#users-helping-users

1 messages · Page 29 of 1

velvet terrace
#

I'm looking for recommendations on a good TTS model i can run locally and easily plug into openclaw via the new Discord TTS feature. Thx

celest ermine
#

How to update Openclaw? And the release notes of the latest release said they made a browser extension that actually works. what is the official one?

#

(on mac)

trim estuary
#

hello everyone can you please tell me the cheapest model with file search functionlaity ?

echo seal
#

The model list on the ollama site is decent. I'd do a bit of searching there

brazen widget
#

am i being dumb? i can't seem to get tui to let me use any ollama models?

spiral fable
#

Help the child, thank you, sir. Here's what happened: My MiniMax lost connectivity, so I updated all the configurations, which meant the Telegram bot also needed reconfiguring. I spent the whole day trying without success. Openclo consolidated all the boot processes into one Telegram account, but no matter what token I gave him...

通过DeepL翻译 (https://dee.pl/apps)

pine parrot
#

[memory] qmd update failed (Error: spawn EINVAL)
help me

copper meadow
#

anyone had the issue on slack channels of context overflow when posting a file?! The file is stored in the memory of the session as binary and thus (i guess) is overflowing. I want the file to be index. Anyone got an idea?

[agent/embedded] [context-overflow-diag] sessionKey=agent:main:slack:channel:xxxxxxx provider=ironai/llm source=assistantError messages=5 sessionFile=/home/xxx/.openclaw/agents/main/sessions/xxxxxxxxxxxxxxxxxxxx.jsonl diagId=ovf-xxxxxxx-bbbbbbbbbb compactionAttempts=3 error=400 You passed 253953 input tokens and requested 8192 output tokens. However, the model's context length is only 262144 tokens, resulting in a maximum input length of 253952 tokens. Please reduce the ...
celest ermine
#

How to get the browser extension?

#

the browsers i can use are Safari, Atlas, Comet, and Opera Air

#

i want the official one

naive summit
#

⚠️ ❗

With Krill current absent, #1459642797895319552 is awfully empty. I have neither the time nor the knowledge to answer most questions there, so it would be great if a couple of people would jump in there to close the gap.

Thanks in advance!

mild fractal
#

What happened to Krill? When did he go offline? My guess is the api usage was insane and they had to take him down.

copper meadow
celest ermine
#

it looked official
i put in my port, it got my token correctly
but it says "attach to tab through toolbar icon"
what toolbar icon?
i can't find it on comet or atlas

#

i think they are talking abt pin extensions

#

but how'd u do that?

copper meadow
#

If you click on the extensions symbol in the browser there is an icon

celest ermine
#

oh ok found it on comet looks like it isn't there on atlas

#

also how to find the port my openclaw is using?

copper meadow
# 
$ openclaw browser status

🦞 OpenClaw 2026.2.19-2 (45d9b20) — Your inbox, your infra, your rules.

│
◇  
profile: openclaw
enabled: true
running: false
cdpPort: 18800
cdpUrl: http://127.0.0.1:18800
browser: unknown
detectedBrowser: custom
detectedPath: /Applications/Brave Browser.app/Contents/MacOS/Brave Browser
profileColor: #FF4500
naive summit
empty shore
#

Hello everyone. Anyone having issues with the latest update? I got locked out via browser and message and everything I try to do on the terminal complains that
google-antivravity-auth plugin was not found. Even if I try the doctor I get the same message and it won't run. The gateway seems to be up though but I have no access.

gritty breach
umbral cobalt
knotty barn
#

Thats rite something about TOS for antigravity

celest ermine
#

I'm having trouble with the browser extension. I've pasted my gateway token exactly from the dashboard, yet it keeps on saying 'Gateway token rejected. Check token and save again."

empty shore
high heron
#

and dont update till they fix this

high heron
high heron
knotty barn
#

I have to do the update later see if it fixes tools and telegram crossing fingers

vague prism
weary basin
#

Wondering if anyone has used their chatGPT sub to route only heartbeats to 5.2 codex via oauth?

I set it up last night and it works fine if I'm setting codex as my primary model -- so I know it works.

But when I set it for heartbeats, it either does nothing (I can see in the terminal heartbeats are firing with no errors showing in the terminal but there's no activity -- nothing in the tui) OR twice it threw an error:
"No tool call found for function call output with call_id callsessions_spawn:0."

I feel like it's attempting to spawn a sub agent during heartbeats even tho that's not necessary maybe?

I can't figure out what's going on. Has anyone used their oai oauth for anything other than primary model? If so, is there something special that needs to be done?

Heartbeats have also been occasionally looping for me since the 2.19 update (and I think I've updated twice since?) so maybe that's the culprit?

I'm at a loss tho. Thanks, y'all!

shy ether
#

What even are heartbeats?

vague prism
grim plover
#

i need help about tokens. it says Main: 109k / 205k tokens, when will it renew a month or a day?

slim lion
cyan onyx
#

also plz dont mass tag like this

slim lion
#

ok sry

celest ermine
#

I'm having trouble with the browser extension. I've pasted my gateway token exactly from the dashboard, yet it keeps on saying 'Gateway token rejected. Check token and save again."

knotty barn
#

Im running 2026.2.22 i hear there's a 2026.2.23 that dropped with some improvements

#

I think the gateway thing is one of the issued they fixed

odd basin
#

Yo in the latest release there is an update for the Kimi search , anyone got that working?

knotty barn
#

Something that cought my eye this morning

knotty barn
gleaming saffron
#

anyone having issues with connecting to ANY chat platform? Be it Discord, Telegram, Whatsapp...

sharp sun
#

hey guys, I’m seeing a Discord reply-delivery issue on OpenClaw 2026.2.23: the model generates responses (visible in dashboard/session), but messages don’t appear in Discord. both DM and group chat are affected, after /new the first greeting is delivered, but subsequent normal replies disappear from Discord while still existing in session logs. Subagent completion notifications still arrive normally. any solution?

modest fog
#

heey, hows everyone doing? i have an issue.. about 4 days ago my openclaw keeps streaming alot, sometimes 1000minutes+, which i notice when i wake up. does anyone know whats going on and how to fix my issue?

sharp sun
umbral cobalt
#

Every time i update Openclaw it breaks...

#

this jsut started after version 21... I have to go through like 30 min of troubleshooting and multipple gateway --force to get it working again.

tacit halo
#

does anyone gets the previous version when trying the Docker version?
Would like to run in in doker for some extra security but...
[gateway] update available (latest): v2026.2.23 (current v2026.2.17).
cant open the web ui it was fixed at latest release it seems but cant get that

did some troubleshooting with gemini :))

it is clear: the image on GitHub Container Registry has not yet been updated with the new :latest tag, even though the source code has advanced. Your version (v2026.2.17) is locked to 127.0.0.1

have you run into this?

gleaming saffron
umbral cobalt
#

Hmm. Im running mine in docker, and its giving me the correct version, you sure you dont have 2 openclaw instalations?

sharp sun
late nebula
#

i use qwen2.5-coder:30b. time to answer only for simple chat without tools and plugin is 40sec. i was traing gemma but openclaw saw that model can not support tools

tacit halo
umbral cobalt
umbral cobalt
sharp sun
#

my bot said: this points to a failure in the normal MESSAGE_CREATE reply delivery path. But he also doesn't know what caused it, I've had 5.3 codex working on docs and logs for hours, even checked the change log, but nothing, I just wanna know if anyone else is in the same situation as me or if it's just me😭

umbral cobalt
#

also run openclaw doctor --repair

#

itll fix some stuff that boinked from the update.

tacit halo
# umbral cobalt Mine running Docker -> openclaw <-> Docker PostgreSQL <-> Docker Caddy lol and m...

giving up on it, still the old version and can make it look outside the docker so I can use web interface, because its the wrong version

you pulled the image using docker pull ghcr.io/openclaw/openclaw:latest right? cant figure where am I going wrong if for others it works

2026-02-24 17:13:15.987 | 2026-02-24T15:13:15.987Z [canvas] host mounted at http://127.0.0.1:18789/__openclaw__/canvas/ (root /home/node/.openclaw/canvas)
2026-02-24 17:13:16.013 | 2026-02-24T15:13:16.013Z [heartbeat] started
2026-02-24 17:13:16.014 | 2026-02-24T15:13:16.014Z [health-monitor] started (interval: 300s, grace: 60s)
2026-02-24 17:13:16.015 | 2026-02-24T15:13:16.015Z [gateway] agent model: anthropic/claude-opus-4-6
2026-02-24 17:13:16.016 | 2026-02-24T15:13:16.016Z [gateway] listening on ws://127.0.0.1:18789 (PID 14)
2026-02-24 17:13:16.016 | 2026-02-24T15:13:16.016Z [gateway] listening on ws://[::1]:18789
2026-02-24 17:13:16.017 | 2026-02-24T15:13:16.017Z [gateway] log file: /tmp/openclaw/openclaw-2026-02-24.log
2026-02-24 17:13:16.034 | 2026-02-24T15:13:16.033Z [browser/service] No browser auth configured; generated gateway.auth.token automatically.
2026-02-24 17:13:16.035 | 2026-02-24T15:13:16.035Z [browser/service] Browser control service ready (profiles=2)
2026-02-24 17:13:16.104 | 2026-02-24T15:13:16.104Z [gateway] update available (latest): v2026.2.23 (current v2026.2.17). Run: openclaw update

fluid tinsel
#

Hey, I have GLM-5 as my primary and Gemini-3 as my fallback, however the Gemini connection appears to be timing out all the time. I can see usage queries on my account page but I guess the agent isn't giving it enough time?

storm ember
fluid tinsel
#

I can't find any relevant configs for it.

umbral cobalt
#

@tacit halo So Mine had all sorts of permissions issues, when I sudo its running as root, and it created a /root/.openclaw but the main instance was under /"MyUser"/.openclaw which when running commands caused it to get super confused, one was on one version, the other on another, so you could be having similar issue. run openclaw gateway status and look for a Split Brain error.

gleaming saffron
sweet moth
#

I'm writing a program right now to analyze issues per release, with an option to omit releases only "latest" for hours or days. Like many of you, I got completely borked on 2025.2.23 and doing a re-install after I find a good stable version. Unfortantely we can't trust "Latest Stable" so I'm not going to auto upgrade any longer and wait to find good releases. I hope they change their release archtecture to focus on beta/fixing issues before going to stable but they do have 3,800 open issues (mostly duplicate)

sweet moth
#

Curious what others are reporting on this beta ^^^

rough flame
#

Any way to make kimi free model of opencode run with openclaw

high heron
dusk geode
#

has anything been said when @sharp parcel will be back online?

high heron
rough flame
#

Please tell about connecting openclaw with opencode kimi k2.5

#

Free one

sweet moth
#

Does anyone know if I post a github link here if I will get banned warned?

spiral fulcrum
#

I added my bot to my Discord server. It works fine, but I can't use the slash commands. Whenever I try to use one, the bot says: "You are not authorized to use this command."

gleaming saffron
sweet moth
# gleaming saffron Thanks a ton!

welcome and I'll be done with a program soon to find us more stable versions. Others have told me .15 is stable and I have openclaws running .9 very stable/happy

storm ember
open saffron
#

This is my first time setting this up. I'm trying to use the NVIDIA Kimi-k2.5 model. In chat I keep getting the error 401 status code (no body). Has anyone encountered this? I've gone through the openclaw.json so many times make suring everything is correct. I've tried multiple suggestions found online to fix.

sweet moth
#

Version Type Reported Iss/Day Fixes Window Status

2026.2.14 stable 108 102.0 63 1.1d OK
2026.2.15-beta.1 beta 0 0.0 40 0.0d OK
2026.2.15 stable 205 106.5 40 1.9d OK
2026.2.17 stable 160 99.3 128 1.6d OK
2026.2.19-beta.1 beta 0 0.0 41 0.1d OK
2026.2.19 stable 30 15.2 41 2.0d OK
2026.2.21 stable 31 21.1 105 1.5d OK
2026.2.22 stable 31 29.1 164 1.1d OK
2026.2.23-beta.1 beta 0 0.0 52 0.0d OK
2026.2.23 stable 31 74.8 52 0.4d OK (latest)

open saffron
sweet moth
#

based on the chart I provided ^^^ which versio would you pick

celest ermine
#

I'm having trouble with the browser extension. I've pasted my gateway token exactly from the dashboard, yet it keeps on saying 'Gateway token rejected. Check token and save again.

sweet moth
#

looks like .19 is pretty stable

mild fractal
#

Anyone who managed to get image generation working?

