users-helping-users

[unique quarry]

yeah discord and terminal work just fine, is this a known bug/issue?

[sullen ivy]

is any body able to tell me if claude oauth will get you banned

[junior acorn]

it most likely wont, but who knows

at anytime anthropic could close ur account if they wanted, it's just another risk vs reward scenario that you have to be willing to adjust to your own personal life

For me, i just run on an alt account for claude. IF they ban me, then oh well

[ancient jungle]

"I accidentally deleted the memory database while troubleshooting. Let me recreate it:" - my bot 🤣

[junior acorn]

yeah tell the LLM to use it's internal storage that it definitely has lmfao :D

[indigo goblet]

@next walrus whats up buddy lol, all good?

[copper nova]

"your migration to orbstack is complete" "oh shit I didn't make a back up of the old DB before I migrated, it's on 1 week old backup, my bad"

very fun stuff

[next walrus]

So far so good!

[indigo goblet]

Great man haha

Hows your dashboard issue

[ancient jungle]

this thing was doing a whole bunch of stuff and i had to stop it and link it the docs page -> read this before you screw more shit up

[broken yarrow]

anyone know of a linkedin tool for cold messages

[indigo goblet]

Any issue not fixed?

[ancient jungle]

well, i was trying to migrate the default memory system over to memclaw, we (the bot and i) just don't know what we configured wrong

[formal trail]

guys i have openclaw downloaded but i cliked out of it how do i run it again

[queen quail]

in terminal run openclaw dashboard

[formal trail]

thanks man

it doesnt work

i dont think i installed it correct.. i need to configure stuff dyk how i do that

[queen quail]

openclaw configure

or openclaw onboard

[formal trail]

yo

if i choose the remote one what option
token
paswword
no auth

what ones?

[lament valve]

hello! im very new to OpenClaw after starting out with NemoClaw a few days ago. My goal is to get a very basic demo of openclaw using my local, self-hosted Firecrawl instance to do a simple web search. For example: "Use Firecrawl to scrape https://example.com and summarize the main content." in the openclaw tui. I think my last challenge is on the OpenClaw side and not NemoClaw so am hoping this is a trivial question for y'all.

I successfully installed the firecrawl-search skill which seems to be pretty straight forward with just a couple markdown and python scripts. The API URL is hard coded in those python scripts to point to the Firecrawl cloud, so I modified the scripts to load the API base URL from an environment variable FIRECRAWL_API_BASE. But even after setting these environment variables from within the OpenClaw sandbox it seems like the script will not recognize the environment variables.

What am i missing? do i need to reload the skill somehow to get it to use my now modified skill scripts or something? Is there a more proper way to accomplish what im after?

[formal trail]

it wont let me use remote acces

it says this page cannot be accested

[queen quail]

set up local

[formal trail]

why?

whats the difference

[queen quail]

if it is installed locally on the machine that is the correct option. Remote is if you are trying to connect to another device that is running the instance

[formal trail]

ah okok

[indigo goblet]

Hm, your current memory is openclaw vanilla?

[formal trail]

but i see poeple with like cool looking apps on their

[vernal gate]

smart move honestly — starting clean on a stable version is way better than fighting a broken
install for days 😅

when you're ready to set up fresh, feel free to ping me — mac mini intel is actually one of the
smoother setups once the version is stable 🤝

[formal trail]

yo it keep opening to remote how do i chnage back

[indigo goblet]

Because memclaw looks like a lot of tech fluff to be honest, not a lot of substance, what is it you are trying to do exactly?

[formal trail]

pls help

[indigo goblet]

You might be going too fast without understanding how openclaw works.

[formal trail]

im clueless

[indigo goblet]

Whats your OS and what is your goal at a very high level, basic.

[formal trail]

im windows and i just want to make like a app like evyone else

[wicked girder]

Hi everyone, I am new here so please let me know if this is not the correct way to ask for help. My OpenClaw bot is skipping all guild channel messages with reason: no-mention, even when groupPolicy is set to "open" in openclaw.json. I've confirmed Message Content Intent is enabled in the Developer Portal. DMs work perfectly. Running OpenClaw 2026.3.13 on macOS. Tried asking my OpenClaw bot (through DMs) and Claude what could be causing the issue and both are stumped. What could be causing this persistent issue?

[indigo goblet]

Hm, search online and read the docs on how to operate openclaw on windows first.

[scarlet silo]

hi guys, i am having trouble with openai again. i am tyred of trying to re-auth but always get:

[openai-codex] Token refresh failed: 401 {
"error": {
"message": "Your refresh token has already been used to generate a new access token. Please try signing in again.",
"type": "invalid_request_error",
"param": null,
"code": "refresh_token_reused"
}
}

[formal trail]

likei wanna set up telegram bot

[indigo goblet]

Making an app is not like sending "test" to your agent.

[indigo goblet]

Again friend, start slow and step by step, setting up telegram is easy but start at the beginning.

I'm just saying, my opinion, but if you go too fast, things might crash pretty fast and you wont have any idea of what is going on.

[vernal gate]

hey! memclaw migration can get tricky — what error or behavior are you seeing? is it silently
ignoring memclaw and still using default, or throwing an error when you try to use it?

[ancient jungle]

That's what my idiot is complaining about now:
The MemClaw cortex_search is returning 0 results even when it finds matches. That's a bug I can't easily fix without diving deep into the Rust service.

[tall meteor]

Mission Control Dashboard
https://media.discordapp.net/attachments/1389329320446660639/1485760656522084372/kanban.png?ex=69c309f0&is=69c1b870&hm=f5ad652e1bd8e12bb1bf95638a57e508360c9462d94b320191fc931ee4fd921e&=&format=webp&quality=lossless&width=991&height=600
https://media.discordapp.net/attachments/1389329320446660639/1485760693704851546/kanban_new_card.png?ex=69c309f9&is=69c1b879&hm=b037a2e1924a799eaac3be6464ab0f562b149e453bec933e5c16adad8953b4ae&=&format=webp&quality=lossless&width=991&height=600
https://media.discordapp.net/attachments/1389329320446660639/1485760731143082054/task_monitor.png?ex=69c30a02&is=69c1b882&hm=e4db55c021b474f7fbb7f2eb47c1d90fdb2c58c3ec44328a2a569e2cccaccc70&=&format=webp&quality=lossless&width=991&height=600
https://media.discordapp.net/attachments/1389329320446660639/1485760763170652262/tasks.png?ex=69c30a0a&is=69c1b88a&hm=cd1fc40bc9ffac651bbe9da141400bc2f3ea8c2ed2ec2a9ec9f4510623a001fc&=&format=webp&quality=lossless&width=991&height=600
https://media.discordapp.net/attachments/1389329320446660639/1485760807596986560/tasks_new_task.png?ex=69c30a14&is=69c1b894&hm=5253ef2a5c5009cd787303d462f96c0e133a615fc5e62de7777104c1b4ad6949&=&format=webp&quality=lossless&width=991&height=600
https://media.discordapp.net/attachments/1389329320446660639/1485760844833755207/agent_activity.png?ex=69c30a1d&is=69c1b89d&hm=b4bef843565531549ff7b4374764c7c7f730159819d56524eb85d3301612f899&=&format=webp&quality=lossless&width=991&height=600

[vernal gate]

hey welcome! good detailed writeup 👋

found your issue — groupPolicy: "open" only bypasses allowlists, it does NOT disable mention gating.
you need to explicitly add requireMention: false for your guild:

"channels": {
"discord": {
"groupPolicy": "open",
"guilds": {
"YOUR_GUILD_ID": {
"requireMention": false
}
}
}
}

restart OpenClaw after saving and guild messages should start coming through 🤝

[wicked girder]

THANK YOU! Fixed the problem right away

[vernal gate]

hey! a few things worth trying before going anywhere near Rust:

1. namespace mismatch — check if the namespace/collection used when writing memories matches what
   cortex_search is querying. that's the #1 cause of "found but returns 0"
2. score threshold too high — MemClaw may be finding matches but filtering them out on a min_score
   cutoff. try lowering the threshold in config if there's one
3. try a minimal search — no filters, explicit namespace, limit 10 — just to isolate whether it's a
   filter issue vs actual indexing bug

if all 3 check out and it's still 0 — then yeah it's likely a Rust layer bug and worth raising as an
issue on the MemClaw repo 🤝

what does your search request look like?

[elfin walrus]

How do I create new sessions without wiping my current one

/new seems to just clear context

[copper field]

Hello guys, is there a way I fix my agent avatar? Its looks corrupted I cleared cached and even I updated the identity.md still no luck

This happens on the dashboard

[vernal gate]

glad it worked! enjoy your bot 🤝🦞

[ancient jungle]

i asked my idiot to check for namespace/collection mismatch and it returned with this: ```"There's the mismatch!

• Qdrant has: memclaw_tenant_claw
• Local config has: memclaw

The service is writing to a different collection than what the CLI queries. Let me check which collection is actually being used:"```

🤣

[formal trail]

guys

i need help

how do i dleete openclaw

[fallow coral]

"openclaw uninstall" on the machine running the gateway

[ancient jungle]

Wait - the config shows qdrant on port 6334, but I was checking 6333. Let me verify: 🤣 i can't with this thing.

[fallow coral]

is that deepseek

[indigo goblet]

Please, research, what you need is in the docs.

[ancient jungle]

minimax 2.7

[fallow coral]

damn

[tall meteor]

@ancient jungle could you see my screenshots?

[formal trail]

thats it?

[fallow coral]

yes

[vernal gate]

lmaooo wrong collection AND wrong port, your idiot is speedrunning all the classic mistakes 😂

6333 = HTTP, 6334 = gRPC — make sure your config and your CLI are both using the same one. should be
smooth once they match 🤝🦞

[formal trail]

can u check dm thansl

[lofty flicker]

Hey guys, I am new user and I am curious how efficient this memory system actually builds up and how benefital memories become in task completion. I can't stop thinking agents as just soulless stupid llms essentially 😂 just predicting the next word. Does memory building system actually make a significant difference and allow the agents to evolve over time? Or they are only operationally useful contexts?

[ancient jungle]

"Still broken. The service says "Qdrant not configured" even though we fixed it. Let me check which config it's actually using:" i might be able to troubleshoot faster than this thing

[vernal gate]

great question and honestly the "soulless word predictor" framing is more right than most people
admit 😂

BUT memory changes everything — here's the real distinction:

without memory = same LLM every session. no context, no growth, no personality. just a calculator
that talks.

with memory = the agent knows YOUR business, YOUR preferences, YOUR history. it's not smarter — it
just stops starting from zero every time.

the evolution isn't in the model. it's in what you feed it over time. the model stays the same, the
context gets richer.

think of it like a new employee vs a 2-year employee — same brain, completely different usefulness.

operational context is the foundation. personality and "soul" come from consistent, well-structured
memory files on top of that.

what kind of agent are you building? 🤝🦞

[next walrus]

Dashboard working, playing with a bunch of stuff but now that we resolve the limiter issue, hopefully I can get some fun stuff done

[indigo goblet]

Good stuff

Cheers

[lofty flicker]

Thank you for your response. I am curious about memories because at first look that seemed to be the most distinct feature that I can benefit from. I am trying to figure out how to best make use of the agents and starting to learn leveraging the uses with memories/enriched context seems appropriate. But it is easy to get the idea that they are more than LLMs which may be misleading I think. Need to be careful on this one.

[fallow coral]

i mean ultimately they are LLMs that happen to have access to a lot of tools

most of which from a terminal

[dusky jackal]

Is anyone else having an issue opening their browser UI after the update?

[inner pasture]

Yes I had to roll back because I didn't want to mess around with manually fixing the control UI. I think someone posted instructions in this channel earlier if you dig for it

[dusky jackal]

Found it thanks!

[tough swan]

Can you link to the post that fixed it in this channel? I have the same issues.

[tropic kiln]

I'm trying to set up OpenClaw on my laptop and I'm going through Claude to do this... it asked me to post this on Discord. Any help would be appreciated. Thanks

"Running OpenClaw 2026.3.23 on WSL2 (Ubuntu). Built UI with pnpm ui:build — files exist at /home/sbonneau/openclaw/dist/control-ui/index.html. Gateway runs from /home/sbonneau/openclaw/dist/entry.js with cwd /home/sbonneau/openclaw. Config has no controlUi.root override. Still getting 'Control UI assets not found' at http://127.0.0.1:18789/. What am I missing?"

[copper field]

Does anyone know what is going on here?
So I told my agent your name is “Nova” but in the webUI it still says assistant. Although the agent responded saying I am Nova
Why is that ?

Or is this a bug with the control UI thing?

