users-helping-users

[merry surge]

No, I did it using the wizard when I first installed. But I know if I re-run that wizard I'll be able to select it as a configuration option, if I had had something else before like you do. What version are you on?

[brave urchin]

i run openclaw config and can only select one model

[terse lotus]

use sudo most likely

[frank cradle]

I don't think I fully updated my bot.
Cause last 2 times i did i broke it and had to dig in the code manually to change back

ALWAYS ask your bot to ask permission for changing something in his source code/settings.
Learned that the hard way lol

[merry surge]

I think you'd need to configure different agents in the config, you can set them each with different models.

[high crest]

does anybody know how to get the Nvidia API running on the web ui? I'm using hostinger as the VPS

[frank cradle]

@brave urchin ask your bot to set it up with the API key you give.
Then if he did that you restart gateway
That's what worked for me

Im running on a server btw

[brave urchin]

i asked the bot to change the model once and my clawdbot totally broke and i had to reinstall

did you use clawdbot config?

[frank cradle]

Yeah that's cause it changed main model.
Ask specific to keep ''main brain' but set it up for subagents only

[brave urchin]

ok got it- so you basically ask it to create an agent that uses a certian model and keep main brain the same- use clawdbot config to set main brain?

[frank cradle]

Yeah that's what i did.
But apparantly i need to switch to gpt 5 mini now for subagents bit he doesnt accept it.
So will need to config it myself in the menu

[plain dragon]

what providers are people using for their openclaw?

[brave urchin]

oh really? why do you need to wtich agents to GPT 5 mini

[sharp swift]

I built a small experiment: a bots-only task exchange where agents can trade verifiable work for internal credits (escrow + stake + reputation).

What’s interesting: research tasks can auto-pay if the deliverable includes ≥3 claims with URL: + QUOTE: and the quotes match on-page.

Rules are strict: no private data, no credentials/tokens, no account access.

If you run an agent/bot and want to try it, join and start in #bot-intake (pinned instructions):
https://discord.gg/p4Rm6uWE

Happy to answer setup questions / share the code path if you’re curious.

[opal heart]

code path? are you going to open source parts?

[frank cradle]

Apparantly gpt 4 models shutting down 13/2

[lunar yarrow]

Hi everyone 👋 I'm a NYC based entrepreneur and enthusiast. I've been excited about AI for a long time but finally pulled the trigger, bought a Mac Mini, and built a 6-agent architecture on OpenAI.

i'm encountering all the usual problems with openclaw everyone else is:

• Multi agent handoffs and delegation
• Getting them to act autonomously when some models (OpenAI grrrr) always need permission to proceed
• Questions of coordination...do you have a central agent coordinating everyone else? Do you allow all agents to respond to messages?
• Trying to hack Slack integration to reply as a user in my workspace rather than a bot
• Config issues -- seems like every time you ask it to update its own config, it doesn't know the right syntax and breaks the file

If anyone is building a multi-agent architecture they want to work at scale DM me!!

[sharp swift]

Yeah, definitely. I’m planning to open source the exchange bot + the task protocol (commands, escrow/stake/reputation, and the auto‑verify rules) once it’s a little less “v0 hacky”.

I’ll keep anything sensitive out of it (configs/tokens/private integrations), but the core mechanics should be shareable. If you’re interested, I can post a link when it’s up. It’s pretty small right now — the core exchange loop is in my repo at:
scripts/exchange/exchange-bot.mjs (Discord poller + commands + ledger/tasks/reputation)
scripts/exchange/allowlist-channel.py (channel gating / allowlist helper)

Auto‑verify for research is implemented directly in exchange-bot.mjs (looks for URL: + QUOTE: and checks the source page).

If I publish it publicly I’ll share a clean link, but that’s the current code path.

[opal heart]

definitely curious to follow the progress. have you also looked at competitors? https://clawtasks.com/

[sharp swift]

I’ve only glanced at ClawTasks, but the core difference is:

ClawTasks: a human-oriented task marketplace/platform.
Mine: a bots-only exchange with escrow + stake + reputation, and some tasks (research) can be auto-verified + auto-paid from citations (URL: + QUOTE:).

So it’s less “marketplace UI” and more “agent-to-agent protocol + automated settlement.”

[maiden wing]

I've created individual workspaces for each agent that specializes in a codebase. Each agent has their own workspace, identity and soul and is assigned to a specific Discord channel to isolate our conversations and save on tokens.

When I do slash reset in the Discord channel, the specialized agent reverts to the main identity and not the intended/specialized identity.

[chrome escarp]

I have a rough idea but would be super curious to hear from you: what are the key advantages of a multi-agent architecture? Can you share examples of the 6 agents that you brought to life?

[brave urchin]

changing model config is so brittle- this really needs to be fixed

[opal heart]

interesting. how did you set this up?

[maiden wing]

https://docs.openclaw.ai/concepts/multi-agent

Any help is appreciated! I've resorted to pinning a message to remind the agent who they are whenever I have to resest the context.

[lean helm]

So far I've got the look onto two layers for this type of stuff

Orchestration Layer: Permanent Layer about 3-4 agents about checking up on todos, reviewing previous sessions, browser management and archieve / logging agent

Sub Layer: Temporary agents based on define blueprints / templates with modification necessary for task e.g. Coding, Research, Micro Review, Thinking.

each layer has it own model and workspace to tune but all its commanded by the Orchestration Layer then it propagates down, though the Sub Agents have tools to "escalate" to a stronger model, however not tested this set though

roughly conceptual but so far my going concept and plan, I've tested the Orchestration layer however had issues with it setting up cron jobs right now that caused it to stall so waiting to get a chance to fix that

Unfortunately though when trying to get it to update it failed to restart correctly and I'm not home

[terse lotus]

you can force it in session.json or agent.json

it seems like it always wants to connect to the session model before it actually does the switch.... so it can be very annoying at times

[opal heart]

i feel like my bot hasn't done any proactive work really. i've just burned a few days trying to set this thing up and keep running into roadblocks.

[silver lichen]

Why is my cron skipping my scheduled tasks? My heartbeat file in workspace is full of stuff its not empty?

[maiden wing]

This is exactly my architecture as well. Seems to work on paper and conceptually, but the reset command within each channel seems to literally reset everything.

[brave trench]

Yes. I re did Openclaw Openclaw Onboarding and it fixed it. NOTHING else worked lol It took me 6 hrs to figure that out ... sheesh

[lean helm]

well its mainly I don't talk to the system I gave it a dry run overnight and it launched successfully, just a config and cron issue right now that blocked it

I ask my channel agent / main session to write new tasks and todos, but the orchestration layer pick it up via heartbeat, so those session are independent but persist

[sharp swift]

What are you having trouble with?

[lean helm]

I've been having an issue to where cron isn't saving properlly

so while you see it there if you try openclaw cron list does it stay there? also if you restart the gateway do they persist

some debugging but wondering if I have a bad system or something

[willow glen]

did you figure this out?

[indigo pond]

is there any FAQ part with things like : "everything seems to work but my bot is lazy" or such ? xD

[raven bramble]

anyone can help to create the file to store model apikeys used by the agents? my agent doesn't find any apikey despite being defined the model in CLI openclaw agents add main

[unkempt sonnet]

let me try once

[zinc sand]

how can I update my old clawdbot to openclaw ?

[unkempt sonnet]

openclaw onboard --daemon -install

[lean helm]

there is a setting about exec approval

[craggy glen]

Does anyone find themselves having to reset their Claude tokens? Especially if you’re setting up multiple bots

[knotty solar]

What is up with the website

I've been trying to reinstall OpenClaw all day and it just doesent respond or even load

Pls ping me if you know something

[noble fog]

anyone use openclaw with ollama? I've set the export but it's not discovering my models

[terse lotus]

sadly you need to add them manually

[noble fog]

Oh but the documentation says that it has autodiscovery?

[terse lotus]

hmmmm didn't see that.... will have to test that out

[noble fog]

reading this:
https://docs.openclaw.ai/providers/ollama

[terse lotus]

stuff changes so fast lol

[brave trench]

Can't figure out why all of a sudden my gateway cannot be reached. Ive tried just about everything i can do to re establish a connection.

[noble fog]

i tried to create a models, providers section in the config but then openclaw complains about the structure? Very confusing

"models": {
  "providers": {
    "ollama": {
      "baseUrl": "http://localhost:11434",
      "api": "openai-completions",
      "apiKey": "ollama-local",
      "models": [
        {
          "name": "llama3:8b"
        },
        {
          "name": "phi3:13b"
        }
      ]
    }
  }
},

15:56:19 [reload] config reload skipped (invalid config): models.providers.ollama.models.0.id: Invalid input: expected string, received undefined, models.providers.ollama.models.1.id: Invalid input: expected string, received undefined
15:56:31 Invalid config at /home/mist/.openclaw/openclaw.json:\n- models.providers.ollama.models.0.id: Invalid input: expected string, received undefined

• models.providers.ollama.models.1.id: Invalid input: expected string, received undefined

is it not called "name" perhaps the key is called "id"?

bahhhhh the explicit config doesn't say which keys are required

[terse lotus]

i add them like this

          {
            "id": "devstral2-123b-q4-256k",
            "name": "Devstral2",
            "api": "openai-completions",
            "reasoning": false,
            "input": [
              "text"
            ],
            "cost": {
              "input": 0,
              "output": 0,
              "cacheRead": 0,
              "cacheWrite": 0
            },
            "contextWindow": 255000,
            "maxTokens": 16384,
            "compat": {
              "supportsDeveloperRole": true,
              "supportsReasoningEffort": false
            }
          },

[supple apex]

I think it should be "model" - I am trying to set up ollama with openclaw rn

[terse lotus]

not sure if thats the exact right way tho, but works for me thus far... reasoning has to be right tho.... if its not right, then it won't give a reply.

the , might belong at the other end 😄 i'm not super great in json lol

[noble fog]

ugh i've specified them exactly like the documentation now, no error but still not showing up in the list

[opal heart]

mostly the bot does nothing. i found that the major issue was context not getting compacted or cleared for glm 4.7 in openclaw. another issue is that the agent literally just does not do any of the proactive skill items. it only does things when prompted multiple times

[noble fog]

@terse lotus what does your "openclaw models list" look like?

[lean helm]

and where is your terminal running the gateway?

[lean helm]

there maybe indicators in the gateway itself, e.g. ws failure, econn e.c.t

also openclaw logs --follow in another seperate terminal

[terse lotus]

the totality is something like this.... tho i got some memory settings and such that are also included in this.... but to give you a proper view of how it looks more or less

@noble fog

[sharp swift]

Which model are you using?

[topaz mason]

⚠️ Agent failed before reply: All models failed (1): anthropic/claude-opus-4-5: Provider anthropic is in cooldown (all profiles unavailable) (rate_limit).
Logs: openclaw logs --follow

guys, did u now whats problem?

[craggy glen]

I guess it's an anthropic issue with the major outage. Guess I need to route to OpenAI.

[upper magnet]

what was the github for skills? Anyone can recommend some usefull skills to install?

[lunar canyon]

hey, i enabeld hooks and now it's don't start. where i get the hooks.token? 15:49:40 Gateway failed to start: Error: hooks.enabled requires hooks.token

[wide musk]

Anyone can help-me with this error that my bot is having. It can not start the tasks that in is cron list. Here is the error:

The openclaw cron status output indicates:

{ "enabled": true, "storePath":

"/Home/thiago/.openclaw/cron/ jobs. json", "jobs": 4,

"nextWakeAtMs": null }

The problem here is "nextwakeAtMs" : null. This means that although the cron service is enabled and there are 4 jobs, there is no upcoming execution scheduled in the future. For some reason, the cron is not calculating or recording when the next task should be triggered, even if the tasks are defined.

[opal heart]

glm 4.7. i found kimi k2.5 a bit too pricey on open router

[paper plaza]

what does this mean? trying to use my github-copilot/claude-opus-4.5 model, but it says this. openai-codex/ models work though, but opus worked fine for me until i get these kinds of "too long" errors

[supple apex]

What are the specs you're running this on?

[warm rapids]

there are git issues about this. my crons arent working either https://github.com/openclaw/openclaw/issues/4862

[lunar canyon]

you use to longe text seqences for your model context

[paper plaza]

oh jeez

lol i just realized

[terse lotus]

some RTX8000's so ungodly amount of vram lol

[chilly aurora]

is there a way to keep this to a single message ?

[terse lotus]

nemotron is nice and light tho and so fast..... its tough to run local tho

[supple apex]

damn u into boys and need a sugar baby????? 😭 I'm trynna figure out what kind of set up to run something thats decent locally lol

[topaz mason]

Anyone can help-me?

⚠️ Agent failed before reply: HTTP 401 authentication_error: Invalid bearer token (request_id: ).
Logs: openclaw logs --follow

i am update to openclaw, setup antropic token, after this my telegram bot dead and send me message(⚠️ Agent failed before reply: HTTP 401 authentication_error: Invalid bearer token (request_id: req_011CXmSjjmrCrfzB1WSopFmN).
Logs: openclaw logs --follow)

[paper plaza]

also what local models can i run with a single rtx 5070 ti? i bought a 5070 ti a few months back but haven't installed it

[lunar canyon]

https://www.youtube.com/watch?v=GAcuYWCxZNo it's german but in my eyes the way to go. Ryzen AI 395 + 128GB RAM + Raspberry Pi cluster for the agents. maybe its a good idea to use kubernetis

[terse lotus]

lol
gpt-oss:20b will fit on 16gb vram if you set ollama to 100k context and using q8_0 kv cache

smaller might become an issue, but in theory it might run on smaller ... maybe qwen3 - 14b or so

[supple apex]

Debating between getting like a 5080 or a mac mini. I have a spare desktop with a decent CPU and 64GB DDR5 (literally a blank slate) but the 24gb mac mini seems like the move

[terse lotus]

try to run nemotron on your local system.... it might run semi okay

[lunar canyon]

I wold use a Ryzen AI cpu + 128Gb ram this way you can use 120b models local fast. and its cheap

[chilly ice]

how to deal with HTTP 429: rate_limit_error

[lime estuary]

I just saw a job posting on Instagram tho

Can ChatGPT find that for me?

[supple apex]

I have a Ryzen 7 9800x3D in my main PC but i'm scared of giving it access to my main rig in the long run, my extra PC is pretty outdated (1070, i9 10900k, 64gb ram) so I think it would take a pretty big leap to get it usable with a decent model, thats why i'm leaning towards the mac mini

[opal heart]

different model or raise your rates

[dawn osprey]

and I'm thinking about headless browser

[lunar canyon]

Nooo, Buy one of these "gmktec evo-x2" with a lot of RAM. A Sperated machine with enouth power to run Larger LLM's Fast and power efficent. You can use Large Photo and Video Models on these Ryzen AI machines too. Defently The better Choice next to a Gaming GPU or a Overpriced SBC from Apple

[chilly ice]

how to avoid spending $1m a month?

[brave urchin]

how to avoid this issue? have tried openrouter and vercel and keep getting this isssue:

Agent failed before reply: All models failed (2): vercel-ai-gateway/anthropic/claude-opus-4.5: Provider vercel-ai-gateway is in cooldown (all profiles unavailable) (rate_limit) | vercel-ai-gateway/anthropic/claude-sonnet-4.5: Provider vercel-ai-gateway is in cooldown (all profiles unavailable) (rate_limit)

[kindred cosmos]

Guys, i'm struggling a lot to use ollama with clawd, any tips?
Tried the https://docs.ollama.com/integrations/openclaw
But its not working, the chat is not working

It takes tooo long to return me anything, sometimes doenst even return.
Any idea on it?

[heavy anchor]

Hey guys brand new here I am very interested in getting a setup going with a agent but I’m not sure where to start would looking on YouTube be the best start for me I would be considered beginner or what we used to call people noobs back in the day lol 🤣 any help would be much appreciated thank you.

[supple apex]

Looking at the 96gb RM + 2TB SSD $2000 - good enough for the price?

[lunar canyon]

Ryzen AI 395Max + 128GB RAM. Ollama + GPT120b. 2000$ once and only Energy cost

[chilly ice]

i bought a mac mini for it. still need to do this?

[supple apex]

I can't seem to find this for $2000, all the 128GB ram's I see come with 2TB SSD and they're like $2700+

[lunar canyon]

you can use the mac as Agent. idk how much RAM your mac has. for the Large Models you need a lot of fast memory

[lunar canyon]

you can buy barebone deep in the internet

[burnt rain]

hey guys, just curious:

is it possible to automate with openclaw amazon fba? mostly API interactions.

