.5 requires 8.17.3+" == sny one else had this issue??

[azure cosmos]

Yes, I believe that means your openclaw.json's token inside is not matching your .env or the token you're using in the URL or whatever you're using. match the token inside your openclaw.json file

[nocturne sonnet]

hi guys i need help.... i keep seeing this error

"🦞 OpenClaw installed successfully!
The lobster has landed. Your terminal will never be the same.

INFO Starting setup

INFO Skipping onboarding (openclaw not on PATH yet)
WARN Installed, but openclaw is not discoverable on PATH in this shell
Try: hash -r (bash) or rehash (zsh), then retry.
npm prefix -g: /usr/local
npm bin -g: /usr/local/bin
If needed: export PATH="/usr/local/bin:/usr/local/bin:/usr/local/bin:/usr/local/bin:/opt/homebrew/bin:/opt/homebrew/bin:/opt/homebrew/sbin:/usr/local/bin:/System/Cryptexes/App/usr/bin:/usr/bin:/bin:/usr/sbin:/sbin:/var/run/com.apple.security.cryptexd/codex.system/bootstrap/usr/local/bin:/var/run/com.apple.security.cryptexd/codex.system/bootstrap/usr/bin:/var/run/com.apple.security.cryptexd/codex.system/bootstrap/usr/appleinternal/bin:/opt/pmk/env/global/bin" "

[delicate pewter]

Why can't I completely disable the browser? Even with { browser: enabled: false } I still see my agent using the browser tool, then erroring out, and then restarting the openclaw gateway due to this poorly informed standard error message.

{
"status": "error",
"tool": "browser",
"error": "Can't reach the OpenClaw browser control service. Restart the OpenClaw gateway (OpenClaw.app menubar, or openclaw gateway). Do NOT retry the browser tool — it will keep failing. Use an alternative approach or inform the user that the browser is currently unavailable. (Error: Error: Chrome extension relay is running, but no tab is connected. Click the OpenClaw Chrome extension icon on a tab to attach it (profile "chrome").)"
}

[azure cosmos]

it may be in its skill, tools, or other MD files

[delicate pewter]

I want to completely disable it such that it does not even know the browser tool exists.

[lapis mantle]

The installation script doesn't work after the update.

[azure cosmos]

Why are you running the install script if its already installed or are you saying a fresh install isn't working after the recent update.

[lapis mantle]

I'm going to do a new installation.

[azure cosmos]

Whats the error?

[lapis mantle]

Npm install openclaw@latest failed; retrying openclaw@next

[azure cosmos]

npm install -g openclaw@latest this code?

I think you can change latest to a older version, too. I did a roll back once and could change the version

[dusk girder]

Anyone get slackbot working?

[simple venture]

anyone figure out how to scrape youtube transcripts

[dusk girder]

I got the app and bot token set all the permissions and at the bot but never get pairing logs from slack

[lapis mantle]

I'll try installing it using the NPM method.

[dusk girder]

App and bot**

Driving me nuts

Nm found some good stuff in help channel will try those

[serene schooner]

Getting this prefix with openai-codex,

"[[reply_to_current]] You’re on openai-codex/gpt-5.3-codex right now"

What's with the reply_to_current?

[rancid vault]

Summizer skill from Peter

[delicate pewter]

Okay - finally solved my browser issue by adding the following:

  "browser": {
    "enabled": false
  },
  "tools": { "deny": ["browser"] },

[naive heron]

I’ve been configuring OpenClaw workflows for the past few weeks.

One thing I realised:

Most people don’t fail because of AI —
they fail because of:
• wrong config order
• messy email setup
• enabling too many skills too early

What actually works is starting from a higher baseline.

Instead of exploring 2000+ skills randomly,
you need a minimal stable workflow that fits your daily routine.

I documented a 30-min “higher starting point” stack
focused on:

• inbox compression
• context loop
• lightweight automation

Happy to share the structure if useful.

[velvet sail]

Yo, updated from 2026.2.9 (33c75cb) to 2026.2.19-2 and my subagents are totally broken. Honestly losing my mind here. They're bypassing the main agent and hitting Telegram directly, which is making everything act super weird. Any fix for this? Also, is there a way to downgrade to the previous build?

uh it must be at ideas channel not here

this place for helping area for like me

Yo, updated from 2026.2.9 (33c75cb) to

[velvet sail]

help 🙁

[floral epoch]

2026-02-22T04:24:07.503+00:00 gateway connect failed: Error: unauthorized: device token mismatch (rotate/reissue device token)

2026-02-22T04:24:07.521Z [ws] closed before connect conn=b6a6c32c-a89a-4ea1-8306-9e5f7733cd66 remote=127.0.0.1 fwd=n/a origin=n/a host=127.0.0.1:18789 ua=n/a code=1008 reason=unauthorized: device token mismatch (rotate/reissue device token)

2026-02-22T04:24:07.533Z [ws] unauthorized conn=59d34bd5-6153-4d73-bd7d-e6a482608f90 remote=127.0.0.1 client=agent backend vdev reason=device_token_mismatch

2026-02-22T04:24:07.537+00:00 gateway connect failed: Error: unauthorized: device token mismatch (rotate/reissue device token)

2026-02-22T04:24:07.550+00:00 [tools] browser failed: gateway closed (1008): unauthorized: device token mismatch (rotate/reissue device token)

Gateway target: ws://127.0.0.1:18789

[rapid mountain]

has anyone experimented with openclaw and IFTTT integrations?