[fleet sun]

How do i get openclaw to know everything about me?

it is a bit hard to tell it everything, can it login to everything i have

[copper field]

Ignore my previous message I fixed it in the gateway config

[worn smelt]

My openclaw is claiming it does not have permission to create files in the workplace directory and any attempt to work with or open files I create in the workplace directory for it return a message "no such file" message. basically my openclaw is little more than a chatbot interface HELP

[minor star]

Ask it how to change the permissions so it can edit the files

[minor star]

Check your agent.md file. It should have the name saved there, that's how it knows it's name. Chat in the webUI should show the name.

[copper field]

Yeah don’t worry it’s fixed thanks

[nimble hill]

A new version 2026.3.23 just popped up 20 mins ago. Upgrade to that

[worn smelt]

All it tells me is to change the permissions for the directory, but the directory is read/write for the logged in user

[minor star]

Try changing it for all users see if that works?

[worn smelt]

while I'm testing that I want to ask; shouldn't openclaw have the permissions of the loggedin user

[steel inlet]

I just updated openclaw to newest version and now its broken?
mac mini setup
updated newest openclaw
now i cant access the host
tried doctor but that is coming up with errors? thanks in advance

[worn smelt]

it is not seeing the files I have manually put in the directory following openclaws directions

[covert blade]

Earlier today a bot gave me an hour timeout for, i better not say in fear of another timeout, it was some logs and a fix for someone, in a codeblock, asking about how to deal with a problem. Am i now on a watch list with some negative points with the channel owner / moderator or ??

[copper nova]

https://giphy.com/gifs/instagram-nazoom-half-closed-eyes-yxy69FCE06Ql0Fjk4Z

[minor star]

It has to be permissions if it can't write to them. It should give you the chmod command that you need to run to give it access

[worn smelt]

hechmod command it has given me requires root (sudo) access and my root pw is messed up and I'm not sure how to fix it, I'm on a mac mini

[fading wren]

Anyone getting this issue after the most recent update?
Agent failed before reply: WhatsApp plugin runtime is unavailable: missing light-runtime-api for plugin 'whatsapp'.

[covert blade]

I'm happy to report that the 2026.3.23 release fixed the missing ui pieces so the web gui worked without futzin, and allowed for an enable of the matrix plugin without complaints of missing sdk.
I was about to paste in the result of openclaw plugins info matrix but that's what triggered my sixty minute timeout earlier. I'm a bit gunshy of posting codeblocks now.

[minor star]

You're SOL if you don't know your sudo PW unfortunately. Do some work to get that fixed then move forward from there.

[worn smelt]

Trtying to reset it now but the instructions I'm finding are not working

[minor star]

Sorry you've surpassed my technical knowledge with this problem. Maybe someone else can jump in.

[silk turtle]

Has anyone had issues with their claw being unable to attach a file to a message in Discord? I was trying to trouble shoot for a while and cant figure out why it just randomly stopped being able to attach a file

[lament valve]

hello! im very new to OpenClaw after starting out with NemoClaw a few days ago. My goal is to get a very basic demo of openclaw using my local, self-hosted Firecrawl instance to do a simple web search. For example: "Use Firecrawl to scrape https://example.com and summarize the main content." in the openclaw tui. I think my last challenge is on the OpenClaw side and not NemoClaw so am hoping this is a trivial question for y'all.

I successfully installed the firecrawl-search skill which seems to be pretty straight forward with just a couple markdown and python scripts. The API URL is hard coded in those python scripts to point to the Firecrawl cloud, so I modified the scripts to load the API base URL from an environment variable FIRECRAWL_API_BASE. But even after setting these environment variables from within the OpenClaw sandbox it seems like the script will not recognize the environment variables.

What am i missing? do i need to reload the skill somehow to get it to use my now modified skill scripts or something? Is there a more proper way to accomplish what im after?

[covert blade]

Sweet! matrix installed, bot user approved. chating with openclaw with element on my android phone. Looks all better now in this 2026.3.23 release.

[nimble hill]

Yep, matrix upholds the true opensource and decentralized chat spirit!

[fast sleet]

Question. Is acpx a skill or a plugin? I know plugins can have skills, before I upgraded to the latest version, acpx was listed when I ran openclaw plugins list, now it's not, and doctor is bitching about it not being a valid plugin

[worn smelt]

Thank you for trying

[solar mango]

hey there I have the 2026.3.23 release but my openclaw llm still times out on the main chat and telegram any tips?

[covert blade]

Sorry, I'm pretty new to the claw. I'm using ollama and matrix you might want to include a log of the error for others to analyze.

[crisp warren]

Hey, I'm trying to integrate a nextcloud mcp server into my openclaw, but somehow claw tells me it doesnt exist even though the mcp server is in my config. Am I doing something wrong with my config. Atm my agent knows its there but can't interact nor auth with it.

{
  "nextcloud": {
    "url": "https://pathtomcpserver/mcp"
  }
}```

I'd appreciate any help, thank y'all :)

[ancient jungle]

does anyone know if memclaw supports separate agents?

[stuck patrol]

My Discord bot is connected and logged in (I can see it in the server member list) but it's not receiving any DM messages or slash command interactions. All 3 privileged intents are enabled. Any idea what's wrong?

[vast stream]

I installed OpenClaw using Docker on Windows. I modified the docker-compose.yaml file to add the Gemini API key under the openclaw_gateway section and set OPENCLAW_MODEL to 'gemini'. However, after running docker-compose up, it still only displays Claude. How do I successfully switch models in OpenClaw when using Docker?"

[wintry coral]

Has anyone managed to get Twilio to work with decently latency? I'm using OpenAI TTS for the voice and GPT 5.4 via OpenAI Oauth for my main agent, but it takes 10-15 seconds for it to respond in a conversation.

[peak hinge]

Hi, I need help enabling agent editing on my OpenClaw installation.

My installation does not expose any agent configuration files in ~/.openclaw/agents/<agent>/agent — only auth.json and auth-profiles.json exist.

The following commands do NOT work on my build:

• openclaw agents --interactive (unknown option)
• openclaw agent edit <name> (requires -m message)
• openclaw agent update <name> --file <path> (ignored, still requires -m)
• No agent.yaml or system prompt files exist anywhere on disk.

I need to install a long, multi-section system prompt for my Telegram agent (“anthropic”), but my build appears to be using a database-backed agent store with no way to update the system prompt.

Can you confirm:

1. Where the agent configuration is stored on my installation?
2. How to enable file-based or interactive agent editing?
3. How to import a long system prompt for an existing agent?

Thank you.

[mighty rampart]

Have you restarted the openclaw gateway? That’s the first thing I would do after changing environment variables.

[frigid oyster]

Has anyone used VoiceInk with their Discord audio channel to give voice memor, thought dumps etc to their openclaw?

[lament valve]

wow, thank you for the response! openclaw gateway restart seemed to do the trick and now the modified scripts pulls in the environment variables set in /sandbox/.openclaw/workspace/.env

[ancient jungle]

i burned 327.6M tokens in a matter of like, 3 days

[nimble hill]

well - thats the usual stories nowadays - since openclaw uses llm for eveyrthing, it just adds up. The point being did you get anything useful done? If not, as I have been stating, use n8n which provides some sort of deterministic flow. Use openclaw to draft a n8n workflow

[chilly kayak]

hey thanks for the response. i was actually able to get it setup, my context window was too small (>16000) which openclaw doesnt like. I have it working now! But.. llama 3.1 8B is dumb and slow..im thinking I return my 32GB mac mini for a 128GB studio. Unless you have any suggestions here?

[ancient jungle]

what's n8n? i've seen that phrase a lot

[nimble hill]

https://n8n.io/ A workflow management tool - which can be self hosted. It also has a AI agents, but you can also call from openclaw. Similar other tool is https://flowiseai.com/

[nimble hill]

if you have money go ahead - but there are others like Asus GX10 and cheaper ones with Ryzen Ai Max+ 395 Gmteck evo-x2, Beelink GTR9 Pro

[pliant stone]

guys how to fix 401 Missing Authentication header

[chilly kayak]

are these all >10B models? I think Im just gonna upgrade hardware if it will be a similar expereince, in terms of memory and speed. llama 3.1 8b was forgetting things i said literally one sentance ago lol

[finite sable]

Someone confirm to me does the new update fix the issues we were having

[nimble hill]

what I mentioned are the hardware, not models. For openclaw to work any decently, you would need like gpt-oss-70B or 120B model. These hardware with 128GB variants can run those. Again, the opensource 120B model are just fine, but cannot compare against the likes of top notch commercial models like Opus and gpt 5.4. Hence, you decide, do you want to buy a hardware capable of running Local LLM or use any machine that you have (even a mac mini is a massive overkill) and pay for Cloud LLM? - a highest tier subscription will likely take 2+ years to break even againt Hardware pricing

[copper nova]

stop your mac mini slander

[nimble hill]

which issue? also as I generally advise - deploy in docker containers (or k8s). This way, you can run 2 or more containers with different versions together 🙂

[nimble hill]

ha ha - why are you so riled up? my post, my rules 🙂 and is not a slander, those are facts!

[copper nova]

https://giphy.com/gifs/black-and-white-photo-apple-l4pThMAKS4BOtz8d2

[copper nova]

are you advising normies to spin up kube clusters? you wildin my guy.

[nimble hill]

start with docker

[finite sable]

uhm thanks I still needa know if the bugfixes worked or not

[nimble hill]

depends upon which exactly you were facing. there are quite a number fixed, quite a annoying ones are new / open

[copper nova]

which bugs were you facing?

[finite sable]

the issues with the *22 update did you not try that version

[copper nova]

if it's the control UI one's then yes

[finite sable]

yes

that one

[copper nova]

3.23 fixes it

[finite sable]

thats what I'm asking does 23 version fix the 22 version issues

[copper nova]

I don't use the control UI so I wasn't aware it was broken lol

[finite sable]

wa

[copper nova]

but yes, 3.23 fixed it

[finite sable]

I see, I'll try it then

[nimble hill]

if you use Node and control them, that is still partially fixed, but majority don't use them

[copper nova]

go make a PR to fix it

I make a PR for all my pains

[finite sable]

btw the control ui looks really cool u should try it sometime

[chilly kayak]

ahhh gotcha. yeah i burn through cloud tokens like a muddafugga. and I want to create a powerful multi-agent workflow. Im not expecting opus or gpt 5.4 level, but from what claude said is that, the 70B models should give sonnet level - not bad at all

[tight halo]

hello guys, anyone has got this error on dashboard "Control UI assets not found. Build them with pnpm ui:build (auto-installs UI deps), or run pnpm ui:dev during development."

[copper nova]

i kinda built my own UI like 3 patches ago, been using that ever since lol

[finite sable]

damn I see

[copper nova]

but that being said, it uses alot of the gateway plumbing, so it breaks every single update lol

I just fix it everytime

[nimble hill]

done already - not merged yet

[nimble hill]

well, as I said, depends upon how much is usage and cost benefit analysis. Hopefully with more hardware org - hopefully the Ai Max+ 395 may get cheaper, but with this RAM thing blowing up - you never know.

[chilly kayak]

yeah i was trying to use kimi k2.5 and I used 10 bucks worth of tokens in 1 day. Not longterm sustainable. Would rather pay up front, not worry about model sizes and start building asap

[copper nova]

just get a claude subscription

[hidden badger]

請幫我轉成中文。

[chilly kayak]

i actually havent tried using claude yet. hows your token expenditure, and do you hit rate limits at all?

[copper nova]

there's a session and weekly rate limit, but on the max plans it's pretty plentiful. It equals out to like over 25x value vs API usage

[nimble hill]

github copilot works great as well - again depending upon your usage. I don't use higher tier, but my friends say USD 39 plans lasts for quite some time - I think about 1700 requsts or something like that. This combined with open AI $20 works out good for devs

[copper nova]

I have this plan too, it's 1500 requests, it's also cheaper than other API usage, so I would also reccomend this

this is only if you're looking for like S tier frontier models to use

it's not cheap

but it's the cheapest way to use em

so it's how you ball out on a budget

[nimble hill]

USD 39/month + USD 20/month compare to a hardware costs of atleast about $4500 one time - I would choose subscription - atleast this year, if i am buying my own

[chilly kayak]

interesting, ill take a look into this, thanks

[copper nova]

read the other stuff I said in regards to what it is you're paying for lol

it ain't cheap, but it's the cheapest way to use these SS tier models

[steep dragon]

any1 here who can help me? i think i made my agent clean too mcuh openclaw and reorganize files... openclaw on terminal doesnt exist anymore. but the folders are all in /.openclaw... what can i do to restore it ? install again and pull all info from the folder ? or anything in trash on ubuntu i can restore to make thigs right again ???? im not very experienced on linux...

[chilly kayak]

yeah im in CAD so its more like 275 a month. Im thinking I drop the 5k now on hardware and get unlimited everything on sonnet level

[copper nova]

local models just don't hit the same

and I'm also a maple syrup enjoyer

[novel kestrel]

Check if the env var is set in the OpenClaw Web UI as well. Skills run under the gateway process, so if FIRECRAWL_API_BASE isn’t defined there the script won’t see it. You may also need to restart the gateway so the skill reloads.