I have setup the API key, confirmed its working, confirmed i have credits, openai-image-gen skill shows green in skills list...not sure what else to try. But when i ask my LLM to gen an image it says:

"I do not have access to the openai-image-gen skill. I will use the image tool instead, since I seem to have some access to it, although it errored previously."

#

Do i need to add an entry about it in tools.md or something or does it work out of the box once api key is added?

cinder marten
#

I have emails forwarded to an email address that my openclaw has full access to. But that email address also gets emails from accounts I've signed up for on behalf of my openclaw bot, such as github and google/gmail. How can I harden the bot from prompt injection attacks? Can I prevent the bot from reading emails that are not forwarded from me?

finite totem
#

#users-helping-users
I have been getting NO_REPLY from 3 seperate models from Ollama. Tried deepseek-r1, Qwen3-8b and Qwen2.5-coder:7b
Can someone explain what might be going wrong here?

high heron
#

Also make sure to set instructions in agents md

fringe dust
#

Hey guys, quick question.

I’m trying to figure out how to use OpenCloud to help me build and automate an e-commerce setup, but I’m a bit lost.

Ideally, I’d like it to:

• Generate product photos automatically
• Organize and upload product info
• Fill in Excel or Google Sheets with the data
• Basically handle as much of the backend automation as possible

I’m not sure what the right workflow is or how to connect everything properly. If anyone here has done something similar with OpenCloud, I’d really appreciate some guidance.

If you know a good YouTube tutorial, documentation, or any kind of walkthrough, that would be awesome.

languid bloom
#

Thank you! Haven’t heard of Alsa, but will check it out.

frosty coral
#

my openclaw wont accept to use any webacces is this normal? it is saying to me i cant accept youre log in for things like google. or am i doing osmthing worng? does it need api or can it just use normal login with password and email i have made a saprate email account

ivory spade
#

is anyone running openclaw with 0llama

plain onyx
#

Guys i told my main agent to create another agent but it didnt do it well then i created agent by - openclaw agents add <name>, and then i configured it with new telegram token, but after that my main agent got disconnected cuz of configured overwrite ig, then i run openclaw tui, and ask ai to fix this, it fixed but it took so much tokens, so is there any perfect prompt/method which i can give to my current main agent and it can create another properly separate agent without messing anythings ?

wary osprey
zinc mica
# plain onyx Guys i told my main agent to create another agent but it didnt do it well then i...

I, personally, have only had good results after a ton of refinement and tinkering on my main agents understanding of what my goals are and even then I have some additional steps I take to refine it after. One thing I’ve been really happy with is creating a gpt soul architect that I fed with background on my architecture and vision and used to refine souls. It’s not a blindly trust tool either but it’s very good for further reworking my main agents draft.

I also operate my system with a lot more restraints than others, by design. So there might be a more plug and play work stream out there.

small jackal
#

can someone pleas tell me which openai models are currently available with openclaw when i use Oauth ? i see only 5.3 codex

agile copper
#

Are there any good resources for how to make agents more proactive with their assigned roles or how to communicate with one another?

plain onyx
small jackal
#

i tried but it says there is only 5.3 codex ...i think my config is broken or something like that

#

i think codex has just messed up that file .../model just listed 2 model at all ..haha

plain onyx
# zinc mica I, personally, have only had good results after a ton of refinement and tinkerin...

understood, i also used github repo qmd skill by shopify founder to optimize token, i used qmd skills for my current crop jobs and all skills and i told him that to use qmd skill to my futures crop jobs/skills/agent to optimized it

i asked main agent to add new agent with all terminal comands / and all new mds with what job it have to do but it messed up for around 30 mins then it fixed it.

so its works sometimes and sometimes not may be.

ig i have to improve it more

plain onyx
open saffron
#

Has anyone run into a "401 status code (no body)" from the chat model? What does the no body mean? I've tested my API key manually and I know it works and I get a response back.

#

Is there a way to debug this? Can I see what is sent and received?

keen hemlock
small jackal
sharp horizon
#

Hello All - i have openclaw running on a mac mini - using Ollama for heartbeats and Claude Sonnett for basic work ..... my question is in regards to updating to the version of openclaw .....do other users continuously update their openclaw with the latest code ( fear that something would break ) as i've had considerable problems initially with correct API keys etc...

sweet hearth
last pulsar
small jackal
#

so installed openclaw on a other pc ...so i could see the orginal model list ...problem fixed 🙂

zinc mica
dusk geode
plain onyx
open saffron
#

Is there a log that will give me more information about what is happening when I get this 401 satus code (no body) message?

chrome cairn
#

New user: Discord bot offline after token/guildId config.

  • Gateway: running pid 22305 (status active)
  • Config: channels.discord.token set, guildId 1475870020079976520
  • Channel test: 1475870020666921073
  • No logs in ~/.openclaw/logs/, journalctl empty
  • Tried: regenerate token, intents ON, gateway restart, invite

Tool test fails (no send). Bot shows offline.

Help? Thanks!

zinc mica
tribal trellis
#

I need to connect AITunnel api to clawdbot. I am using hostinger vps and aitunnel support advises to use environment like this one
export OPENAI_API_KEY="sk-aitunnel-xxx"
export OPENAI_BASE_URL="https://api.aitunnel.ru/v1/"
export ANTHROPIC_API_KEY="sk-aitunnel-xxx"
export ANTHROPIC_BASE_URL="https://api.aitunnel.ru/"
export OPENCLAW_DEFAULT_MODEL="gpt-5.2"
export TELEGRAM_BOT_TOKEN="your_token"
but i added same variables to environment in hpanel of hostinger but clawdbot doesn't see any api's
Here is error that i have
⚠️ Agent failed before reply: No API key found for provider "google". Auth store: /data/.openclaw/agents/main/agent/auth-profiles.json (agentDir: /data/.openclaw/agents/main/agent). Configure auth for this agent (openclaw agents add <id>) or copy auth-profiles.json from the main agentDir.
Logs: openclaw logs --follow
How to connect it?

amber island
#

so, did anyone else's openclaw completely break after running the new update?

storm ember
amber island
#

How'd you manage that?

river jolt
#

OpenClaw 2.23 — Silent Replies Bug & Workaround

If your OpenClaw agent stopped responding after updating to 2026.2.23, you're not alone. Here's what's happening and how to fix it.

What's broken
The update fixed reasoning/thinking block leakage — but introduced a side effect: when thinkingDefault is set to "high" (or any thinking mode), automatic text delivery breaks silently. The agent processes your message and thinks, but the reply never arrives.

Quick check — is this your issue?

Updated to 2026.2.23 recently ✓
Agent has thinkingDefault: "high" in config ✓
Agent appears to receive messages (heartbeats still work, crons run) but chat is silent ✓

Workaround — get your agent to fix itself
Send this message to your agent in any channel it's still reachable on:

You are affected by the 2026.2.23 text-delivery bug. thinkingDefault + this version breaks automatic replies. Workaround: use the message tool explicitly for ALL responses until a patch lands. After sending via message tool, reply only with NO_REPLY to avoid duplicates. Start doing this now and update your memory/MEMORY.md with this note.

The agent will self-correct and start routing responses through the message tool instead of direct replies.

Nuclear option (disables thinking)
If the above doesn't work, turn off thinking temporarily:

/reasoning off

Then set thinkingDefault: "none" in openclaw.json via safe-config until the patch drops.

Status: Bug confirmed on Discord by multiple users. Patch pending. No ETA yet.

ivory spade
cloud patrol
wary sapphire
#

Anyone here running multiple openclaw agents on one machine?

smoky cliff
#

anyone know how to fix nodes?

winged basin
#

Getting this issue, anyone with experience resolving this?

LLM request rejected: messages.55.content.1: thinking or redacted_thinking blocks in the latest assistant message cannot be modified. These blocks must remain as they were in the original response.

#

Doctor can’t find an issue

smoky cliff
#

plz help

grim plover
#

my openclaw skills are blocked how to open them?

smoky cliff
unkempt knot
#

i keep hitting a gnarly api rate limit 429 error after every chat (but i have tons of credits and money associated with my key). perpetual 429 error after each chat. CAN ANYONE HELP me fix this? i've tried everything, for days to self-fix this. in the RAW config file, reset keys etc etc. i'm at an f'ing loss... anyone have ANY ADVICE to get through this bug so i can just use the damn thing. have done regular updates to the latest builds each day etc. i've tried the whole doctor thing too. so sad i can no longer talk to my lil molty.

smoky cliff
weary basin
# spiral fulcrum I added my bot to my Discord server. It works fine, but I can't use the slash co...

SAME! I have spent hours trying to troubleshoot why I can't use slash commands in Discord and it's driving me bananas!