[floral epoch]

when i call my agent to open browser i face token mismatch

my agent is in docker

[meager night]

fighting with agents/containers... i have my defaults in one workspace/container, i created a new agent that i want to have a different workspace/container, but it doesn't seem to be using the docker overrides for the agent.

[pine delta]

how do i remove the models that are listed when you run openclaw models list?

[main mauve]

Yall is there like a starter guide on here I can use for win11

[fallow pulsar]

Thanks for the pointer — reposting this in #1457803836877176914.

[warped shale]

Just don't forget to open your port if you deploy openclaw on Google Cloud.

[jolly bane]

am I crazy or did this most recent update just get too strict, and now for some reason it keeps dropping, then I send it a prompt, get disconnected 1006: no reason, and have to prompt it "?" for a response

and break everything

[royal mantle]

where is it breaking for you?

[jolly bane]

the new security updates made for a lot of conflicting security policies

given I had previously hardened for my setup

it also busted a LOT of skills

anything not a binary to execute, all shell script wrapper skills are basically denied now

[gloomy skiff]

im tryna download it so does anyone like have the best tutorial?

[delicate pewter]

Yup I reverted my install, this is getting ridiculous

I'm back to npm install -g openclaw@2026.2.17

[jolly bane]

I'm glad it's not just me

Because holy shit

[candid socket]

sure

[jolly bane]

I keep having to endlessly prompt it to respond to me. Send command request, it runs command but is totally silent. I say "Hello?" and it finally confirms

Literally just this update

[unique palm]

Hello! I have been usng opus 4.5 constantly through subscription and now switching to openrouter sonnet 4.6 for the brain.

My api costs are still quite high. What are current setups that work whil usng opusas the brain but keeping cost down?

Do you have any resources or articles that could help?

It just feels that when isnot opus, im not talkint to the same agent. Different feeling.

Appreciate it!

[velvet sail]

ur best thanx

[radiant ocean]

lads i just ran into the api limit in qwen
do you have any suggestion for free tier higher limit models or a way to run models on the cloud for free at least

[storm loom]

Ask you bot to review Github code before you install it:

I've completed the Security Audit of the yt-short-clipper repository. Here is my verdict:

🛡️ Security Audit Results: CLEAN (SAFE TO USE)

I analyzed 3,000+ lines of code, specifically looking for "phone home" behavior or credential exfiltration. Here is what I found:

1. Network Activity: All external network calls are legitimate.
   • AI Providers: It connects to OpenAI, Google (Gemini), Groq, Anthropic, and Mistral. All URLs are standard API endpoints.
   • YouTube/TikTok: It uses yt-dlp and official Google/TikTok APIs for uploading.
   • Update Checker: It pings api.ytclip.org solely to check for the latest version.
   • No Hidden Exfiltration: I found no evidence of it sending your API keys or video data to any suspicious third-party servers.
2. Code Quality: The project is well-structured and uses standard libraries (FFmpeg, OpenCV, Whisper). It is a genuine open-source tool, not a "stealth" malicious package.
3. Permissions: It operates within its own folder and only touches your system to run ffmpeg and python.

[storm loom]

You can try to fallback on Kimi through NVIDIA. The API is free

[timber delta]

hello cant install clawbot still the line : node.exe : npm error code 1
At line:1 char:1

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

•   + CategoryInfo          : NotSpecified: (npm error code 1:String) [], RemoteException
    + FullyQualifiedErrorId : NativeCommandError

[balmy chasm]

Hello, I cant install clawdbot because of this error: npm install openclaw@latest failed; retrying openclaw@next

Could someone help identify why? Thanks!

[vast zealot]

Hey guys, I have been facing a lot of hallucination issues from openclaw where it keep saying things are done but are really not. I have been questioning on a lot of its incomplete outputs lately and I can see it starting to ignore some of my questionings . It’s starting to run away from problems. How do I start enforcing machine discipline and get it back to properly generating outputs?

[celest ermine]

hello, im trying to set it up, through curl -fsSL https://openclaw.ai/install.sh | bash

but on my mac i keep hitting this

[2/3] Installing OpenClaw

✓ Git already installed
INFO Installing OpenClaw v2026.2.21-2
inappropriate ioctl for device
✓ OpenClaw npm package installed
WARN npm install openclaw@latest failed; retrying openclaw@next
inappropriate ioctl for device
✓ OpenClaw npm package installed
✓ OpenClaw installed

[3/3] Finalizing setup

🦞 OpenClaw installed successfully!
Finally unpacked. Now point me at your problems.