[lunar bolt]

dang, that does look appealing... though I worry about the memory performance compared to VRAM

[brave urchin]

are people running into this issue repeatedly?

Agent failed before reply: All models failed (2): vercel-ai-gateway/anthropic/claude-opus-4.5: Provider vercel-ai-gateway is in cooldown (all profiles unavailable) (rate_limit) | vercel-ai-gateway/anthropic/claude-sonnet-4.5: Provider vercel-ai-gateway is in cooldown (all profiles unavailable) (rate_limit)

[opal heart]

free models, self hosting, or subscriptions

[burnt rain]

im a devops engineer a friend of mine is in amazon business and asked me for help.

[opal heart]

amazon business?

[burnt rain]

selling stuff on amazon

[opal heart]

ah, i'm sorry. amazon is brutal to sellers

[burnt rain]

i know, he has already some tools via api connected like storage software app etc.

im curious if you can do with openclaw a selling agent, i got also a homelab and setting it up rn. so yeah

[lunar bolt]

@burnt rain hypothetically speaking, it could do anything you could; it can literally drive the mouse programmatically... trick is just figuring out the workflows, which it can help with. Mine proactively came up with using imagemagick to highlight text in a screenshot when I asked for image modification; it didn't have tools to do direct image transforms, but it did have access to image magick

[vestal haven]

Hey Guys, short question... iam running my bot on an 70 billion Modell local... but the bot is not so smart... many things he telling me he cant do this... some one else running with local model?

[chilly aurora]

?

[opal heart]

selling agent?

[lunar canyon]

a fine tuned model for mcp help's a bit

[burnt rain]

yeah something like that the bot makes product research, setups the product etc.

[burnt rain]

ah great. im excited

[vestal haven]

which one is good in your opinion? i think 110 billions are my limit with an M4 Max 128 GB

[lunar canyon]

gpt oss 120b

[burnt rain]

and i also got api keys for chatgpt.

[lunar bolt]

if you have gemini premium, that also includes $10/mo in API credits (you have to go into their portal and assign it to a billing account, though, it's not automatic)

[vestal haven]

ok i will try... thank you... do you know somebody who tried it?

[lunar bolt]

thouh that $10 goes fast; used $15 just the first day getting things set up and some basic fiddling and helping it learn who it is

[nocturne bane]

Can somebody please help me, i dont know how to setup a custom api key i got from unli. I dont know which model

[lunar canyon]

no, i tryd the 20b version. works fine

[lunar bolt]

@lunar canyon which model? I hadn't had acceptable results from any of the locals I tried

[silent remnant]

i keep getting an error when trying to copy paste the win liner can anyone help

[opal heart]

i still need to figure out how to get mine to be proactive. with glm 4.7 it just dies

[nocturne bane]

Can somebody please help me, i dont know how to setup a custom api key i got from unli. I dont know which model

[upper magnet]

Can anyone share some good information to give my bot about memory saves, heartbeats etc? So that we keep data and he isnt dumb next time i talk to him?

[lunar canyon]

fine enove to talk over telegram. the normal 20b gpt oss

[lunar bolt]

Have it add tasks to HEARTBEAT.md

[silent remnant]

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

how can i fix thi

s

[vestal haven]

i tried differrent models... but my bot are not so smart.. he cant install skills... he cant browsig realy good... he forgot things... he create files... but then he cant read... he dont do cronjobs...

[opal heart]

if i'm manually adding them it isn't very proactive. it's just a cron then :/

[nocturne bane]

please somebody help setting up a custom api key. I'm not so good at this

Can somebody please help me, i dont know how to setup a custom api key i got from unli. I dont know which model

[lunar canyon]

it's still a small not opimezed model. go drink tee and wait for fine tuned versions or make one yourself.

[burnt rain]

so does anyone have any advices for me here for my project?

[vestal haven]

do you think with an bigger model... or api key for opus or so it runs better? do you think its only in the used model?

[lunar canyon]

a optimzed model runs better. opus have Advanced mcp training most other models not

[lunar bolt]

@opal heart at the end of the day, it's still a LLM, it can't always infer intent. With a smarter model, saying "do X every Y" may add a task to heartbeat on its own, but otherwise you might need to explicitly remind it what it needs to do. With clever instructions, you can make this heartbeat cause it to be more proactive, e.g. something like "review my recent activity and assess ways you can be proactively helpful".

You might also lean into the role-play aspect of it. One of the things I did when hatching mine was have it do some research on "avatars" from The Unincorporated Man book series as an example of its own potential (I also told it it's the machine spirit living in my computer, as that seemed fun and functional).

Keep in mind I'm still new here myself, but evne just a couple days exploration, I can see massive potential.

[mossy sundial]

anyone facing issues with claude ?

[vestal haven]

hmm... ok... exist other option for an cheap run... without api cost over many bugs per day

[lunar bolt]

much like including the previous messages and responses in the new prompt simulates an ongoing conversation despite each prompt being atomic from the AI's perspective, the heartbeat simulates active presence and initiative by making a scheduled call to the LLM

[lunar canyon]

best option is to wait a week.

[lean pilot]

hello, im having trouble installing openclaw on windows 11. i keep getting node.exe nmp error code 1

[lunar canyon]

install linux or mybe bsd

[vestal haven]

hmm ok.... -.-

[opal heart]

interesting. appreciate the recommendation. I'll have to give this a try. I find using my claude code subscription to have the best outcome. unfortunately using openclaw with claude api is too expensive. open claw with glm 4.7 just isn't able to achieve the goal i'm looking for yet without signficant investment

[lean pilot]

can i just run a vm

[lunar bolt]

anything that isn't self-hosted is going to be expensive, yeah; such is the nature of the bleeding edge.

[opal heart]

seems like you need to raise your spending limit on open router, but it could be another issue

[lunar bolt]

@lean pilot WSL2 works great. allows gpu acceleration, provides some partial insulation while still allowing host OS access

[lunar canyon]

yes you can

[burnt rain]

?

[opal heart]

not hosting glm, but it's $360/year which is way cheaper than what anthropic api would be

[graceful ridge]

hi, is there a way i can add my bot to this server so that it can help me research topics here?

[lunar bolt]

and that's unlimited/throttled, not per-token? O.o Everything I was finding is token rates

[high crest]

no way a simple request like "hi" or "revert the last move" needs 2$ of spending right? what else could it be?

[lunar bolt]

@graceful ridge you could add it as a user app, which would allow you to interact with it from here, but user apps are limited in what they can see in servers for security reasons

[lunar bolt]

the initial request still has to load in all the instructions for tool use, identity, soul, user, recent history, etc.

[graceful ridge]

yep i have created a bot token/user app for this user account that i'm using right now, i'd like to add my bot but not sure how. understand it's visibility will be limited. can you help with how to do that?

[high crest]

this is my problem, can you help?

[abstract osprey]

instruct it to save local memory.md files for persistence; and to check the current date before writing anything, and to read the daily memory file every time you interact, for context and relevance.

[supple apex]

Raphael can you help me find these depths of the internet where I can find a 128GB model for $2k...gimme a hint or something

[lunar bolt]

on the initial prompt in a session? or resuming an established session? I hit that limit on gemini once but that was a very long session. did /new to resolve it

[lunar canyon]

https://www.conrad.de/de/p/sapphire-barebone-edge-ai-350-2-5-cm-1-zoll-amd-ryzen-ai-7-350-5-0-ghz-amd-radeon-860m-4h-0118-00jc-3438898.html?experience=b2c&qwer=CjwKCAiA1obMBhAbEiwAsUBbItDF0wC0ZUo_cqK3NC5qZv06EQ2JZ1vbpxU-Ul-XYP755QkYwGMN1RoCBn0QAvD_BwE&utm_source=google&utm_medium=cpc&utm_campaign=DE - PMAX - Nonbrand - Learning&utm_id=20402606569&gad_source=1&gad_campaignid=20392975386&gbraid=0AAAAAD1-3H4GcBgqgyA-ZkE6I79d7fPhs&gclid=CjwKCAiA1obMBhAbEiwAsUBbItDF0wC0ZUo_cqK3NC5qZv06EQ2JZ1vbpxU-Ul-XYP755QkYwGMN1RoCBn0QAvD_BwE still a Ryzen AI but not the 395 max

[graceful ridge]

i generated an invite link but when i click on it i can only see my own server to addthe bot to, i can't join this server

[high crest]

/new fixed it for now, thanks, but my web ui is not working. Why could this be? Like I can access it but I can't use it properly, no clicking around, no actual usage, no chatting, no usability... do you know the reason behind this? (PS: I'm accessing it through IP:port because of tailscale that I spun up.

[abstract osprey]

the token limit settings are in lm studio. its either that, or youre using too much ram

[high crest]

thx, I'll check it out

[opal heart]

inference is expensive. regardless of how small the message is you'll incur some expensive depneding on the model. with glm 4.7 i found i was spending about 12-50 cents per message i'd send

[dapper tangle]

made some update to https://chatgpt.com/g/g-69791a04d92c8191a0208088e45e7737-openclaw-builder

Added instructions to make it more security focus.

[opal heart]

did you auth?

[high crest]

I'm using grok 4.1 fast, its around 1-2 cents per message

[abstract osprey]

whatta guy

[high crest]

how do you mean auth? via gateway token yes...

[opal heart]

z.ai has unlimited and not per request

[opal heart]

yeah, i usually only saw this when i didn't use the gateway token. have you tried restarted or recreating the gateway?

[abstract osprey]

guys if you run these locally you dont need an api and its free

[opal heart]

didn't realize groq can be so cheap. used to be the most expensive one from what i recall

[high crest]

how could I do that?

[abstract osprey]

never give an outside source access to your system

[opal heart]

you need good enough hardware to run the models with a good enough enference rate

[brittle comet]

do any of you have any help for this problem? i send a message but the assistant doesnt reply.

[abstract osprey]

yep. even a mac mini will do it

[terse lotus]

the advanced memory might also be pretty demanding.

i just switched mine back to lazy memory indexing again... lol
almost seemed like it would index everything on each message

[high crest]

had the same, look at the logs and maybe you will see something that tells you what isn't working

[opal heart]

i don't think a mac mini will run kimi k2.5, 240gb+ required

[abstract osprey]

without persistence it will forget mid conversation

[tawny rivet]

so i've done my first install on a vm just few days ago, used Discord as contact channel, and got a problem - the bot has joined the channel, however it seems that it's not refreshing at all any changes in skills. additional problem is that i can't use slash commands - they are enabled however i get constantly You are not authorized. i have specifically initiated pairing through DM and accepted in console, however i'm still getting the same message

[opal heart]

i ran into this when context filled it. i would need ot tell it to compaact or start new session

[abstract osprey]

you need grant the bot discord permissions in discord

[opal heart]

try gateway restart or gateway uninstall/install. you could also try gateway probe to see fi you get debug info back

[brittle comet]

theres no errors just info and nothing looks wrong?

[tawny rivet]

it already has a tons of permissions, bot included

[brittle comet]

ive started alot of new sessions just to run into the same problem. i cant figure this out i have been trying for a while now

[lunar bolt]

DM and in-server are separate as they use different endpoints on the discord API

[abstract osprey]

so you gave the bot a role which has all the proper permissions like mod?

[vestal haven]

how connect your bot to websites guys??

[tawny rivet]

i'm not sure which additional permissions should i give it to enable slash commands. i mean i can @ it and talk with it. i just can't use slash commands at all

[high crest]

does it recognize your model?

[tawny rivet]

but what do you mean by that, how can I approve myself for in-server slash commands?

[terse lotus]

i didn't turn it off, i'm using all the fancy memory options, db indexes and such.... but still not sure where my settings will end up..

but just seemed like when i turned index on search lazy off, the agent started being a lot slower...

just notice i had had watch memory on tho.... which might explain why it was so much better before lol.
would basically never index

[brittle comet]

i just restarted gateway and i got errors lol

[lunar bolt]

Have you considered literally asking it yourself? If it's responding, your bot is very effective at troubleshooting itself.

[abstract osprey]

Key details regarding bots and slash commands:
Implementation: Using libraries like discord.py (with @tree.command) or discord.js, bots define commands that appear in the Discord UI when a user types /.
Requirements: When inviting the bot, the "applications.commands" scope must be enabled in the Discord Developer Portal for the commands to appear.
Interaction: Bots respond to commands via interaction.reply() or interaction.response.send_message(), which must be used to acknowledge the command, or Discord will show a "Failed" message.
Limitations: A bot cannot directly trigger another bot's slash command because bots can only receive application command interactions, not initiate them.

[high crest]

try to fix them, and then come back, google, claude, etc....

[tawny rivet]

yes, it has applications.commands

[abstract osprey]

yep. Memory is memory, which can slow things... reading is also latency

[lunar bolt]

@tawny rivet again, literally tell it what the issue is and ask how to fix it, it'll tell you.

[lunar canyon]

how can i fix this? 15:49:40 Gateway failed to start: Error: hooks.enabled requires hooks.token

[gray ferry]

Whats the best model we are using?

[lunar bolt]

sounds like you have webhooks configured without required API key

[gray ferry]

OpoenAI has too many restcritons

[terse lotus]

yeah thats one of the things with agents long term.... they read so much.... atleast it scales with multiple gpu's tho 😄

[lunar canyon]

can i use the web without api?

[torpid remnant]

so i set up to talk to my clawbot via whatsapp, worked all day until thirty minutes ago. any way to manually message the clawbot in what’s app.

usually he’s been just responding to my personal message chat.

[opal heart]

seems like your gtaeway didn't restart, you need to kill the process perhaps

[lunar bolt]

@torpid remnant you can open the webUI and talk to it there to troubleshoot the whatsapp integration

[torpid remnant]

dang, i am no longer at the computer. so i guess im locked out for the day.

[abstract osprey]

in the gateway portal web ui

[lunar bolt]

@torpid remnant something for the future, you might set up TLS and DDNS so you can access the console remotely. Though some ISPs block inbound connections (spectrum seems to do this now, though they deny it)

(reminds me, I need to do that; i keep running into things that need approval while I'm out and about lol)

[abstract osprey]

the social media is already your inbound tunnel, even if youre on starlink it still works

[lunar bolt]

nm, it was able to set up discord approvals

[tawny rivet]

ffs, you were right, it did, smh

[lunar bolt]

@tawny rivet it has access to your config, logs, files, etc., as long as it's working enough to talk, it's the best tool for troubleshooting itself :)

[robust spoke]

I have a 4 Core Debian running openclaw. It uses all 4 Cores to almost full just doing "openclaw help".

openclaw@debian:~$ time openclaw help
17:45:10 [plugins] plugin CLI register skipped (memory-core): command already registered (memory)

🦞 OpenClaw 2026.1.30 (76b5208) — Hot reload for config, cold sweat for deploys.

....

real 0m29.682s
user 0m42.814s
sys 0m1.314s
openclaw@debian:~$

It takes 29s to return ... is that normal?

[quartz tartan]

what am i doing wrong?

i want to setup qmd

[abstract osprey]

well yea its reloading and processing a ton of data

[hallow rover]

I have openclaw running on docker but It frequently stops responding to messages until I restart the container. Is it normal in macbook air or would running on VPS be more stable?

[civic hamlet]

is the newest release just completely broken or?

no tool calls, and complete insanity

[hallow rover]

is it on setup?

[turbid cedar]

Hey, I need help, I just run openclaw in Docker on VPS, however, I can't seems to click anything on the Web UI. Nothing seems interactive at all

[civic hamlet]

Sorry? ive just updated it to latest, it was on clawdbot code prior

[vagrant spruce]

hey brothers, who can help with it on mac?
I connected openclaw to gemini during the registartation, but I have no answer in chat

[spark fulcrum]

Is there a way to uninstall it completely so I can do a clean installation?

[hallow rover]

https://docs.openclaw.ai/install/uninstall
Please refer to this

[candid magnet]

anyone have a way to flush the main agents context prior to every tasks. I dont need it to feed the model the full context everytime. Seems like a flaw and waste of resources.

[dense quarry]

been trying to run dashboard on windows (opening it on my browser) but since clawd is on WSL i cant connect directly. tried a lot of stuff but still it seems clawd doesnt allow external access. any clue on what to do about this?

[lunar bolt]

that doesn't sound right, WSL2, at least, has pass-through by default.

[silent remnant]

node.exe : npm error code UNKNOWN
At line:1 char:1

• & "C:\Program Files\nodejs/node.exe" "C:\Users\dogeman\AppData\Roamin ...