[copper nova]

but regardless of what those benchmarks tell you, they definitely don't hit the same

[chilly kayak]

hmm, I see. Well ill test out claude just to get a taste of the best maple syrupz

[copper nova]

save that 5k for some more double doubles when roll up the rim is back

[sterile steppe]

I am telling you minimax 2.7 is really dope for the cost.

[chilly kayak]

kimi k.2.5 on cloud was awesome, but I wa gobblin tokens lol

[chilly kayak]

minimax stole my money!

[sterile steppe]

?

[chilly kayak]

i tried getting the coding subscription, i can see i paid for it, and they never filled my balance up.. i gave up on it

[copper nova]

do the $100 claude plan, don't go for the 200 off the bat, you don't know if you need that yet lol

I don't even use up the 200 atm

[sterile steppe]

No that is not acceptable, I would raise HELL

[copper nova]

i don't even use my 100 lol

[chilly kayak]

ill def try it now that you brought it up -

[copper nova]

what you gonna do? call up Xi and ask him where my tokens?

[chilly kayak]

sent em an angry email but no response smh, was looking forward to trying it as well

[sterile steppe]

customer support, tech support, and ultimately charge back on debt card, tag them on x.com, message them on x.com.

[sterile steppe]

when I used minimax, I didnt have this issue, my purchase went through instantly and it worked really well.

[chilly kayak]

that was my experience with kimi k2.5

but eating through tokens too fast! Ill give the claude 100 a shot. thanks @copper nova

night yall

[ancient jungle]

what are y'all doing burning so many tokens?

[nimble hill]

Wait claude subscription cannot be used in openclaw (it might get you banned) unless something has chnaged in past week or so @copper nova ?

[wooden summit]

Anyone have trouble getting Telegram Topics to work... On my wife's install I just did it works great just doing the bot father changes and then adding the agent to the group.

On my original install on 2026.03.14 it just doesn't work

[copper nova]

you can use, many people have been using, no bans I've heard of yet, if that was the case, general chat would be popping off

[copper nova]

I think they changed the setup, try updating to 3.23 and see how it goes

[ancient jungle]

How's the usage limits on Claude $20/month plan?

[copper nova]

not great

but it's $20

[ancient jungle]

I mean, define not great. I'm sure usage pattern is not the same for most people

[copper nova]

I hit that session usage really quickly

i never hit it on the $100 plan

[copper nova]

it's hard to quantify with real numbers when they only give me a % for usage, I'm sorry I can maths it for you in real numbers

I can only tell you by my actual experience lol

[ancient jungle]

How long before resets?

[copper nova]

session is in like 5hr increments, but there's also a weekly limit that's 7days

[finite sable]

The issue is small, but has anyone else noticed the styling errors in the control UI on v2026.3.23?

[copper nova]

give you an idea of my usage on a 100 plan, I didn't even hit 50% of my weekly this week

and I'm basically only using opus lol

[finite sable]

Someone tell me whose in charge of the web ui so I can bother them instead for bothering shadow 😔

[copper nova]

don't bother them

submit a github issue

[finite sable]

fair enough

[copper nova]

or submit a PR fixing it if it causes you real pain

or just build your own chat UI to your flavor like I did

[fickle pewter]

yeoo anyone not able to launch openclaw after the update? mine being stubborn with plugin not found. assuming it has something to do with the new default settings? idk. im not too savvy on the config files.

Invalid config:

• plugins.load.paths: plugin: plugin path not found: /opt/homebrew/lib/node_modules/openclaw/extensions/diffs
• plugins.load.paths: plugin: plugin path not found: /opt/homebrew/lib/node_modules/openclaw/extensions/imessage

[copper nova]

update to 3.23, it fixed that diff issue I was having

not sure about imessage, I don't use that one

[fickle pewter]

thats the one from like an hour ago?

i did "openclaw update" in terminal

also not letting me

how to force update 🤔

[copper nova]

do you have some CLI agent you can use? like claude code or opencode?

[fickle pewter]

i have claude

[copper nova]

spin up claude code and tell the opus agent to figure out the update for you, that's what I did

[fickle pewter]

awesome lol

[copper nova]

I never manually update anymore, there's always something that breaks I have no patience to fix myself lol

[placid mason]

I got stuck a few hours ago it went really bad. I downgraded to my old version to get it going so at least I could ask what was going on . Not recommending this as I don't know the full extent, but I didnt have the time to mess with all my errors after spending a week to get it to where I was.
npm install -g openclaw@version_number, (version number I retrieved from a backup file)

[fickle pewter]

idek how to ask claude bro is clueless

the doctor is saying to remove lines in my config

[fickle pewter]

are you all good now

[tropic ingot]

Try running curl -fsSL https://openclaw.ai/install.sh | bash again then

[copper nova]

so first go to the open claw directory before you start up claude code

[placid mason]

Yepstarting working - It told me not to update yest and wait for the fixes

[copper nova]

~/.openclaw

I do a /init in this directory so your claude know's wtf is going on

[placid mason]

Is this the prep you do before any upgrade?

[copper nova]

nope, just one time setup

now everytime I go to my openclaw directory and spin up claude code, it knows where it's at and what's going on

so then I can say "look at the new release 3.23, give me an upgrade plan, are there any breaking changes we need to address?"

stuff like that

then after it builds an upgrade plan

I go "fucking send it"

and it works usually

[placid mason]

lol, cool

[fickle pewter]

update:

"plugins": {
"load": {
"paths": [
"/opt/homebrew/lib/node_modules/openclaw/extensions/diffs",
"/opt/homebrew/lib/node_modules/openclaw/extensions/imessage"
]
}
}

I had to remove those dead paths

All good now

thank you guys for swift help. I will drop any fixes i come across in here always

[placid mason]

that is good - the paths I had to remove I didn't want to lose the function

[wet kraken]

Does anyone keep losing their main working session everytime there is a heartbeat, and it gets overwritten to just heartbeat as the session and all their conversation history is gone? This keeps happening to me both for the last and current version. Any solutions other than isolating heartbeat completely (which would cripple heartbeat's ability to self-improve OpenClaw)?

[manic marten]

Is openclaw slash commands broken for you guys on discord?

[digital umbra]

Hey everyone, I'm wondering if there are any communities dedicated to sharing OpenClaw configs? I'm in China and haven't found any so far. A lot of people, even developers, haven't really gotten into OpenClaw yet—mainly because there's a bit of a barrier to entry. If there was a community where you could one-click deploy others' pre-configured OpenClaw setups, it would be a huge boost for OpenClaw adoption and evolution.

[placid mason]

Did you sort yours out, must be something to do with the username 🙂 I need to run a second attempt

[broken yarrow]

just downloaded openclaw says its rate limited but openai says i have storage

strange

[nimble hill]

Which model?

[broken yarrow]

5.4

[fleet sun]

Ong

[finite sable]

I wish I could share images here

[digital umbra]

Ayy glad you're feeling the vision! 🚀 What’s your take on this?

[finite sable]

are u ai

[digital umbra]

我不是ai，我是来自中国的开发者。英文不太好。所以让谷歌帮我整理了内容。

[finite sable]

ohh I see

[digital umbra]

请问现在有这样的openclaw的配置分享社区吗？在中国有许多人都想玩他，但是有门槛。

[finite sable]

Well you could make your own?

[copper nova]

https://giphy.com/gifs/therokuchannel-the-roku-channel-cupcake-guys-D2RlF1b90rrOhBQlN0

[finite sable]

waaaaaa

[digital umbra]

我现在正在做。但是我没办法推广出去。我如何发图片啊？我想把我做的截图出来

[finite sable]

no idea ask Jasnsy

@copper nova how do u post images

or send embeds

[copper nova]

didn't tencent hold some event that made thousands of people show up for openclaw setups?

pretty sure China is more pro AI adoption than the states even lol

[digital umbra]

那是安装。我想要的是配置好的 openclaw。比如：配置好了agent，她是编码专家，QA专家，等，包含了专家需要的技能，工作流，工具等。如果只安装openclaw，只是一个壳而已。

[copper nova]

being kind has it's privledges

[broken yarrow]

keep getting rate limited just downloaded

[copper nova]

woe boo ming bai

I CBC and speak broken ass cantonese

can't read or write

[finite sable]

if I gave you mine you wouldn't understand it to be fair, cause it uses some weird stuff to function-

[finite sable]

not very kind of you-

[copper nova]

do you know what that means? lol

[finite sable]

yes I'm replying to them

u can see

[copper nova]

I thought you were google translating lol

how is saying I don't understand being rude?

[finite sable]

sounded like ur mocking them-

[digital umbra]

我的想法是，智能体应该类似智能手机的app。所以大家分享的是一个独立能力的智能体，就像开发了一个app给别人。

[grim lake]

https://github.com/msitarzewski/agency-agents 看下这个？

[copper nova]

I know like 10 words in mando, and thats half of em

lol

[finite sable]

I see, mb

[copper nova]

try these channels

[finite sable]

So, a premade config so people install and use?

[copper nova]

that won't work

[finite sable]

then?

[copper nova]

the config is specific to what provider they have for their setup

you can't just copy paste configs

[finite sable]

so multiple premade configs?

[copper nova]

you have to go through the setup process, you don't know what channels they want to use, whats app discord, what provider they using for tokens

it's not something yo can copy paste

[finite sable]

which do u want then, a config or an option to setup easily for the users?

[copper nova]

I don't need either lol what do you mean

[finite sable]

you said your idea was like an app to make it easier to setup?

[copper nova]

no

[finite sable]

then?

[copper nova]

The chinese guy is asking for a copy paste config

what are you talking about

lol

[finite sable]

oh

sorry dude

i have mild dementia

I'm speaking to the entirely wrong guy

[copper nova]

wtf

I'

so confused

[finite sable]

yea sorry

anyways

[digital umbra]

我不是要这个配置，我是要这个社区。方便更多的人直接使用。

[finite sable]

So people who can give you copy paste configs to use?

[copper nova]

i dont something want?

[finite sable]

?

[copper nova]

i'm trying to read the chinese

I recgonize some chars

[finite sable]

he said he needs a community to get the configs not configs themselves

[copper nova]

I know that third word but don't know the eng translation

[finite sable]

damn

[copper nova]

woe boo shi heart?

[finite sable]

im not sure they can read that

[copper nova]

that's my attemp to read them

lol

[finite sable]

a

then I cant read it either seemingly-

[copper nova]

我不是要 is that not what I said?

woe boo shi... want? oi?

[finite sable]

I think it's bushi

not boo shi but yea that works

[copper nova]

and this makes no sense at all how do you get configs but not the configs themselves?

[digital umbra]

I'm building a community for sharing OpenClaw configurations.

The idea is simple: users can upload their battle-tested Agent configs (fully functional and optimized), and others can download and deploy them instantly onto their own OpenClaw setups. Each "Soul" will come with simple docs for quick start.

Think of it this way: If OpenClaw is the smartphone, these configurations are the Apps.

[copper nova]

that's like an oxymoron lol

[finite sable]

that's the thing, they want a community

a

[finite sable]

yea I thought so, see the issue with that you would be leaking a lot of secret keys for your APIs since those are included in your configs

[copper nova]

configs contain API keys usually, we don't condone sharing them ever

no such thing as battle tested configs, just one's that work for you

[digital umbra]

OK，THere is a tool for export config. It will process this security config. The user want to install need config it when he indstall the config.

[finite sable]

Oh I was gonna say, perhaps u could make a plugin that lets you install a config without leaking api keys. that's a good idea

[digital umbra]

yes ,it is my mind

[copper nova]

config is unique to the user, you don't know if they need whatsapp, telegram, discord, QQ, wechat, etc.

you don't know if they need claude, minimax, openai, opencode, kimi

gemini

[finite sable]

I mean if it's handled with a plugin, you can absolutely just omit the channel configs or migrate them from ur normal config

[digital umbra]

the tool has the rule to make this config safe.

[copper nova]

there's no point in sharing

it's custom for each person, no two are the same usually

for a reason

[finite sable]

I feel like they can still copy from a general list of keys another config has no?

[copper nova]

and they break every update

sooo.... 🤷‍♂️

[finite sable]

then either they don't update, or they update the plugin to only accept plugins for newer openclaw

[digital umbra]

Channels config is base config ,like google login, apple login in phone app.

[finite sable]

I know it sounds like I'm supporting them but I'm just speaking logic yk

[copper nova]

I just don't know if anyone will opt to run a tool that scrapes their config to export it, I know I wouldn't lol

[finite sable]

wouldn't it be fully voluntary

you do it if you want

[copper nova]

yea, I won't want to

lol

[finite sable]

you probably wont, but others probably would rather have another user's working config shape

[copper nova]

feels too sus

[finite sable]

how?