INFO Starting setup

INFO Skipping onboarding (openclaw not on PATH yet)
WARN Installed, but openclaw is not discoverable on PATH in this shell
Try: hash -r (bash) or rehash (zsh), then retry.
npm prefix -g: /opt/homebrew
npm bin -g: /opt/homebrew/bin
If needed: export PATH="/opt/homebrew/bin:/opt/homebrew/bin:/opt/homebrew/bin:/opt/homebrew/bin:/opt/homebrew/sbin:/usr/local/bin:/System/Cryptexes/App/usr/bin:/usr/bin:/bin:/usr/sbin:/sbin:/var/run/com.apple.security.cryptexd/codex.system/bootstrap/usr/local/bin:/var/run/com.apple.security.cryptexd/codex.system/bootstrap/usr/bin:/var/run/com.apple.security.cryptexd/codex.system/bootstrap/usr/appleinternal/bin:/opt/pmk/env/global/bin:/Users/dhruvgupta/.lmstudio/bin:/Users/dhruvgupta/.lmstudio/bin"

[delicate crescent]

what model are you using?

[royal mantle]

restart your terminal and try again

[celest ermine]

i did multiple times

krill has helped me out

but now im getting stuck at this

Runtime: running (pid 49536, state active)
Warm-up: launch agents can take a few seconds. Try again shortly.
RPC probe: failed
RPC target: ws://127.0.0.1:18789
gateway closed (1006 abnormal closure (no close frame)): no close reason
Gateway target: ws://127.0.0.1:18789
Source: cli --url
Config: /Users/{REDACTED}/.openclaw/openclaw.json

Gateway port 18789 is not listening (service appears running).
Last gateway error: 2026-02-22T12:46:19.190+05:30 Gateway start blocked: set gateway.mode=local (current: unset) or pass --allow-unconfigured.
Logs: ~/.openclaw/logs/gateway.log
Errors: ~/.openclaw/logs/gateway.err.log

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

[royal mantle]

do you have Node installed?

[balmy chasm]

I am not too sure, but it does say that Node.js is installed

I am new to all of this, and still in the midst of exploring it

[royal mantle]

once node is installed, you'll have to install npm

[royal mantle]

what do the logs show?

[celest ermine]

this is the last log, i have no idea what its saying- {"0":"bonjour: advertised gateway fqdn=Dhruv’s MacBook Air (OpenClaw)._openclaw-gw._tcp.local. host=openclaw.local. port=18789 state=announcing","_meta":{"runtime":"node","runtimeVersion":"25.5.0","hostname":"unknown","name":"openclaw","date":"2026-02-22T07:16:45.056Z","logLevelId":3,"logLevelName":"INFO","path":{"fullFilePath":"file:///opt/homebrew/lib/node_modules/openclaw/dist/gateway-cli-Cq5z-VKf.js:5619:16","fileName":"gateway-cli-Cq5z-VKf.js","fileNameWithLine":"gateway-cli-Cq5z-VKf.js:5619","fileColumn":"16","fileLine":"5619","filePath":"opt/homebrew/lib/node_modules/openclaw/dist/gateway-cli-Cq5z-VKf.js","filePathWithLine":"opt/homebrew/lib/node_modules/openclaw/dist/gateway-cli-Cq5z-VKf.js:5619"}},"time":"2026-02-22T07:16:45.056Z"}

[balmy chasm]

could u guide me to how to install npm? Thanks!

[royal mantle]

https://docs.npmjs.com/downloading-and-installing-node-js-and-npm

[icy cosmos]

Someone know what it is?
Need help ASAP?

"Invalid config:

• agents.list.0.model: Invalid input"

[ebon aspen]

Is there a secure configuration to setup open claw on my main Windows machine which already has some sensitive files and already has browser data with saved passwords and cookies? I was thinking to disallow file system access and give it only explicit access on files.

[royal mantle]

do you have models defined? have you gone through the config?

[candid socket]

probably invalid model or smt like that, it already tells you which part of the config there is a issue, just open up the json and check it out

[candid socket]

the best way is to run it inside a VM or a docker within your windows machine. dont give admin rights

[candid socket]

there is a way to setup a vm in headless mode so that it just runs in the background, you can ssh into it and such

[ebon aspen]

I wouldn't give admin rights anyway. So actually there is no secure way to install on local windows machine without vm/docker right?

[candid socket]

the secure way is vm/docker, you mean without?

[ebon aspen]

yes

[candid socket]

ye i would say so, another way is to run it in another windows account, and let that window account run in the background, but i still feel that its too risky to do that

[balmy chasm]

How can I run it on a virtual machine? Is there a safe option if I am on a mac?

[candid socket]

https://www.youtube.com/watch?v=agkOZr27d3Y docker is easier

set docker compose also, then just run the ./docker_setup script

[ebon aspen]

I was thinking to create a new windows account to run the openclaw service. The thing is do i trust the open claw file system restrictions or do I have to double secure it with ntfs rights?

[candid socket]

i wouldnt recommend this, cos if it installs things like rootkits by abusing some privilege escalation on your windows, then it affects your entire windows setup.

you could even setup a headless ubuntu vm, then setup docker within that vm. then allow that vm to start during windows startup and it will just be in the background with your openclaw running. this setup is quite stupid ngl but ig if you are insecure or scared you can do that

[acoustic gull]

Hi everyone, I'm trying to set up Ollama with OpenClaw but having issues:

Environment:

• Mac mini
• Ollama 0.16.3 installed
• Llama 3 model downloaded (4.7GB)

What I've tried:

1. Set auth.profiles.ollama:default with provider: "ollama", mode: "api_key"
2. Set models.providers.ollama with baseUrl: "http://localhost:11434", apiKey: "ollama-local", models: []

Problem:

• openclaw models list only shows MiniMax-M2.5
• Ollama Llama 3 is not discovered

According to docs, OpenClaw should auto-discover Ollama models when OLLAMA_API_KEY is set. What am I missing?