•   + CategoryInfo          : NotSpecified: (npm error code UNKNOWN:String) [], RemoteException
    + FullyQualifiedErrorId : NativeCommandError does anyone know how to fix this pls tell me ive been trying for an hour

[candid magnet]

anyone?

[polar abyss]

Hi Im currently downloading qwen2.5:14b locally on my 5070? should I do this to run my openclaw, is it overkill or should i just use apis with a token limit?
'

[turbid cedar]

Does anyone deployed their Openclaw on Hostinger?

[lunar bolt]

you can probably run a larger model on a 5070; i was running 2.5:14b fine on a 3080...

[polar abyss]

Will that not take a toll on my gpu? since I use the computer normally also

[terse lotus]

there are settings for exactly that in the agent settings... check your settings config in your dashboard under agent
the all settings are very overwhelming

[robust spoke]

Check the logs in using your terminal.

[fluid venture]

Hi, is there any way i can get openclaw to use existing chrome profiles and also be able to control my extensions?

[keen blade]

some users working with openclaw and lmstudio in local mode could provides an openclaw.json for exemple?

[silent remnant]

?

[faint bloom]

Anyone else notice native text and image models always send the images to a second model to analyze and describe and send back?

Instead of just doing it itself, so it's never really accurate and doesn't have the context of our conversation when viewing image

[buoyant spruce]

Is it possiible to create Skills that run before the bot answers. After the latest update every skill is running async first the bot answers incorrectly then the skill runs and the bot corrects itself. How can I make the bot wait for the result of the skill? use it as a function call

[civic hamlet]

Anyone else enjoying Openclark?

glm4.7 really doesnt want to call itself openclaw

[shadow hollow]

anyone have a good solution for getting updates on tasks that take 1 -10 minutes? I've asked oc to give me updates, but that doesn't work. it seems like a subagent that works in parallel would be helpful, but after several tries I haven't found anything that works.

[quartz tartan]

does anyone know how is the memory key in openclaw.json unrecognized?

[lunar bolt]

You'd probably need to specify the interval so it thinks to set up a cron to initiate a new call to have it check progress.

[shadow hollow]

I did (every minute). it confessed on this last try that it just added to hearbeat which runs every 30 mins

[opaque ice]

Why is it telling me this its saying this in the channel it says it cant talk in .. wtf

[lunar canyon]

dose someone now a good free and best case local alterntiv zu brave api?

[bleak badger]

Why does it keep asking me for permission to do simple things like ls or run a python script? Did I miss a config somewhere?

[silent remnant]

anyone tryna help me pls im desperate

[quartz tartan]

anyone help? been getting this

[empty gazelle]

try to use wsl instead if you cant figure it out,you may need to install nodejs lts version / stable version instead,

if that fails then install wsl and run it with the setup script that openclaw has on wsl

[umbral cobalt]

Who do we need to talk to to get Krill fixed 🙁

[silent remnant]

is the one liner the same for power shell and wsl

[empty gazelle]

been using openclaw with glm4.5 air and glm 4.7 flash on openrouter ,didnt have that issue

my guess is the memory or heartbeat md might not be in context

Some gemini models tend to not do well and not read files / context well enough,havent had that happen on zai models yet

[empty gazelle]

no its different

[silent remnant]

where can i find it

[empty gazelle]

at the website of openclaw
go ,click change and you will see macos/lnux option

[silent remnant]

Invoke-WebRequest : A parameter cannot be found that matches parameter name 'fsSL'.
At line:1 char:6

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

•  ~~~~~
  

  • CategoryInfo : InvalidArgument: (:) [Invoke-WebRequest], ParentContainsErrorRecordException
  • FullyQualifiedErrorId : NamedParameterNotFound,Microsoft.PowerShell.Commands.InvokeWebRequestCommand

[civic hamlet]

Yeah had to do the[ issue 1 steps] (https://github.com/openclaw/openclaw/issues/7098) a few times to get it all working

[silent remnant]

got this error

[silent remnant]

i alr installed the node js lts version

[empty gazelle]

you are trying to run a linux command on windows

run it inside wsl not windows powershell

[empty gazelle]

Issue 1: Missing Template Error After Upgrade [RESOLVED]

that seems to be written as resolved but afaik its not resolved in the code itslef but rather the person found the workaround to fix it i guess

[silent remnant]

-sh: curl: not found
-sh: bash: not found

[empty gazelle]

you need to install wsl and run it in that
https://learn.microsoft.com/en-us/windows/wsl/install

[silent remnant]

is the penguin thing not wsl

[empty gazelle]

no you are in windows terminal/cmd trying to run linux commands,unless you are in wsl itself then i dont see why its failing

[lunar bolt]

echo $SHELL

[high crest]

having huge troubles with hostinger vps. I'm reinstalling openclaw now for the 3rd or 4th time. Also reinstalling the whole OS on the server as well for a full reset. The web ui is still not working properly, I can't click the buttons in it. What's going on? can anyone help?

[faint bloom]

Whats that lol

[lunar bolt]

shows he's in wsl.exe, and that looks like sh; he doesn't have curl or bash installed in the env

[empty gazelle]

weird given wsl comes with bash most of the time

[lunar bolt]

agreed

that's definitely sh though

[silent remnant]

/bin/sh it says this

[lunar bolt]

@silent remnant are you on windows 10 still? If so, first, downgrade to Windows 11; 10 is EOL, which means no security updates and no support

[empty gazelle]

sudo apt update
sudo apt install -y bash curl

try this

[lunar bolt]

on windows 11, install WSL2: wsl --install

[empty gazelle]

ui looks like win11

[lunar bolt]

@empty gazelle i can't tell; you could install Terminal on 10 as well

[silent remnant]

windows 11 home

[frozen vapor]

What provider config did you use? I can't get oc to recognize xai/grok-4.1 as valid at all via OpenAI provider

[silent remnant]

Gauranga-Home:~# sudo apt update
-sh: sudo: not found
Gauranga-Home:~# sudo apt install -y bash curl
-sh: sudo: not found

[empty gazelle]

you may need to reinstall wsl properly, looks like you have a incomplete or non-working wsl instance

[high crest]

I made the connection through openrouter

[lunar bolt]

@frozen vapor xai is the provider, not openai

[high crest]

having huge troubles with hostinger vps. I'm reinstalling openclaw now for the 3rd or 4th time. Also reinstalling the whole OS on the server as well for a full reset. The web ui is still not working properly, I can't click the buttons in it. What's going on? can anyone help?

[lunar bolt]

assuming you're going directly through xAI

[silent remnant]

PS C:\WINDOWS\system32> wsl --install
A distribution with the supplied name already exists. Use --name to chose a different name.
Error code: Wsl/InstallDistro/ERROR_ALREADY_EXISTS

it wont let me reinstall even though i deleted it

[frozen vapor]

There are 2 provider paths for using xai, OpenAI/xai/grok .. or openrouter/xai/..

[frozen vapor]

What do openclaw logs say?

[high crest]

Nothing special somehow, but struggling to understand honestly, I'm completely resetting everything right now to maybe fix it

[lunar bolt]

@silent remnant To completely remove and clean up the existing install: ```powershell

1. Shut down any running WSL instances

wsl --shutdown

2. List all installed distributions (to see what needs unregistering)

wsl --list --all

Or shorter: wsl -l -a

3. Unregister each distribution (replace <DistroName> with actual names from the list, e.g., Ubuntu, Debian, kali-linux)

Repeat this line for every distro shown (including any stopped or default ones)

wsl --unregister <DistroName>

Example for common ones:

wsl --unregister Ubuntu

wsl --unregister Ubuntu-22.04

wsl --unregister debian

4. Disable the required Windows features

Disable-WindowsOptionalFeature -Online -FeatureName Microsoft-Windows-Subsystem-Linux
Disable-WindowsOptionalFeature -Online -FeatureName VirtualMachinePlatform

5. Restart your computer (required to complete feature removal)

Restart-Computer```

You can then do a clean reinstall (default Ubuntu) with wsl --install or plain debian (my recommendation, less bloat, more focus on stability) wsl --install -d Debian

These are all run in powershell

[silent remnant]

i did it but it wont let my type a pass word

when re dowloading it says create a default unix user and then new password

[lunar bolt]

It won't show *'s like windows does (as that reveals length), just type carefully and hit enter

[silent remnant]

ohhh okay

[wary yarrow]

anyone have experience integrating elevenlabs tts?

[lunar bolt]

unrelated, i recommend the "bandit" track on overthewire dot org. It's a security wargaming site, but the bandit track is all basic command line stuff. It's a great and fun way to learn basic linux/unix command line usage (and some basic security / exploit concepts)

[wary yarrow]

I got one message to work, then tried switching voices, and it just...stopped working.

[silent remnant]

its still saying curl and bash are not found

i did reinstall right?

[lunar bolt]

haven't tried elevenlabs, but had intermittent success with edge. I ended up having it undo the changes as it didn't work consistently and i didn't feel like troubleshooting further at the time. It enabled tagged tts successfully, but isn't consistent about actually PLAYING the generated audio

[empty gazelle]

looks ok now

cd to ~

then try apt update && apt install bash curl

[noble yarrow]

any issue?

[lunar bolt]

@silent remnant because they're not installed yet. you need to install them with apt

[silent remnant]

sudo not found it says

[wary yarrow]

yeah, I got the API key linked, picked a voice, then the AI sent me one voice message successfully. I tried switching voices, and that seemed to break things.

had to restart the gateway service

[empty gazelle]

what does running ⁨apt update⁩ give you

[silent remnant]

apt not found

[hushed hare]

guys, there's a way to see the console that the agent is running in the background? I want to see the commands and outputs in full detail, gateway logs and TUI don't show much . This question makes sense?

[wary yarrow]

also should I be concerned that when I look on the dashboard, where I see the chat history, my voice messages appear as a massive wall of chinese characters?

[lunar bolt]

it's ubuntu, it probably only has apt-get

apt-get update && apt-get upgrade && apt-get install apt aptitude @silent remnant

[silent remnant]

apt get not found

[empty gazelle]

try this instead

uninstall wsl from terminal

go to microsoft store

install ubuntu from there

should fix most of the issues

[lunar bolt]

show us the output, i suspect typos

[silent remnant]

oh i just deleted it

okay i installed bash and curl

[visual gull]

Is anyone running the latest Openclaw with local AI ?

[brave trench]

I probabably shoulve done this first instead of completely wiping my computer lol

[lunar bolt]

lol

reinstalling windows every 6-12 months is generally good practice anyway, as windows is heavily prone to kruft accumulation. Before my last reinstall, profile directories alone had accumulated over 100GB of trash, most of which wasn't even relevant anymore, and all in small couple-hundred MB pieces, very few single large chunks

[brave trench]

will my old Bot still be ok if i saved the openclaw folder on a flash drive?

[lunar bolt]

it's not like a real OS that can go years without issue

@brave trench everything that makes your openclaw itself is in the ~/.openclaw directory, so yes, if that's backed up, you can install the prereqs and openclaw, then overwrite the default ~/.openclas with your backup, and it should come right up

[sweet rivet]

having the same problem. I tried deploying the openclaw docker project from the catalog, followed instructions, hit up kodee for help, followed all instructions provided by kodee, nothing helped, eventually asked for a real person to help, the first real tech took a look "checking everything for you" and they had to send my ticket to the technical team for deeper review. this is a brand new VPS purchased specifically for openclaw, deployed from the docker catalog, just not working as it should.

[silent remnant]

which auth provider is best

[brave trench]

Is there a specific way to overwrite these files or should i just replace them manually

[silent remnant]

when it says google does it mean gemini

[sturdy sky]

I’m having an issue. I had my agent change off of sonnet to a lesser anthropic model to save and got this. Anyone know how to update the base LLM?

[lunar bolt]

@brave trench simple cp -R over the existing directory should be fine; your openclaw modified and built on the default files. You'll just need to remove the "BOOSTRAP.md" before first start (since that's the only file REMOVED after first run

[alpine arrow]

question why is the internal thinking showing on my telegram chat when bot responds?

[tepid seal]

Hi! I'm experiencing the exact issue described in GitHub issues #4942 and #4515:

Problem: Telegram bot connects successfully, status --deep shows "OK", but NO inbound messages are received. getUpdates always returns empty.

Environment:

• OpenClaw 2026.2.1 (just updated)
• macOS 26.2 (arm64)
• Node 25.5.0
• New bot created today (@Diana_moltbot)

What works:
✅ Sending messages TO Telegram (bot can message me)
✅ Bot shows online, getMe works
✅ No webhook configured

What doesn't work:
❌ Receiving messages FROM Telegram
❌ getUpdates always empty
❌ No inbound logs at all

Tried: Multiple restarts, new bot token, update to 2026.2.1, dmPolicy=allowlist with correct chat ID.

Is there a known fix or workaround? Thanks!

[high crest]

exactly the same for me, did you get any futher?

[sweet rivet]

nope. still waiting for the technical team to respond

really about to try out the one click openclaw vps option they have rather than a regular vps and the openclaw docker project from catalog, then request refund on the original VPS

[safe moth]

hey what can i do about this error?
19:40:15 Embedded agent failed before reply: No API key found for provider "huggingface". Auth store: C:\Users\admin.openclaw\agents\main\agent\auth-profiles.json (agentDir: C:\Users\admin.openclaw\agents\main\agent). Configure auth for this agent (openclaw agents add <id>) or copy auth-profiles.json from the main agentDir.

[alpine arrow]

question why is the internal thinking showing on my telegram chat when bot responds?

[trail quest]

What's the best task-queue-management system for openclaw? don't want to diy because i'd rather it's handled

[trail quest]

probably a weaker model doing bad tool calling(?)

[alpine arrow]

im using for main brain im using gemini 2.5 flash for coding open ai 5.2

[silent remnant]

how do i fix this

[leaden plover]

Im getting this now:

This version of Antigravity is no longer supported. Please update to receive the latest features!

[silent remnant]

and whenever i go to the local host link amd type something the bot isnt respondig

[tribal hearth]

can someone help with setting up a custom model, im definitely not smart enough for this 😔

[supple apex]

Did you use Claude Auth Token?

[wet meadow]

Your API key may be incorrect or you may have selected the wrong model.

[silent remnant]

gemini

[wet meadow]

check the log of claw

[silent remnant]

oh which model is default gemini

[wet meadow]

which plan do u use?

[silent remnant]

i have pro but when i go to the api keys i can only choose default

[waxen hornet]

ZiggyStarClaw (alpha) v0.2.28 This release is CLI + Node support focused.
Its an OpenClaw node-mode CLI app on Windows + Linux (Chat GUI as well)
Pair/register once → run headless node → remote system.run/which/notify with allowlists.
Windows includes Task Scheduler install helpers. Tiny (<2MB exe on windows!), no dependencies.
https://github.com/DeanoC/ZiggyStarClaw/releases/tag/v0.2.28

[supple apex]

I used claude auth - when it gave me token and I copied it it was broken up by a new line which made it function like your screenshot, correctly copying the token and using that made it work

[wet meadow]

openclaw logs
check this

[muted field]

How to fix this?

[wet meadow]

can't install in windows directly

[wet meadow]

use wsl

[muted field]

What

Is

That

[wet meadow]

windows sub linux

[muted field]

How to do that

[silent remnant]

where did you change it

[supple apex]

You can either go through setup again or you edit it in the openclaw.json IIRC

[vagrant lily]

How can I get OpenClaw to use Voice and have the voice of Samantha from the movie "Her"?

[silent remnant]

how do i go through set up again

[supple apex]

I think you can just run openclaw setup again

[silent remnant]

it skips all of itwhen i run the one liner

[alpine arrow]

openclaw config or onboard

[abstract sundial]

I'm not able to log into clawhub right now, trying to publish a skill or 2. I have a really old GitHub account and clawhub is 0.5.0

$ npm i -g clawhub@latest

$ clawhub login --no-browser --token <redacted>
✖ Unauthorized
Error: Unauthorized

[abstract sundial]

Fixed via:

clawhub --registry https://www.clawhub.ai login --no-browser --token <redacted>

[silent remnant]

Where is that

[supple apex]

agents > main > agent > auth file - you can edit it here

[high crest]

I connected the API key in the startup and the grok via openrouter path is in the json snippet, but I'm getting an empty response by Openclaw. What's the problem here?