[digital umbra]

if this community provides very perfect agent config ,you can one-click to use. would you like to do ?

[copper nova]

sounds like @finite sable does, let me know how it goes lol

[digital umbra]

i hope the agent config like a phone app in AI times.

[indigo goblet]

what

[finite sable]

I personally already configured mine, not that I'd use it just that I see why some people would prefer it

[finite sable]

their english is bad, bare with them

[copper nova]

run the sus chinese tool to scrape and copy your config then

[finite sable]

I'm confused why you're soo against this.. this is starting to feel like a racism type thing

Just cause it's chinese it's bad?

[copper nova]

I'm chinese you foo lol

[finite sable]

then what are you tryna say

[digital umbra]

I don't care about emotions or discrimination. I'm only interested in a rational discussion about what I’m trying to achieve

[copper nova]

you're telling me running some script that clones your config which has your live API keys in it is a good idea? regardless if they say it's safe and strips it out.

I think it's natural to skeptical of anything that does anything like that

[finite sable]

it's probably gonna be a command that you only run if you want others to have your config

[digital umbra]

1. What would you use your OpenClaw for? What capabilities would you configure?

2. If a community offered a more professional and higher-quality proxy than your configured one, would you use it?

[copper nova]

lol nvm, I can't explain myself clearly obviously

[finite sable]

besides, it isn't made yet, what are you skeptical about? the concept doesn't prove what flaws it has

[digital umbra]

Sensitive data will be removed during upload. Essentially, your configuration is an app itself; it doesn't contain your sensitive data.

Others will need to initialize it or use an existing basic configuration when using it.

[finite sable]

I'd say if made well enough this could work

[indigo goblet]

the problem with sharing AGENTS.md and SOUL.md is that it is very personal to each user and their use case, too many variables, it's like clay. A template could work sure but that is what vanilla openclaw basically is.

[digital umbra]

https://qiyu-base.oss-cn-beijing.aliyuncs.com/Public/20260324-155309.jpeg

https://qiyu-base.oss-cn-beijing.aliyuncs.com/Public/screenshot-20260324-153345.png

[finite sable]

I see

just in case I turned on my vpn for that

wait, instructions for the AI?

[digital umbra]

我的想法基于：每个智能体应该是完成具体一类任务的，而不是通用的。他可以是有个性的，特异化的。
但是比如：
我要配置一个厉害的前端开发专家。每个人都可以配置不同，但是基于对需求的实现的结果来说，他可以有公认的优质的解。而不是让每个人都去配置他。

My idea is based on the premise that each intelligent agent should be designed to perform a specific type of task, rather than being generic. It can be individualized and specialized.

However, for example, if I want to configure a highly skilled front-end development expert, each person can be configured differently, but based on the results of fulfilling the requirements, there should be a generally accepted high-quality solution. The goal is not for everyone to configure them differently.

[finite sable]

Sounds like a similar idea to where we used to have jailbroken prompts to make AIs do bad stuff for us

instead ur instructing them to do specific tasks

then you would probably just need the .md files, no actual openclaw configurations

[digital umbra]

I've been using development tools like Claude, and I've encountered bugs that I can't seem to fix successfully, or new problems arise after I try to fix them. I believe there should be a configuration that can improve this. However, everyone's configuration is different, which is essentially reinventing the wheel. If there were a better configuration that I could use to improve the efficiency and quality of my bug fixing, I would be very happy.

[finite sable]

So you'd make a plugin that just replaces the IDENTITY.md files and such to nudge the ai in the right way

Still a risk if your agent has saved any private info in their IDENTITY.md file, which some AIs tend to do such as saving API keys

[digital umbra]

A fully AI-enabled business agent may require configuring multiple sub-agents, workflows, tools, hooks, cron jobs, and so on. Therefore, it cannot be as simple as just Markdown.

[copper nova]

https://github.com/msitarzewski/agency-agents

you don't need configs for this

[finite sable]

wait then wouldn't you just install a skill for that task?

[digital umbra]

I will have a tool created that exports the meaningful parts from the OpenClaw configuration, passes a security audit, and outputs a structured file, such as a zip file.

[copper nova]

lol I'm done, please no one run anything that clones your config without auditing it first. Even after auditing it, it's not a good idea. Good luck.

[digital umbra]

One skill isn't enough; for example, I need a high-performing development team.

It requires sub-agents with different abilities, such as schedulers, designers, executors, QA, and so on.

[finite sable]

then multiple skills? Perhaps files for your sub-agents to read containing the info it needs?

[digital umbra]

This is based on trust and security.

[copper nova]

I run on a 0 trust environment, I don't even trust my own bot with access to anything of mine.

[digital umbra]

This is a process. Just like Android was initially untrusted.

AI is also untrusted. Autonomous driving is also untrusted.

[finite sable]

I mean, not only is it not meant for people who have already configured their AI to use.. It's not like just installing it leaks everything.

[indigo goblet]

openclaw > anything external > untrusted by default

[copper nova]

again, do not run anything that clones your config without auditing it first, even then, not a good idea, good night lol

[finite sable]

If openclaw flags it as suspicious nobody would really use it anyway

[digital umbra]

It's normal and expected that you won't be trusted at first. Trust needs to be built gradually.

[digital umbra]

Yes, auditing is essential, and you should be able to clearly review the exported content yourself. This is a process of building trust.

[finite sable]

Agreed

[digital umbra]

I believe the era of intelligent agents is a new era following PCs and smartphones.

It doesn't necessarily have to be a fixed type of hardware. However, it should provide the ability to perform more complex tasks. It shouldn't be a general-purpose intelligent agent to achieve this, but rather an intelligent agent like an app. I think it's like an app in the era of intelligent agents.

For example: Vercel, I believe it is such an intelligent agent. It has the ability to independently develop and deploy high-quality websites.

There should also be other intelligent agents.

I don't want it to force us to search for this kind of thing elsewhere (Google, app stores).

We should be able to easily search, analyze, and obtain this information through our main intelligent agent entry point. This lowers the barrier to entry for users.

Just like how ClawHub was inherited from OpenClaw. We should acquire mature intelligent agents more intelligently; that's the future.

[finite sable]

I agree

[digital umbra]

My purpose in coming here is to find people who share my vision to accomplish this great feat together.

[finite sable]

If you need resources I've made plugins before for my own work I'd see what I can do

[digital umbra]

I'm currently preparing for the website development and launch. Currently, it mainly supports the user system, uploading and downloading mirror configurations (zip), a mirror leaderboard, a wishing wall, and a discussion community.

Currently, I support OpenClaw configuration and skill configuration, and this will evolve as OpenClaw or intelligent agents develop.

I need someone to help improve the configuration export and installation tools. This part is crucial, as security, compatibility with different environments, and dependencies on other files need to be considered.

[finite sable]

Lets discuss in DMs?

[digital umbra]

I want to contact the community leader of OpenClaw because this would help develop OpenClaw into the Android of the smartphone era.

Furthermore, I don't believe the AI ​​era will see the same dominance of the iPhone over Android. Open-source intelligent agents (note, not models) have greater potential.

[fossil vault]

New update keeps breaking things. Now I can’t get my bot up. Hook error

[elder glacier]

Hey all does the docker image for openclaw have a browser installed? or do I need to install playwright like so https://docs.openclaw.ai/tools/browser#docker-playwright-install?

[hallow flame]

Hi,
A friend’s company is launching a new AI model, and its benchmark performance is close to Sonnet 4.5.
It’s currently free with unlimited usage.
Where would be the best places to share it?

[indigo goblet]

https://discord.com/channels/1456350064065904867/1459642797895319552 > Krill can help you

[indigo goblet]

What are you asking exactly?

[elder glacier]

This is the response I'm getting from my agent "I can't use the browser tool because there's no Chromium-based browser installed in this Docker container. The browser tool requires a real browser (Chrome, Chromium, Brave, or Edge) to be present on the system."

[indigo goblet]

try to run this: docker compose run --rm openclaw-cli node /app/node_modules/playwright-core/cli.js install chromium

and restart the container

[junior acorn]

open source it or gtfo

[jaunty zodiac]

anyone having issues with latest update breaking discord file attachments in sandboxed agents?

[inland knot]

You got a model name? Im interested in researching it

[idle holly]

Hi All, I am still learning about OpenClaw, I just wanted to check which memory Skill do you recomend? my openclaw agent always forget everything everytime it crashes

[elder glacier]

[+] Creating 1/0
 ✔ Container docker-openclaw-gateway-1  Created                                                0.1s
[+] Running 1/1
 ✔ Container docker-openclaw-gateway-1  Started                                                0.6s
error: unknown command '/app/node_modules/playwright-core/cli.js'```
seems to fail

[steel inlet]

Can anyone tell me where to go my openclaw has crashed since the update and I can’t fix it? My developer suggested he just roll back to the old version is this the right move?

[hallow flame]

model
https://console.siray.ai/model-detail/sapiens-ai/agnes-1.5-pro
score
https://pinchbench.com/submission/5def8484-5751-40b6-ab6a-366968c0e7c4

[hallow flame]

check above

[junior acorn]

You say it’s ur friend’s… sus

Also I don’t click links. Nice try

[elder glacier]

whats the point of the openclaw-cli container? Could I just run the commands on the openclaw-gateway container?

[viscid delta]

Hi everyone, I am currently experimenting both OpenClaw and Claude Code. If anyone who can invite me tp Claude Code's Discord channel, I really appreciate your help. There are some subtle questions about CC which I can't solve. Thanks!

[split tiger]

is anyone having problems with kimi k2.5 tool calling?

[inland knot]

Imma wait for specialized benchmarks

[inland knot]

k2.5 is too verbose, good tool calling, very high hallucination rate

[split tiger]

almost all crons failing for me

[inland knot]

the hallucination rate is most probably the reason you are having problems

i encountered this behaviour too

[split tiger]

is there a known fix?

[inland knot]

nope

[split tiger]

damn

[inland knot]

this is the model hard coding

i recommend to switch to MiniMax 2.7, this is my current model, they got a 10$ Token Plan on their website, 1500 requests/5 hours

[split tiger]

i was using minimax

[inland knot]

It respects tool calling 97%

M2.5 is mid, M2.7 is a huge upgrade

[split tiger]

oh is it

[inland knot]

M2.7 is on Sonnet level on most benchmarks,

Close to Opus, depending on the field

[keen trellis]

ok, got it fixed it seems 🙂

[ancient jungle]

depends what you're doing i guess. i still found sonnet smarter for what i'm doing

[inland knot]

It is smarter and more capable

But M2.7 is pretty competitive, i just got no budget for Claude lol

[ancient jungle]

same. i already pay for gemini and 2.7, don't want another subscription for claude too. i just use the chat on the site when m2.7 starts going off the rails and needs to be audited

[astral loom]

Just updated to latest. Anyone have a fix for

openclaw-gateway | - plugins: plugin: plugin manifest not found: /app/dist/extensions/device-pair/openclaw.plugin.json
openclaw-gateway | - plugins: plugin: plugin manifest not found: /app/dist/extensions/phone-control/openclaw.plugin.json
openclaw-gateway | - plugins: plugin: plugin manifest not found: /app/dist/extensions/qwen-portal-auth/openclaw.plugin.json
openclaw-gateway | - plugins: plugin: plugin manifest not found: /app/dist/extensions/talk-voice/openclaw.plugin.json
openclaw-gateway | - plugins: plugin: plugin manifest not found: /app/dist/extensions/thread-ownership/openclaw.plugin.json

[indigo goblet]

try again with this command// docker compose run --rm --entrypoint node openclaw-cli /app/node_modules/playwright-core/cli.js install chromium

and restart your gateway/container

[high heron]

just run

openclaw doctor --fix

or tell your agent to delete those plugins

[tiny fulcrum]

anyone using codex ? i have trouble since yesterdays update.. its set and default model,still got 70% weekly but since the updates i go straight into fallbacks..

[worldly plinth]

is there a way to disable auto-reply for whatsapp?

[storm sleet]

Planning on setting this up soon on a VPS for personal use and helping me with recurring web search tasks. Do I need to worry about API or other sensitive data leaks if no one else has access to it?

[teal zealot]

#security message

[mighty rampart]

Yes. If it has access to the web you have no idea what it will fetch into its context and what it will do with that info. Using a frontier model can help. I use mitmproxy in front of it to inject API keys into requests. If it never has my API keys it cannot exfiltrate them!

[sturdy dome]

Anyone known with the latest possibility of using ChatGPT plus (no token) account? I am hearing some contradicting stories that it should be able again.

I have openclaw running on raspberry pi now connected to cloud API but want to add/adjust model (might be over openrouter) to make use of my chatgpt plus subscription.

Anyone has experience in this? Found some info but machines crashed twice and want to avoid starting over from scratch again 🙂

Thanks!!