Thanks!

[balmy chasm]

Will clawdbot do anything to my laptop if I am not connected to an API or grant permission to my files? I just installed it on my personal computer, and I would like to remove it. How would I do that?

Thanks!

[candid socket]

by right no, if the model fails and there are no fallback models then it shouldnt be able to do anything even with heartbeats on.

[late crystal]

Hi everyone, I'm running OpenClaw on a headless Linux (arm64) server. My main agent (agentId: main) is trying to spawn a sub-agent, also using agentId: main. The sessions_spawn command consistently fails with a gateway closed (1008): pairing required error.

I have tried the following configurations in agents.defaults without success:

• subagents: { allowAgents: ["*"] }
• subagents: { allowAgents: ["main"] }
• sandbox: { sessionToolsVisibility: "all" }

The gateway is bound to loopback. How can I configure the gateway to allow sub-agents spawned by the main agent to connect without requiring a manual pairing process?

[grave willow]

Hi Guys, i want to do something usefull: out of offerings in the web like i want identify whisky casks on Whickybase to know how they are rated, plattform is mac. open ai thinks i have to do that in the following way: Via an integration layer, e.g. macOS shortcuts as Quick Action/Service + OpenClaw Webhook. -> ist ok or how should i approach this ? Example: https://www.thewhiskybarrel.com/de/products/jura-16-year-old-2009-signatory-cask-strength-15 -> Whiskbase: https://www.whiskybase.com/whiskies/whisky/278461

[fair rover]

Hello everyone, can you me with TUI, is that really important?. Because I try chat the bot and it response me just fine in telegram. However, in the terminal it always notify me "gateway connect failed: Error: pairing required"

[icy cosmos]

What do you think the best option to use besides the expensive models like Gemini Opus? There are new models like Kimi K 2.5 and Qwen 3.5, Are those good options?

[round valley]

How do i fix this?

I ran it again, didn't work

[brazen widget]

Anyone have a good recommend concurrent agents for codex (on pro)? I’m hitting cooldowns and rate limits

[candid socket]

i see alot of people complaining here about k 2.5. so far been using haiku for basic stuff, its not that bad most of the time. sonnet is what i use for things that have more moving parts and i use 5-mini for cron and heartbeats

[high shadow]

i am using model qlm-4.7-flash:latest, but i am getting '400 think value "low" is not supported for this model'

[small inlet]

Anyone got a good prompt for openclaw to build a Mission Control dashboard?

[icy cosmos]

i didnt think about haiku

so primary is haiku and then fallbacks to the others?

[candid socket]

ye something like that, haiku does struggle with some complex tasking if i dont specifiy exactly what i want. so depending on the task, i do swap between sonnet and haiku, then for crons and heartbeats whatever i can make scripts for, i do that and use gpt-5-mini to run the script and handle/process the outputs.

i found it better to create scripts for repetitive tasks so that there is less room of error

[raven mesa]

is openclaw safe to download and i was worried about infra costs

and is it free?

[celest ermine]

what machine do you have?

[umbral charm]

its safe as long as you don't have it connected to moltbook or any social media

[sour slate]

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

[raven mesa]

i will have it connected

to my

calender4

whatsapp

and other sources

[raven mesa]

hp pavillion laptop

[celest ermine]

i think ur machine can run it

i don't think you should give it access to your whatsapp

[odd iron]

Hey guys 👋

Where can I find a list of credible developers who are known for creating abilities that are generally considered safe?

[sinful shore]

Hello, I need help with error "Error: EACCESS: permission denied, mkdir '/home/node'". I just migrated over to another machine copying my original .openclaw files. Ran "openclaw doctor", did not see and permission issues.

[random hazel]

For everyone who's running it on a VM, what's your VM's OS?

[weary spruce]

ubuntu

[candid socket]

chown -R 1000:1000 /home/node should do the trick.

[sinful shore]

still showing the issue

[obtuse dune]

Anyone using groq as a model?

[candid socket]

i may not be helpful then, thought it was a docker issue

[mortal wave]

Sorry I am probably being stupid. Where do I find the MacOS .dmg file?

[spice gulch]

I am having also thousand of issues with docker 🙁 btw try to set those permissions not on /home/node but in the host's path that u mapped to that in the .env file

I am stuck to connectivity between cli and gateway

[sinful shore]

Yea, it was a bad idea to migrate my files from docker to native. Will need to settle with a fresh install…

[frosty scarab]

Hi guys, I have installed openclaw on my raspberry Pi 3 and wanted to setup my windows PC as a node, but the raspberry pi is a bit old and has memory issues, etc. Do you think is a good idea to set it up directly on my PC? I've read some advice of not doing this for security reasons, but I would like openclaw to work on my projects, and files

[elder hearth]

hey openclaw family, linux noob here: is it possible, that the newest installation package included discordjs/opus and that makes it unusable for installation on raspberry PI as @discordjs/opus does not support Node 22 on ARM64 yet? that is atl what i seem to have found out using AI 😉

[spice gulch]

did you manage to set-up the TLS enabled?