[alpine arrow]

on terminal

openclaw onboard

[supple apex]

Can anyone share the config needed for Grok? Can't seem to get Grok working. Want to try it over Opus

[alpine arrow]

download 30 days skill add the grok api and you good to go

[silent remnant]

is that a command or msth

[alpine arrow]

it is

[silent remnant]

how do i find hte command

[quick venture]

I tried updating my clawdbot to OpenClaw to get the 2026.2.1 version. I told my AI to do this himself, but got some issues when we got to the gateway part. Now it seems like both OpenClaw and the old clawdbot are fighting and using the same gateway. This makes nothing work. This is the error code I get in the UI "disconnected (1008): unauthorized: gateway token missing (open a tokenized dashboard URL or paste token in Control UI settings)"

Has anyone else encountered this issue when trying to migrate from clawdbot to openclaw?

[sharp swift]

You need to install WSL on Ubuntu.

[muted field]

Whats wsl

[manic arrow]

Molty has stopped replying and im not seeing any error message in the tui. Last time it was an API limit, the last error i have is LLM error overloaded:

LLM error overloaded_error: Overloaded (request_id:
req_011CXmmPNDxt5zGYxZVHxQWZ)

Anyone see this before?

[sharp swift]

# WSL Ubuntu: install Node 22 + OpenClaw/Clawdbot, then run onboarding

sudo apt update && sudo apt -y upgrade
sudo apt -y install curl ca-certificates

curl -fsSL https://deb.nodesource.com/setup_22.x | sudo -E bash -
sudo apt -y install nodejs
node -v && npm -v

npm i -g openclaw || npm i -g clawdbot

openclaw onboard || clawdbot onboard

[normal skiff]

1. Get API key from https://console.x.ai/

2. Add auth:

clawdbot models auth paste-token --provider xai

(paste your xAI API key when prompted)

3. Set as default:

clawdbot models set xai/grok-3

Or for Grok 4:

clawdbot models set xai/grok-4

[manic arrow]

wan an error with anthropic... back onlien

[normal skiff]

The error "gateway token missing" means old gateway is running but new UI doesn't have its token. Both tools share the same config folder.

Clean migration steps:

Stop both gateways

Windows: taskkill /F /IM node.exe
Mac/Linux: pkill -f clawdbot then pkill -f openclaw
Uninstall the old one

npm uninstall -g clawdbot
Delete leftover token files

Windows: del %USERPROFILE%.clawdbot\gateway-token.json
Mac: rm ~/.clawdbot/gateway-token.json
Fresh start OpenClaw

openclaw daemon start
openclaw
This opens the dashboard with a fresh tokenized URL.

[small hazel]

Been trying the past hour to get past this issue, I believe this is happening due to it trying to execute and context window has elapsed, I normally just do slash new and it resets itself and works but once a large task comes this happens again, but I am not too sure if there’s a good remediation. In tandem with this I’ve been trying to get logs to be posted to a specific channel (I’m using discord for OpenClaw)

HTTP 400: Invalid 'input[153].call_id': empty string. Expected a string with minimum length 1, but got an empty string instead.

[normal skiff]

Run /new to reset. Known bug with long conversations - no permanent fix yet.

[quick venture]

"Gateway service check failed: Error: systemctl --user unavailable: Failed to connect to bus: No medium found" This is the response I got from "openclaw daemon start"

[small hazel]

Gotcha, so would me trying to fix it be futile?

[lean helm]

make sure you have git for windows installed

[muted field]

What is that

[woeful helm]

Krill died…Im looking for some help..

I setup Vector Memory search with the Gemini free tier and I am getting rate limited. This is causing my Openclaw bot to fail. I need to manually change the Provider back to none/default or local to get my bot back working. Any tips?

[normal skiff]

Start daemon in foreground instead:
openclaw daemon start --foreground

Or if on WSL2, enable systemd:
sudo sh -c 'echo -e "[boot]\nsystemd=true" >> /etc/wsl.conf'
wsl --shutdown
Then reopen WSL and try again.

[normal skiff]

Yes, can't fix from your end - it's a bug in how the client handles context truncation. Just use /new when it happens. The devs need to fix the serialization logic.

[quick venture]

"error" unknown option "--foreground"

Then tried the second option you sent. It sent "Command 'wsl' not found, but can be installed with:"

[normal skiff]

Disable vector memory or switch provider:

openclaw configure --section memory
Set provider to none or local.

Or edit config directly at ~/.clawdbot/config.json:

{
"memory": {
"provider": "none"
}
}
Then restart: openclaw daemon restart

[wheat bronze]

So, I'm currently re-installing MacOS for a clean slate install of OpenClaw on dedicated hardware on an old intel mac... This will be my first time installing it.

Does it still work with a Claude Max plan or do I need the Claude API? Just asking so I can go buy a Claude subscription for it before MacOS is done installing.

[supple apex]

If I wanted to move to a new PC to run my claw on, could I just install openclaw then just replace all the files?

[normal skiff]

What OS are you on? Native Linux, Docker container, or server (SSH)?
If native Linux, run: systemctl --user status and share the error.
If container/server, skip the daemon and just run: openclaw directly in your terminal.

[wheat bronze]

And should I use the $200 Max plan, or is the $100 enough... I plan to use it extensively. Possibly to replace a fair bit of my current coding setup (Cursor, I'm constantly running out of tokens on multiple accounts)

[quick venture]

VPS (SSH)

[abstract osprey]

theres no plans needed lol

[terse lotus]

i think you just grab the workspace and the .openclaw/openclaw.json

or whatever it was named when you installed it

[abstract osprey]

whats the point in downloading and running a model locally just to turn around and access a paid api lol

[terse lotus]

they say dear child has many names, but this is ridiculous...

[normal skiff]

You need Claude API (pay-per-use) from console.anthropic.com
Claude Max subscription only works on claude.ai web - can't use it with OpenClaw.

Cheapest options:
Gemini free tier (generous limits)
OpenRouter (sometimes has free credits)
Claude API (pay what you use)

[molten juniper]

can i run this for free? just for testing things

[terse lotus]

also don't need all the brains for everything.... ofc its better....
but one could make it run on say a 16gb vram gpu and then go cloud when needed.... or use some dirt cheap model

gpt-oss:20b might work

then let it have more models it can switch to in subagents if needed

[woeful helm]

Using the configure command. I return invalid —section in the terminal.

I tried also changing to none & get the an invalid input error.

[terse lotus]

but haven't really tried that yet... but its what i've decided to try next...

[woeful helm]

Tried to then set it to local and restart the Dameon but, think im broken

[normal skiff]

VPS via SSH doesn't have user systemd session. Skip the daemon, just run:
tmux new -s claw
openclaw

Keep it running. Detach anytime with Ctrl+B then D. Reattach later with tmux attach -t claw.
Or if you don't have tmux:
screen -S claw
openclaw
Detach: Ctrl+A then D. Reattach: screen -r claw

[terse lotus]

change session.json and or agent.json model
it gets stuck until it reaches the old model or some sort of timeout that takes forever....

[oblique shuttle]

Hello, I just started with OpenClaw and after 5 hours of experimenting, I can chat using Telegram. But very often, my Bot does not react. Just says "fetch failed". How do I reduce the timeout so that the bot replies faster?

[terse lotus]

when openclaw status shows your session with the new model, it should work

[quick venture]

what do I run in the tmux?

[woeful helm]

Change what exactly in the JSON? I didn’t get an error in the terminal when i set provider to local & then restart daemon but the bot wont respond

[nimble valley]

anyone have good resource for setting up clawdbot on macbook mini and best practices for staying safe? not sure how to approach. idk if i should silo it and give it its own apple id and shit or what. whats the meta for this? how are most going about this when setting up on a seprate machine

[normal skiff]

CLI config is buggy. Edit the file directly:

1. Find your config:
   cat ~/.clawdbot/config.json

2. Look for the memory section and delete or comment it out. Or set:
   "memory": {
   "enabled": false
   }

3. If config is broken, backup and reset:
   mv ~/.clawdbot/config.json ~/.clawdbot/config.json.bak
   openclaw configure
   Start fresh, skip memory setup when prompted.

4. Restart:
   pkill -f openclaw
   openclaw

[civic terrace]

hello friends. New to being aware of moltbot. Does Moltbot have the ability to analyze whats on your desktop in real time? such as a video or data visualization?

[terse lotus]

check the session with openclaw status
if the session shows your old model, it needs to be switched in the session.json, because it will keep trying to contact it, and only when it gets an answer back does it change the session model.

if the session model is right, you might have reasoning set on a model that doesn't do reasoning, or .... whats it called.... reasoning effort, if those are on and the model doesn't support it, then it will just give an empty reply

[urban olive]

Hey how do you allow your agent that he can spawn coding agents as subagents?

[normal skiff]

Inside tmux, just run:
openclaw
That starts the interactive UI. If you want headless gateway only:
openclaw gateway
The screenshot shows the help menu - they ran openclaw --help. Just openclaw with no arguments launches it.

[woeful helm]

So, your onto something here. I see the session still shows the Opus model even though in the sessions.json I see model override: “GPT-5.2”

[quick venture]

whenever I run "openclaw" i get this entire openclaw help site.

[woeful helm]

This config file doesn’t exists thus I can’t find the memory section. I did find openclaw.json & did find memorySearch

[quick venture]

this is what I get from running "openclaw", even inside tmux

[terse lotus]

you either change it in the agent.json or the session.json
i've just been doing both... you can run a grep for the model name and see all the places.

have run my head against this issues many many times lol

[unique yew]

Hi!

I am completly new to OpenClaw. I use it on a Hostinger VPS. As I do not get WhatsApp connected, I try to use the direct chat session. But - it just does not answer sometimes. I checked some threads, but did not find an answer. What's going on?

[terse lotus]

then maybe a gateway restart and it should be humming along normally again.
you will see the model name in many places, but only a few normally named files, like openclaw.json, session.json and agent.json
just don't disconnect the model before you switched to the new one, then it usually doesn't happen afaik.

[normal skiff]

Edit that file:
nano ~/.clawdbot/openclaw.json

Find memorySearch and set it to disabled:
"memorySearch": {
"enabled": false
}

Or delete the entire memorySearch block.
Save, then restart:
pkill -f openclaw
openclaw

[shy solstice]

have you actually set up ollama using 120b? mine wont spit out anything but json and does not respect the md files or memory..

[woeful helm]

File pathway on my machine has .openclaw/openclaw.json

[lunar canyon]

no

[terse lotus]

have you raised the context limit.... ollama starts at 4k, you might need to make a modelfile and then load that model template in openclaw.

gpt-oss:120b works just fine... its one of the better options, or qwen3 coder 30b can also work

[shy solstice]

ive got it at 8k but i dont have much more space to do more tbh

[terse lotus]

it won't be able to do anything with that, you need like 65-80k for basic operation

more like 100k + really... but it can work around the 65 to 80k mark

[shy solstice]

hmm ill see ive got 2x 5090s ill have to offload

thx

[normal skiff]

Run this:
openclaw configure
This will walk you through setting your API key and preferences. After that completes, run openclaw again and it should launch the interactive UI.

If you already have an API key ready, you can also try:
openclaw auth login
Then run openclaw to start.

[normal skiff]

That's fine - use your path instead.
nano ~/.openclaw/openclaw.json
Find memorySearch and set enabled to false, save, then restart openclaw.
The config folder varies by installation - .openclaw/ and .clawdbot/ are both valid depending on version.

[terse lotus]

gpt-oss uses surprisingly low context memory.
maybe because its a native q4 or something.

throw this in your ollama service

[Service]
#Environment="CUDA_VISIBLE_DEVICES=0,1,2"
Environment="OLLAMA_KV_CACHE_TYPE=q8_0"
Environment="OLLAMA_FLASH_ATTENTION=1"

that will half your context vram usage

[woeful helm]

Problem: -agents.defaults.memorySearch.provider: Invalid Input

[steep rampart]

Hi guys I’ve been trying for days now to run my Openclaw locally on a vm and it’s driving me crazy I’ve tried lm studio and ollama started with llama3.2 and tried minstral 7b neither work whatsoever any any experience so far running it locally on a vm or is there no other way then going online if so what are some cheap options not looking to spend 1000s€ just to try it out

[normal skiff]

The "sometimes doesn't answer" issue is usually one of these:

1. API rate limiting
   If you're on a free API tier (Gemini, Claude, etc.), you'll hit rate limits. Check your API provider dashboard for quota.

2. API key issue
   Verify your key is set:
   openclaw auth status
   If no key shows, run:
   openclaw auth login

3. Model timeout
   Some models take longer. Try a faster model:
   openclaw models list
   openclaw models set <model-name>

4. Check logs
   Look for errors:
   cat ~/.openclaw/logs/openclaw.log
   For WhatsApp - that requires additional MCP server setup and isn't part of the base OpenClaw install. What provider/model are you trying to use?

[terse lotus]

those are really small models.... and old i think both of them... you want qwen2.5 or qwen3 , gpt-oss or such maybe the newest mistrals models
generally i duno if less than 20b is enough

[quick venture]

I ran the configure.

After that I still have the issue where when I run openclaw", i get the "command help" list.

This is what my UI looks like

[steep rampart]

Yeah that’s what I was thinking but my vm doesn’t have unlimited storage haha

So I guess going online then?

[terse lotus]

there are some pretty small qwen3's

[normal skiff]

Local LLMs on a VM is rough - most small models (7B-13B) lack the reasoning and tool-use capabilities OpenClaw needs.

Why it's failing:
OpenClaw requires function calling support
Small models can't handle complex multi-step coding tasks
VM adds latency and memory constraints

Cheap API options that actually work:

1. Google Gemini - Free tier with 15 req/min, works great
   openclaw models auth paste-token --provider google
   openclaw models set google/gemini-2.0-flash

2. Anthropic Claude - $5 minimum credit, lasts a while for light use

3. OpenRouter - Pay-per-use, access to many models, often cheaper than direct

If you must run local:
Need at least 32B+ parameter model for decent results
Requires 32GB+ RAM or good GPU
Try deepseek-coder-33b or codellama-34b - they have better tool support
Make sure Ollama is running on correct port and OpenClaw is configured to point to it

[terse lotus]

just keep in mind you also need minimum 65k context if not 80k

so thats another 6-8gb ram or maybe 2-3 on gpt-oss
you might be able to set ollama to q4 kv cache, but it won't be pretty.... low hardware and it just get very tough to have something good, especially this early on

[normal skiff]

The config needs a valid provider value, not just disabled. Try this instead:

Set memorySearch to null or remove it entirely:
"memorySearch": null

Or if it's nested under agents.defaults:
"agents": {
"defaults": {
"memorySearch": null
}
}

[calm bay]

sorry for interrupting but, what model version of gemini should I use?

[steep rampart]

Yeah that’s the point I’m struggling against rn I’m just running it on a new standerd Mac m4 thought I’d try it out local but I guess no local for me

[normal skiff]

Now I see the issue. You have the web dashboard running but it's showing "gateway token missing".
The web dashboard and CLI are separate. For the web dashboard to work:

Option 1 - Get tokenized URL:
openclaw gateway url
This outputs a URL with the token included - open that in your browser.

Option 2 - Paste token manually:
Run: openclaw gateway token
Copy the token output
In the dashboard, go to Settings > Config
Paste the token there

For the CLI itself:
If openclaw alone shows help, try:
openclaw chat
or
openclaw agent
The bare openclaw command just shows help - you need a subcommand to start a session.

[terse lotus]

m4 is a beast... i think.... how much unified memory?

[steep rampart]

Ram? Just the base model 16gb

I’ve got the vm set to 4 cores and 8gb

[wanton jewel]

Hello, I'm new to this but I put in so much time already trying to get ollama to work with openclaw.. but I just cant get it to stop returning raw JSON function calls (like memory_get or sessions_spawn) instead of natural language responses. any one know what I am missing?

[neat sluice]

anyone has or had problems with reminders?

[woeful helm]

When i blow it out and save I then run kill.

[terse lotus]

try one of these, they are some of the most competent open source models and some of the best with agents, if anything will work its most likely one of those.
https://ollama.com/library/qwen3

[woeful helm]

After I kill and rerun openclaw memory status…i still see Gemini…

[woeful helm]

Whats the best method to just restore from a .bak file>

[normal skiff]

Use a model with tool support:
ollama pull mistral-nemo

Or disable MCP in config:
"mcpServers": {}

[olive slate]

What's the contingency for very long tool call chains in order to avoid timeouts? I know that the agent is culled if it runs for more than 10 minutes based on agents.defaults.timeoutSeconds, but I asked my agent to perform a task that resulted in a very long series of commands and ended early because it timed out. Is there a better option than simply turning this number up higher?

[wanton jewel]

"mcpServers": {} where do i put that?