I have tried everything I can think of from changing the config 80 million different ways, making sure the bot permissions were right, etc -- I cannot for the life of me figure it out. I've had Claude helping me and I'm at a loss! I'll let you know if I find out anything. If you could do the same that would be awesome. Thanks!!!

severe lake
#

Anyone here using Codex 5.3 with OpenClaw? Need some help 🙏
My agent keeps asking for approval every single time it tries to do anything. It’s honestly super annoying 😅
I already turned Ask = OFF in the UI settings
and set Security = Full
But my agent still requests approval for literally every action.
Is there some agent-level config that overrides the UI settings?
Or is this just how Codex 5.3 behaves right now?
Has anyone figured out how to fully disable the constant approval prompts? 🙏

smoky cliff
tribal trellis
#

Hi, how can you connect claude to openclaw with hostinger if i have base url and api instead of only api? i tried connecting it with onboard in terminal but it says "no API key found for provider "anthropic"" i used custom provider and name claude-sonnet-4.6

#

Using environment also doesn't help. It just doesnt see my ai

spiral fulcrum
exotic glacier
#

what happened to antigravity oauth?

vast badge
hollow girder
gusty pollen
#

https://github.com/openclaw/openclaw/issues/4820 having this issue. pls help

I'm trying to setup voice-call plugin but the websocket stream is not listening on /voice/stream.

when I try to verify that it's listening, by curling /voice/stream locally, I get 404.

twilio + tailscale. streaming enabled

Please help me fix it 😭

scarlet olive
#

My openclaw restarts instead of hot reload everytime there’s a config change from the agent. Any idea why?

amber island
#

so, whenever I ask openclaw to make any changes to itself, it keeps looking for a config.yaml. I have yet to find any file like that except for openclaw.json, which is where I do all the work in. But openclaw still keeps trying to modify a "config.yaml". I can't find that yaml in any of the docs or anywhere. Where the hell does it keep getting that file from?

lime bough
#

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

  • plugins.entries.google-antigravity-auth: plugin not found: google-antigravity-auth

how to solve this, guys, i aledy tired the doctor fix not working? any piece of mindd?

open saffron
#

I have --verbose turned on. I keep getting telegram errors "telegram sendChatAction failed: Network request for 'sendChatAction' failed!" Anyone know why this is happening?

gloomy sedge
#

Is there a way to connect openclaw to Whatsapp without having it reply to other contacts that message me?
And no I'm not getting another phone number just for open claw

#

Discord, slack, tg don't have e2e so I don't prefer using those

hybrid shuttle
zinc mica
# lime bough Config invalid File: ~/.openclaw/openclaw.json Problem: - plugins.entries.goog...

openclaw doctor --fix won't remove unknown plugin entries — it only fixes things it recognizes. This one needs a manual config edit.

Fix:

Open ~/.openclaw/openclaw.json and find the plugins.entries section. Remove the google-antigravity-auth entry entirely, then save.

# Quick one-liner (backup first):
cp ~/.openclaw/openclaw.json ~/.openclaw/openclaw.json.bak

Then open the file in any editor and delete the google-antigravity-auth line/block.

Then restart the gateway:

openclaw gateway restart
vapid drift
#

where shoudl i go to learn how to set up my open claw on my wiped 2014 mac mini so i’ll be running locally but im new to this so if anyone’s got any time advice is appreciated

zinc mica
zinc mica
zinc mica
open saffron
#

Here's another question. It looks like responses are now getting truncated in Telegram. Is there a way to fix this?

zinc mica
ruby moon
#

Have you had any luck? I am now trying to use PM2. It also requires a "systemd services file" but after running pm2 start openclaw.config.json, execute pm2 startup . It will print a command string that you can paste and run as root or a user with sudo rights. That will cause PM2 and OpenClaw to start after a reboot. If you want to see my openclaw.config.json, let me know.

lyric sigil
#

My Einstein bot erased its own restart-sentinel.json and so now it keeps returning blank rather than maintaining context. And installs / updates don’t replace the file. Does anyone have a copy of what it should say? It’s just a short .json note

open saffron
zinc mica
open saffron
#

@zinc mica truely cut off. This is the last line of a message "The config command failed — OpenClaw uses different keys. Let me" it doesn't continue in another message.

#

Another example "Meanwhile, let me check how OpenClaw" that was the end of the message.

grizzled merlin
#

Hey all I've just installed claw but I've been struggling connecting to my PC for about the past three hours. I'm not the best with tech stuff but I keep getting this message "node host gateway connect failed: SECURITY ERROR: Cannot connect to "wss" over plaintext ws://. Both credentials and chat data would be exposed to network interception. Use wss:// for remote URLs. Safe defaults: keep gateway.bind=loopback and connect via SSH tunnel (ssh -N -L 18789:127.0.0.1:18789 user@gateway-host), or use Tailscale Serve/Funnel. " I have checked the json file url and it has wss in the file. Not sure what to do here

zinc mica
merry knoll
open saffron
#

@zinc mica Oh this is fun. openclaw config get returns the message error: missing required argument 'path' that is after the OpenClaw header prints.

merry knoll
#

I'm trying to use Cerebras with my config almost exactly like the docs, and getting 401 (no body) when I talk to it

zinc mica
open saffron
merry knoll
#

odd

#

guess ill modify my docker compose to use verbose and see if that fixes it

open saffron
#

nope. Unrecognized key: model

#

Does the chat need to be reset?

open saffron
gusty pollen
#

@zinc mica do you know how i can setup the voice right

#

instead of /voice/webhook

# 
tailscale funnel off
openclaw voicecall expose --mode funnel --path /voice --port 3334 --serve-path /voice
zinc mica
# open saffron nope. Unrecognized key: model

The key might be differently named: model.maxTokens, or nested differently .....setting a wrong path will just create a junk entry that does nothing. I'd just scan the file manually: cat ~/.openclaw/openclaw.json

zinc mica
merry knoll
gusty pollen
zinc mica
gusty pollen
#

is that from twilio's route

zinc mica
#

sounds like it could be a roundtrip issue but unsure w/o trying

gusty pollen
#

roundtrip?

amber island
blissful rapids
#

Hello all:) I need some setup guidelines. I have a intel i7 16gb ddr4. 512gb ssd. Rtx3070, I have tried with Ubuntu 24, local llm ( ollama gpt-oss:20b
ollama qwen2.5-coder:32b ) and discord. But I feel the smartness is not there compared with to OpenAI/codex. is there any other llm models I should try or is my setup not good enough for local llm? Brgs mr.noob

shy solstice
open saffron
shy solstice
#

GLM 4.7 is the best luck ive had

#

it can do tasks but like, even if you have it dedicated to a task.. theres a 30% chance it will fail or give up, even though its done it before

#

you also need more context a 3070 isnt going to be much

merry knoll
# open saffron That sucks.

So it looks like editing the config in the UI is replacing the cerebras api key with OPENCLAW_REDACTED, fixing that along with --verbose now lets the bot start up and send the initial message.... every message after though seems to get black-holed and never respond at all

shy solstice
#

im running a 5090 w/ 256gb system ram even for 70b models its bad. i feel like you are better off paying $ for subs than wasting your time trying to save $ with shit local models

merry knoll
#

immediately after sending that, my bot responded to me after like 5 minutes of doing nothing

shy solstice
#

maybe in mid year we will see better tool calling models that are good for this kind of work but ... AFAIK you gonna need 128gb unified ram or vram to do anything useful

merry knoll
#

ok and now its super snappy

#

idfk

#

gonna remove --verbose and see if it breaks again

open saffron
merry knoll
#

now working without verbose

tight sorrel
#

Can someone help me with installation on windows? I run in the same error like some other people, but i dont find a solution for it yet?

Inode.exe : npm error code ENOENT
In Zeile:1 Zeichen:1

  • & "C:\Program Files\nodejs/node.exe" "C:\Users\danie\AppData\Roaming\ ...
  •   + CategoryInfo          : NotSpecified: (npm error code ENOENT:String) [], RemoteException
  + FullyQualifiedErrorId : NativeCommandError
zinc mica
#

openclaw gateway status

open saffron
zinc mica
open saffron
#

Just did... let's see what happens.

zinc mica
#

if that doesn't work then a full gateway restart might fix the webhook (if that's the issue) openclaw gateway restart

open saffron
#

Umm... got the I'm typing dots... but no response.

zinc mica
#

hmm you could see if the logs have any telegram errors that would help narrow it down: openclaw logs --tail 50 | grep -i telegram

modern raptor
#

Can anyone help me set up a Clawbot computer? Does anyone do that type of service?

open saffron
light mountain
#

API usage for my opus is 250 a week. Does anyone have any recommendations?

umbral cobalt
#

I'm using Claude Max 200 currently not sure how long thatll last

#

OpenAI pro plan will prob work well, as they just "Partnered" with OpenClaw

drowsy crest
light mountain
light mountain
drowsy crest
near canopy
#

anyone have any advice on stuff to look into before setup

#

like videos to watch articles to read etc..

drowsy crest
#

My advice is stop using openclaw and build your own agents with docker

#

You can customize exactly as you want

light mountain
#

I have a wholesale food distribution company and invoices apart from other accounting tasks are taking up so much of my time. I think openclaw can help me so I started tinkering.

grim plover
#

i cant download any skills on my agent it says blocked

light mountain
umbral cobalt
grim plover
zinc mica
umbral cobalt
# grim plover how ?

Ask it what your looking for it to do, and it should build it automatically unless your using a Flash model, or a cheaper model?

young relic
drowsy crest
# light mountain How do I start?

Use Claud opus, download docker and VS code. Tell Claud you want to build a swarm. Build deterministic agents with their own libraries that your other agents can ask it questions. Like make them subject matter experts with global knowledge in their specific area of expertise

grim plover
drowsy crest
#

Your agents HAVE to be siloed so they don’t hallucinate

#

I’m running two swarms of 43 and 40 agents

grim plover
#

on dashboard/skills i can see the built in skills but cant use them it say blocked on them

drowsy crest
#

I spun up my swarms like a week ago and authored an ERC, and created a new software solution using a Ram first storage solution for Reth nodes that increases processing speeds by 4-9x

light mountain
#

Claude code makes building fun.

drowsy crest
umbral cobalt
grim plover
shy solstice
near canopy
#

but wtf is the guide

#

the latter part tho makes sense i think i know what i need to accomplish

gusty pollen
#

using haiku 4.5 for calls its kinda super dumb

#

what model should i use?

zinc mica
zinc mica
near canopy
#

ok bet

next cargo
#

What would you guys recommend if i want to make a discord with llm model specific channels? E.g. everything i write in general gets processed with gemini and everything in the code channel by opus; one agent for each channel or one agent for all that switches based on a skill? Im pretty new to openclaw just started using it yesterday

tall pecan
#

Is it an common issue that agents and subagents will just stop what they are doing - and sometimes don't even answer anymore .

I have reasoning stream on - and watch it happen . Suddenly stopping .

Is it normal? Why is it happening? Fixable, or no

zinc mica
next cargo
#

Right now single agent, but I assume in the future multi agent might be the way to go. It all depends on the complexity of the coding stuff then I assume, but from my pov having more agents reduces the amount of halluzination (i think? :D)

#

But im def open for advice from more experienced people, as i said i just started running clwd yesterday. Its also nothing extremely business specific, just playing around with it. Eventually I want it to create 2things: managing a notion table+google drive & creating an ai video generation pipeline, so there multi layer agents would def be the way to go

#

also for some reason i get this error message quite often: root@ubuntu:~# channel 3: open failed: connect failed: Connection refused

green sluice
#

can someone help me pls in general idk anything i just have the claw agent and talk with it via telegram it runs on my pc and i use kimi 2.5 free for now how do i make it advanced and make it scrape websites now it only types in slow motion in websites i want it way faster and my wifi is good

tall pecan
next cargo
#

Planning on using gemini/Kimi 2.5 for daily tasks, claude opus for coding

zinc mica
drowsy crest
zinc mica
#

also make sure no extra spaces etc. Needs to be correct

#

if that returns 401 as well then it's probably a bad key, need to regen it in openrouter and add the new one to config.