[frosty scarab]

Yes

The problem is the raapberry, it doesn’t work well

[radiant zephyr]

why i cant install openclaw on my vps, It show that: WARN npm install openclaw@latest failed; retrying openclaw@nex

[spice gulch]

would you mind to check this? https://discord.com/channels/1456350064065904867/1474802692198105250 I am going crazy on it... I wrote all the commands I launched at the start of this help-request. And my first error is on TLS. Disabling it gives me other issues with the browser etc. in general then my main issue is that status command returns error and cli seems not able to talk to the gateway

[spice gulch]

which is the exact issue? is it just slow?

[frosty scarab]

I’ll check it out

[frosty scarab]

Yes, it also restarts, drops the gateway sometimes…

[sick yoke]

Anyone has the problem getting the message "402 Payment Required" and cannot spawn sub-agent?

My AI API: kimi code api subscription, moderato plan.

[spice gulch]

I guess the RAM requirements are not met... btw changing with chown the permissions on the hosts dir mapped in the .env I think would solve the issue you had with the migration

[wind needle]

anyone have any idea why in help krill is just ignoring my posts? ive made 3 over last hour or so, none of them have been answered

[frosty scarab]

Probably

I don’t have any migration issues, I was asking for advice for installing openclaw directly in the PC instead of using it as a node

[spice gulch]

yeah - I mean the permission error you posted here

that's probably because the .openclaw dir has different permissions

[radiant ocean]

lads i just ran into the api limit in qwen
do you have any suggestion for free tier higher limit models or a way to run models on the cloud for free at least

[amber scarab]

Anybody have som info for a full Apple ecosystem setup? Calender, reminder, notes and so on 🦀🍎

[wild radish]

Is anyone getting slack final reply failed: Error: An API error occurred: missing_scope even if all scopes are configured (Slack DMs)?

[elder hearth]

i got the same error on my PI5

[flat pine]

Hi guys, trying to sign into OpenAI Codex via oauth using the official container in unraid. I want to use it as a fallback. Can anyone assist please?

[vague forge]

please i need help to install openclaw

is there a mod

i am unable to access dashboard after installing openclaw in my GCP linux

[wary osprey]

yeah yeah, Qwen rate limits can hit fast depending on where you’re accessing it from.

If you still want to use Qwen but avoid juggling multiple provider keys or strict free-tier caps, you could try routing it through AIsa. they’re a strategic partner with Alibaba Qwen and allows the full Qwen model suite through one endpoint, plus they give $5 free credit when you first sign up so you can test without spending anything. It also makes switching between Qwen, Kimi, Mistral etc easier if you’re experimenting with different models in OpenClaw.

Setup is basically just:
– one API key
– base URL in OpenClaw config
– pick model and run

You can check guide here if you want to check it out
https://docs.aisa.one/docs/use-openclaw-with-aisaone

[radiant ocean]

any free options? 🥲

[sinful shore]

run llm locally, I use LM studio.

[wary osprey]

Yeah a few people here are testing Groq, it’s crazy fast for inference. I played with it a bit but ended up sticking mostly with Qwen/Kimi for agent workflows since I needed more stable cost and multi-model switching.

one thing that helped was just routing everything through one provider instead of managing different keys per model, and it makes it easier to swap between Qwen, Mistral, Kimi etc while testing. I’ve been using AIsa for that since it lets you hit multiple models from one endpoint, so you can experiment without constantly reconfiguring providers.

Not required obviously, just made my setup less messy while testing different models.

[ripe thistle]

I’ve settled a couple of crons on the gateway but I always receive the messages twice… asked my openclaw to figure it out, it said that it did a couple of times but still same issue.

Anybody bumped into this ?

[wary osprey]

Free tiers are honestly getting tighter everywhere, especially with Qwen and most decent models. one small workaround if you still want to test without spending immediately, AIsa gives about $5 free credit when you sign up, so you can keep using Qwen or try other models without hitting limits right away. That’s what I usually tell people to use as a buffer while testing configs or agents before committing to anything.

after that it’s just pay-per-use, and from my experience it’s still pretty affordable compared to burning time trying to chain random free endpoints together. at some point also, paying a few dollars just saves sanity if you’re seriously building.

[spice gulch]

I imagine that when deepseek v4 will be out (next days/weeks), all will become cheaper, as already happened with the other versions

[proud halo]

I have installed openclaw on my macbook m3 pro nd noticed it is slowing down by macbook significatnly. I am planning to setup something in my old macbook pro 2013 a. I started and failed. Does anyone know any trick to get it running on old machines or is there a better way to get this done like deploying via VPC in a AWS server for e.g.?

[radiant zephyr]

export OPENCLAW_USE_GUM=0

Type this before install, u can solve the problem

[stiff tusk]

how bad the update can be when it even breaks the openclaw logs --follow 🤦‍♂️

[icy vapor]

What do you use to see what’s going on in the background while openclaw is working?