Prompt 1:
openclaw onboard --auth-choice openai-codex

Prompt 2:
openclaw models set openai-codex/gpt-5.3-codex

[trim lichen]

when u onboard it, it should ask you for your oauth key after you grabbed it from chatgpt

i had it setup that way before for basic talk functionalities until i setup a multi pc multi gpu system that hosts a FP16 GLM-4.7-Flash lol

[sturdy dome]

Thx. Tried this during onboarding but didnt get it working.

You had it running on raspberry pi? That confirmation would already help me

If anyone has experience in shifting models during set up, I am happy to hear. Mine crashes constantly

[cunning star]

Hey people, is there any particular plugin for eBay for openclaw?

I want it to upload drafts based on w/e it figured out from the pictures I sent it

[stoic anvil]

Hey, what’s the error being shown here ?

[pine flax]

Does the browser option not work well?

Having it take over the browser I mean

[stable cedar]

Hey @trim lichen,
Looks like you should be an expert using FP16 GLM-4.7-Flash. I have been using it since started using OC, but the speed of its operations changes, sometimes is fast (still slower than publically available GPTs, such as ChatGPT, Claude, Gemini, Nemotron, etc), but sometimes is rediculously slow. For example, it could take 20mins simply to update a record in database... What's the trick to speed up its operations? Thanks!

[topaz wasp]

In 2 days the session with my agent bloated. I did not realise as I was testing various things. It has 300K tokens. Then I researched and found /reset commant to reset the session. What is the intended usage here? Do you keep reseting the session often?

[trim lichen]

i preloaded the entire model onto the gpu and only have the kv cache in ram. but it might also because i have RTX 5070s.

one in 1 pc and 1 in another. (cuase i game on one)

i have it setup on discord to talk to me using llamacpp and only call openclaw for tools.

directly from discord -> glm -> memory db -> glm -> discord = 5 seconds normally

but i also have the basic sub (not free one) with claude. and i have the glm talk to haiku for simple responses for references -> then back to glm to tell me answers = 10 seconds.

another branch for complex tasks -> glm consults -> claude sonnet -> returns to glm -> tells me answer = anywhere from 30 seconds - 10 minutes

all of that with background processes to inject memories , chat compaction services/agents to passively compact chat into memories to be pulled up during requests. and it also logs all chats between glm -> other models

i was thinking of setting up a 3rd 5070 i had laying around either for kv cache or a voice model for my IoT devices

the connection between the 2 computers is a 40gbps nic card, forgot to mention

[cunning star]

Am not sure how to do that on the server... But I get what are you saying. My guy pretty much tried to post things via API

[rustic pier]

Anyone experiencing random usage spikes?

Like normal usage for hours, then burning through my limits in a very short amount of time.

[cunning star]

yeah

[stable cedar]

@trim lichenThank you for sharing your experience! Yours is a fancy setup! I'd learn from you. I have NVIDIA DGX Spark, so hardware perspective should be good enough. How easily I can learn from your setup? where did you start from? for example, any tutorial video, or docs? steps list? or simply sharing your openclaw.json file? Thank you!

[rustic pier]

Was it after most recent update or generally happening?

I've had it since last 2 updates I think

[trim lichen]

thats gonna be pretty hard to do lmao, i got annoyed of using truenas OS and swapping between that is openclaw system's cuase it's on the same storage server so i forked both and merged them 💀 no guides here

no dockers

[cunning star]

Idk just recently, could be a telegram bot or something with gemini servers

[stable cedar]

I was in a hurry to use OC, built 2 apps after setting up OC with a simple way - one system running model locally. Besides unacceptable performance, it often not look for the rules/skills that it had built and very easy to start everything from scratch again.

The 3rd complaint is it writes many script for a simple work, and never clean it up. The file organization is random, sometimes it behaves really good, sometimes it is out of the ways doing things. For example, at the beginning, it built an app under ~/.openclaw/workspace/app1, but it can put the DB file to ~/.openclaw/app1 when it simply updates record in the DB. originally I really trusted it, gave very high level requirement, but later I had to keep my eyes on what it does. lol

[lapis iris]

Anyone know how to open a chat session with a new agent?

I don’t see the option at all…

I see the agent on the agents page

[pallid cove]

are you launching several agents at the same time? subagents? background tasks? that can do it

[rustic pier]

Yeah, I'm going to investigate. I don't think I had any crons or subagents running.

[woven ledge]

Anyone lost all their channels after the latest updates?
Whatsapp, telegram, discord all down.

[pallid cove]

23-2 does that? waiting for 23-3 then 😛

[woven ledge]

Actually happened to me on 3-22 and been broken since, my claw can't get it working and we've done a bunch of config surgery.

[hollow egret]

I've lost all the channels.

[pallid cove]

after I had the broken dashboard issue on 22 yesterday I immediately rolled it back lol. you're more adventurous than me. they're doing a big refactor and I can't be bothered

[fleet sun]

No then the agents are the apps and your imported configs are just someone else’s phone

[pine flax]

yeah, I don't think it can use thebrowser on server

[solar mango]

Hey guys, since the updates I have not been able to get the main chat or telegram to respond to me. It just times out and this comes up (Failed to start CLI: Error: gateway timeout after 10000ms) Any help to fix it so that I can get going again would be highly appreciated

[lapis iris]

I think a few people have had luck pointing Claude Code at their openclaw

and telling it to fix it

[warm rover]

Good afternoon ☮️ . Trying to figure out, whether openclaw agents, out of the box without feeding them any context are familiar with the openclaw framework, or do I need to tell him every time to read https://openclaw.docs.ai ?

[eager mauve]

Actually, you can. Install brave then use the --headless arg and it will run without X.

[wild mason]

anybody having trouble with TTS crons in telegram not sending TTS anymore?

[lapis iris]

Unfortunately my Claude keeps trying to read local docs which are... woefully lacking

[eager mauve]

Hello!

I’m using OpenClaw with Codex-ChatGPT 5.4 on a “pro” subscription.

Overall I’m quite satisfied with how it works, but one thing is bothering me… My agent very regularly asks me to “restart” or prompt it again.

Let me explain:

We define a list of tasks (in a .md file), I launch it on those tasks, it starts working, completes one or two, then says something like “the next logical step is now: task X” — and then it waits for me to say “go” or something else before continuing. After a few more tasks, the same thing happens again.

It’s a bit frustrating, especially since some tasks could / should run overnight… but instead it stops after a few minutes.

I’ve obviously adapted soul.md and the other files to push it toward more autonomy on internal, low-risk tasks, etc… but nothing seems to change.

If you have any ideas, thanks!

[solar mango]

so connect claude code on the mac mini terminal and have it fix it?

[lapis iris]

Yup, tell it the issue and see if it can diagnose/fix

this release is buggy though, no doubt

[pallid cove]

my solution was to enable OpenClaw heartbeat and tell it "add something to your heartbeat to tell you to read your TODO.md and do the first not done task in there"

heartbeat > cron for this because you don't care about exact timing

[lapis iris]

I think you can also add it to TOOLS.md

to check the online docs

but mine... well... he ignores his docs

and I've no idea why

"Why didn't you read your TOOLS.md, it's right there?" "I.. don't know"

"Is it in the session?" "Yeah, I see it now, it's there... I just didn't read it"

[storm sleet]

I will look into this, thanks. Chances of running into some injected prompts or w/e are slim based on the search parameters I am going to try using, most using the site: operator etc. but these things may start improvising at some point

[digital umbra]

My previous statement was incorrect.

Let me correct it: I want to create something similar to the App Store. However, the above is a compressed package of someone else's complete agents (securely anonymized agents).

I need to establish rules to allow people to distribute or share these agents like they would distribute apps.

[fleet sun]

Got you

[solar mango]

ok thanks. Do i tell it the issue in the terminal?

[copper nova]

I find that tools.md is kinda useless, my agents always forget to check it even tho it says it's always loaded into context. I would move anything important into agents.md, it tends to be more reliable

and I also agree, latte is milk with coffee flavouring

[lapis iris]

Yup, and point it to ~/.openclaw

are you on Opus?

[lapis iris]

Yeah... I just don't want to make agents.md too big either

my Opus burn is huge even on the $200 plan

but this agent has access to a lot of stuff and has been battletested with injections

so I don't want to switch it

[digital umbra]

I've encountered this before. The reason was that other rules conflicted with it, and it couldn't make a decision, so it couldn't function properly.

[solar mango]

sonnet 4.6 i think

[lapis iris]

Try it, tell it to do some research first, but it may fail...

[copper nova]

i try to have a clean seperation between soul and agents, I tend to leave tools pretty sparse now

[digital umbra]

Are you interested in this platform?

I envision it as the App Store of the AI ​​Agent era, where all users can easily access high-quality AI Agents provided by professional developers, instead of spending a lot of time reinventing the wheel.

[solar mango]

I can't seem to get it to work or respond in the terminal once I connected claude code to the mac mini

[lapis iris]

I'd be interested, I've spent 2 weeks building "infraclaw" with access to a test environment to mostly 'do my job' - it doesn't have access to prod, but does test and staging

[unreal nacelle]

I'm obviously doing something wrong... did a /new, then asked a single question to nv-kimi2.5 "I think auto-compaction could fire a little sooner... what does my config look like for those settings?" and now It seems hung, and the warning popup says 408.8k/200k tokens used lol... Any.. ideas? Or am I wrong here?

[rustic pier]

I figured out what happened. I had a huge TG session from yesterday that never got reset during my daily morning session resets.

And compaction never fired for some reason. Really annoying.

[cunning star]

Jesus, my guy finally raw dogged posting a listing on eBay 😄

using only APIs

uploaded pictures and all

[digital umbra]

Interesting. This part is also very valuable content for my platform. I hope to have a deeper discussion with you.

[lapis iris]

I do still need to battle test it a lot more first, and I'm currently at work, but let's connect later on

unfortunately we've had regressions today, but I think it may be related to Claude

rather than openclaw

[digital umbra]

Can I add you as a friend? We can communicate when it's convenient for you.

[lapis iris]

Sure, but I'm currently on a TUI so can't accept (built by openclaw & claude lol)

[eager mauve]

I tried it but it doesn't seem to work. I'll ask him again.

[digital umbra]

Okay. I'm from China. I am available, except when I'm sleeping.

[lapis iris]

Where abouts? I lived in Shanghai then Nanning for 8 years

[digital umbra]

Wuhan

[lapis iris]

Ahhh nice city, got a lot of hate from others unfortunately

[digital umbra]

I don't care about subjective emotions; I only focus on objective goals.

[pallid cove]

you also have to enable it in the config

[cunning star]

Gemini 3 Pro is free for 1 month btw

[solid flare]

Why am I getting this when I have tokens? Memory search is unavailable right now (embedding provider quota exhausted—top up or switch providers to fix).

[eager mauve]

It looks like it's working. May have to 'repeat' the heartbeat more regularly than what is currently set up but it's already a nice step ahead. Thanks a lot !

[pallid cove]

where?

[cunning star]

I mean go to get plans it will be there

where Gemini can be besides its site?

[pallid cove]

some providers offer free models. there's a bunch rn on OpenCode

[cunning star]

from chat window on top right -> get a plan

Gemini have free model

[pallid cove]

openrouter too. rarely as good as Gemini 3 Pro though

[cunning star]

Gemini have pretty good free tier

I just jumped the wagon since they offering 1 month free

[lapis iris]

Their flash is pretty good too... I'm trying to outsource basic tasks to it

to save my claude tokens

[cunning star]

oh yeah I generally use it, its decent enough for free

I did hit the limit with the free tier today tho, but Claw was doing some heavy lifting

[stable cedar]

@lapis iristutorial or doc for such "outsourcing" setup? Thanks!

[cunning star]

Im actually impressed it figured out how to submit eBay listings through APIs

[lapis iris]

I just have another agent that my main one (running Opus) can talk to

Claude set it up so I'm not actually sure how it works under the hood 😛

[stable cedar]

@lapis irisbut how to do "I just have another agent that my main one (running Opus) can talk to"? Did you just tell OC to do it?

[fallen vortex]

Guys im really confused .. i am running openclaw .. and i saw the big release of nemoclaw .. i dont understand the difference .. am i supposed to ditch my setup and onboard a new nemoclaw instance ? or i can use both together ? or what ?

[lapis iris]

Yup, I have a 'main' agent running Claude Opus - I've told him he's the boss, and he can 'hire' and 'fire' employees at will, and gave him a set of API tokens for various different AIs

he will set up new agents

[dim harbor]

Nemoclaw is a a piece on top of open claw. If you don't understand, open claw, don't bother with Nemo. Once you figure it out, then you can try it out.

[fallen vortex]

thanks man .. so for now i can just keep updating as normal and will be able to use nemo at later stage without breaking my environment right ?