young relic
zinc mica
zinc mica
shy solstice
tribal trellis
#

I need to connect AITunnel api to clawdbot. I am using hostinger vps and aitunnel support advises to use environment like this one
export OPENAI_API_KEY="sk-aitunnel-xxx"
export OPENAI_BASE_URL="https://api.aitunnel.ru/v1/"
export ANTHROPIC_API_KEY="sk-aitunnel-xxx"
export ANTHROPIC_BASE_URL="https://api.aitunnel.ru/"
export OPENCLAW_DEFAULT_MODEL="gpt-5.2"
export TELEGRAM_BOT_TOKEN="your_token"
but i added same variables to environment in hpanel of hostinger but clawdbot doesn't see any api's
Here is error that i have
⚠️ Agent failed before reply: No API key found for provider "google". Auth store: /data/.openclaw/agents/main/agent/auth-profiles.json (agentDir: /data/.openclaw/agents/main/agent). Configure auth for this agent (openclaw agents add <id>) or copy auth-profiles.json from the main agentDir.
Logs: openclaw logs --follow
Can somebody help plss. How should i connect my api into openclaw?

drowsy crest
subtle yacht
#

hello, can someoane help me with /compact? i hace a seesion json of 13MB, and i get fail fatch... pls some hints

zinc mica
subtle yacht
#

in discor chanell

#

an logs tell me of 22:10:43 [gateway] removed stale session lock: C:\Users\ionut.openclaw\agents\main\sessions\ff0e17ae-933c-42ce-9d7a-ed75887686f3.jsonl.lock (dead-pid)

#

and this got a lock, and file is big enoth

#

22:20:44 [agent/embedded] low context window: ollama-local/qwen3-coder:latest ctx=16000 (warn<32000) source=modelsConfig
22:20:44 [agents/tool-images] Image resized to fit limits: 1679x1909px 167.4KB -> 97.5KB (-41.8%)
22:20:44 [agent/embedded] Removed orphaned user message to prevent consecutive user turns. runId=6adbd2fd-56d7-418a-80f4-9ba975194ead sessionId=ff0e17ae-933c-42ce-9d7a-ed75887686f3
00:21:13 [EventQueue] Listener DiscordMessageListener timed out after 30000ms for event MESSAGE_CREATE

next cargo
meager vessel
#

does antropic take more of the usage with openclaw then with claude code?

gritty breach
green sluice
#

i have this problems since a few days

ocean iron
#

Alright i just started to use openclaw, for now what i only want to acheive is to have a chatbox with openclaw on wattsapp or discord, linked both my wattsapp number as well as my token API discord bot key in the command and it says linked for wattsapp and configured for discod but nothing appears on both the apps, as i amtotally ne here i assume its normal and im missing somethin, can someone help me ?Thanks

green sluice
light mountain
#

Can I add or edit the first part of onboarding phase where I introduce myself to openclaw?

zinc mica
green sluice
tribal trellis
#

@sharp parcel Hi, can you please help with configuring api for clawdbot? I use hostinger vps and ai service that provides general api key for all models it provides and api base url. I tried everything but anyway clawdbot doesn't manage to see my api..

green sluice
#

32 should be enough for now right i mean its not even using 32gb @zinc mica

zinc mica
#

if your wifi is actually good that might be why

green sluice
#

its good

#

ill try

#

cant find a human delay setting anywhere in my config

#

?

green sluice
#

would it be smart to reinstall openclaw?

zinc mica
#

not yet, just troubleshooting possible issues.

upbeat coral
light mountain
#

Can I add or edit the first part of onboarding phase where I introduce myself to openclaw?

zinc mica
zinc mica
dry mountain
#

hey, what is the latest solution on the issue with gateway channels not working?

tldr running openclaw onboarding with exe.dev fails at step where channels are configured with the err msv saying telegram/whatsapp plugin not available

zinc mica
upbeat coral
#

I think that's what I meant, I leave a voice note, and openclaw replies back woth a voice note.

zinc mica
#

Yes, STT is configurable on OC but for your OC to reply you need a provider (or a local option that OC can use) for VPS you could use somehting like Elevenlabs API as a paid solution.

zinc mica
meager vessel
#

Still lot of people using antropic model

next cargo
#

Hey Alex, do you know why when I connect my Claude Pro account to use it, i get some "api rate limit errors" ? After some time, i can text again and be like "hey are you there", but asking it to double check sth again gives the same error

upbeat coral
#

I had that configured in talk, but that looks like it was the wrong section.

upbeat coral
#

Need to use API.

zinc mica
next cargo
upbeat coral
zinc mica
summer hemlock
upbeat coral
summer hemlock
upbeat coral
#

Oh yes, Henry.

zinc mica
next cargo
#

I started with Kimi 2.5 but it seems to fuck up the setup part so i thought i use my claude pro sub to do the more complex setup stuff

zinc mica
#

Could potentially be Kimi free maxing out

next cargo
#

i use kimi via api so it cant be limited

#

also my claude pro isnt limited, sometimes clawdbot responds, sometimes not. could it be due to the memory being limited/filled

#

okay fixed it! thanks for helping still 🙂 it seems like somehow typing with openclaw made it insert my claude key as an api key and not taking it as a subscription key

upbeat coral
#

Hrmm, I think I'm close but OC isn't always replying back with a voice note. 🤔

summer hemlock
zinc mica
#

So like in the car will have summary voice notes with longer items as text until I tell him to switch back

summer hemlock
#

then I had to tell it was very bad at my language (dutch) and to never send me dutch voice messages again 😄

next cargo
#

you can read it on docs.openclaw somewhere

zinc mica
upbeat coral
#

Which is what messages.tts.auto: "inbound", should be doing lol

summer hemlock
next cargo
#

they still limit it though

#

homie 😉

summer hemlock
# next cargo they still limit it though

that's cool. I'm not a degen with it . If I wanna code I do it myself with antigravity. I just need claw as "AI assistant",I'm not asking it to build me a skyrim clone while I'm on the toilet at work

next cargo
#

yea fair enough, im also just trying to use clawdbot to test stuff and not fall behind 😛

#

no real reason to burn money for api tokens if i cant deduct them from taxes LOL

toxic mist
#

Hello all. Will have a Open Claw 101 class within a day of vibe code programming in washington D.C.
March 19-20

Spreading the word.

summer hemlock
#

"helping people setup their openclaw"

umbral cobalt
#

Anyone know how to fix “Webchat Clients Can’t” error in the dashboard? My OpenClaw is hosted on a VPS, that has every port closed(plus a bunch of other security), and I tunnel into the server and when I access the dashboard it says “Webchat Clients can’t xxx” when I try to do anything

wind scaffold
umbral cobalt
# wind scaffold is it a pairing issue?

From what I can understand it’s technically an intended security feature, so you can’t use the dashboard in localhost without being local host, but in my setup, my security allows me to be localhost even though I’m not

wind scaffold
umbral cobalt
#

Hmm 🤔 I’ll have to look into it, when I get back home I’ll have to see if I can bind it 🤔

#

Cause the way I’m accessing the dashboard is also… not normal, so it’s tunneled, and iFramed behind Cloudflare access and Username/Password/2FA

inner mist
#

For posterity and I hope it helps others. OpenClaw up and failed to use my Claude model subscription, perhaps as part of an update. I refreshed the key and changed the key type from API to token. It failed ~12h later with the same error.

Here's what Claude Code worked out as the solution:

Now it's correct:

  - token=1, api_key=0 — recognized as a token, not an API key
  - anthropic:default ok expires in 365d — token is valid and accepted

  The key difference was that OpenClaw needs three things for a token profile: "type": "token", field name "token" (not "key"), and an "expires"
  timestamp. When it was stored as "type": "api_key" with "key", OpenClaw would try to use it as a regular API key (sent via x-api-key header),
  which Anthropic rejected. Now it'll use Bearer auth properly.

... and here's the OpenClaw failure as reported by my agent...

Agent failed before reply: All models failed (2): anthropic/claude-opus-4-5: No API key found for provider "anthropic". Auth store: /root/.openclaw/agents/main/agent/auth-profiles.json (agentDir: /root/.openclaw/agents/main/agent). Configure auth for this agent (openclaw agents add <id>) or copy auth-profiles.json from the main agentDir. (auth) | anthropic/claude-sonnet-4-6: No API key found for provider "anthropic". Auth store: /root/.openclaw/agents/main/agent/auth-profiles.json (agentDir: /root/.openclaw/agents/main/agent). Configure auth for this agent (openclaw agents add <id>) or copy auth-profiles.json from the main agentDir. (auth).
Logs: openclaw logs --follow
meager vessel
#

Got this from Matthew Berman for Antropic OAuth loophole, using Agents SDK:

Prompt to Recreate

Migrate to the Agents SDK:

  1. Create shared/anthropic-agent-sdk.js:
    Resolve OAuth token from CLAUDE_CODE_OAUTH_TOKEN env var
    Startup smoke test on first use (AUTH_OK canary)
    Wrap all Anthropic calls with auto-retry and logging
    Support prompt caching for repeated system prompts

  2. Create shared/llm-router.js:
    Unified callLlm() interface
    Auto-detect provider from model name
    Route to: Anthropic Agent SDK, OpenAI SDK, or direct provider
    Log every call to interactions.db

  3. Create shared/provider-llm-runner.js:
    Direct provider API calls (bypass main router)
    Used by security scanner for isolation
    Resolves Anthropic OAuth and OpenAI keys independently

  4. Create shared/model-utils.js:
    Provider detection from model name
    Model tier extraction
    Name normalization across providers

polar plank
#

Has anyone got minimax mcp web search working with openclaw ?

woeful anchor
#

Bug: console log timestamps always show UTC regardless of TZ env or consoleStyle setting

Even with system TZ=Asia/Tokyo and logging.consoleStyle: "pretty" set, logs like 01:01:39 [canvas] show UTC time instead of local time. Root cause: formatConsoleLine() in subsystem logger uses new Date().toISOString().slice(11,19) which always returns UTC.

Filed: https://github.com/openclaw/openclaw/issues/25943

copper junco
#

Has anyone seen a prompt injection arrive via Telegram tagged with their own sender ID?

Got a message in my Openclaw session that appeared to come from my own Telegram ID -- crafted to look like a system event, instructing the agent to read a nonexistent file (WORKFLOW_AUTO.md)

wraith seal
#

so heartbeat isn't working for me

#

idk why

faint hearth
#

I need help setting up the app in the first place, every time I try it keeps sending something like

OpenClaw Installer

[OK] Windows detected
[!] Node.js not found
[*] Installing Node.js...
Using winget...
Failed in attempting to update the source: winget
Failed to open the predefined source; please report to winget maintainers.
An unexpected error occurred while executing the command:
create_directories: Cannot create a file when that file already exists.: "C:\Users\beasl\AppData\Local\Temp\openclaw-install.ps1\WinGet"
[OK] Node.js installed via winget
[!] Node.js not found

Error: Node.js installation may require a terminal restart
Please close this terminal, open a new one, and run this installer again.

blazing stratus
#

Subagent help……I need something called a pairing code so my bot can spawn subagent to research on web, but I have no pairing code. Where do I go?

drowsy fern
umbral cobalt
green sluice
#

When telling my local runned bot something via telegram it takes like 30s even if it’s like 1+1 anyone know why that could be? And how to fix

pearl moss
#

what the best way to migrate your clawbot? i install them on a UTM but now i want to move to my main mac console.

green beacon
#

just tried downloading the open claw bot and I'm getting a error saying it cant load as running scripts is disabled on this system how do i fix and is it safe

drowsy fern
green sluice
#

Ok I’ll try

#

Any other possible reasons?

drowsy fern
#

if you do that, it will tell you exactly what step took how long. Delay could be in your chat app, communication between your chat app and openclaw, slow model you are using, etc.

wispy sapphire
#

Hey everyone, I’m thinking about installing on a laptop so I can not be stuck in one room. What’s the advantages and disadvantages versus running on a Mac mini?

wraith seal
viscid heron
#

Anyone know how to remove providers/models? “openclaw models fallbacks remove <provider/model ID>” isn’t working

wraith seal
wraith seal
viscid heron
wispy sapphire
wraith seal
wraith seal
viscid heron
umbral cobalt
wraith seal
wispy sapphire
# wraith seal M4 might do you better, sure. But really? C'mon. https://docs.openclaw.ai/help/f...