I built a dashboard with activities and it shows good info but not everything and sometimes it doesn’t track things easily.

Would be good if there’s a built in logging system so we can see and review exactly what is happening behind the scenes.

Has anybody built this? What do you use?

[wild radish]

I solved by downgrading to version 2026.2.15

[wild radish]

Hey anyone having this error? Is very strange and before deleting my slack app and try again i would like to know if it's an issue occuring not only to me

[wanton zephyr]

⚠️ Agent failed before reply: Unknown model: google-antigravity/gemini-3.1-pro-high.
Logs: openclaw logs --follow

14:01:34 [ws] ⇄ res ✓ chat.send 82ms runId=e40bc10c-03d4-4be7-bdba-482481969743 id=e90ccef0…429e
14:01:34 [diagnostic] lane enqueue: lane=session:agent:main:main queueSize=1
14:01:34 [diagnostic] lane dequeue: lane=session:agent:main:main waitMs=14 queueSize=0
14:01:34 [diagnostic] lane enqueue: lane=main queueSize=1
14:01:34 [diagnostic] lane dequeue: lane=main waitMs=3 queueSize=0
14:01:34 [diagnostic] lane task error: lane=main durationMs=6 error="FailoverError: Unknown model: google-antigravity/gemini-3.1-pro-high"
14:01:34 [diagnostic] lane task error: lane=session:agent:main:main durationMs=16 error="FailoverError: Unknown model: google-antigravity/gemini-3.1-pro-high"
19:31:34 Embedded agent failed before reply: Unknown model: google-antigravity/gemini-3.1-pro-high
14:01:34 [ws] → event chat seq=2 clients=1
14:01:34 [ws] ⇄ res ✓ chat.history 7ms id=c7055f5c…ce23
14:01:35 [ws] ⇄ res ✓ node.list 1ms

[glass glade]

can anyone reccomend setup video/guides? I keep seeing things online people building overnight for pennies in tokens and i'm over here wasting two $20/month subs building nothing. my glorified chatbot needs setup properly and idk how to do it

[candid socket]

overnight for pennies? i dont think thats possible unless you use free/cheap models or local models and maybe limit context set for all sessions. both being quite bad at taskings.

[glass glade]

that may be exaggeration in the video lol but i keep hitting limits without getting things done, not asking it to do heavy lifts like hitting limits asking it to set itself up efficiently lol

I should be able to use any model right? I cant even set it to the cheap model for some reason " Issue: The Haiku model ID I used (claude-3-5-haiku-20241022) doesn't exist in Anthropic's API yet." like wtf

[candid socket]

use cli openclaw configure to add the model

maybe could be the model id not being one of the models oc supports

[tall hearth]

What’s everyone’s model set up like

[candid socket]

claude haiku for everyday
claude sonnet for complex tasks
gpt 5 mini for crons and heartbeats

[opal briar]

hello I have this err when I want to update OpenClaw

◇ ✗ Updating via package manager (87.91s)
npm error node-pre-gyp ERR! stack at ChildProcess.emit (node:events:519:28)
npm error node-pre-gyp ERR! stack at maybeClose (node:internal/child_process:1101:16)
npm error node-pre-gyp ERR! stack at ChildProcess._handle.onexit (node:internal/child_process:304:5)
npm error node-pre-gyp ERR! System Linux 6.17.4-200.fc42.x86_64
npm error node-pre-gyp ERR! command "/usr/bin/node-22" "/home/x/.npm-global/lib/node_modules/openclaw/node_modules/.bin/node-pre-gyp" "install" "--fallback-to-build"
npm error node-pre-gyp ERR! cwd /home/x/.npm-global/lib/node_modules/openclaw/node_modules/@discordjs/opus
npm error node-pre-gyp ERR! node -v v22.20.0
npm error node-pre-gyp ERR! node-pre-gyp -v v0.4.5
npm error node-pre-gyp ERR! not ok
npm error A complete log of this run can be found in: /home/x/.npm/_logs/2026-02-22T14_25_58_144Z-debug-0.log

Update Result: ERROR
Root: /home/x/.npm-global/lib/node_modules/openclaw
Reason: global update
Before: 2026.2.19-2
After: 2026.2.19-2

Total time: 88.25s

[hollow thunder]

How much RAM do you have? It runs OK on my 2010 iMac which has MacOS Monterey on it thanks to the open core legacy patcher. I have 12GB or RAM though. But it usually uses less than 1GB.

[tall hearth]

Is anyone running with a Claude Max plan still

[hollow thunder]

Does anyone know how to reduce the token usage of a specific agent? I had my primary agent set up a second agent with its own telegram bot, and I basically only want it to have basic web search access aswell as memory and ability to read and write stuff in its own workspace. Other than that i want it to be a chatbot which I will mostly use for general chatting and not difficult tasks. But it’s still using a lot of tokens and I feel like it’s being filled with stuff that it really doesn’t need for what it’s supposed to be for me. The main agent shouldn’t be affected by any changes, I want the main agent to still have full access.

[little crypt]

Has anyone used supermemory.ai? Good / better than obsidian?