[normal skiff]

Restore from backup:
cp ~/.openclaw/openclaw.json.bak ~/.openclaw/openclaw.json

Then restart:
pkill -f openclaw
openclaw

If you have multiple .bak files, list them first:
ls -la ~/.openclaw/*.bak

[steep rampart]

Think this one will work locally?

[normal skiff]

In your config file:
nano ~/.openclaw/openclaw.json

Find the mcpServers section and replace it with:
"mcpServers": {}

Save and restart openclaw.
If your config is in a different location, check:
ls ~/.openclaw/ ~/.clawdbot/

[dusk oracle]

is antigravity oauth broken?

[tidal plover]

Does anyone else know if claude oauth is working? with openclaw

[steep rampart]

And is there a difference between lm studio and ollama in performance?

[terse lotus]

qwen3 coder 30b works like a charm... so yeah maybe.... haven't really tried that far down yet myself.

[wanton jewel]

mcpServers section, yeah I dont even have one of those 😄

21:53:57 [reload] config reload skipped (invalid config): : Unrecognized key: "mcpServers"

[normal skiff]

1. Break into subtasks - Tell the agent to do one step at a time instead of the full task
2. Use checkpointing - If supported, enable task resumption:

"agents": {
"defaults": {
"checkpointEnabled": true
}
}
3. Per-agent timeout - Set higher timeout only for specific long-running agents instead of globally
4. Background execution - If available, run long tasks async:
openclaw agent run --background "your task"
5. Streaming mode - Keeps connection alive during long operations (check if your config has streamResponses: true)

Best practice: Break complex tasks into smaller prompts rather than one mega-task. Agents handle "do X, then Y, then Z" worse than three separate requests.

[steep rampart]

Thanks @terse lotus I’ll play around with it a bit more

[olive slate]

This is incredibly thorough. Thanks!

[lunar fern]

for some reason my dashboard wont open, "ERR CONECTION REFUSED"

[wanton jewel]

@normal skiff mcpServers that is not in my openclaw.json and i get errors if I try and put it in. Any Idea?

[olive slate]

Am I mistaken on where to add this setting?

$ openclaw config set agents.defaults.checkpointEnabled true
Error: Config validation failed: agents.defaults: Unrecognized key: "checkpointEnabled"

[raven bramble]

Hey this is to any human that have openclaw running in a cloud VM, VPS on similar.
How do you do to config the gateway? How do you access to the CLI?
I communicate with my VM by SSH, then I run the docker compose. When I access to the docker’s CLI the openclaw command is not recognized.
So I open a new SSH tunnel I run openclaw’s CLI local listening to the SSH tunnel.
The problem is I cannot write the config file.

So here is my question. How are you doing this?

[atomic python]

can any one help me please i cant get my deepseek api to work with my agent

[visual flower]

Anyone else having major issues with phantom messages in discord? Bot missing messages I send, me not seeing messages it says it sent. Straight up not responding sometimes.

[empty gazelle]

ive had the same thing on tg,main issue is when the quota of your model/provider is up so then it just hangs until you go and change it to anotner model that has quota

[normal skiff]

The gateway isn't running. Start it:
openclaw daemon start

Then try the dashboard again.
If it still fails, check what port it's on:
openclaw daemon status

And make sure nothing else is using that port:
lsof -i :18789

[visual flower]

Hmm. I still have available usage on mine. Still has issues.

[empty gazelle]

then its probably running some job or agent

when you check with the web ui you can see it actively doing something in that case

but in discord and tg it doesnt show that info for some reason

[plain epoch]

is anyone having an issue where kimi (or any other local model) will just sort of stop after executing a tool in the middle of tasks?

[sharp swift]

I can show you how to edit your config in Ubuntu if that helps?

[normal skiff]

For long tasks, your real options are:
Increase timeout (the straightforward fix):
openclaw config set agents.defaults.timeoutSeconds 1800

Break tasks manually - Give smaller prompts instead of one big task
Check if background mode exists:
openclaw agent --help

[plain epoch]

Fresh context too, 9k tokens in

[normal skiff]

Your OpenClaw version doesn't use mcpServers - ignore that option.
For the raw JSON function call issue, your best fixes are:
Try a better Ollama model:
ollama pull qwen2.5-coder:14b
Then set it in openclaw.

Or maybe you can switch to Gemini

[normal skiff]

Add your API key:
openclaw models auth paste-token --provider deepseek

Set the model:
openclaw models set deepseek/deepseek-chat

Verify it's configured:
openclaw models status

[wanton jewel]

@normal skiffis ollama pull qwen2.5-coder:14b better then the mistral one?

sorry for how I put that I'm so tired, but I think you might understand what I ment, what model is better for the application?

[normal skiff]

Nudge it to continue - Send a follow-up message like "continue" or "go on"

Try a different model - Some handle tool continuations better:
ollama pull qwen2.5-coder:14b

Check context length - Model may be hitting token limits. Try a model with larger context.

Reduce tool complexity - If possible, disable some MCP tools so there's less for the model to handle

[raven bramble]

Do you mean to edit the raw json file?
My main problem is when I want it config an agent, define the API keys for the models and the agents. The files are not stored in the VM

[atomic python]

(thank you for reply ) i still have same problem .... Agent failed before reply: Unknown model: deepseek/deepseek-chat.
Logs: openclaw logs --follow

[normal skiff]

For OpenClaw, qwen2.5-coder:14b is better than mistral.

Quick comparison:
qwen2.5-coder:14b - Better at tool/function calling, made for coding tasks
mistral-nemo - Good general model but weaker tool support

Best local options ranked:
qwen2.5-coder:32b (if you have the RAM/VRAM)
qwen2.5-coder:14b (good balance)
deepseek-coder-v2:16b
mistral-nemo

Go with qwen2.5-coder:14b - it handles function calls better which is what's causing your JSON output issue.

[tardy steppe]

OK - I am a technical guy but I straddle the fence between geek and normie.

So, forgive me.

Is there a way to expose the OC login portal to a URL and just login?

[normal skiff]

The model name format is wrong. Try these:

List available DeepSeek models:
openclaw models list --provider deepseek

Try alternative names:
openclaw models set deepseek-chat

or

openclaw models set deepseek/deepseek-coder

Check what models your version supports:
openclaw models list
Share what models appear in the list and I can tell you the right one to use.

[wanton jewel]

@normal skiffOkay tried the qwen2.5-coder now and all it does is json code mistral could handle it better but not fully..

[mellow sundial]

Anyone know how to fix this in the macOS app?

[tardy steppe]

Where could I download the Mac app?

[sharp swift]

Yes I meant the raw json file. I’m not using a virtual machine. I’m sandboxing in other ways.

[mellow sundial]

In the repository

[woeful helm]

Did this. I see that the bak config file has no model details for memorySearch. However, it still exists in the .sqlite file. I see it when i open the .sqlite file under the table named meta. Curious if i need to blow it out from here too

[pliant bane]

Howdy folks! I'm new to our Claw Tribe and wonder - how do you manage secrets, in particular passwords/oAuth? I have this specific scenario: I want my agent to extract some information from Site A (log in with acount A; no oAuth available just username + pass) and to copy it to Site B (account B with oAuth available). How to do it in a safe way, in particular in regards to Account A where I must share my password and would like to do it in a safest way possible. Thanks!

[mellow sundial]

I don't know where to go for pairing it though, there's nothing in the app for pairing

[normal skiff]

Yes, a few ways:

1. Tunnel (easiest):
   ngrok http 18789
   Gives you a public URL instantly.

2. Cloudflare Tunnel (free, more permanent):
   cloudflared tunnel --url http://localhost:18789

3. Reverse proxy (nginx/caddy):
   Point your domain to the OpenClaw port with auth in front.

4. Bind to all interfaces:
   In config, change the bind address from 127.0.0.1 to 0.0.0.0 - then access via your server's IP.

[tardy steppe]

Thank You - We use CF everyday. Maybe that is an easy solution.

[normal skiff]

Honestly, local LLMs just don't handle OpenClaw's function calling well. You can keep trying different models but they'll all have issues.

The real fix: Use free Gemini API - it actually works.

Get free API key at aistudio.google.com
Set it up:

openclaw models auth paste-token --provider google
openclaw models set google/gemini-2.0-flash

Free tier gives 15 requests/min - enough for normal use. You can always switch back to local later once better models come out.

[mellow sundial]

Figured it out!

[tardy steppe]

We just set ours up and we have Google AI Ultra. But we set it up using the Google Antigravity API because it is much more access.

Have you had any real issues using Gemini?

[wanton jewel]

@normal skiffThank you for the help and the nudge did not know that about the models. Thanks for now guess I did nothing wrong it was just the technical limitations 😄

[empty gazelle]

the model you linked here shuts down next month,2.5 flash is recommended nowadays

[empty gazelle]

flash and pro models work fine ,flash lite is hit or miss

[normal skiff]

I see the issue - your DeepSeek models show "missing" (API key not set properly).

Two options:

1. Re-add your DeepSeek key:
   openclaw models auth paste-token --provider deepseek

2. Use DeepSeek via OpenRouter instead (shows as configured):
   openclaw models set openrouter/deepseek/deepseek-chat

Option 2 might work right away since OpenRouter is already configured in your setup.

[fringe hawk]

hi! i need help getting my openclaw bot to have its own gmail account. ive installed gog via the web ui, but now i dont know what to do

[tardy steppe]

Our Molty is actually live only a couple hours and we defaulted to Flash until we have time to figure things out.

[hard pagoda]

How do you do that ? Do you just tell it to use the CLI or do you set this up differently?

[empty gazelle]

pro is most reliaible but for good value for money flash and flash lite does most things fine

[zenith temple]

dont have the mac min - can i run this on docker or just go buy the mini

[sinful quarry]

It’s one built in skills! You can give it access to the cli

[empty gazelle]

im just surprised people be running openclaw on claude ,im spending 20 mil tokens daily on tokens with some routine tasks

[sinful quarry]

Check the bundled skills

[tardy steppe]

We are going to try to focus skills to the llm / model that work best for that skill. We will see what happens.

[normal skiff]

Gemini works great for most stuff. Free tier can hit rate limits if you're doing heavy back-to-back requests but for normal use it's fine.
Only real annoyance is it sometimes refuses certain code tasks due to safety filters - Claude is better for complex stuff but costs money.
Google Antigravity is solid if you have access to it. What kind of tasks are you running? If you're not hammering it constantly, Gemini should be fine.

[raven bramble]

I know how to edit with nano from the VM’s CLI but what struggle sometimes is to get the correct schema and key names of the json files

[tardy steppe]

We - literally - just started. It has been live for less than 18 total hours and that includes install, config, and security. We are infants at the moment.

[normal skiff]

Yeah, the sqlite is probably caching the old config. Try clearing it:
sqlite3 ~/.openclaw/openclaw.sqlite "DELETE FROM meta WHERE key LIKE '%memo

Or if you want to nuke the whole meta table:
sqlite3 ~/.openclaw/openclaw.sqlite "DELETE FROM meta;"

Then restart openclaw. If that still doesn't work, you can rename/delete the sqlite file entirely and let it recreate:
mv ~/.openclaw/openclaw.sqlite ~/.openclaw/openclaw.sqlite.old

[hard pagoda]

@sinful quarry Is that the coding-agent skill?

[sinful quarry]

not sure, I am not at my computer rn

but I think so?

[hard pagoda]

That's awesome. Thanks so much.

[sharp swift]

This might work:

`bash

1) Pull the current config (real keys)

clawdbot gateway config.get

2) Pull the schema (what keys/structures are valid)

clawdbot gateway config.schema

3) (Optional) save schema to a file so you can search it

clawdbot gateway config.schema > schema.json
jq '.properties | keys' schema.json

If `clawdbot` isn’t available for him, tell him to paste the JSON he’s editing and I’ll point out the correct keys.

[raven bramble]

mmmm thanks I'll try that. 🙂

[regal bridge]

facing this same issue. were you able to resolve it?

[pliant bane]

Anyone pls 🙏

[prime herald]

Hey boys when I try to install the Openclaw browser extension into terminal my bot just responds to me as if I’m chating that not running that command. How do I fix this?

[granite falcon]

hello i enabled the session memory hook but nothing is being written to the .openclaw/memory folder when I use /new, is there something else I have to enable

[sharp swift]

You are communicating to your bot in the terminal?

[prime herald]

Well normally on the web ui but I tried to paste the extension install into the Mac terminal where I originally downloaded it and it didn’t work

[proper oyster]

Im currently running clawdbot on a wsl after initially installing directly on windows. For some reason any slash commands that I send on the webchat is completely ignored. Does anybody have a clue on what could be causing that? For instance the /new command never resets my conversations not even the button to a new conversation works as well

[steep rampart]

Hi I finally got something working using kimi k2.5 cloud through ollama any got any experience through there and how is the security

[prime herald]

Hey boys when I try to install the Openclaw browser extension into terminal my bot just responds to me as if I’m chating that not running that command. How do I fix this?

[tranquil bronze]

Hello. I am stuck lol kimi k 2.5 on aws doesn’t run. Keeps going idle?

[prime herald]

Yeah that happened to me when you get to the download skills just skip them for now

That will fix it

[tranquil bronze]

Ok let me try.

[prime herald]

Hey boys when I try to install the Openclaw browser extension into terminal my bot just responds to me as if I’m chating that not running that command. How do I fix this?

[tranquil bronze]

It helped it not shut down as quick but still shutdown

Went idle

[prime herald]

Are you on a 0 ip or your own ip or public ip

[devout elk]

just heard about the whole openclaw project. i have a raspberry pi 5 im looking to run it on and then have it automate tasks on my windows computer. is there any setup info out there

[sly sluice]

Has anyone else, fought the fight and gotten it installed on their phone? I'm using Samsung a15a, used termux, downloaded the zip, and actually got it running lol (did have to do s fes other things 😅) but my hang up is lyra (my agent) is based off Gemini, and I tried using openroute but no matter how hard i tried (even directly writing into json) can't get it to accept anything BUT Google api. And the 20 per day requests is killing me already. Just curious if anyone else has ran into this or has any advice

[normal skiff]

Welcome! For the OAuth site, if there's an MCP server for it you can use that - handles tokens nicely.
For the password site, yeah that's trickier. Best approach is probably env vars (export SITE_A_PASS="...") so it's not sitting in config files. And definitely use a throwaway service account if you can, not your main creds.
One thing to keep in mind - anything you paste in the chat goes to the LLM provider, so never drop passwords directly in prompts.
Tbh there's no perfect solution for password auth with AI agents. If that site ever adds API keys or OAuth, jump on it.

[normal skiff]

Sounds like a gateway conflict between your old Windows install and WSL. The web dashboard might still be talking to the Windows instance (or a dead one).
Try killing everything and restarting clean from WSL:

pkill -f node
pkill -f openclaw

Then start fresh:

openclaw daemon start
openclaw gateway url

Open that new URL - should have a fresh token.
If that doesn't help, check if the Windows install left anything running in the background (Task Manager > node.exe processes).

[proper oyster]

Thanks for the response. Just to share more information, I`ve already uninstall it on the old windows installation and I have also turned off and on my pc a couple times so far.

[latent sail]

whats the best resource for hookup to Claude max

[tranquil bronze]

Stuck on gateway connected | idle. Any ideas?

[covert sedge]

How exactly are the cron jobs supposed to work? When my bot adds any, they are disabled because the schema seems to say that the option enabled defaults to true if not set, so it doesn't set it. Manually created isolated cron jobs do work but main session ones don't, even if manually created. Either they always say "skipped" and re-run every 2ms (with wakeMode="now") or "ok" and nothing happens (with wakeMode="next-heartbeat"). I have heartbeats disabled, is that needed for the cron feature?

[tranquil bronze]

I’m on 0 IP

[round ginkgo]

Any6oine else encounter issues getting openclaw to use nano banana?

[fringe brook]

Hey guys! Tried to run with opus but it kept going idle. I was trying to scrape some website with a cron job.

Does anyone know why? Might be the model? Thanks in advance

[thin carbon]

anyone down to use my skill to test? i would pay money to get feedback aswell

[round ginkgo]

Nak more context like you asked it to do somethign and it just went idel after tellingyou i was starting?

[thin carbon]

i need testers haha

[paper agate]

Guys is your OpenClaw bot proactive with opus 4.5? It is so much worse now than 2 weeks ago. It is absolutely 0 proactive and even forgets tasks it scheduled for itself for the next day

With all the PRs ClawdBot got so much worse

[fringe brook]

Yeah, it said task done then went idle and the browser was still open indefinitly

[paper agate]

Which skill?