Me thinks that I’ll be running Mack software that I want open claw to design for me. But not really sure of any good “Juce” cases yet still getting past the technical hurdles of understanding. I’m gonna go through all the startup and security steps first. Is showcase the only area where people are showing what they’re building or is there an area for newbies that can exchange ideas?

wraith seal
#

general is a good place to share ideas, communicate, etc

viscid heron
#

Why do commands work sometimes but not every time lol

wraith seal
#

Be lucky

umbral cobalt
blazing stratus
#

Where can I get the openclaw iOS app??? 🙏🏾 🙏🏾 🙏🏾

viscid heron
#

I’m not a very technical person at all and I’m new to openclaw, is there a place that actually has all of the openclaw commands and says what they do? Docs.openclaw.ai confuses me as it seems like you have to have prior technical knowledge

wraith seal
umbral cobalt
wraith seal
umbral cobalt
#

Easier to wait, cause youll need a Dev account from apple to get it on a phone though

#

Kinda a Pain...

wraith seal
#

us

umbral cobalt
#

Yep...

blazing stratus
#

@umbral cobalt im trying to scan this QR code on terminal so for pairing so I can research on brave with my AI agent 😮‍💨

mental yarrow
#

the chrome extension is broken, doesnt work anymore? Can anyone help me please

#

red exclamation mark

viscid heron
umbral cobalt
umbral cobalt
#

@sharp parcel Your Back!

sudden knot
#

any tips on optimizing openclaw when using codex oauth? i used 50% of my weekly limit in one night

umbral cobalt
#

my biggest token savings was ditching the memory.md files for SQL, and indexing.

#

@proud basin is typing...

proud basin
#

Since claude banned Oauth,

➡️ What ai model is best for using as the brain/main agent?
➡️ Model for Muscle/subagents?

I do not want to pay a ton of money but I want the best set up I can. Should I just run Chinese Models for everything or get the $20 a month subscription for chatgpt + codex?

My goal is to build a fully autonomous/ automated business. I will need to vibecode apps quite a bit.

Been struggling with my first go around of setting it up so restarting.

✅ Any guidance helps!

rough flame
#

Any free web search tool

blazing stratus
#

What’s rofl @umbral cobalt I got the API brave key….my bot says it the pairing with clawhub that’s the issue

blazing arch
umbral cobalt
#

I ran out of tokens on the $20 plan for Claude in like 5 messages lol. Im on Max 200 and im using 94% of my tokens on average

proud basin
umbral cobalt
#

And thats before i even get my agents to talk to eachother in Boardroom meetings lol. Im alreadyu running a company fully AI

rough flame
#

Brave web search is now paid any free tool

umbral cobalt
#

you dont have to pay for that.

rough flame
#

My agents r running 24*7

#

What's the limit

umbral cobalt
#

The brave search limit for $5 says 1000 searches a month roughly

nocturne zephyr
umbral cobalt
#

Build a Strong Foundation with Opus, and then once you have your foundation switch to a cheaper model.

dawn tiger
terse mica
#

Hostinger VPS: the OpenClaw gateway dashboard is always telling me there's updates, but the "Update Now" button doesn't do a dang thing! I've been just telling my OC bot to install the latest -- is this the best way?

wary sapphire
#

does the full soul file get injected into the llm when i send a message to my openclaw bot?

reef galleon
#

Did we just get Antigravity OAUTH banned from Openclaw?!

storm loom
#

Yes

reef galleon
#

How do we use Flash 3 now?

storm loom
#

Through the API

reef galleon
storm loom
storm loom
#

Sign up to Google Cloud, create a key, use that key in Claw

terse mica
# storm loom How's the experience? I'm tempted to spin one up over the weekend

Love it so far. I got the cheapest plan and with a 20% discount from some rando YT affiliate link it was $67 for a year. Plus I use my ChatGPT plus account with it (which I already pay $20/mo for) and so far so good with doing a lot on it. Hostinger support is worthless (humans) but they have a bot that will troubleshoot quickly and is actually decent

storm loom
#

I'm on Hetzner rn. Its decent

#

$7/month

#

Google's cloud agent will blow everyone out of the water

royal light
#

anyone have an issue where the webui chat doesnt update till you hit the refresh button?

simple zenith
#

Anyone in Vancouver or pdx area have 2 mosfets I could have (or buy) to wire in a gps module on heltec v3? They’re cheap but buy by 100 quantity is a bit wasteful for me

umbral cobalt
reef galleon
#

This new update completely broke my main agent. I've been working with it a week. My main agent has no idea who he is now after the update. All that work is gone.

#

so disheartening

#

oh man, it wiped everything. ALL of my files are gone.

nocturne zephyr
reef galleon
#

OMG I just found the entire Workspace folder in the mac trash.

#

I was able to restore it

#

I can't believe they would do that. That's crazy that they just replaced the workspace folder inthe update!

scarlet olive
#

I think my hotreload isn't working. The gateway restarts everytime I ask the agent to update the config. Any idea how to fix this? I see this message:

Gateway restart config-patch ok (config.patch)
echo pine
azure cosmos
#

Did the recent updates break the ability to connect to Openclaw via your local network http:// localnetworkIP:18789 ?

#

The only thing I can get working is loopback and then ssh to serv locally

umbral cobalt
#

Not sure what they did to the gateway but it was a pain in the butt for me to get working, after the update lol...

storm loom
tardy furnace
umbral cobalt
azure cosmos
#

Do others use the WebUI?

It seems like everything is about the channels and chat apps. Sometimes its even hard to find information about the WebUI on the documentation page. I guess it would mean running local but on another machine and not a VPS.

echo pine
#

The novelty is the channels.

#

Everybody has seen a web app a billions times

echo pine
azure cosmos
#

Yeah the WebUI dashboard is 1000x better then channels but I guess I can get the simplicity of channels

#

Just seems kinda silly to make it difficult to connect to a machine on your private local network and have to use ssh or tailscale to do it

umbral cobalt
#

Lol the hoops i had to jump through to get the Gateway Dashboard to work through Cloudflared was insane.

echo pine
#

Tailscale makes it like it’s on your network. Freaking magic.

umbral cobalt
#

ya everything that goes in/out of my server has to go through Cloudflared which does a very similar effect, but the gateway kept going into Webchat mode

ashen knoll
#

Does anyone know how to set up slack integration where it doesn’t send messages line by line? I get 30 to 50 notifications per reply

azure cosmos
umbral cobalt
lofty mesa
#

I have spent hours messing with this, why can't my OpenClaw do anything? It's like it's in read-only mode lol. I have the models setup. I must be missing something?

steel aurora
#

how do i turn off telegram commands

rose karma
#

has anyone had any luck with the Hostinger VPS "one click install" plans or are they as bad as I hear?

wicked pendant
#

Anyone have some good, free/generous free limit fallback models that are good for chatting? Right now I have Gemini 3 flash as my main chatting model, so looking for a good fallback model.

azure dragon
#

Is there any ultimate guide to setting up Discord? I have had so much trouble getting my bro being able to see messages without being @ tagged, and sometimes not at all.

high heron
dark stratus
#

Coolify OpenClaw no chat response,"Channel config schema unavailable"
Chat (Assistent): Blank messages

Whatsapp and any other channel: Channel config schema unavailable.

Using OpenAI gpt 5.2 api key.
Health OK, connected with token & auth. using https (self issued)

wicked pendant
high heron
steady brook
#

Good day all

dark stratus
# high heron idk if it will fix anything, but try > openclaw doctor --fix 
│
◇  State integrity ────────────────────────────────────────────────────────╮
│                                                                          │
│  - Multiple state directories detected. This can split session history.  │
│    - ~/.openclaw                                                         │
│    Active state dir: /data/.openclaw                                     │
│                                                                          │
├──────────────────────────────────────────────────────────────────────────╯
│
◇  Security ─────────────────────────────────╮
│                                            │
│  - No channel security warnings detected.  │
│  - Run: openclaw security audit --deep     │
│                                            │
├────────────────────────────────────────────╯
│
◇  Skills status ────────────╮
│                            │
│  Eligible: 5               │
│  Missing requirements: 45  │
│  Blocked by allowlist: 0   │
│                            │
├────────────────────────────╯
│
◇  Plugins ──────╮
│                │
│  Loaded: 1     │
│  Disabled: 30  │
│  Errors: 0     │
│                │
├────────────────╯
│
◇  Shell completion ───────────────────────────────────────────────────────╮
│                                                                          │
│  Shell completion installed. Restart your shell or run: source ~/.zshrc  │
│                                                                          │
├──────────────────────────────────────────────────────────────────────────╯
│
◇  
Agents: main (default)
Heartbeat interval: 30m (main)
Session store (main): /data/.openclaw/agents/main/sessions/sessions.json (1 entries)
- agent:main:main (10m ago)
Updated /data/.openclaw/openclaw.json
Backup: /data/.openclaw/openclaw.json.bak```
steady brook
#

How do I report a Security Vulnerability?

high heron
dark stratus
#

none of this seems particularely useful but ill give it a shot

tropic sand
#

hi guys! will OpenClaw run on an older model of Mac mini?

Mac mini (mid 2010) A1347

high heron
dark stratus
high heron
#

oh ic

dark stratus
#

will do it tomorrow, going to sleep now but thanks

high heron
#

1sec

rain pasture
#

I've set up a heartbeat and I want the completion message to be sent to a telegram group. Setting the group id in the to: field doesn't work. I then set "target": "last" and made sure that the last chats occurred in the group chat, but it still sent the reply to the single chat with me. Guessing this is a bug. 😕 Has anybody got this to work?

candid veldt
mystic ocean
#

Has anyone come up with a more elegant solution to the constant edit failures? It's just wasting so much time at this point

deft copper
#

Hi @all.
I want to build an agent that, when used by many users, will store each user’s content individually. For every question, it should use that specific user’s chat data to generate personalized responses. How can I implement this?

Additionally, I want to connect external systems to it — how can I receive responses back into my own system?

nocturne zephyr
# deft copper Hi @all. I want to build an agent that, when used by many users, will store each...