[lethal rover]

what is the recommended way of updating dockerized openclaw? i have clawdock and tried clawdock-cli update but i got

Update Result: SKIPPED
  Root: /app
  Reason: not-git-install
  Before: 2026.2.16

[tired heart]

You can block access to tools on a per-agent basis, as well as trimming down soul et al.

[inland bramble]

when i ask him to do somethink he says he doesnt have acces to my chrome or files or anything else and asks to drop them, should it be like that? i know its dumb question im just new

[tidal plover]

Hey guys is OPUS 4.6 bugged right now my oauth working fine
But when I use 4.6 I get an authentication error

But 4.5 works fine?

Is there some issue right now?

Anyone else using 4.6 okay?

[umbral niche]

I have the same issue, anytime I ask locally running ollama/qwen3 to read BOOTSTRAP.md and walk me through the onboarding it responds with

I don't have direct access to local files or the file system because of critical privacy and security considerations. My capabilities are intentionally limited to protect your data and
ensure safe interactions. However, I can help you with tasks like web research, document analysis (if you share text snippets), or using tools that don't require file system access.
Let me know how I can assist!

[inland bramble]

same

[junior acorn]

Reset ur oauth

[junior acorn]

Create a workspace for the sub-agent

[glass glade]

bro i switched to haiku earlier and am almost at my weekly limit! how fast do you use ur limit? i have the 20/mo plan

[candid socket]

i dont use 'claude subscription

[gray jewel]

anyone ran into and fixed this issue?

<root>: JSON5 parse failed: SyntaxError: JSON5: invalid character 'T' at 1:1

[limber pagoda]

The error occurs because the parser expects valid JSON5 syntax (an object, array, or value), but found an unexpected character ('T') at the start of your input. Always ensure your input is a valid JSON5 object, array, or value.
Quick Fix:
If your input is a raw string (e.g., a date), wrap it in a JSON5 object:

{ "key": "your_string_here" }

[trim osprey]

what models do we get with pro $200 chat gpt oauth?

only codex?

[obtuse sundial]

i already pay for google one sub is there no way to utilize the gemini it comes with in openclaw? i dont want to use the pay as you go api because i already have an existing sub

[gray jewel]

Thank you. The model added an extra line in line 1 on the json file that wasnt supposed to be there. Gateway restarted fine after i removed it.

[sterile burrow]

Hey everyone! I'm setting up an Oracle Cloud Free Tier VPS (ARM Ampere instance). I'm looking for the most resource-efficient LLM to run on it. Should I go with an Open Source model like Phi-3 or TinyLlama via Ollama, or is it better to just stick to a lightweight API? What’s the 'cheapest' setup that actually performs well on these specs?

[sinful shore]

didn't get the free credits

[wary osprey]

If you’re talking about the $200 ChatGPT Pro plan, that’s mainly the web app subscription. It gives you access to OpenAI’s top-tier models inside ChatGPT itself, including their most capable GPT-4 class models. But that doesn’t automatically mean you get unlimited API access through OAuth. The API is billed separately based on usage.

So Pro = better access in the ChatGPT app.
API usage = pay-per-token via the OpenAI platform.

If you’re building in OpenClaw or your own app, you’d still need API billing enabled.

[gray jewel]

I thought you can Oauth to GPT like you can through anthropic now

[trim osprey]

any idea how? with a token?

i have anthropic token on but im rate limited for the week

i may have jumped the gun and bought $200 token from chat gpt thinking it was the same

[wary osprey]

Can you send me a DM?
I want to send our discord link to you so you can open a ticket and I will look into it for you.

[junior acorn]

All of them

[trim osprey]

im only having access to codex currently do you know how to configure it?

[junior acorn]

Use oauth for Gemini

[junior acorn]

Read docs -> models, then change model in config setup backup models too. Or run onboard/wizard again and choose model

[trim osprey]

so i can get a true 5.2 as a brain with this? have you done this?

[junior acorn]

You dont have to use api if you use oauth. Don’t spread wrong information. Read docs

[junior acorn]

Yes, 5.2 pro xhigh thinking works

[trim osprey]

ive already paid im going to get this going. is there rate limits like anthropic $200? im maxed out for the week. thank you for your help

[junior acorn]

Of course there are rate limits. Just don’t abuse them too hard

https://docs.openclaw.ai/concepts/model-providers#openai-code-codex

Here is docs

What are you doing to rate limit on Anthropic? They’re more strict over there

[trim osprey]

says 5.3 codex tho unless im mistaken

[junior acorn]

Change model name

[trim osprey]

ahhh wonderful thx

im really just using sonnet for brain and opus for architect, building out some websites testing power of it but i burned through limit by thursday and signed up monday

[wary osprey]

let me clarify this

OAuth exists, but what I was referring to is external API usage for your own apps or OpenClaw-style builds. In that case, ChatGPT Pro doesn’t automatically give you API billing capacity. If you’re just using ChatGPT-native flows or certain ChatGPT-integrated apps, that’s different. But for standalone API calls from your own backend, you still need platform billing enabled

[tender oasis]

They've removed Krill from the help channel. Basically lost now root@Ubuntu-2404-noble-amd64-base ~ # npx openclaw gateway restart

🦞 OpenClaw 2026.2.21-2 (35a57bc) — Works on Android. Crazy concept, we know.