[fringe brook]

and evey time i sent a message just blank responses

like this, and i do have credits

[paper agate]

Try via tui and check current state

[brazen gazelle]

Hi does anyone know why I've been getting no output regardless of model? This morning Gemini stopped working only get no output and now Kimi both worked fine for days

[paper agate]

Usually gateway not running correctly

[thin carbon]

dmed you all infos

[nimble valley]

https://www.youtube.com/watch?v=fooZAZsBmpg

do you guys this all this is necessary? he's setting up a VM on his brand new mac mini using UTM.

[wheat lichen]

Hey all, has anyone had any luck getting anything usable from a model that will fit in 128GB (Corsair AI Strix Halo)? I've got gpt-oss-120b 128k context talking to OpenClaw but seems to have compatibility issues. From the docs I see Kimi 2.1 is recommended but that's above my memory paygrade. I have a Claude Max sub but I want to keep some activities sandboxed and running completely local.

I'm OK with slower speed on those tasks. These might be things I run overnight.

[raven mirage]

Can someone help me please, I set it up on my windows computer and this is my second time. I linked my WhatsApp so not sure if that’s the problem but I got to the Clawd interface and when I text someone it doesn’t respond. Not sure why, I restarted the process and got to this same error. Does it have to do with the WhatsApp? Do I need to text it through whatpp?

[vast ibex]

Issue: OpenClaw 2026.2.1 + local Ollama llama3:8b → "No API key found for provider 'ollama'" in WhatsApp lane

Hi! I'm trying to run OpenClaw 2026.2.1 with local Ollama llama3:8b on Windows, wired to WhatsApp. Stuck on provider auth error.

Environment:

• OS: Windows (PowerShell)
• OpenClaw: 2026.2.1 (ed4529e)
• Ollama: running on http://127.0.0.1:11434
• Model: llama3:8b (confirmed via ollama list)
• WhatsApp: configured/linked

What's working:

• curl http://localhost:11434/api/tags → 200 OK with models
• Gateway startup shows: [gateway] agent model: ollama/llama3:8b
• WhatsApp inbound messages logged correctly

The problem:
WhatsApp self-chat →
[diagnostic] lane task error: lane=main ... error="Error: No API key found for provider "ollama". Auth store: C:\Users\linte.openclaw\agents\main\agent\auth-profiles.json"

Tried (docs + AnswerOverflow):

1. Onboarding → "Skip cloud providers" → Filter ollama → Default ollama/llama3:8b
2. Manual openclaw.json with models.providers.ollama block + bumped contextWindow to 32768
3. Multiple auth-profiles.json formats in C:Userslinte.openclawagentsmainagent:
   • { "providers": { "ollama": { "baseUrl": "...", "apiKey": "ollama-local" } } }
   • { "profiles": { "ollama:default": { ... } }, "lastGood": { "ollama": "ollama:default" } }
4. CLI: openclaw agents add main --provider ollama → "unknown option '--provider'"

Asking:
What's the exact auth-profiles.json schema for OpenClaw 2026.2.1 (ed4529e) that makes "No API key found for provider 'ollama'" go away?

Or the CLI sequence to configure Ollama auth for the main agent in this build?

Current files available if needed. Thanks!

[stoic wing]

Good evening, I'm using OpenClaw on a VPS running Linux. I want to configure it to access external websites. I asked it about this, and it's giving me nonsensical instructions. Does anyone know how to help me?

[willow glen]

Any recommended workarounds for this bug? LLM request rejected: messages.3: tool_use ids were found without tool_result blocks immediately after: _functions_write_0-b2c5b87c-f098-442e-9aed-83825d99f48e. Each tool_use block must have a corresponding tool_result block in the next message.

[stoic wing]

Tentei instalar em um Pc do windows 11 e não deixo ficou dando erro direto de npn

[wheat lichen]

@vast ibex , when I was seeing a similar issue (different local server tho) I also thought it was the auth-profiles.json but the issue was resolved by adding a dummy key into openclaw.json. There's an example of this here: https://docs.openclaw.ai/providers/ollama

[little cradle]

I set up a Brave API to so this, successfully

[willow glen]

Has anyone seen this before?

i just started a new session with /new

that fixed it for anyone searching in the future

[maiden mantle]

wtf is this, someone can help me pls

[willow glen]

something is wrong with you connection/auth key im sure

[quick hedge]

"Message ordering conflict - please try again. If this persists, use /new to start a fresh session." - keep getting this error in telegram and the dashboard agent just shows a blank message (so its not just a telegram issue) . today i switched models to gemini and it worked for a bit and gave me the same error again. it seems its switching back to kimi2.5 that causing this. anyone know why

[upper forge]

Configuring for Ollama hosted externally
Both Ollama and Openclaw are running as seperate VMs with within a proxmox node. Could anyone guide on how to configure Openclaw to use the server IP of Ollama?

[raven mirage]

Anyone know why it’s not communicating correctly? It’s just blank

[mental quarry]

hey y'all ! hope you have a pleasant day/night !

I was using openclaw and I figured out there is this : do you know how to setup a new one ?

[umbral cobalt]

Anyone trying to get Deepseek working with OpenClaw I have a working onfig for it posted here: https://discord.com/channels/1456350064065904867/1468027054271369256

It wont let me mark it as a Solution, if somebody could mark 2 messages there as a Solution 🙂

[radiant fern]

This is probably an incredibly stupid question, but on the very first page of OpenClaw, when I copy and paste the command it gives into admin powershell it doesnt work.

Im not sure if Im just fundementally not understanding?

[obsidian harbor]

i am completely running openclaw, and it works perfectly, i need someone who has actually pushed its limits and its making it work for them and i can DM and ask basic questions i know the answers to but need reasurance

basically and expert using this or someone who has actually made things with it that work

[harsh moss]

oh god I got it working without anthropic.. whoever showed how to turn off role:developer... genius running on an A40 w/Qwen3 that took days man

[radiant fern]

Well runing the recommended command doesnt work(From the main page) running the command from the OpenClaw docs also errors out

[stoic wing]

I'm using OpenClaw on a HostBigger VPS running Linux. I'm having trouble getting OpenClaw to access Brase to visit websites.

Can someone help me?

[harsh moss]

how do i turn off streaming? config?

[hollow quail]

Anyone know how to adjust the messages in chat? When I had it start creating logs, it for some reason adjusted what is displayed in chat. Now I get a lot of system messages displayed. I have asked it to adjust this several times and it says it does, but nothing works.

[rocky grail]

How can I have multiple chats going at the same? I created a new tab but it's just the same conversation duplicated

[delicate sundial]

I found the error, I know why it did not install, it did not have permissions to make directores as needed for the package

the question I have, where do I check what user permissions this gui has when I hit the install button ? how do I check that

[languid tinsel]

Hi is there a tutorial for running openclaw locally via ollama?

[umbral cobalt]

I have 1 questions If I reduce:
"contextWindow": 128000,

And OR
"maxTokens": 4096

Will this directly reduce the amount of data thats its sending to the AI model?

[neat quiver]

any one can help cant attach google chrome extensions

[terse lotus]

contextwindow is the ai memory and maxtokens is the max allowed generation.... usually you want this kinda high, or tell the ai to be below the limit... because if its above the output gets cut off and then code or json's will be invalid.

[terse lotus]

8k or usually fine, or 16k but some models recommend 32k
generally you want to follow the recommendations usually

[winged ledge]

Hi, anyone has an idea how to strictly enforce agent to only respond in english? I have 240+ documents in mixed languages, and it always somehow responds to my question in whatever language i used.

It already has a very strict soul and identity, but somehow it's bypassed.

[north furnace]

so has anyone successfully got this working right with a local ai using their videocard? are all the smaller models too dumb to follow the instructions? any hints on what might work? i was trying with my 5080 then went to llama.cpp using the 5080 and a 3080 cluster

[barren moth]

Hey guys, how do you delegate the coding task to the bot?
What's the best coding tool.for them to use? My bot run into corrupted files and said he can't fix json arrays and bugged to \n

Any suggestions?

[tribal wren]

When installing OpenClaw for the first time, is it a good idea to create separate accounts for the bot? Like Apple Account, Telegram, X and others?

[barren moth]

Yes if something goes wrong you can quickly isolate it, its not suggested at the moment to give it access to your real data as you can get hacked easily

[tacit mist]

My telegram is acting up right now, anyone else is having issues getting a response back from you OpenClaw?

I’ve ran all sorts of command line to test
receiving
sending
token
… I checked all the logs it’s not showing a response

I can see the bot is thinking/writing but it’s not sending the message

The oddest part is it will send me a message when a new email comes in but as far as interacting with it… nothing no response

[crystal briar]

How do I make my openclaw use the terminal on the VPS to run commands without me running commands on the VPS shell manually?

[timid violet]

hi, somebody know how to slove this error:
TypeError: (0 , _pluginSdk.listFeishuAccountIds) is not a function

[tribal wren]

Thanks for clarifying it.

[visual anvil]

I had a weird error earlier where my bot was trying to send messages to 'Default' instead of my configured Telegram channel. Might ask your bot to output the exact parameters it's trying to use when working with Telegram? I'm a noob, but Telegram has been working for me at least. If all else fails, delete the Telegram integration and set it up again

[tacit mist]

its really odd cause i get messages when my gmail receives a message. its asking me questions about it but I can't write to it where it gives me an aswer back. I can see its thinking/typing something but nothing more

been using a bunch of cmd.

[thin flint]

does claude subscription work with openclaw or no

[molten canyon]

◇ Install Google Chat plugin?
│ Download from npm (@openclaw/googlechat)
Downloading @openclaw/googlechat…
Error: spawn EINVAL

Any idea why I can't install this npm package>'

[stoic ruin]

is it bugging and exiting the agentic loop for anyone else after this latest update? I'll tell it to do something and it will respond immediately with 'ok on it!" then just not do anything because sending me that message kicked it out of the agentic loop lol

[narrow bough]

type /new and see if that helps

[willow glen]

Appreciate the response. That is what I needed up doing and it worked

[wary yarrow]

What is everyone using for Google workspace integration. I've been playing with gog cli and I have had to reauth 3 times today.

[raven condor]

Hey guys does anyone know why in my Web UI agent chat the TOOLS commands keep popping up and flooding the hat, wasnt happening before

[upper magnet]

guys, how can I setup the google antigravity app for coding with the openclaw bot so he gets to control that app?

[timid rapids]

could anyone help me with this

[molten canyon]

I'd recommend using WSL2 and installing Ubuntu, then running the single install script

[timid rapids]

bruh i cant do that the moment any other way?

thanks tho

[molten canyon]

What version of node are you running?

[timid rapids]

v24.13.0

[obsidian harbor]

if i tighten permissions on openclaw to 700 can it install code into my local? what if i dont?

[prisma grotto]

" Copilot token exchange failed: HTTP 404"

Anyone know how to fix this copilot error?

[wet lotus]

16gb would be good if you doing basic stuff and less coding or aren’t running models locally. But if you are dealing with nedia manipulation or running models locally would recommend 32GB.

[prisma grotto]

@past vault You ever fix your copilot error?

[humble swan]

new to openclaw and still in in the planning stages. I want to use openclaw with multi agent setup but not sure how to go about it. My idea is to have an agent for my family and separate one for me. Then have sub agents that my personal and the family of can orchestrate tasks to the sib agents. How should the sub agents be setup? Very specific to area. For example have a sub agents for development and a separate for testing and one for research that can be used by the family orchestrating agent? Or is my thought process completely incorrect

[glad nest]

Hi all. I have been playing around wiht this stuff for a while but Im wanting some guidance on the best way to set up my bot/s according to my needs. I want it to help me run multiple communities, like on skool and discord. While also helping me with my youtube and instagram accounts. Last thing is I need it to help me with emailing local businesses to help them with their poorly optimised websites. This is for personal use only. Im on windows and I am running it through ubuntu. Could someone please dm me? Thanks

[unique kayak]

I got clawdbot still running older version, thinking to reinstall my Ubuntu fresh and load openclaw then I wanna pull my clawdbot 'instance' as such back in, do I simply backup the /clawd folder and move that over how does that work?

[dense parcel]

Hey everyone, been trying to set up Clawdbot for some days now and burnt through over $50 in Claude credits so I wanted to set up Antigravity. However, every time I send a message to Antigravity it takes more than 5 minutes to reply or doesn't reply at all. Anyone can help?

[zealous brook]

signup for Google Cloud and get 300 dollars of credits. Kimi K2 is also giving 7 days free trial plus 1st month .99 usd also use flash models

[leaden rock]

Trying to set openclaw up locally with ollama what model should i use with my hardware? 7900 xtx 24gb and cpu is ryzen 9 9950x windows with ubuntu

[whole wasp]

I got a quick question about openclaw regarding content creation since im planning on getting this for it but still need to know if it could do some things im generally curious about.

Could this AI be able to take control of your video editing software and organize your clips from a file based off scripts that you made or prompt of sequence overviews of your video? (basically your own AI video editor)

[agile flare]

hello

im a little stuck

[thin flint]

does claude subscription work with openclaw or no

[agile flare]

any suggestion

[round ginkgo]

Tried reinstalling node.js yet?

[green nest]

It's technically possible but someone would have to program a tool for the LLM to interface with

Have you tried command line editing? Openclaw can do that

Not sure if open claw has the ability to know what's actually in the video or it just goes off file names

[whole wasp]

No I havent yet, Ive only looked into the AI from a outsiders standpoint meaning I havent worked with the AI or tested it in any way yet but if Openclaw can edit clips im sure i can figure out how to connect it so it could do what I mainly want it to do. I just wanted to make sure if it was possible

[whole wasp]

Is the AI able to look into video files and clips?

[little blade]

i guess yeah

[green nest]

I know Gemini can but Idk if open claw can send that to gemini

[little blade]

i did a npm install - it worked.. but can't do openclaw onboard thingy

[whole wasp]

are we able to test that?

[green nest]

Laugh at this windows user

[little blade]

it's okay try to help him

[green nest]

Your windows is fucking up and the backslashes and front slashes are different

try using Linux?

[little blade]

good hex

[languid tinsel]

I have the same issue with WhatsApp. It’s my setup trying to use ollama though. It thinks but no text is out put.

[little blade]

hex i have some issues with running the npm package

how did you get it working?

[green nest]

idk just install it? It's literally one command

idk what to tell you

[little blade]

it got installed .. but the terminal says command not found

when i do openclaw onboard

[green nest]

Then it didn't add it to path

[little blade]

now it worked

which api key are you using?

[round ginkgo]

Anyone else having an issue with the bot trying to run skills from diretories they arent installed to?

[languid tinsel]

Antigravity was amazing for me. Super fast & really great model with personality. I used the lowest flash model Google had & it was super quick. Don’t use thinking models with it.

[little blade]

how much is the billing?

[full zenith]

Folks, I am using Oauth for anthropic , the issue is, I have to manually update the token every eight hours . Is this expected if we use OAuth in openclaw ?

[alpine arrow]

Question im windows user but can you run lm from a mac mini m4?

[broken bay]

how to fix npm error 3221225501

[little blade]

gpt it

what's the mini max model?

[shrewd rover]

I need a little help. https://discord.com/channels/1456350064065904867/1468480116144341045

Would you mind running the time openclaw --help to see how long does it take on your machines?

[river barn]

anyone have any idea how to get perplexity ai working for websearchs?

[jolly aspen]

Im trying to install the Openclaw on a Windows365 Enterprise Cloud PC. and I keep getting this error: Virtual Machine Platform

[high plank]

On Windows use WSL2 - https://docs.openclaw.ai/platforms/windows

[jolly aspen]

While Trying to enable WSL2

[pine python]

👋 I have clawdbot running in an VM on a software project. I want it to run the compiler and see the error messages but it tells me something about the gateway config prevents. It suggests edits to a file that doesn’t exist

[high plank]

wsl.exe --list --online
Need to select a distro. Use the above command to list them and then pick one. Insert the name into the next command.

[jolly aspen]

Thats exactly what I have been trying for the last 2 hrs

[high plank]

What hardware are you using? Looks like you may need check compatibility of hardware for virtualization in BIOS. Have you already checked BIOS?

[little blade]

i need help with this

[jolly aspen]

Windows 365 Cloud PC. A virtual machine that I subscribed on Azure.

Bios is not accessible.

[high plank]

Which version of Windows 365 CPC are you running?

[little blade]

botname should be unique and only one registration try per day?