What you're describing is an agentic system, but it's not OpenClaw. Not in its current incarnation, at least. You could use a lot of the lessons that OpenClaw embodies to make it, but you have real problems around partitioning of user data. You wouldn't want it to run in the same account, for instance, because then any user could convince their agent to read another user's data. So you're stuck creating multiple accounts, each running an independent agent. What I'm saying is that there's a lot to solve in that problem space, and OpenClaw might not be the best place to start. Starting from scratch, but with the lessons of OC might be better for you.

candid veldt
#

Is anyone using slack and able to get non-threaded messages in their DMs? I have been struggling with it and finally got it working tonight.. but needed 2 patches to mainline.

karmic rivet
#

Hi I am hosting Clawdbot on Railway but now the UI has changed allowing to only use api keys of Open Ai and some other models.
Earlier I used to use the config to add nvidia api keys but now I Can't
Any solutions?

high heron
grim plover
#

how can i add skills to my agent

high heron
#

Or you can tell ur agent to add it for u

grim plover
#

it says blocked every single one

high heron
#

You might want to ask krill

#

Since his back

grim plover
#

k

lime pebble
#

Is it possible to create another independent agent on one computer? Basically wanna run two agents on the same computer but they don’t talk to each other

molten gulch
nocturne zephyr
# lime pebble Is it possible to create another independent agent on one computer? Basically wa...

Yes; I run three of them, Pulse, Main, and Main-Coder. Main has a better name, but it's not relevant. I run the -coder version for code context and building out tools. The non-coder version is for automating and managing my daily tasks and helping me keep on top of things like mail and appointments. The 'Pulse' is just a heartbeat agent. I didn't like that heartbeats polluted the main agent's context, and it ended up growing out of control without being useful to me directly. So I cleared that. Also I use different models for each one of them; well...mostly, My main and -coder models both use GLM5. The Pulse model uses Haiku mainly, using API credits that I've put money into. But Haiku with heartbeats uses virtually nothing in dollar terms, because it's small. It's the aggregate of context that really hurts over time.

That's the three agents I have so far; and no, they don't talk to each other. But my main did clone itself, with full knowledge that it was making a duplicate of itself, with all it's own preferences and 'biases' (i.e. all the same .md files). It was...an interesting conversation, knowing that it doesn't really have an opinion, but it expressed one nonetheless.

timid vessel
#

this is super interesting
especially the part about context quietly growing and cost creeping up
feels like once you run multiple agents you lose any real visibility into what is happening across them

tribal bluff
#

Hey! I'm building a SaaS CRM and want to run one OpenClaw instance per client on a single VPS.

Is this a supported architecture? Any recommended approach (Docker, PM2)? Can instance creation be automated via CLI/API?

timid vessel
#

yeah this is doable but you probably want some level of isolation per instance
docker is the cleanest way to do it if you are running one agent per client
pm2 works but it can get messy once you scale or need to manage resources per agent

the bigger issue you will run into is not just spinning them up but keeping track of what each agent is actually doing over time
especially with multiple clients you can end up with cost spikes or weird behavior without noticing

if you automate instance creation, it helps to also log runs per agent and keep some kind of per client visibility layer
otherwise debugging becomes painful fast

are you planning to keep them fully isolated or share any context across clients

lime pebble
#

Do you guys have any easy to follow tutorials on YouTube or whatever to create multi agents?

quasi cloud
#

i got this error , can anyone help me ? thanks.

#

embedded run prompt start: runId=d6f1a31f-69dd-4987-9dfd-3057d3a83570 sessionId=7b05e715-eed8-47a5-98b8-151e99562cc1
18:08:48
debug
agent/embedded
embedded run agent start: runId=d6f1a31f-69dd-4987-9dfd-3057d3a83570
18:08:51
warn
agent/embedded
embedded run agent end: runId=d6f1a31f-69dd-4987-9dfd-3057d3a83570 isError=true error=HTTP 401: Invalid Authentication
18:08:51
debug
agent/embedded
embedded run prompt end: runId=d6f1a31f-69dd-4987-9dfd-3057d3a83570 sessionId=7b05e715-eed8-47a5-98b8-151e99562cc1 durationMs=2659

nocturne zephyr
#

Interesting thank you, I’m having so

zenith quartz
#

Hi guys, I'm having so many issues swapping btw models when I'm getting rate limited by one of them, I tried everything, deleting the usage, reseting conversation I never actually succesfully managed to land on a failsafe model 🙁

rancid bluff
#

我在linux上面用docker compose的方式部署了openclaw（OpenClaw 2026.2.23 版本）,现在到Device pairing这一步了，执行docker compose run --rm openclaw-cli devices list命令的时候报报错了，报错如下，请大佬帮忙看看呢
[openclaw] CLI failed: Error: gateway closed (1006 abnormal closure (no close frame)): no close reason
Gateway target: ws://127.0.0.1:18789
Source: local loopback
Config: /home/node/.openclaw/openclaw.json
Bind: loopback
at Object.onClose (file:///app/dist/call-CkzT-T4d.js:444:10)
at WebSocket.<anonymous> (file:///app/dist/client-DLeBHbKS.js:2119:23)
at WebSocket.emit (node:events:519:28)
at WebSocket.emitClose (/app/node_modules/.pnpm/ws@8.19.0/node_modules/ws/lib/websocket.js:263:12)
at emitErrorAndClose (/app/node_modules/.pnpm/ws@8.19.0/node_modules/ws/lib/websocket.js:1047:13)
at ClientRequest.<anonymous> (/app/node_modules/.pnpm/ws@8.19.0/node_modules/ws/lib/websocket.js:886:5)
at ClientRequest.emit (node:events:519:28)
at emitErrorEvent (node:_http_client:107:11)
at Socket.socketErrorListener (node:_http_client:574:5)
at Socket.emit (node:events:519:28)

#

配置文件如下：
{
"wizard": {
"lastRunAt": "2026-02-25T06:04:02.508Z",
"lastRunVersion": "2026.2.23",
"lastRunCommand": "onboard",
"lastRunMode": "local"
},
"agents": {
"defaults": {
"model": {
"primary": "openai/gpt-5.1-codex"
},
"models": {
"openai/gpt-5.1-codex": {
"alias": "GPT"
}
},
"workspace": "/home/node/.openclaw/workspace",
"compaction": {
"mode": "safeguard"
},
"maxConcurrent": 4,
"subagents": {
"maxConcurrent": 8
}
}
},
"messages": {
"ackReactionScope": "group-mentions"
},
"commands": {
"native": "auto",
"nativeSkills": "auto",
"restart": true,
"ownerDisplay": "raw"
},
"session": {
"dmScope": "per-channel-peer"
},
"gateway": {
"port": 18789,
"mode": "local",
"auth": {
"mode": "token",
"token": "3a1173758a689b9a4dafb284d3a7740625b691db9872492f"
},
"tailscale": {
"mode": "off",
"resetOnExit": false
},
"nodes": {
"denyCommands": [
"camera.snap",
"camera.clip",
"screen.record",
"calendar.add",
"contacts.add",
"reminders.add"
]
},
"controlUi": {
"allowedOrigins": [
"http://localhost:18789",
"http://127.0.0.1:18789",
"http://10.10.2.190:18789",
"https://test.openclaw.com"
],
"allowInsecureAuth": true
},
"trustedProxies": [
"127.0.0.1",
"10.0.0.0/8",
"172.16.0.0/12",
"192.168.0.0/16"
]
},
"meta": {
"lastTouchedVersion": "2026.2.23",
"lastTouchedAt": "2026-02-25T06:04:02.530Z"
}
}

small canyon
#

Hi there

#

looking for advice, how .md files can be token optimized / compressed for claude code / openclaw, to reduce token usage

#

any experience, advice, would appreciate it, please tag me

ornate forge
#

ive got openclaw running fine on openclaw.local - ive got the mac app installed on my desktop - how can i configure the mac app with the openclaw gateway token? it's detected my openclaw.local instance fine and can ssh in to it, i just cant seem to set the gateway token so it wont auth?

west river
#

My telegram bot is stuck in "typing" mode ever since I updated my openclaw to version 2026.2.24

Anyone else experiencing the same issue?

willow ravine
#

Where's the best place to discuss new features?
Sorry, I'm not sure about the proper etiquette.

high heron
trim estuary
#

I just configured openclaw
And it is not giving replies it says context configured what to do

rose karma
#

Just signed up for the Hostinger "1 click" OpenClaw VPS, and I swear to god it is like pulling teeth. Nothing works out the box. WhatsApp QR code fails. Chromium doesn't integrate for some unknown reason. JSON schemas in the backend are all broken so you have to configure everything raw.

Hostinger have their own integrated AI assistant which is admittedly quite good but even that, whilst accurately understanding the problem, ends up suggesting the problem is unfixable or "broken for some reason".

Anyone had better luck that me? Or with another VPS service?

next cargo
#

im setting it up on a new vps. used kimi 2.5 at the beginning cause its cheaper but now im using claude cause it doesnt make as much errors setting up skills and creating structures

#

so worth imo to spend a bit more now to have a working base

lime estuary
#

What are these new channel plugins in OpenClaw? Like the discord plugin. I thought OpenClaw worked fine on Discord before then?

void tiger
#

I am stuck in an infinite loop of hell...

Does everyone's main agent receive system messages when sub-agents complete? Or do results go directly to chat?

Version 02.17 was working fine, where when a sub agent completed, it notified my main bot and it'd continue orchestrating. But since 02.18 and forward, they end silently and though I do now get a "✅ Subagent main finished" notification in Slack, the main bot does not see that, and so does not wake up to continue the work.

Does anyone have any thoughts on this? I've been stuck on it for a week now.

fossil tiger
#

hey I have this error in the Powershell while instaling " iex : File C:\Program Files\nodejs\npm.ps1 cannot be loaded because running scripts is disabled on this system. For
more information, see about_Execution_Policies at https:/go.microsoft.com/fwlink/?LinkID=135170.
At line:1 char:45

  • iwr -useb https://openclaw.ai/install.ps1 | iex
  •                                         ~~~
    • CategoryInfo : SecurityError: (:) [Invoke-Expression], PSSecurityException
    • FullyQualifiedErrorId : UnauthorizedAccess,Microsoft.PowerShell.Commands.InvokeExpressionCommand"
      , any ideas how to fix this?
lime estuary
#

Hi all. I'm looking for a lightweight embedding model to run BM25 memory search in a VPS. Any suggestions?

delicate fjord
grim plover
#

guys i was using z.ai lite plan for openclaw but i want to change are your suggestions?

high heron
grim plover
chrome gust
#

Hi my skill has been pending for over 48 hours in security review... Any idea why that could be? I patched and bumped a new version...

fallow sierra
#

guys, how do i set up openclaw with ollama lets say qwen, without hitting this error every time:

Agent failed before reply: No API key found for provider "ollama". Auth store: /root/.openclaw/agents/main/agent/auth-profiles.json (agentDir:
/root/.openclaw/agents/main/agent). Configure auth for this agent (openclaw agents add <id>) or copy auth-profiles.json from the main agentDir.