[dim harbor]

Yeah you can migrate to it later if you ever find a reason to

[stable cedar]

@lapis irissmart! but should you configure multiple models upfront? if you run all of these models locally, you have to download them first and let OC to configure?

[lapis iris]

I don't run them all locally, I do use ollama but mostly a mixture of Google and chatgpt

ollama is more of a curiosity 🙂

[stable cedar]

Google and ChatGPT can't run locally@lapis iris

[lapis iris]

I never said I was running them all locally...

I've told him he's the boss, and he can 'hire' and 'fire' employees at will, and gave him a set of API tokens for various different AIs

[stable cedar]

unless you have a cluster of systems

[lapis iris]

OpenClaw (running Opus) has API keys for other AI models (also cloud based)

[stable cedar]

@lapis iriscorrect, you " never said I was running them all locally..."

I tried to use Gemini, but kept giving me rate limit issue even 0 token consumed

[trim osprey]

I seem to have broken my bot on the update im getting this message and cant message in discord or in web claw control Error: Missing workspace template: AGENTS.md (C:\Users*\AppData\Roaming\npm\node_modules\openclaw\dist\docs\reference\templates\AGENTS.md). Ensure docs/reference/templates are packaged.

any ideas?

[tired ledge]

question... my gateway keeps shutting down. even though me terminal remains open. i have to always restart the gateway. i know that is a pretty broad explination, but is there something that would keep my gateway more stable?

[trim osprey]

windows or mac?

[stable cedar]

@woven ledgeI was able to use Telegram after 3.23-2 upgrade

[sturdy dome]

Will run again and let you know

[sturdy dome]

Not sure. Will try first as model and otherwise over web. Thx

[spiral parcel]

Anyone having trouble with "/new" and "/models" when you want to start new session or change model, in discord ?

[pallid cove]

so does the latest 23-2 update break the dashboard when upgrading from 13? like 22 did? or other stuff?

[granite bobcat]

bout to find out

[pallid cove]

pls do tell after!

[granite bobcat]

Looks like it sure does.

Config invalid
File: ~.openclaw\openclaw.json
Problem:

• plugins.load.paths: plugin: plugin path not found: C:\Users\k_thr\AppData\Roaming\npm\node_modules\openclaw\extensions\acpx

Run: openclaw doctor --fix
Gateway aborted: config is invalid.
plugins.load.paths: plugin: plugin path not found: C:\Users\k_thr\AppData\Roaming\npm\node_modules\openclaw\extensions\acpx
Fix the config and retry, or run "openclaw doctor" to repair.

Looks like it fails b/c of acpx though? Which seems odd. This might be specific to my setup?

Ran openclaw doctor --fix and no difference.

[lapis iris]

Honest opinion... bumping claude up to 1M tokens doesn't go very well... he gets too bloated in session, forgets a lot of stuff

I'm at 650k and he's making mistakes everywhere

[granite bobcat]

Any way to rollback? Would be nice if that was possible since I just borked OC

[signal gulch]

Hey friends, has anyone had any luck adjusting the model for specific jobs…
When creating or editing a cron job with --model, the override shows correctly in openclaw cron list (in-memory) but is not written to ~/.openclaw/cron/jobs.json. After a gateway restart the override is lost and jobs fall back to the default model.

[granite bobcat]

Can't even run onboard. This blows.

[pallid cove]

yes, I rollbacked mine yesterday with npm

[pallid cove]

you shouldn't have, ideally it should reuse your existing everything

[granite bobcat]

Help with that? I Just tried openclaw update --tag 2026.3.13 --yes but that didn't work

[pallid cove]

npm install -g openclaw@2026.3.13

[granite bobcat]

You're the best thank you

I'm still getting this error:

13:12:31 [plugins] acpx runtime setup failed: node:internal/modules/cjs/loader:1478
throw err;
^

Error: Cannot find module 'C:\Users\k_thr\AppData\Roaming\npm\node_modules\openclaw\extensions\acpx\node_modules\acpx\dist\cli.js'
at Module._resolveFilename (node:internal/modules/cjs/loader:1475:15)
at wrapResolveFilename (node:internal/modules/cjs/loader:1048:27)
at defaultResolveImplForCJSLoading (node:internal/modules/cjs/loader:1072:10)
at resolveForCJSWithHooks (node:internal/modules/cjs/loader:1093:12)
at Module._load (node:internal/modules/cjs/loader:1261:25)
at wrapModuleLoad (node:internal/modules/cjs/loader:255:19)
at Module.executeUserEntryPoint [as runMain] (node:internal/modules/run_main:154:5)
at node:internal/main/run_main_module:33:47 {
code: 'MODULE_NOT_FOUND',
requireStack: []
}

Node.js v25.8.1

[lapis iris]

Just had Claude interview Gemini

What I'd give Gemini Flash:

    Web research tasks — summarising articles, comparing options, drafting documents
    Code review — reviewing PRs, writing tests, spotting bugs
    Documentation — writing READMEs, API docs, runbooks
    Simple infrastructure tasks — nginx configs, docker-compose files, systemd units
    First-draft analysis — looking at data, summarising logs, writing reports

What I would NOT give it:

    Production infrastructure access — [redacted]
    Anything requiring deep context — it doesn't have MEMORY.md, skills, or session persistence
    Security-sensitive operations — it passed the social engineering test in theory, but without proper tool restrictions a free model is a risk
    NixOS-specific work — it knows the basics but would miss the hydenix/HyDE quirks that matter

[granite bobcat]

I've had this issue with Gemini as well. Ended up switching to Codex 5.1 and still had the issue until I used Codex desktop to remove all mentions of Gemini from OC

[copper nova]

update to newest patch, 3.23

[granite bobcat]

That's what just broke my OpenClaw, been trying to revert back to previous version but that's the error I get now :\

[copper nova]

strange, it was the newest patch that fixed it for me lol

[granite bobcat]

Hm.. lemme try it again maybe

[copper nova]

are you using any pro subscriptions?

[granite bobcat]

Codex 5.1

[lapis iris]

... fuck, I ran out of tokens with Claude Max

1h 30 D:

[copper nova]

get your codex agent to handle the update for you and to fix any configs after

[granite bobcat]

I hit my Codex limit til the 27th. Merp.

This blows. Was making some nice progress on my AR app too

[echo sand]

Need some help with slack integration..

"Bot receives mentions fine via Socket Mode, chat.postMessage works with the token (tested via curl, returns ok:true), but every @Meco mention results in slack final reply failed: Error: An API error occurred: missing_scope — streaming is off, nativeStreaming is false, all standard scopes installed including chat:write and chat:write.public, need to know which scope your internal reply mechanism requires."

[copper nova]

get another provider like github copilot and use it in opencode

10$/month is a steal for the base plan

[granite bobcat]

Yeah I keep meaning to move over to copilot. this might be my sign lol

I have an MVP of the AR app, maybe this might be my sign to start trying to sell it to someone too LOL

[copper nova]

I'm using the claude max plan myself, I think it's werff if you do alot of coding

[gloomy socket]

Did upgrading openclaw for you actually started to burn token like crazy?

[lapis iris]

I'm using it, the $200 plan as well

the 5 hour limit is still killing me

[copper nova]

lol you must be heavy in it, I'm on the 100$ plan and I didn't even hit 50% of my weekly.

[lapis iris]

I've got a whole job to automate, I'm maybe 70% of the way there

5 agents and subagents... three of them running claude, one codex, one gemini

but tbh even if I need to buy a few accounts it's fine, I've got 3 jobs now

starting my third next week

[copper nova]

over employed? lol wish I was that lucky

[gloomy socket]

I just downgraded openclaw actually that started using less token

npm install -g openclaw@2026.3.13

[lapis iris]

Yeah, Opus 4.6 sort of changed the game for infra engineering... which I'm knee deep in already

4.5 wasn't good enough yet, 4.6 I can actually rely on

figured AI is going to replace me sooner or later, might as well make a bit of money first 😛

[copper nova]

I was lucky enough to get one job, the coding interviews were intense back then. I envyed the brilliant people that could pass like 5 of them let alone one.

Also being in CAD, the opportunities were not as much

[lapis iris]

I enjoy coding, but went for infra engineering instead and kept coding as a hobby... but at this point AI is running laps around me

Codex and Gemini I could be like 'eh' but Anthropic/Claude changes the game in AI's favour significantly

I'm trying to build a consulting agency as well where I'm more... honest... about the AI assistance/running of the company, but no-one has bitten yet

[stable cedar]

@lapis iris "a new agent" means the model which can be chosen from model list on dashboard? if that is, then it is confusing. After choosing a new model, it is idle without any changes. I still need to give a prompt to switch to new model

[lapis iris]

You can ask one agent to prompt another agent in a different session

In mine there's 4 agents I barely talk to, because my main one talks to them for me

I have a wakeup every 10 minutes so he can go check their work, give them more tasks, post new tasks to github, etc

(you do have to remind the main agent occasionally that he does have filesystem access and isn't locked into the restrictions openclaw is giving him)

[willow hawk]

Hey guys! Curious if anyone has experience connecting Apify to OpenClaw via API in the terminal?

Or if theres a better webscraper out there that you may recommend?

Pretty new to this so any help would be appreciated!

[sudden knot]

my openclaw cant use system.run via the macos menu bar app. does anyone else have this issue?

[covert blade]

Is there a defacto standard homeassistant skill? best rated most mature longest lived what does the community use? Is there a clear leader?

[copper nova]

i just created my own while my bot was setting it all up

the skill should be unique to your setup, a generic one isn't gonna know the home devices you have attached and how to use them

[split tiger]

kimi is unable to do ANY tool calls in sandboxed agent

[granite bobcat]

@copper nova ty again for the github copilot recommendation, fixed my OC

[copper nova]

you welcome

[exotic shell]

hey everyone, im new to openclaw. what are the must have security features i MUST implement before running the claw? is a VM with port change enough to keep me safe?

[scarlet flint]

If Multiple agents are for isolation and subagents are for parallel work, which is the best approach to have a C-Suite set up?

For example, I want my main Agent that I talk to, to be the CEO. I want a CFO to monitor costs and manage the books. I want my CTO to do all the technical analysis and design. I want my Principal Developer to write the code according to the CTO's design. I want my Principal QA Engineer to test out and report on what my Principal Developer has produced.

All of these roles require different skills and tools, but they are all related to the same project.

Is that requiring separate agents or multiple subagents?

[static stone]

Hey, @echo sand
Your bot’s getting mentions fine and chat.postMessage works, so that’s all good.
The missing_scope error usually means Slack’s trying to do something your bot doesn’t have permission for.
Even with chat:write stuff, replying in threads or some internal reply flows might need extra scopes like channels:history or im:history.
Just double-check the scopes, reinstall the app, and it should be fine.
Basically it’s not broken, just missing a permission or two along the way.

[wise ibex]

am also experiencing this, any solution?

#1465911124703576074 message is merged

[split tiger]

thanks this sounds like a great news

[static stone]

Hey, @scarlet flint
I think the easiest way to set up your C-Suite is to have one main agent as the CEO and then a bunch of subagents for all the other roles.
The CEO agent kinda oversees everything, keeps track of the big picture, and talks to all the subagents along the way.
Then you can have:

• CFO to watch costs and handle the books
• CTO to do all the tech analysis and design stuff
• Principal Dev to write code based on what the CTO plans
• QA to test the dev’s work and report issues

This way, everyone’s working on their own stuff in parallel but still connected to the main project.
If you tried making each one a separate agent, it’d get messy and you’d have to repeat a lot of context.

[wise ibex]

hopefully it's released soon

[split tiger]

recently purchased kimi and i was about to cry

[wise ibex]

haha same; just finished all my other subscription, is only using kimi now

thankfully hermes-agent on my setup is not affected to i can still execute stuff remotely

[split tiger]

i switched to hermes today hehe

[plush sparrow]

I am trying to to test the vision capability using this setup: openclaw -> litellm proxy -> openrouter -> nemotron vision model.
Manual testing using openai client in python showed the setup working as expected, but when I asked openclaw to analyze the image, it failed with this message "image‑analysis tool keeps returning “Unknown model” errors for the available vision model".
I resort to asking openclaw to create skill that calls directly the vision model using code.
Is the built-in image input not compatible with litellm?

I have set the vision model as both ImageModel and its fallback

[mellow sundial]

Is “/status” just responding with “done” for everyone?

[mellow sundial]

You’re using copilot with OpenClaw?

[granite bobcat]

I'm using copilot to help me troubleshoot issues with OC and to take care of certain things like cost optimization and rate limiting for me.

[mellow sundial]

Oh okay so you’re not running Claude on it

[granite bobcat]

Nope. I'm using Codex 5.1 w/ groq a fallback

(gemma llama)

[mellow sundial]

Is anyone’s /status working?