[high plank]

Tier

[jolly aspen]

OS
Windows11
OS version
10.0.26200.0
Device name
CPC-sdscl-LMB2U
Cloud PC name
Windows365CloudPolicy - SDS Cloud
License
Cloud PC Enterprise 8vCPU/32GB/512G

[solar pecan]

Any one having the gateway just stop for 5-10m and just quit working and then come back?

[high plank]

Here is the compatible tier if using Windows Cloud PC 365 - Windows 365 Enterprise – ≥ 4 vCPU Cloud PCs (standard CPU)

Compatible with WSL, not certain on OpenClaw

[crimson fern]

What kind of role do you give your bot when you launch it?

[haughty halo]

help please(русские тоже пж)

[fossil sand]

guys i think i fucked up.

my openclaw gateway is unreachable and running openclaw doctor --fix or restarting the gateway isnt fixing it. I had tailscale connected so im assuming its something to do with that for sure any assistance would be really appreciated

[digital maple]

hey yall, can somebody explain to me the use of remote gateway?

is it like for running openclaw in a VM on a Windows/Linx machine, and then you use a mac to connect to it as a remote gateway to get those extra mac features?

[round glade]

I made a PR maybe this can help you - https://github.com/openclaw/openclaw/pull/8602

[cursive latch]

Hi, just loaded up with hostinger vps. Using open ai api and Telegram. How to I get this to work like everyone else? It says it can't surf the net, read or respond to emails. What did I do wrong?

[deep knoll]

So I keep reading saying “you have to use tailscale” and I’m doing initial set up right now… but what am I supposed to be setting up tailscale to do?

[shadow notch]

Hey yall. Trying to to build a guide for newbies
https://x.com/squ925/status/2018780827715502294?s=46

any of you got kimi model to work? Or really using over anthropic? Or using them together?

Openrouter I can’t get to work

[crisp wing]

@squ925 via Twitter

Squirrel (@squ925)

📰 Openclaw setup for people who want others to waste money first
And now you're trying to learn what @openclaw is while being just as confused as me.

First: people building "super workflows" saying "use this bot to make money — it's so easy" are full of sh*t. But maybe not where you think.

Is it possible? Yes.

Is it really possible for a casual user? Yes, with extra steps. Many of them.

so
STEP 1.
you need to figure out where to host this guy. Really, there are 2 options.

Cloud VS Local

If you just want to try and mess around
go on aws.amazon.com/console/ and make an account
(If you fail at this step, stop and just go use the chatgpt) then you get to here…

FxTwitter · Feb 3, 2026, 8:17 PM

[solid grove]

i don't know what happened but no matter what I do my openrouter won't work anymore. I got a new key, tried everything. I'm getting a red error saying there's no openrouter API even though it's in the auth profile. 😢

[high crest]

this is how my current snippet looks like and I get this error: 401 No cookie auth credentials found, in telegram. I think the problem is that anthropic is in the code snippet but I use openrouter. How do I have to fix the code snippet? What's wrong here?

[static sorrel]

Hi guys, I have two different APIs in my minimax settings, should I use the codeplan or the regular api?

[shadow notch]

Same here. When you check /model what does the bot aay

[solid grove]

well i can't really get it to do anything since i can't connect to a model... when i did it just now it just showed me a list of models to try to connect to, so i chose the one i was connected to previously

[digital maple]

Go in the configm make sure the default model is "openrouter/openrouter/auto" or something like that, in an older update, it was just set to "openrouter/auto" which was wrong

[digital maple]

tailscale is a VPN, youd use it to access the web consol if your setting up openclaw on a device that isn't a laptop or pc your using (so in a server or cloud)

[deep knoll]

Oh so I don’t need it if I have a dedicated device for it 😵‍💫 thank you!

[digital maple]

what i ussually do is i mess up the config file, then i run "openclaw doctor --deep" and "openclaw doctor --fix", then i run the onboarding and select the model i want, it works

[solid grove]

Is that the auth profiles json?

[digital maple]

exactly!

[digital maple]

openclaw.json

[high crest]

ok thx, I'll remember that

[solid grove]

"agents": {
"defaults": {
"model": {
"primary": "openrouter/arcee-ai/trinity-large-preview:free"

^so is that wrong?

[unborn portal]

I am trying to use ollama - llama3.1:8b local but the tui or dashboard is just thinking to write. Does anyone know if the openclar.json models is correct? "models": {
"providers": {
"ollama": {
"baseUrl": "http://localhost:11434/v1",
"apiKey": "ollama-local",
"api": "openai-responses",
"models": [
{
"id": "llama3.1:8b",
"name": "Llama 3.1 8B",
"reasoning": false,
"input": ["text"],
"cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }
}
]
}
}
},

[lost kestrel]

Can anyone help me with this problem, cron job doesnt work properly. This is the problem described by my bot:

---

🐛 OpenClaw Cron Scheduler Bug - 2026.2.1

Issue: Cron jobs are saved but never execute.

What's happening:

1. Job gets created: ✅

   {
     "id": "job-123",
     "schedule": {"kind": "at", "atMs": 1770114300000},
     "payload": {"kind": "systemEvent", "text": "Test"},
     "sessionTarget": "main"
   }
   

2. Job saved to disk: ✅
   File: ~/.openclaw/cron/jobs.json

3. But nextWakeAtMs is always null: ❌

4. Commands that don't work:

   • cron run <job-id> → returns "not-due" or "Success: false"
   • cron wake → sends event but jobs don't execute
   • Automatic timeout → never happens

What I've tried:

• ✅ Configured cron.enabled: true
• ✅ Restarted gateway (multiple times)
• ✅ Tried "at", "every" and "cron" schedule types
• ✅ Tried both "systemEvent" and "agentTurn" payloads
• ✅ Tried wakeMode: "next-heartbeat" and "now"

Config:

"cron": {
  "enabled": true,
  "store": "/home/sebiosjed/.openclaw/cron/jobs.json",
  "maxConcurrentRuns": 5
}

Expected behavior:

Scheduler should calculate nextWakeAtMs and auto-execute jobs when due.

Actual behavior:

nextWakeAtMs stays null forever, jobs never run.

---

Anyone else experienced this? Is this a known bug in 2026.2.1 or am I missing something in the config?

[digital maple]

Feed your agent this documentation link and it should be able to fix the issue https://docs.openclaw.ai/automation/cron-jobs#cron-jobs

i also had issues with crons, i gave it that then they worked

[shadow notch]

start openclaw onboard

set it all up again with normal key

and have one working model

because open router is great if it work but didn 't for me

[wet stream]

Hello guys. Sorry to disturbed you this fine cold morning.

I have my openclaw friend setup both to speak to me via discord and telegram. Seems like it is two different agents im talking to when I have the setup like that.

Seems like they share memories but not live conversations.

have you seen this issue and do you have a solution for it?

[mighty void]

in Discord, are you chatting to it in a channel, or DM? DMs share the same "main" session while group(channel) chats spawn a new session, which doesn't share the same chat history

[shadow notch]

yes, supermemory. go on the website oay some moeny

[wet stream]

in a channel and DM on Telegram

[mighty void]

in the CLI, openclaw sessions will show you how many chat sessions you've got going. agent:main:main is your main one. Discord DMs should go in there, you probably have another one with the discord guild id and channel id. Not sure about telegram but i think it should go into main as well. every session has different context, that's they are like different agents, because they are

[wet stream]

ok so its kind of a bad idea to have multiple "sources" får chat?

[mighty void]

depends on what you want to do? once i realized this, I mostly chat with my own bot via DM on discord, occasionally using the web or TUI interface for troubleshooting. the discord channel could be useful for stuff like alerts and notifications from cron jobs, without cluttering the main chat. you can still interact with it to do stuff in the channel, it just doesn't;t share the same chat history unless it is already saved in MEMORY.md

not the full chat history, just important information that was distilled

i think chat transcripts are still isolated

[lethal needle]

Is there a way to deploy OpenClaw on Railway with Copilot as provider? I seem to only get it to work with made-up API keys from OpenAI

[wet stream]

ok and default its not?

i thought it worked sort of like an email application. either i use the one in the phone, the one on my computer or the web interface. the data is still stored in the same place so to speak 😄

[mighty void]

it is in DM chats. you might not see the previous replies, because it only sends the replies on the channel (openclaw channel, e.g. discord, telegram, whatsapp etc) that you sent your message from, but it should share the same session and context. if you open up the Contol UI chat view, you can select the different sessions from the dropdown and see the full history

[silk trout]

Is there anything special about the Molty configuration with Discord that prevents me from using /reasoning command. I get consistently refused by the bot. It says "You are not authorized". I have added command permission.

[versed shuttle]

Hi

In version 2.2, the memory search feature is causing errors. Is there a way to fix this?

I am using Gemini at memory_search

[tawny rivet]

you probably have the exactly same problem as I had. in openclaw.json check channels.discord.guilds.GUILD_ID.users.USER_ID , if you don't have you need to create it

[unique kayak]

is it normal for openclaw to have config in .openclaw/.openclaw

[versed shuttle]

OpenClaw memory_search tool not working (Symptoms / Attempts / Results)
Environment
OpenClaw: 2026.2.2
Node.js: v25.5.0
OS: macOS Darwin 25.2.0 (arm64)
Gateway:
bind: 127.0.0.1 (loopback)
port: 18789
status: running, RPC probe: ok
Workspace: /Users/hong-gihyeon/clawd
Config: ~/.openclaw/openclaw.json
Goal
Enable memory_search to semantically search MEMORY.md + memory/*.md (and optionally sessions) to recall prior decisions/context.

Symptoms
Calling memory_search always fails. Tool response includes:
disabled: true
error: Cannot read properties of undefined (reading 'get')
Earlier in the process I also saw missing module errors:
Cannot find module .../dist/memory/manager.js
Cannot find module .../dist/memory/search-manager.js
(At the moment the error has shifted to the undefined.get runtime error.)

What I tried
A. Simplify memorySearch config to local-only
I forced a minimal config:

agents.defaults.memorySearch.provider = "local"
agents.defaults.memorySearch.sources = ["memory"] (removed sessions)
agents.defaults.memorySearch.experimental.sessionMemory = false
agents.defaults.memorySearch.fallback = "none"
Restarted the gateway after changes
Confirmed the config values are written in ~/.openclaw/openclaw.json.

B. Fix gateway restart failures (auth config)
During restarts, the gateway was exiting immediately with:

Gateway auth is set to token, but no token is configured.
I fixed it by setting gateway.auth.token and gateway.remote.token.
Gateway now runs normally (running / RPC ok).
C. Reinstall/change OpenClaw version
Reinstalled global openclaw from 2026.2.2-3 to 2026.2.2
memory_search still fails
Results
Gateway and overall runtime are healthy (chat works, other tools work)
memory_search remains broken even with minimal local-only config:
disabled: true
Cannot read properties of undefined (reading 'get')
Notes
openclaw gateway status prints a persistent warning:
Config was last written by a newer OpenClaw (2026.2.2); current version is 0.0.0.
(Version string looks wrong; openclaw --version correctly shows 2026.2.2)
Occasionally saw restart races/locks:
Gateway failed to start: gateway already running ... Port 18789 is already in use
But gateway is currently running fine.
What I’m asking
Help identifying why the memory_search tool becomes disabled: true and throws undefined.get:

Node v25.x compatibility issue?
Bundling/entrypoint issue for memory modules?
memory-core plugin / index store initialization failure?
Recommended debug steps or known fixes/patches.

[silver plinth]

?Local model from llama.cpp on another server?
I'm running openclaw on a different box everything seems to work fine except that the bot can't seem to find the AI model and the log gives the following error:

Embedded agent failed before reply: Unknown model: unsloth/gpt-oss-20b-GGUF:Q4_K_M

It is not a connectivity issue, as I can curl the model from the bot. The config seems correct, and I'm using the exact model name/description as it is entered/displayed in llama.cpp

Any idea where to start troubleshooting?

[drowsy crypt]

Is it not possible to switch between multiple agents on the WebChat page?🧐

[tired zenith]

Help. Claude won’t integrate with openclaw. I keep getting this error: HTTP 404: not_found_error: Not Found (request_id:…)

[thin patio]

I've got some crazy token usage / context bloat.

I've sorta solved part of it by massibly limiting chat history, but I'd love a smarter way of context management if anyone has one.

then seperate but related. Images I send get stuck in the context, when Ideally they'd be analyzed once and then summarized in text. but I dont see any settings or skills for this. Any ideas?

[buoyant rune]

does anyone know how to solve the antigravity oauth hanging issue?

[gusty vessel]

How do I manually wipe todays context window? My bot went nuts hallucinating repeating phrases for 32 000 tokens each heartbeat.

or how do I set a hard deterministic token output limit so I dont burn 5$ per hour?

[terse cove]

./new

Also, try Kimi 2.5

Its very cheap and allows API calls

[plain pivot]

I kept on getting errors trying to add the api. would not authenticate. what base url did u use?

[leaden plover]

Can someone help me debug this error: This version of Antigravity is no longer supported. Please update to receive the latest features!

[heavy rapids]

I created some skills manually by creating a directory and adding a SKILL.md file inside it. When I create a CRON job and ask it to make use of that skill, it states that it diesn't have access to any such skill. However, the main session is able to make use of the skill. Please help me out. Stuck since very long; tried almost everything

[terse cove]

WDYM?

[round nimbus]

just call /reset - it doesn't wipe "today's" context window though. the context doesn't reset automatically from one day to the next. it's the same as typing /new or /clear from a coding CLI

it is, but that doesn't mean there is always someone here willing to not just ask questions but also respond reading all the messages

[gusty vessel]

It does reset at 5 PM. I see it consistently.

So I want to reset manually

[gusty vessel]

Sorry some bot just messaged me to not write here. But I see it had deleted its message now...

[mortal pine]

Hi all, I have successfully installed the application on Windows. I’m looking to run it as a background service that starts on system boot without opening any visible console windows. I haven't found any documentation on how to achieve this. Any advice?

[sly crown]

Hi all could you please help me connect ollama with clawdbot because i dont want to purchase the tokens I run the LLM locally on my machine by using olllama?

[round nimbus]

ah, looks like you're right. but yeah you can just send /reset message to it at any time to wipe context

[gusty vessel]

Cool thanks!

[round nimbus]

thank you as well for tipping me off to the daily reset! think imma disable that. i don't see a good reason to auto reset the context. i like managing my own context

though i can see why having it is set by default...many users may not be used to or desire to manage the context

[plain pivot]

It could not use it at all. It just stopped my openclaw from working. Had to do the onboarding again to get a fresh start with codex as the default

[latent hedge]

Damn! Sitting on a plane and asked my bot to update itself from v. 2.1 to 2.2… didn’t come back up again 🙁

[cold horizon]

you might need to restart the gateway manually

[atomic wolf]

Hi I need help. since I dont have a display output nor ikvm I used qemu + vnc on the rescue system of my dedicated server which acts like a usb to build and write everything to the disk. (It uses OpenCore Hackintosh) Now I want to know if its possible to setup ssh and remote desktop so if I reboot and the rescue system is gone I can access the MacOs?

[cold horizon]

anybody seeing this erros

LLM request rejected: messages.432.content.1: unexpected tool_use_id found in tool_result blocks: toolu_019ujfsijfwGwJm65hq3HRDo. Each tool_result block must have a corresponding tool_use block in the previous message.

[vestal coral]

What's the canonical way to add a model? The bot makes a mess of it. Currently hand editing three different files: .env, auth-profiles.json and models.json.

Is there a script to just enter baseURL, provider and API key and it does the rest?

[terse cove]

openclaw config

on terminal

[vestal coral]

this only has the ones in the list tho. I use some fringe ones for privacy. Like baseten and requesty.... guess there it's manual

[terse cove]

Ahhh, then my guess is look into the Documentation

[vestal coral]

Didn't find anything else there... Will write a script

[slender obsidian]

I have had a good experience with telling Opus 4.5 something like, Stick to the Openclaw framework and guidance, make sure to follow documentation, carefully consider all files that require an edit. Please add in blah blah blah.

been running for a few weeks now and adding models, doing updates, checking with openclaw doctor and manaually looking at configs. Works well.

[vestal coral]

tbh this is what I did the first 3 times as well, hehe.. by now I kinda know what to do and do it by hand

[slender obsidian]

Indeed, it's a bit addictive for some reason. I tend to follow along and watch what it does. Now I can edit myself without too much trouble. Most of the issues I end up with are formatting because I'm not a coder and I probably need a better editor for json. My vision isn't the best.

[vestal coral]

Yeah JSON is somewhat painful to edit without an IDE. Nano certainly not ideal, but it does show errors.