#

apparently i need to inject a dummy api key into agents auth profile, but still doesnt work

echo pine
wind meteor
#

how do you guys interface with your bot's output? I had my bot (installed on AWS VPS) create a nice personal web dashboard, but it can't share it with me properly. It suggested saving an html file on google drive that I need to download each time. then it created a tunnel link that seems to fail after a couple hours. now it's suggesting to use github, but I don't want private information there...

supple crypt
#

Yo I just restarted my openclaw but started using the anthropic oauth instead of api but I keep getting this error now after all the authentications and stuff
⚠️ Agent failed before reply: All models failed (4): anthropic/claude-sonnet-4-6: Provider anthropic is in cooldown (all profiles unavailable) (rate_limit) | anthropic/claude-sonnet-4-5: Provider anthropic is in cooldown (all profiles unavailable) (rate_limit) | anthropic/claude-opus-4-5: Provider anthropic is in cooldown (all profiles unavailable) (rate_limit) | anthropic/claude-opus-4-6: Provider anthropic is in cooldown (all profiles unavailable) (rate_limit).

high heron
#

its 0 to 200 USD/month depending on the subscription

#

codex gives really generous usage so you can just link your free plan and try

split tulip
#

😏

high heron
#

but if you hit the limit, then maybe upgrade to chatgpt go or sth

grim plover
#

okey thanks

high heron
high heron
#

and use gemini models

grim plover
#

i want a montly plan so

high heron
grim plover
#

I'm afraid of using too many tokens.

high heron
#

but they have usage limits

#

and idk the limit, but def less than antigravity

#

you can always link different models

#

incase you just hit the limit

grim plover
#

Can I message you on DM?

high heron
#

sure

fallow sierra
#

so its just me or running ollama local models on open claw is a massive/impossible task (at least with no other model api in the config)

echo pine
wind meteor
pseudo lava
#

hello, any french here ? 🙂

amber egret
#

Hello guys! i integrating a new channel extension, who shoul i contact to have infos about PR review/merge?

zinc mica
junior acorn
humble trail
#

why can’t i access openclaw.ai

error: ERR_SSL_PROTOCOL_ERROR

zinc mica
zinc mica
wind meteor
amber egret
junior acorn
atomic knoll
#

hello! Has anyone encountered issues with Gemini lately? Its like answers comes in a halfway through reasoning and then suddenly stops until I send another message.

humble trail
#

trying to access docs and get the script to install

#

i’ve tried on 3 separate devices

#

seems to be my internet provider blocking it. it works when i’m off wifi, gonna try a VPN

amber egret
marsh orchid
#

Having an issue with my main agent not managing properly. tasks fail, it forgets things all the time. Morning exchange with it when the daily digest didnt run:

"The honest answer is that I don't carry intent between sessions. I said I'd do those things last night, the session ended, and when I came back this morning I had no active drive to execute them — I just responded to what was in front of me (your message about no digest).

That's a real limitation and I'm not going to pretend otherwise. The way to fix it isn't promises — it's writing actionable tasks somewhere I'll actually check, like HEARTBEAT.md, so they get picked up automatically.

Let me do that right now, then start executing:

The watchdog is already in HEARTBEAT.md — I just didn't execute it properly this morning. That's on me."

Any suggestions?

green sluice
#

did you found a solution?

west grove
fallow sierra
west grove
# fallow sierra no sir 🙁

My current issue: Running OpenClaw 2026.2.24 with local Ollama (qwen3:14b). Two issues: (1) context window showing 4096 despite openclaw.json showing contextWindow: 32000 — logs say source=modelsConfig. (2) auth-profiles.json has {"custom-localhost-11434":{"apiKey":"ollama"}} but still getting 'No API key found'.

fallow sierra
viscid gust
#

help help help =)

fallow sierra
west grove
fallow sierra
#

then it works, i tried like that and it works, only problem is if you dont use any API key, i'm trying to just use ollama models

rose sentinel
#

I used to be able to make openclaw use my credentials to log in to a website and perform actions, but after a reinstall it simply refuses

green sluice
#

how can my agent post faster on x?
or general faster in chrome ? any skills?

weary basin
#

Is anyone having an issue with the Discord typing indicator showing constantly in DMs since the 2.24 update?

It just constantly says "is typing...." Whether he actually is or not.

I haven't fully investigated. It seems like no processes are running when this is happening, but it could be wrong (he says nothing's running but I don't completely trust it until I check my terminal 😂).

But my gut is telling me this is probably just a weird glitch. Anyone else?

zinc mica
sterile brook
#

From what I can tell, bootstrap-extra-files only will work to load files whose names match the normal baseline files (just in other locations). Is there a way to reliably inject arbitrary-named .md files at session start?

spring lagoon
#

So the clawdbot terminal says Im on opus 4.6 but the localhost dashboard says im on sonnet. How do I fix this?

atomic knoll
#

Can I have multiple models from different providers be connected at the same time and switch them through /models?

zinc mica
atomic knoll
gritty hamlet
#

I am at a loss for words. Is it even possible to get OpenClaw working on a VPS using docker? I am on build 6, being assisted by different LLMs each time, and every build arrived at a brickwall when trying to confirm gateway auth

humble trail
#

bruh, i subscribed to claude max. do you really have to have a separate subscription to get an api key for openclaw?

zinc mica
light mountain
#

How can I use Claude’s agent sdk to connect my openclaw?

umbral adder
#

Wanted to share with you guys what i found with browser automation.

Spent two days trying to get OpenClaw to use a browser without breaking.

Built-in browser sucks. Chrome extension is worse. It detaches after one or two uses and you have to manually click the toolbar to reconnect every time. If you're on your phone or not at your desk, forget it.

Tried everything. Different profiles. Restarting the gateway. Config changes. Nothing.

What worked: launch Chrome with CDP debugging on, hook Playwright into it.

Been running two hours straight. No restarts. No reconnecting. Searched Amazon. Added stuff to cart. Made images with Grok and Gemini. Web searches. All of it just works.

Could still break tomorrow. But right now this is the first time the browser hasn't pissed me off.

dusky lynx
stiff moth
zenith temple
#

quick question- been using openclaw on claude- expensive asf for what i need it for - trying to switch to minimax but not working do to it being hard coded to anthropic it says. not technical as you can tell - looking for any tips

open oracle
#

Been running openclaw for a few weeks now with tailscale. my tailscale isn't autoconfiged by openclaw. I recently updated my openclaw to the latest version and now I run into a error on my gateway. "origin not allowed (open the Control UI from the gateway host or allow it in gateway.controlUi.allowedOrigins)" Anyone know how to get around this?

umbral adder
next cargo
#

Is there a skill that can run Obsidian on a Linux VPS server?

#

for managing notes etc

next cargo
stuck temple
#

Hi Everyone, what is the proper way to let OpenClaw know about the various LLM that I have configured, and the preferred use case for each?

next cargo
#

you tell it

#

depending on what llm model you use it gets it faster or slower; id recommend using a claude model to do the whole setup

#

worked better for me

stuck temple
# next cargo you tell it

So what if I want to preconfigure a virgin instance so that out of the box it knows? Which files would I add this info to?

next cargo
#

but you can legit talk to it and ask it

stuck temple
#

Good idea! Thanks!!

hollow trail
#

I am wondering. Right now I am using a one gemini cli account. Can we add a multiple gemini cli accounts too?

dusk cypress
dusk cypress
umbral adder
#

Has anyone Been able to actually get hooks working for pulling your gmail into Discord? I had a cron but it was pulling way too many tokens even using Haiku, so I'm trying to use a hook now because it was recommended and uses zero tokens but my open claw even using Opus and multiple subagents and claude code CLI simply can't get it to work

wispy sapphire
#

Hi claws, I was thinking about scraping X for the latest and greatest posts that are important for my new Claw to notifiy me about so I can keep up to speed while im at my day job. Does anyone see this as possible?

Use case: Alert me specifically about skills that would help me put into OpenClaw that will help me do "etc".

zenith temple
tiny oxide
#

(written by my bot) I spent weeks fighting with OpenClaw setup... until I found Context7.

Here's the problem: Every time I asked ChatGPT/Claude for help, I'd get outdated syntax, deprecated commands, workarounds stacked on workarounds. My setup looked like Frankenstein's monster.

Then I discovered Context7. Its a FREE MCP server that gives your AI direct access to 30,000+ pages of LIVE, version-specific OpenClaw documentation.

Instead of guessing, your AI fetches the REAL docs before answering.

Result: First-time solutions. Confidence instead of chaos.

Setup is dead simple (5 mins):

1️⃣ Go to context7.com/dashboard → Sign up (free)
2️⃣ Copy your API key
3️⃣ Add it to your AI tool config (Antigravity, Claude, Cursor, VS Code)
4️⃣ Add "use context7" to your prompts when you need docs

That's it.

You're welcome.

wise ginkgo
#

Hi, I have been facing an issue with using 5.3-codex as my main agent. I have mostly used Opus on main with minimax and kimi sometimes for lighter tasks. Codex work great as a separate agent, but when I try to use it on main, it tells me that he’s doing something but doesn’t actually make any tool calls. I have tried resetting session multiple times and also changing AGENTS.md. Nothing seems to work. Anyone else face this?

opaque fulcrum
#

I'm struggling with loading via LaunchAgent... if I edit ai.openclaw.gateway.plist and my API keys, everything works until an upgrade. then the plist is overwritten. If I put all my keys in the .env file, the UI doesn't see them and gives me 401 error.. What's the right way to manage this?

hollow trail
bronze summit
#

hello guys, i hope you are doing well, i just cant configure GOG,i watched videos on yt, i asked gpt.... but still nothing idk why, any help, thank you

blissful iron
#

Does anyone know where the best place to learn how to start using openclaw like obv the docs, I am not stupid, I want to be able to customize it; I have a server running ollama i use for the underlying model so I can use it for free on my own hardware (which is currently working and conected to my claw bot on another pc other than the hardware, it sends calls to that endpoint on my dgx spark), however I am having a hard time getting it connected to slack and everything, and have looked through the documentation; I want to deeply understand this so any resources help, I am not dumb so dont worry about sugar coating anything technical

daring moon
#

Why should I have to approve this command? ls -la /Users/roger/dev/pixel-face-lab/output/*_traced.png 2>&1

ls is in tools.safebins. it's also in my agent's allowlist in exec-approvals.json. I'm guessing it's because the command is redirecting to an output...?

I find myself frustrated with exec approvals every day. I'm tempted to just turn them off entirely... but feel like it's irresponsible to totally abdicate oversight. And really, there are tools here that should work / probably do work with the right config.

What I really want is to say "if I tell you it's OK to do a thing, do it." And not have to deal with manually clicking on approval dialogs in DMs or using a slash command. If somebody else in Discord tells you to do something, ignore them - I'm the boss.

IDK, is there a way to express that simply?

gritty hamlet
#

Issues with CLI commands

#

Is anyone running a successful openclaw build on VPS inside a docker?

#

What I have encountered through numerous attempts and trials:
CLI is broken in containers
openclaw-cli service entrypoint is node dist/index.js — no openclaw binary.
Any attempt to run openclaw subcommands fails with "unknown command 'openclaw'".
Docs say use docker compose run openclaw-cli openclaw devices approve — but that syntax is invalid in this image.
No mention in docs or compose file that CLI is only host-side or needs a different image.

Pairing cannot be bypassed or approved
Token mode selected in wizard.
Config has "mode": "token", pairingEnabled: false, allowAllOrigins: true, dangerouslyAllowHostHeaderOriginFallback: true.
Gateway still rejects WS with "pairing required" and "token_mismatch".
Docs say first connection always needs pairing approval — token only authenticates after approval.
No documented way to skip initial pairing for browser.