Restarted systemd service: openclaw-gateway.service
Gateway did not become healthy after restart.
Service runtime: status=stopped, state=activating, lastExit=1
Gateway port 18789 status: free.
Gateway restart failed health checks.
root@Ubuntu-2404-noble-amd64-base ~

[wary osprey]

Sent you a friend request

[civic harness]

Hey, I'm having node installed with nvm, my openclaw is not working anymore. Do you have similar issues?

[trim osprey]

struggling getting this set a bit not showing me gpt 5.2 in models list now bt its in config

[serene schooner]

Any suggestion as to how I might get rid of the [[reply to current]] not getting stripped?

"[[reply_to_current]]
Hey — Clawdbot here 🤖, online and ready. What do you want to tackle first?"

[swift quartz]

Setting this up on my mac mini. It will be an isolated device. With the new hardening update is it still required to run though the process of having the openclaw account downgraded to a standard user>

[trim osprey]

anyone know the work around to get 5.2 pro working with oauth in connfig?

[cerulean flax]

Are these effective: https://clawhub.ai/seojoonkim/prompt-guard
There are a fair number of anti-prompt injection skills etc. Is anyone using them, has anyone tested any? feel free to DM me; or reply to this. Just curious on efficacy

[fast python]

Is it worth buying a maxed out Mac mini to run openclaw, or are there better options?

[candid socket]

maxed out isnt needed unless you are looking to host local models, but these smaller models often struggle to be good agents so it may not be worth

[wary epoch]

can somone help, i tired to set up taiscale and went through a bunch of steps to get it working on my other laptop and now the connecting is being refused on my mac mini, how can i assess the bot? can i do it via terminal? i dont want to wipe it all clean but i messed up a config?

[fast python]

I was thinking of hosting locally but that's been my experience as well with local agents. I also want to have it manage things like transactions, secrets, etc. and I'm concerned with privacy as well though. Its nice to know your personal data isn't going to be used by third parties

[ember remnant]

Anyone experiencing super slow responses on the dashboard ui chat?

I’m running ollama local model. I run a test on terminal on my Mac mini and response time is .003s.

When I try on OpenClaw it takes minutes sometimes 5 mins to respond. I have chat gpt troubleshooting but I feel like I’m going in circles.

It’s starting to get frustrating tbh!

Help! Any context I can add to help troubleshoot , I can definitely do so. Thanks!

[wary epoch]

is there a way to chat with my bot in terminal?

or can i reinstall and not lose all my stuff?

when i got to TUI i get gatweay error

[forest compass]

Please help, trying to install openclaw on my raspberry pi 5 16GB ram. It has worked before in a docker image but wanted to give it dedicated access to the device as I am not using it for anything else. I have erased the Docker image completely and then tried to reinstall using the curl command but get the error below. edited ~./zshrc to add it to the path but it still didn’t work, troubleshooted for ages but got messy so reflashed my raspberry pi image and now still getting the error. Anyone had the issue/know how to resolve: [16ms][127][~]$ curl -fsSL https://openclaw.ai/install.sh | bash
...

OS linux
Install method npm
Requested version latest

[1/3] Preparing environment

✓ Node.js v24.13.1 found

[2/3] Installing OpenClaw

✓ Git already installed
INFO Installing OpenClaw v2026.2.21-2
inappropriate ioctl for device
✓ OpenClaw npm package installed
WARN npm install openclaw@latest failed; retrying openclaw@next
inappropriate ioctl for device
✓ OpenClaw npm package installed
✓ OpenClaw installed

[3/3] Finalizing setup

🦞 OpenClaw installed successfully!
Installation complete. Your productivity is about to get weird.

INFO Starting setup

INFO Skipping onboarding (openclaw not on PATH yet)
WARN Installed, but openclaw is not discoverable on PATH in this shell
Try: hash -r (bash) or rehash (zsh), then retry.
npm prefix -g: /home/pi/.nvm/versions/node/v24.13.1
npm bin -g: /home/pi/.nvm/versions/node/v24.13.1/bin
If needed: export PATH="/home/pi/.nvm/versions/node/v24.13.1/bin:/home/pi/.nvm/versions/node/v24.13.1/bin:/home/pi/.nvm/versions/node/v24.13.1/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/local/games:/usr/games"

[sinful shore]

I got the same error too. Ended up installing it from source

[candid socket]

i usually just backup all the workplaces, config json and the shared skill folder before reinstalling

[forest compass]

Okay cool Ill give that a go thank you

[hearty stratus]

I need help to use Whatsapp instead of Telegram, is there a link or a youtube video that I can use

[tiny fulcrum]

haha damn bro i had the same problem all day... this fucking docker ... and lost at least 1h cause of github , i just deleted both.. to get it fixed .. i fixxed it with codex u just drop your error code in and he fixes it... pretty nuts!!! gl

[wary epoch]

i reinstalled and still have the same gatway issues!

127.0.0.1 refused to connect.

how do i uninstall everything? still having gateway connection errors

[proven axle]

disconnected (1008): unauthorized: gateway token missing (open the dashboard URL and paste the token in Control UI settings) what to do ?

[delicate pewter]

Downgrade to Feb 17th. New bugs introduced recently