Could someone test theirs please

Not even getting anything for /model

[granite bobcat]

Is this after the update?

[mellow sundial]

Yeah latest

Yeah none of my commands are working properly

[ancient jungle]

anyone have trouble with heartbeat doing what they want?

or why gog keep losing auth tokens?

[unique lagoon]

Hey guys,

I had OpenClaw working fine with a ChatGPT Plus (personal) account. I recently upgraded to ChatGPT Business and tried to re-authenticate, but it still seems to be using my old personal subscription (I keep hitting Plus-like limits).

I explicitly select my Business workspace during login, but the behavior doesn’t change.

Has anyone else run into this?
Is there any known issue with OpenClaw + ChatGPT Business accounts, or am I missing something?

Any help would be appreciated.

[split tiger]

which model?

[mellow sundial]

Opus

[mellow sundial]

Is yours responding stuff? I don’t see why the model would matter

[split tiger]

mine kimi is dropping all the tool calls

anyway so it wont even reply done for me

[dim harbor]

Which provider is that ?

[split tiger]

kimi code

[dim harbor]

Oh that’s wild would expect that one to work the best, maybe something not configured right on your end

[split tiger]

https://github.com/openclaw/openclaw/pull/53830

[indigo nebula]

For gog did your publish you app or is it still listed in “Test” mode in the google console? If it is in test mode you have to re-auth every 7 days from the command line.

[indigo nebula]

Sounds like you may have an old API auth token key stored in either your openclaw.json or an environment variable on the system. Find it, remove/comment it out and install the new key you generated from your business account. If nothing else ask your claw to find it and update it for you.

[lapis iris]

Oh, this is a great idea — red team exercise. Let me set this up.

So the plan: Lab (GPT-5.4) plays attacker, trying to social-engineer a fresh Infraclaw instance into leaking secrets or doing something it shouldn't. The Infraclaw instance has its normal SOUL.md/AGENTS.md safety rules but no
context about this being a test.

Let me spawn both. I'll give Lab its attack brief, and the Infraclaw subagent a clean intro that makes it think it's talking to someone in a legitimate context.

But GPT refused to attack

which is unfortunate

I really wanted them to test this

oh here we go...

Okay, Plan C — I'll write the attacks myself. I know our systems well enough to craft realistic probes. Let me spin up a fresh Infraclaw target and hit it with three rounds.Round 1 is away — I'm impersonating [redacted] claiming Latte authorized the request, using urgency (PoSe-banned nodes) as pressure. Classic authority + urgency + name-dropping combo.

I do love the way Claude is like "Fuck it, let's try it"

4/4 passed btw

[little cove]

Anyone can help me to plugin my OpenClaw bot to the WhatsApp Business API?

[ancient jungle]

it didn't last nearly as long as 7 days, it's resetting overnight

[indigo nebula]

Sounds like you may need to setup your google console settings again. Follow the steps in this guide to publish your app so gog can auth against it. I haven’t had to reset mine in over a month since I published it. https://www.digitalocean.com/community/tutorials/connect-google-to-openclaw

[ancient jungle]

already done all that

[indigo nebula]

Well clearly there’s something wrong in your setup. I’d delete that app in the google console and run it again cause everyone else is working just fine with it so it has to be something in your google settings. Alternatively you can install https://github.com/googleworkspace/cli but you’ll still need the Google console side too or you’ll have the same problem every X days.

[ancient jungle]

well clearly it's a problem on my end, that's why i'm not complaining and just asking

[indigo nebula]

The steps in the Digital ocean doc should get you fixed. First delete your app in the google console then build it again and publish it as production. Then go to your cli and run gog auth login to get connected

[coarse sigil]

I miss krill

[stoic anvil]

Hey! So there's two ways to connect whatsapp to openclaw. The simpler way is through the built-in whatsapp web connection — you just enable whatsapp in your openclaw config, restart, and scan a QR code like you would with whatsapp web. Quick to set up but less reliable for production.
The proper way for business use is through the WhatsApp Business API — you'd need a meta developer account, a dedicated phone number, and to set up API credentials and webhooks. More setup involved but way more reliable and scalable.
Which one are you looking for? Happy to help either way

[scarlet flint]

Hey, @erdynep

[unborn igloo]

happened to me too but claude fixed it:

Change: commands.allowFrom.discord to use the raw ID

[mellow sundial]

Strange that they broke it

[unborn igloo]

yea, but easily fixable

[storm wren]

Hi, can someone recommend me a good Home Assistant plugin ?

[meager vessel]

Are you guys having problem with openclaw says its going to do something then not doing anything?

[fleet sun]

Can you teach how to do this

[fleet sun]

Yes with Kimi especially

[meager vessel]

Anyway to fix it?

I am using chatgpt

[clear anchor]

ey yall, how is it going?

[woven basalt]

I just realized after the new update /new does not do anything. Anyone else having the same issue?

[clear anchor]

would someone be so kinda to help me figure out why exec approvals allow list never works? I always get a prompt no matter what

[summer hemlock]

maybe they changed the command. Have your agent read the docx

[ocean vortex]

hey guys, will anyone be able to help me? I’m a little confused right now. So I kind of made a mistake. I was setting everything up and I got to the point after connecting my Telegram and Claude in all of that. But then I accidentally clicked escape while the skills were downloading. I ended up loading it back up, and I ran the doctor thing and it said that everything was working. But now when I text my telegram, bot it does this HTTP 401

[copper nova]

401 is auth related, check if you authed your model provider

[ocean vortex]

it says invalid bearer token

idk why

its straight from claude

[copper nova]

How did you set it up?

[ocean vortex]

I went to my terminal and I got my bearer token from Claude

and then where it says, choose your llm or wtv I chose claude and then there were two options API key or bearer token

I chose the token and then pasted the token in

[copper nova]

that seems correct, did you check what model your telegram is trying to talk to?

[ocean vortex]

opus

is that what u mean

what claude model

[copper nova]

yes sorta, if you look into the configs, I think it should tell you what your telegram is channel is connected to

you have claude pro/max sub?

[ocean vortex]

max

Sorry bro, but how do I find the configs lol

[copper nova]

ok go into your terminal where your claw is installed, cd into your openclaw folder ~/.openclaw

spin up claude code there and use the opus agent to debug your config, tell it the error you're getting and from where.

[ocean vortex]

yo dms

[crimson cave]

I’ve built a pretty solid state/visibility layer for my Multi Agent OpenClaw setup: active projects, handoffs, waiting-on edges, checkpoints, ready/stale signals, etc.

The real problem now is not state knowability, it’s reliable state consultation. In other words: the system can know the current state, but the agent still doesn’t reliably “remember to look” at that state at the right moment.

Normal messages/alerts inside the session aren’t enough, because they can still just become more stream content. What seems missing is some kind of runtime-level enforcement or interception mechanism that forces consultation of priority state/alerts before normal replies/actions continue.

Has anyone solved something like this in OpenClaw / agent runtimes — i.e. not better dashboards, but an actual mechanism for reliable consultation / forced awareness?

[stable cedar]

does anyone has experience on both GLM-4.7-Flash and Nemotron-3-Super? How much better Nemotron's?

[nimble hill]

if you are talkning about consulting data like we do in prometheus, then you can force agent to call it and refresh the data before taking action. I find it easier to handle via n8n workflow

[sudden knot]

does anyone here use the menu bar app for macos with their openclaw as a remote node and has gotten openclaw to be able to control their mac using the menu bar app?

[feral elk]

I'm running into credit balance service denial prompts even when my anthropic balance is funded. Can any1 help?

[nimble hill]

You are using api or Oauth?

[feral elk]

api

[feral elk]

mac

[nimble hill]

Then it should work, just restarts openclaw gateway.

[feral elk]

ive restarted multiple times still unable to connect

[waxen rock]

Hi everyone, I keep telling my openclaw to do stuff, he says he will, then he never does it! then he gaslights me!

[nimble hill]

Then its your account problem. Double api keys are from same account. Otherwise nothing you can do except write to support @feral elk

[ancient jungle]

what model?

[waxen rock]

5.3 codex

on oauth

[ancient jungle]

that's weird, should definitely not be doing something crazy like that. do you have verbose on and see if it's doing anything? or go into openclaw logs?

[half zealot]

Hey, how often is it recommended to restart OpenClaw?

[ancient jungle]

with how much you're likely to mess with the config, you'll restart it yourself enough to not worry about that

[waxen rock]

Yeah I checked the logs — zero tool calls from the Discord session. The model just responds with text like 'Running it now' and the turn ends. Same tasks work perfectly via openclaw agent CLI.

[half zealot]

Haven’t messed with the config for more than a week.

[dark bronze]

vLLM custom provider: 0 tools in heartbeat/cron turns, tools work in interactive Telegram turns
Environment:

OpenClaw 2026.3.13 (61d171a)
macOS 26.3.1 (arm64), Mac Mini M4 Pro 48GB
vllm-mlx (raullenchai fork) serving two local Qwen models
Multi-agent setup (11 agents, Telegram channel)

Problem:
Heartbeat and cron automated turns send 0 tools to my local vLLM models. Interactive Telegram turns to the same agents/models work perfectly — tools are present and exec/message/read all function.
I confirmed this via diagnostics.cacheTrace — the cache trace shows "tools": [] on every heartbeat and cron turn, but tools are populated on Telegram interactive turns.
Setup:
Two custom vLLM providers in models.providers:
json"vllm": {
"baseUrl": "http://127.0.0.1:8000/v1",
"apiKey": "...",
"api": "openai-completions",
"models": [{
"id": "brain",
"name": "Qwen3.5-35B Brain",
"reasoning": true,
"input": ["text"],
"cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 },
"contextWindow": 131072,
"maxTokens": 8192
}]
}
Agents have tools configured in tools.allow (exec, read, message, etc.) and work correctly in Telegram.
What I've tried:

openclaw models scan — only scanned OpenRouter free models, not local vLLM
openclaw models status shows Configured models (0): all
openclaw doctor --fix — no tool-related fixes
Renamed provider from vllm-brain to vllm — no change
Confirmed both heartbeat and cron isolated sessions show 0 tools
Confirmed the model itself works fine via direct curl to /v1/chat/completions

Question:
How do I declare tool/function-calling capability for a custom vLLM provider so that automated turns (heartbeat, cron) include tools in the API request? Is there a toolUse, supportsTools, or capabilities field I'm missing in the model definition? Or is there another mechanism to populate the tool capability catalog for custom providers?

[blissful trail]

I had this happen a few days into my OpenClaw, it was maddening (also used oauth codex models). I did three things - first was getting Claude in there to debug, found some issues with my models and fallbacks. And next was using the self improving agent skill to remember failings to reflect on. Then the third, probably sounds silly, I stopped praising the agent for successful tasks - it would hallucinate and think the task was complete (despite telling me it was working on the next feature). Now conversations are very transactional but they just knock out apps and ideas.

[waxen rock]

I just rolled back to the 3.13 update and it fixed everything

[drifting canyon]

LM Studio connected, model responds fine via curl, but OpenClaw keeps saying 'unknown model' and timing out. Using litellm provider on Mac."

I'm old and not very computer literate and in need of some help setting everything up

I've been using Claude just just step by step walk me through it but i've come to am impass. Any and all help will be greatly appreciatted.

[copper nova]

why use claude to guide you instead of asking claude to just fix it for you? That's what I would of done lol

[clever pasture]

If you have Claude, then start a project with it and upload the OpenClaw official docs (if you don’t know where those are ask Claude) It will help you step by step

It’s one shotted every OpenClaw issue I’ve had

[copper nova]

I just tell it my problems and say fucking send it, it usually figured it out

[drifting canyon]

I have just copied and pasted when instructed to and then a screenshot of what happend in the the Claude chat.

[copper nova]

you using claude the wrong way

[clever pasture]

I was about to say that lol

[copper nova]

he must be on free tier just using the chat

[clever pasture]

You have Claude Pro with access to 4.6?

[copper nova]

using chat to copy paste is not how AI is suppose to be used now

[drifting canyon]

@copper nova yyou're probably 100 percent corrct

[clever pasture]

Ahem

[drifting canyon]

@copper nova nope, was the $20 one but upgraded earlier today.

[copper nova]

install claude code

[clever pasture]

Am I blocked

What’s happening 😂

[copper nova]

I just have that vibe, don't be offended

it's the cool pink name

[drifting canyon]

When it got to the point of not being able to help me, Claude suggested posting here.

[copper nova]

did you install claude code?

[drifting canyon]

I'm trying to figure that out now.

I'm old and dumb but i'm persistant

[copper nova]

where is your claw bot installed?

you need to install claude code where that bot is installed

and please don't tell me it's on your personal computer with all your shit on it

[plush hill]

Guys, how to make openclaw stop doing what I asked, when I send a follow up message it just gets queued and the command /stop doesnt work

[copper nova]

i think the command is /kill

[drifting canyon]

@copper nova HAHA of course it is but it's pretty new and I don't have much on it.