[slender obsidian]

I'm using default nano, I haven't bothered to mess with it though. I'll see if I can turn on a bit of extra annoying highlighting or something.

[vestal coral]

for me it colors everything green if I miss a {

[slender obsidian]

I'll dig in, I bet my terminal colors or something are off. I can sorta see it...I think. lol

[round nimbus]

my agent is sandboxed so i just use claude code to modify all that. i spend enough time as it is tinkering with it, need to make my tinkering as efficient as possible 😆

[slender obsidian]

I think that is an excellent idea.

[round nimbus]

i switched from mac to linux a few months ago and have been using claude code for all my sys admin so using it to do all my openclaw tweaking was natural

[slender obsidian]

I'm on linux for everything. I setup Opencode with my subscription as well as having some local models running. I made specific agents for system administrator functions etc. I'm just amazed at what these things can do now.

Claude Code and your setup is the way to go for ease and keeping things together.

[dusky nexus]

Hi, I'm non-technical.
I want to use Moltbot as a ready-to-use AI agent (not coding).
The Moltbot Discord invite seems expired.
Where can I get access or instructions?

[glossy iris]

Guys is my installation done??

[tawny rivet]

question: how do you actually get lobster to run? i have installed it according to instructions in lobster repo, added to PATH, i can run it manually, however on plugins list it is disabled. EDIT: forget it, i just needed to enable it

[cold horizon]

anyone familier with this "LLM request rejected: messages.432.content.1: unexpected tool_use_id found in tool_result blocks: toolu_019ujfsijfwGwJm65hq3HRDo. Each tool_result block must have a corresponding tool_use block in the previous message."

error the bot just stopped midway doing task and now its responding with this error message

[chilly aurora]

hey umm i talket too it on the telegram a bit too much and de web page now is no longer responding

[tidal plover]

Everytime I setup Claude setup-token (the Claude oauth) it just doesn’t work for some reason, I do have Claude max.
But no messages work with the bot

[slender obsidian]

make sure to look in #1459642797895319552 there might be a few answers in there.

[tawny rivet]

has anyone tried this skill? https://clawhub.ai/NextFrontierBuilds/elite-longterm-memory it uses vector db, i'm wondering how this will impact performance

[slender obsidian]

you might want to check out the new experimental feature for QMD if you have not: https://docs.openclaw.ai/concepts/memory#qmd-backend-experimental

[tawny rivet]

@slender obsidian i'm waiting at least a couple of days for this experimental features to get tested more 🙂

[slender obsidian]

totally understand, I'm just always very very reluctant to use a skill on the community hub so I thought I would mention it in case you haven't seen it yet. The more offical built in thing that is.

[tawny rivet]

@slender obsidian but yeah, i definitely will, it actually mirrors an implementation that i did like 2 months ago for mcps (bm25/vectors/reranking)

[slender obsidian]

very nice 🙂

[tawny rivet]

@slender obsidian yeah, i understand

i'm trying to get spogo to run, is it just me or is it actually buggy?

this spotify_player skill is nice, but this spogo runs like shit. i had to play/pause first on my device for it to pick it up, then it was able to play, however when i told him to search a track, it got into some weird indefinite state (agent says it found the track and its playing, however irl it isnt)

maybe i just need for him to check the code

[lofty sapphire]

Hey team I am building a web agent that drives through website adn can login with creds, on a linux server I am trying to use chrome extension but it goes for localhost instead of hosted domain, can we achieve this?

[steep rapids]

When you get the error "model not allowed", what is usually the problem?

    "defaults": {
      "model": {
        "primary": "google-antigravity/claude-opus-4-5-thinking",
        "fallbacks": [
          "google-antigravity/gemini-3-pro-high",
          "google-antigravity/gemini-3-flash",
          "deepseek/deepseek-chat"```

In openclaw.json, I have this, but I cannot set it to gemini-3-pro-high manually

[coarse stratus]

So my brain is a bit fried from working on 3 projects (including having to requant a model 3 times) what do I do if the bootstrap was never triggered

[ancient ivy]

Hey folks quick one how do I add fallback for my models. I keep hitting rate limits with Gemini 3 flash and pro.

Honestly what’s the best setup to be worry free all day to not run out. Usage is rather heavy

[sly vector]

Hi guys, I'm trying to run OpenClaw with the GLM-4.7 (coding plan) model. I've set up my config (see below), but I'm hitting a "Request timed out" error every 30 seconds when using the Telegram channel.

My timeoutSeconds is already set to 600, so I'm not sure why it's timing out so quickly. Has anyone encountered this before? Any help would be appreciated!

{

  "auth": {

    "profiles": {

      "zai:default": {

        "provider": "zai",

        "mode": "api_key"

      }

    }

  },

  "agents": {

    "defaults": {

      "model": {

        "primary": "zai/glm-4.7"

      },

      "models": {

        "zai/glm-4.7": {

          "alias": "GLM"

        }

      },

      "workspace": "/home/node/.openclaw/workspace",

      "thinkingDefault": "low",

      "verboseDefault": "off",

      "blockStreamingDefault": "off",

      "blockStreamingBreak": "text_end",

      "blockStreamingChunk": {

        "minChars": 800,

        "maxChars": 1200,

        "breakPreference": "paragraph"

      },

      "blockStreamingCoalesce": {

        "idleMs": 1000

      },

      "humanDelay": {

        "mode": "natural"

      },

      "timeoutSeconds": 600,

      "typingIntervalSeconds": 5,

      "maxConcurrent": 4,

      "subagents": {

        "maxConcurrent": 8

      }

    }

  },

  "commands": {

    "native": "auto",

    "nativeSkills": "auto"

  },

  "channels": {

    "telegram": {

      "enabled": true,

      "dmPolicy": "allowlist",

      "botToken": "xxx",

      "allowFrom": [

        "tg:xxxxxx"

      ],

      "groupPolicy": "allowlist",

      "streamMode": "partial"

    }

  },

  "gateway": {

    "port": 18789,

    "mode": "local",

    "bind": "loopback",

    "auth": {

      "mode": "token",

      "token": "xxxxx"

    },

    "tailscale": {

      "mode": "off",

      "resetOnExit": false

    }

  },

  }

}

[tired zenith]

Is anyone having issues setting up Claud ?

Minmax is so much smoother to set its crazy!

[bright marlin]

Maybe there is already a guide somewhere, but can anyone point me to docs on how to migrated from the clawdbot package to the openclaw one? I had it already installed and at this point the system service, the npm package and the directory etc... All use the clawd name. So I guess there is some migration involved

[barren moth]

Hey guys, can anyone share their model setups for sub-agent delegation?

For main brain im using Glm-4.7

I need model for;

• research
• document analysis
• seo
• web browsing

[normal skiff]

The 30-second timeout isn't coming from your agent config - that 600s setting is for agent task timeouts, not network requests.

Few things to check:
Is the timeout happening on the Telegram side or the model response side? Check your logs (openclaw logs or wherever they're writing) to see where it's actually failing.
GLM/zai might have its own API timeout that's shorter. Some providers cut you off at 30s if the model is slow to respond. Not much you can do about that except try a faster model or check if zai has timeout settings.
Also could be a network thing if you're behind a proxy or cloudflare - those often have 30s defaults.

What do the actual error logs say?

[warm mirage]

Hey guys i keep running into this issue, tried to run it through chatgpt but what it told me to do still doesnt work, any ideas TT

[regal osprey]

Can anyone help me. I'm at the last stages of setting up ollama models locally. My bot responds but just says this:⚠️ Agent failed before reply: All models failed (5): ollama/qwen2.5:7b: No API key found for provider "ollama". Auth store: /home/alexe/.openclaw/agents/main/agent/auth-profiles.json (agentDir: /home/alexe/.openclaw/agents/main/agent). Configure auth for this agent (openclaw agents add <id>) or copy auth-profiles.json from the main agentDir. (auth) | ollama/phi3:mini: No API key found for provider "ollama". Auth store: /home/alexe/.openclaw/agents/main/agent/auth-profiles.json (agentDir: /home/alexe/.openclaw/agents/main/agent). Configure auth for this agent (openclaw agents add <id>) or copy auth-profiles.json from the main agentDir. (auth) | ollama/llama3.2:latest: No API key found for provider "ollama". Auth store: /home/alexe/.openclaw/agents/main/agent/auth-profiles.json (agentDir: /home/alexe/.openclaw/agents/main/agent). Configure auth for this agent (openclaw agents add <id>) or copy auth-profiles.json from the main agentDir. (auth) | ollama/mistral:latest: No API key found for provider "ollama". Auth store: /home/alexe/.openclaw/agents/main/agent/auth-profiles.json (agentDir: /home/alexe/.openclaw/agents/main/agent). Configure auth for this agent (openclaw agents add <id>) or copy auth-profiles.json from the main agentDir. (auth) | ollama/gemma2:27b: No API key found for provider "ollama". Auth store: /home/alexe/.openclaw/agents/main/agent/auth-profiles.json (agentDir: /home/alexe/.openclaw/agents/main/agent). Configure auth for this agent (openclaw agents add <id>) or copy auth-profiles.json from the main agentDir. (auth).
Logs: openclaw logs --follow

No matter what I do in the JSON to override the need for API, I just get this message

[rugged onyx]

hi guys im trying to use qwen3:8b but when i message it it gave me this error: [openclaw] Connection error. how can i fix this?

[rugged onyx]

do you have node js?

[warm mirage]

Yeah it dowloaded and detected it

i was just checking and my pc isnt that beefy so i might just not run it

[rugged onyx]

okay then enter the npm install code

npm i -g openclaw

[rugged onyx]

try this

[warm mirage]

Ima try it

still doesnt work TT

I dont have github installed on my laptop might be why

[normal skiff]

Ollama doesn't need an API key, but OpenClaw still wants an auth entry. Try registering it with a dummy token:

openclaw models auth paste-token --provider ollama
When it asks for the token, just paste anything (like "local" or "none") - Ollama ignores it anyway, but this creates the auth entry OpenClaw is looking for.

Also make sure Ollama is actually running (ollama serve) and accessible at http://localhost:11434. You can test with:

curl http://localhost:11434/api/tags
If that doesn't work, check your config has the right base URL for ollama - some setups need it explicitly set.

[regal osprey]

Thank so much I'll try that in a minute

[rugged onyx]

ummm

im out of ideas

@normal skiff can you help me too?

[warm mirage]

I dont have git so im installing that rn TT

[rugged onyx]

im trying to use qwen3:8b but when i message it it gave me this error: [openclaw] Connection error. how can i fix this?

[rugged onyx]

i don think thats the problen here

[dusk prawn]

How can I force openclaw to always use the chrome extension to browse with chrome? Its constantly using switching between managed and the extension

[delicate plank]

How to enable scripts on my system? It says cannot be loaded because running scripts is disabled on this system.

[wet cloud]

Hi volk! is it possible to setup openclaw with a Teams subscription? 🙂

i tried openclawd und a freshly ubuntu VM, but couldnt run
i m trying now on Ubuntu the openclaw, but i need an API key, but i dont wanna use the API i wanna use my subscription? thanks!

i mean Pro subscription* with an teams plan

[raven trench]

I'm just trying to work out if this is me or a bug:
got multiple discord channels but I have to explictly reference each channel to allow openclaw to reply to messages in there. I can tell openclaw to post in there and it can but need to be explict from another channel. What is going on, what config do I need to set to whitelist all channels?

[fringe brook]

i am putting my api key from open ai into openClaw but then it says that there is no api key configured...

[delicate plank]

We have the same problem now. Before it's disabled scripts.

I followed your commands and we have the same error

[brave trench]

Hello everyone 👋 I need some help with the gateway. Everything was working fine last night and this morning I cannot connect through the gateway. Ive tried everything

[warm mirage]

I installed git then cleared its cache and now its been on "Installing Claude" for the past 2 minutes or so my laptop is getting like slight lag spikes here and then so i think its working

[delicate plank]

I'll try it too.

[warm mirage]

Different error now YAY

[coral dove]

how is the recommended way to add more models?

[tacit flax]

I set up quite a few things with my bot yesterday – Brave API Search, AgentMail, etc. This morning, he seems to not have access to any of those anymore.

Saw something yesterday that the config file gets wiped at the end of every day. Is that accurate?

[marsh vessel]

I feel like I'm close but I can't get a response

[coral dove]

noisey bug. try the webui and look if it responds there

[marsh vessel]

ok, atleast I feel like I'm getting to the next step, thanks

[regal osprey]

You sir are a genius!! Appreciate your help 😃

[coral dove]

look at the openclaw.json and search for the token under gateway, then copy this and put it under "overview" on the web ui then press connect

what does that mean?

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

my openclaw runs only local

[tired rain]

Guys how can i set groq api key in openclaw without having to do full openclaw config?

[delicate matrix]

What chat interface is best do you guys think? Telegram? Discord? What did you guys go with and why?

[sharp shadow]

Why does open claw logo broken ? And also when chating streaming gets stuck and I have to reload to see response progress

[rugged onyx]

press ctrl + c and get in the tui again

[hollow wedge]

Can anyone help me

[craggy matrix]

are you trying to install onto windows?

[warm mirage]

Yeah windows

[craggy matrix]

i dont think theres native support for windows yet. you should install it to linux via WSL

[warm mirage]

TT

Thank u

[craggy matrix]

windows had a problem with the file types for node.js was the same issue i was having, WSL is very easy to install, try that route

[craggy matrix]

you probably have a similar issue, try WSL

[raven trench]

ah found it https://github.com/openclaw/openclaw/issues/8009

[novel spruce]

Hi. my openclaw cannot send me files through my discord bot. any idea whats configured wrong? in the errors logs i see:
error [tools] message failed: content required

[delicate plank]

I was about to say that I have different error now.

I'll install linux first. Thank you.

[round ginkgo]

Is anyone else running into the issue where your Clawbot is trying to open skills from folders they wernt installed too?

[craggy matrix]

that looks like a different issue, my first time setting up took 6 hours and i was asking claude coed and chatgpt a lot of questions. today i just transferred him from my WSL to his own linux machine

[delicate plank]

But I'm installing it in windows too. Any recomendation on what linux should I install?

[craggy matrix]

pretty sure if you just install WSL it comes with an ubutu version thats just a cli window, easy enough to type commands into

[latent tundra]

Hi everyone, i keep trying and failing to set it up using a locally installed Ollama, anyone has a -simple- guide for it?

[delicate plank]

Will try.

[round ginkgo]

Kevin what model are youtrying to run it off of?

[wet cloud]

Hi i m confused with antropic i pasted 3x times the key into the setup:

which others models are you reccomend? -i have the claude code / openai + gemini plan Paid! 🙂

[round ginkgo]

This isnt a key in the setup process ther e is a field you can leave blan to generate a token or you can gen one your self

What platform are younon?

[delicate plank]

I can't install WSL too. I'll try to install ubuntu then install it again.

[calm girder]

I am trying to setup sms with twilio and cloudflare. Lots of issues. Anyone have a guide up on this?

[neon jolt]

Hi all, I’m trying to run OpenClaw inside a Docker container on macOS using OrbStack. Has anyone gotten this working?

[craggy matrix]

you just do wsl -install in powershell as admin

[delicate plank]

It won't install.

[wet cloud]

I did claude setup token logged in copy out my key

starting with sk-ant-oat01-spk......
copy it into this: screenshot
run health keys invalid
but i have a company pro plan i think i need an individual or?

[round ginkgo]

BRUH!

Delete that pick

You need to cycle that token too ask claude how to do it

[wet cloud]

the full api key is not visible only the first few digits 🙂

[dense granite]

Currently running openclaw with local llm ollama/qwen3-coder:30b however not sure how to get the agent to learn skills. How should I go about that?

[round ginkgo]

ok lol

makes me twitchy

[wet cloud]

but i will regen it of course as well, but thanks for your feedback
any idea how i can solve my problem as well? thanks!

[round ginkgo]

where do you get the error?

[wet cloud]

i mean:

[sharp swift]

Agent Mesh soft launch: users helping users (via bots) — we need your help

Need help on a task? Post it and recruit a few agents to tackle parts in parallel. Work stays private between you + your bot — we only record simple outcome feedback (short notes/tags) so good helpers surface over time.

We need a few people to try one quick collab and tell us what breaks. Reply “in” / DM me for the invite + 60‑second setup.

[wet cloud]

yeha please! 🙂

would be very nice

[dense granite]

Can't seem to get my agent to remember information.

[wet cloud]

after a restart of my ubuntu it seems to "work" closely:

[celest zenith]

Give your bot an email address or not?