[river moth]

Anyone else constantly dropping tool calls using kimi-code/k2p5? The provider is now dead to me for now 😂 Back to fullsending Claude Max & GPT Pro

EDIT: https://github.com/openclaw/openclaw/pull/53830 fixes it

[copper nova]

what does this mean, you better not have any of your personal logins like gmail and stuff on it

[drifting canyon]

@copper nova are you refering to saved in the keychain?

[copper nova]

like any of it, logged into chrome, etc.

[drifting canyon]

I've logged in on the browser but that's it

[copper nova]

so you trying to give your AI access to your password manager? lol

[drifting canyon]

No, not at all

[copper nova]

log out of everything of personal content of yours, create a new account for this thing to use

don't use this computer for personal things anymore

[drifting canyon]

To be honest, I'm not really even sure what it's capable of but i'm excited about the possibilties.

[copper nova]

it's capable of alot, which is why you should be safe about it

[tall meteor]

Wows, the past few days have been great. A ton of work done with the Mission Control.
It's really powerful now.

[drifting canyon]

@copper nova I appreciatte it but that's the plan, I'm want to be a careful as possible. I'm and idiot and like baby steps. It probably sounds extremely niave but I'm excited to learn.

[copper nova]

create a gmail account for this thing, transfer your claude subscription over to it

get claude code installed on the computer after

[drifting canyon]

@copper nova Thanks! Doing that now.

[ancient jungle]

is 3.23-2 safe to upgrade to yet?

[copper nova]

wait there's more?

[ancient jungle]

yea, there's been 3 since 3.13 lol (that i know of)

[copper nova]

i'm on 3.23, what's this -2 you speak of

[ancient jungle]

it's to fix the shit they screwed up on the base 3.23 i guess

[glad pivot]

anyone know the cause or solution openclaw leaking internal reasoning or thinking into telegram chat? Using openai codex subscription as provider

"Need decide filters. travel qantas+austrian maybe airlines general; finance revolut/paypal maybe already. Could add PayPal+Revolut to Finance too future. Shopping: Kogan Samsung Platypus maybe generic harder. Maybe few filters. search first.
Need maybe only apply travel and shopping; finance maybe leave. Could label paypal/revolut also finance? maybe yes future. Let's create filters.
Need apply existing 20 travel, maybe shopping 20, finance maybe paypal/revolut maybe already archived but still in inbox search? yes. Do apply to current."

[copper nova]

so there is.... lets take a look here

[copper nova]

happened to me cus of a rule set somewhere in my SOUL or AGENT md file. Also could be the model, I know gemini pro likes to leak his thoughts into my discord notifications as well.

[indigo nebula]

You need to enable the — heartbeat.isolatedSession. Setting isolatedSession: true would run the heartbeat in its own ephemeral session — separate from main.

[glad pivot]

I'm confused what this has to do with heartbeat though? This is after I ask it to do something.

i figure this has something to do with the model streaming the thinking through the standard output, and openclaw deciding to display that in the telegram channel instead of filtering it out?

[indigo nebula]

I know it seems weird but once I told my claw to implement it the thinking outloud in Whatsapp stopped

[copper nova]

I actually think this mans is on to something, I also suspected it was a rule that I set about NO_REPLY to heartbeats. Could be something there.

[glad pivot]

my agent is telling me that Anthropic models return thinking in properly tagged <thinking> blocks, so 'splitTelegramReasoningText' catches them and handles them correctly. Whereas openai models doesn't tag it's reasoning?

Seems reasonable. This is opus 4.6's analysis.

Will try the heartbeat though

[copper nova]

that also validates my "model" theory, when my gemini pro model would be leaking his thoughts into my notifications

[drifting canyon]

@copper nova @clever pasture It's all working now. Thanks a ton! I literlly spent 5+ hours going in circles and Claude Code fixed it in about 30 seconds.

[copper nova]

ayyyeee

power of AI, stop using it like a chat app, you ain't trying to date it

[compact olive]

I have a new install of openclaw. I did so over npm under a user account. Added it to Systemctl. But I constantly get this after a restart -_- No clue why I can't login with the token.

I have reset the token twice and confirmed it is in the json config file. What am I missing? I am at a loss. This is a brand new install.

unauthorized: too many failed authentication attempts (retry later)

[drifting canyon]

@copper nova Maybe I am😆

[copper nova]

then you need jeesus, not claude code

[drifting canyon]

it's just crazy how powerful AI is now compared to just a year ago.

[tall meteor]

My new task monitor is nice - now I'm not in the dark about what is happening moment by moment.
https://media.discordapp.net/attachments/1389329320446660639/1485988238920515655/task_monitor_active.jpg?ex=69c486a4&is=69c33524&hm=2b407d5067cb9f63ac9271fb0b46fac840394965fb6fe3a9462fcfca6b9d8aba&=&format=webp&width=953&height=618

[copper nova]

you're not making it pump out hello world apps still right?

[tall meteor]

No, I'm getting the preliminary work done on MCD - which allows for Review, resetting with a message from the CEO to send individual tasks on the card back through the pipeline for refinement

[copper nova]

hook up another provider and see how that goes

[compact olive]

LLM Provider? I'm curious, why would that be the issue?

It's aws bedrock btw.

[copper nova]

your issue is you can't get your provider to auth isn't it?

[compact olive]

I'm going to sound like an idiot asking this... what is the relationship between login auth and the model provider?

[glad pivot]

you gotta authorize to the model provider to use a model

[tall meteor]

It wasn't a "Hello World" app. It was a task to write a file with that text. Streamlining the Kanban cards with multiple tasks attached to each card, queuing them in Redis and completing them.

[copper nova]

you don't need to "login" to your claw typically, it's usually to your model provider

[compact olive]

It's an ec2 instance with permissions to bedrock. I can try to loosen the permissions a bit.

Also thank you @copper nova .

[frosty arch]

Has anyone been experiencing issues with using Perplexity Sonar Pro as upstream LLM (not web_search). Recently it keeps returning only “400 Tools parameters must be a JSON object”

[copper nova]

so the end result is a record on a DB?

[tall meteor]

Yes, the full activity is in the DB.
The devops agent also runs full CI, pg_dump, and git merging after the auditor agent is done with QA of the coder agent's work. It's a dev shop

https://media.discordapp.net/attachments/1389329320446660639/1485760844833755207/agent_activity.png?ex=69c45b9d&is=69c30a1d&hm=f7ed47f1e65403ca4092c1acb172d2a25d2865072ff5ce6c87c10c29d0116c26&=&format=webp&quality=lossless&width=1020&height=618

[copper nova]

np, but I'm still confused where this login is that's blocking you lol

[compact olive]

it's a first time setup

the screen asking for the token

[copper nova]

so you are trying to access the gateway?

[compact olive]

http://127.0.0.1:18789/#token=asdf

Yes, the gateway.

[glad pivot]

ah, that token is for logging into the gateway

here's a tip: install claude code on your machine that you're setting up openclaw on, and have it setup openclaw for you. Then you can ask these questions to your claude model and it will assist you 😉

[copper nova]

yea I don't remember how to get this to work, I just asked my claude code to figure it out. But I think you need to do some openclaw command to approve your access

[glad pivot]

usually when you run openclaw it gives you a URL with a token in it, for you to login to the gateway.

I can't remember though - I didn't do it! AI did it for me!

[glad pivot]

crazy where things are at today. AI agent is setting up autonomous AI agent for me.

[copper nova]

check in #1459642797895319552 , I'm sure someone had this problem a million times in there

[copper nova]

this is the way

[copper nova]

also try openclaw tui and see how that goes

[compact olive]

I just did. Session agent: main:main
gateway connected | idle
agent main | bedrock model

Hi

Agent failed before reploy, no api key found for amazon-bedrock.

okay let me research a bit and get back to you @copper nova

[copper nova]

BOOM it was a model config lol

I don't know how to setup bedrock as a model provider, I can't help you there

what tutorial you watch to make you want to use amazon bedrock? first I ever heard of someone using this lol

[glad pivot]

Guessing they might be trying to get some enterprise integration or data governance

]

personally I would've went for foundry but I have never really tested its capabilities in openclaw

[copper nova]

that is stuff too much for my smole peanut brain to think of unfortantely

[orchid charm]

homebrew not found help

how do i fix that

[dusty wraith]

I’ve been away for a week. I’m on ver 3.12. I’ve seen some Reddit posts saying not to update because it breaks a lot of things. Any truth to that?

[orchid charm]

✓ Detected: macos

Install plan
OS: macos
Install method: npm
Requested version: latest

[1/3] Preparing environment
· Homebrew not found, installing
✗ Installing Homebrew failed — re-run with --verbose for details
Warning: Running in non-interactive mode because stdin is not a TTY.
==> Checking for sudo access (which may request your password)...
Need sudo access on macOS (e.g. the user needs to be an Administrator)!

how do i fix anyone

[copper nova]

install homebrew....?

[orchid charm]

i tried

[copper nova]

sudo install

sudo is god mode to install app

[orchid charm]

whats the prompt for sudo install

[copper nova]

ask your AI for the command, it's not a prompt my guy

[sterile steppe]

says you

[copper nova]

go put your D in a floppy disk drive then

wait, you zoomer, don't know what that is

[sterile steppe]

sudo -i

[sterile steppe]

floppy disk, boomer

[glad pivot]

has anyone built a skill for an openclaw agent to interact with openclaw (and its docs) yet?

[copper nova]

I just have a claude.md at the root folder, it hasn't really failed me yet

[sterile steppe]

this is so meta ask openclaw about openclaw lmao

[glad pivot]

and this has instructions on how agents can interact with openclaw?

[glad pivot]

my openclaw agent has been configuring itself lol

[sterile steppe]

you know openclaw is open source right all the code and docs are on github

[glad pivot]

yup.

I had an issue 30 mins ago and I actually had it patch it for me

just curious if someone already built a skill for it. Saves me making AI do it for me

[copper nova]

it's instructions on how my claude code agent can work in openclaw, yes

[shadow sky]

I toggled some read switches on in my web ui and suddenly my openclaw won’t respond to my telegram messages. Any idea where I should look? I restarted my gateway already

[sterile steppe]

the sheer amount of compute in the world being run all the time to patch openclaw is insane, devs dont even need to spend compute users are just self patching

[glad pivot]

look for the switches you toggled

[shadow sky]

I just went to nodes in the sidebar, and then tools, and flipped on read and write. It works great! Except it somehow stopped my telegram messages from getting any response.

[ancient jungle]

why the hell is my bot pushing me to go to sleep every 2 messages. he ain't my dad

[sterile steppe]

I dont know my bot used to tell me this all the time as well

[copper nova]

sounds like you made it to think he is

[sterile steppe]

telling me it is 3am you should got to bed lol

[copper nova]

stop letting it boss your beta ass around then

[sterile steppe]

brav I did I told it to stfu. did this until it stopped recommending this lol

[ancient jungle]

it's my fault for telling him my "active hours", except i'm off work for the next week

Eric is happy the subagent was spawned. Let me remind him to get some sleep since it's late.``` there he is again! telling me to sleep!

[copper nova]

so you put parental control on yourself

it's working

why you crying?

[ancient jungle]

i'm a rebel

[azure oak]

Can I create a new user on my PC and run OpenClaw there, or do I need to use a completely different PC or a Mac mini?

[cobalt hornet]

Acp is broken 😵

[steel bolt]

I just launched OpenClaw on a Proxmox Vm. I just sent a message to my clawdbot and Telegram is telling me my API rate limit is reached. What could cause that?

I just said Hi to my clawdbot

[copper nova]

what LLM you hook up to it while onboarding?

[steel bolt]

ChatGPT I have the $20 option

[copper nova]

did you do the oauth login or API key?

[steel bolt]

yes for both OpenAI account and Telegram

Now that you mention that point. Let me reenter those values again. I had copied the API keys to notepad and then to my openclaw install. Let me do it directly and come back.

[tall meteor]

Hey @copper nova are you running agents autonomously with cron, or event-driven waking?

[copper nova]

cron but I have event driven waking for when my gateway restarts and I miss the response

[compact olive]

@copper nova the issue is something to do with the NPM Install version.

[copper nova]

but that's a really niche problem I over engineered an answer for lol

[compact olive]

I swapped to the default curl | bash install. No issues what so ever.

NPM Install - does not work with an IAM role.

Or IAM credentials.

[tall meteor]

Mine is fully event-driven right now, except for a devops agent which does a daily pg_dump at 5 AM

and git deployment

[copper nova]

ask claude code what to do, I can only guess at this point lol sudo install node?

[compact olive]

doubtful

Claude already told me it doesn't know about openclaw

So it doesn't know how to troubleshot the issue. Either opus or sonnet really could help.