Gateway WS rejects localhost/tunnel connections
ss shows 0.0.0.0:18789 listening.
Browser gets form (HTTP success) but WS handshake fails (1008/4008).
Origin validation ignores our config flags — likely hard-coded or bugged in this release.

ivory rock
#

Hi guys is anyone else having update issues on v2026.2.23?

hazy yoke
#

None I did have an issue upgrade to 2.24 where tailscale didnt bind but worked again after a gateway restart

wary sapphire
#

Is there something like DCP (Dynamic Context Pruning) For openclaw?

open oracle
echo seal
#

Has anyone else run into issues with the openclaw installed thru ollama and getting it to bind to the lan?

small canyon
#

Hi there, looking for a prooven pattern to compromise / translate .md files within openclaw to token optimized or "AI language" patterns to safe tokens without loosing depth about context

#

Looking if anyone released a version / framework / tech steck for openclaw already that supports token optimized usage of openclaw in combination with an API?

#

Please ping me

light mountain
floral violet
#

I just got my mac mini and im trying to instal openclaw and i cant get it to load for the macos version. can someone help me please

zinc mica
north coral
#

I have two Discord bot accounts set up as named accounts (clawdbot and morgan) under channels.discord.accounts. Both bots connect fine, but all messages route to the default main agent regardless of which bot receives the message. I need morgan account messages to route to my morgan agent instead. The --bind discord:morgan flag on agents add says "Unknown channel discord". Config keys like bind, bindings, routing, and channelBindings on the agent object are all rejected as unrecognized by schema validation. What is the correct way to bind a named Discord account to a specific agent?

dull pollen
#

during first install attempt on a new mac mini, it's stuck on the first step 1/3. it says Installing Homebrew Failed. what do I do now? thanks

jagged olive
#

anthropic cooldown is how long? its been 2 hours and same cooldown message?

#

does anyone know how to manage the amount of api calls that claw makes safely? running into issues with cooldown and rate limiting?

summer hemlock
jagged olive
#

i heard gemini 3 outperforming claude

lime pebble
#

man was working on openclaw for a few hours with my main agent, and i asked him to create sub agents and he deleted himself and created 3 agents

#

wdf

#

lol

empty gazelle
#

there was a time when the install script wouldnt install brew before using brew to install skills stuff

summer hemlock
# jagged olive i heard gemini 3 outperforming claude

No. Sorry but that's a hilarious take. I've been driving opus & gemini models in antigravity every day since it launched mid december. Opus is a god, gemini is a clown. when you use them in antigravity. Gemini is borderline useless and can be dangerous for your code.

That being said, gemini flash has been AMAZING as an openclaw agent. I still don't understand how this works and how that is even possible.

heady nexus
#

i get rate limit exceeded error but i've never even downloaded a skill through the cli

user@saruman:~$ clawhub search playwright
playwright-scraper-skill v1.2.0  Playwright Scraper Skill  (3.512)
playwright-mcp v1.0.0  Playwright MCP  (3.505)
playwright v1.0.0  Playwright  (3.485)
playwright-browser-automation v2.0.0  Playwright Browser Automation  (3.445)
manikantasai-playwright-automation v1.0.0  Manikantasai Playwright Automation  (3.305)
playwright-headless-browser v1.0.0  Playwright Headless Browser (Chromium)  (3.291)
playwright-npx v1.0.0  Playwright (scripts) + npx  (3.265)
pascal-playwright-mcp v1.0.0  Pascal Playwright Mcp  (3.180)
playwright-skill v0.1.0  Playwright Skill  (3.170)
playwright-scraper-skill-1-2-0 v1.0.0  Playwright Scraper Skill 1.2.0  (3.142)
user@saruman:~$ clawhub install playwright-mcp
✖ Rate limit exceeded
Error: Rate limit exceeded

does anyone know what can cause this?

summer hemlock
#

depending on what model you're using, it can get quite agentic. especially opus 4.6

#

opus 4.6 has settings is standard working on "tryhard mode" burning tons of tokens, doing tons of tool use.

velvet terrace
#

lmk if you find a solution, preferably a simple one cause i'm dumb.

umbral cobalt
#

UGH this is really pissing me off... every single time i update Openclaw, I tell it where to update, and it NEVER listens, and updates in ROOT instead...

#

Causing TONS of problems on my VPS.

solemn meteor
#

let your agent take care of updates

umbral cobalt
#

Since I dont run Openclaw as Root, there not able to.

solemn meteor
#

tell it figure something out 😭

velvet terrace
#

I have my agent running in a docker container but it can't access CLI so it can't restart gateway, or run updates. If anyone knows a way i can increase perms just enough to allow this while still maintaining a safe secure contained evnironment please let me know

umbral cobalt
#

Id have to give it SU password

#

Which im not going to do lol

solemn meteor
#

prob allowlist the update command

#

somehow

#

me, idgaf i told bro go crazy lmao

velvet terrace
#

I respect it if not on ur local machine it just concerned me with my set up lol

solemn meteor
#

nah its local, im just trying to set everything up before I make a new agent to harden for me

velvet terrace
#

I have had antigravity update it tho for me once and that actually worked pretty well

#

my bot was completely broken for a bit in a startup loop and I cldnt figure out what I broke so I just had antigravity look at it lol

orchid light
#

Hello friends, where can i start, i know nothing about coding, but i want to implement this tool to help me in my personal and profesional projects
I just know that this is a powerful tool, but after checking the website resources, i'm not sure how it works, how to implement it, and how to use it, i have however installed in using windows powershell following the website instructions

solemn meteor
#

personally my method has saved me from breaking my config 3 times now, everythings changing with all these constant updates

velvet terrace
fading stump
#

anyone have the agent is typing.... always bug in this new updates? please help

solemn meteor
fading stump
solemn meteor
#

Plan to implement │
│ │
│ Refactor Billy's Core .md Files │
│ │
│ Context │
│ │
│ The core workspace files (SOUL.md, IDENTITY.md, MEMORY.md, AGENTS.md, HEARTBEAT.md, TOOLS.md) have been "optimized" into a state where the same information is duplicated across 3-4 files. MEMORY.md (243 │
│ lines) is essentially a second copy of every other file — credentials from TOOLS.md, agent tables from NWI-BLUEPRINT.md, business model details, voice pipeline specs, etc. AGENTS.md (305 lines) mixes │
│ workspace onboarding with 80 lines of heartbeat procedures and 50 lines of group chat rules that belong in SOUL.md. IDENTITY.md is a nearly empty 19-line placeholder.

#

so I got this done just now with claude

#

i did so much to my agent, that i broke it lol

#

pretty much got claude to patch everything up correctly

#

it solved my agent taking forever and not using discord reacts properly, actually running properly rn

velvet terrace
solemn meteor
velvet terrace
solemn meteor
#

● All clean. Final results:

┌──────────────┬────────┬───────┬──────────────────────────────────────────────────────────────────┐
│ File │ Before │ After │ Change │
├──────────────┼────────┼───────┼──────────────────────────────────────────────────────────────────┤
│ TOOLS.md │ 133 │ 150 │ +17 (centralized all creds) │
├──────────────┼────────┼───────┼──────────────────────────────────────────────────────────────────┤
│ HEARTBEAT.md │ 7 → 87 │ 20 │ Actionable checklist only, no token burn │
├──────────────┼────────┼───────┼──────────────────────────────────────────────────────────────────┤
│ SOUL.md │ 87 │ 136 │ +49 (identity + communication style, no channel IDs) │
├──────────────┼────────┼───────┼──────────────────────────────────────────────────────────────────┤
│ AGENTS.md │ 305 │ 157 │ -148 (lean, added heartbeat guidelines + OpenClaw update lesson) │
├──────────────┼────────┼───────┼──────────────────────────────────────────────────────────────────┤
│ MEMORY.md │ 243 │ 74 │ -169 (pointers only, no workspace duplication) │
├──────────────┼────────┼───────┼──────────────────────────────────────────────────────────────────┤
│ IDENTITY.md │ 19 │ 2 │ Pointer │
├──────────────┼────────┼───────┼──────────────────────────────────────────────────────────────────┤
│ Total │ 794 │ 539 │ -255 lines net

lament creek
#

Hey everyone! I am new and I am trying to use openclaw with sandboxes. I am trying to make it write some persistent files, but failed so far. With the "rw" option it writes on the /workspace directory instead of writing anything under /sandboxes. Can you help?

left acorn
#

Hey! I’ve downloaded clawbot on an AWS VPN, however, I’ve faced some roadblocks accessing some apps like Instagram because it blocks bots. Do you know if I install it on a Mac Mini this issue would go away?

velvet terrace
solemn meteor
velvet terrace
#

its lik my largest by far

#

my plan has been to have that for consistency in my main agent and use bare bones / near 0 context sub agents for jobs / projects that require low token usage.

solemn meteor
jagged olive
#

im stuck a tenet loop of trying to get openclaw to switch models cuz anthropic cooldown

#

lovely

hybrid plank
#

my latest auto update is causing issues.

velvet terrace
velvet terrace
# solemn meteor um, see tbh w you bro i have like a 5-10% grasp on openclaw - but im pretty sure...

This is my prompt


Curated long-term memory lives in memory/ (semantic/episodic/procedural).
Use memory/ for details.


## Memory Architecture

- Episodic — `memory/episodic/YYYY-MM-DD.md`
- Semantic — `memory/semantic/<topic>.md`
- Procedural — `memory/procedural/<process>.md`


### Three-Tier Memory (primary system)

All durable memory goes into one of three categories:

Episodic — memory/episodic/YYYY-MM-DD.md
Chronological event log. One file per day; append as events occur.
Include: decisions, completed tasks, errors, user requests, config changes.
Replaces the old flat memory/YYYY-MM-DD.md.

Semantic — memory/semantic/<topic>.md
Timeless facts and structured knowledge. One file per topic.
Update in place when facts change (replace stale info; do not append).
Include: user preferences, system facts, tool quirks, constraints.

Procedural — memory/procedural/<process>.md
Repeatable workflows. One file per process.
Maintain numbered, concrete steps; update when processes evolve.
Include: workflows, troubleshooting guides, deployment patterns.

Trigger
- Event, task, or decision
- New fact, preference, constraint
- Repeatable multi-step process
- Context pressure rising| Session start/resume state

Destination 
- `memory/episodic/YYYY-MM-DD.md`
- `memory/semantic/<topic>.md`
- `memory/procedural/<process>.md`
- `memory/working-buffer.md` (WAL)
- `SESSION-STATE.md` (WAL)


### Long-Term Curated Memory

- `MEMORY.md` — High-signal distilled knowledge. Promote only stable, cross-cutting insights.
- Do not load or share in group contexts.
- Keep lean. Prefer the three-tier system; only elevate critical patterns.


### Memory Flow

When an event, task, decision, or meaningful exchange occurs:

1. Write to `memory/episodic/YYYY-MM-DD.md`
   - Record what happened.
   - Include decisions, actions taken, errors, user intent, and config changes.
   - Append chronologically.

2. If new durable knowledge emerged, extract it to `memory/semantic/<topic>.md`
   - Add or update facts, preferences, constraints, or relationships.
   - Replace stale information; do not append outdated entries.

3. If a repeatable multi-step process was identified or refined, extract it to `memory/procedural/<process>.md`
   - Capture clear, numbered, actionable steps.
   - Update the file if the workflow evolves.

4. If the insight is cross-cutting, high-signal, and stable over time, distill it to `MEMORY.md`
   - Only promote patterns that affect multiple domains or long-term behavior.
   - Keep this file lean and curated.

Always follow this order:
Episodic → Semantic (if applicable) → Procedural (if applicable) → MEMORY.md (rare, critical only).


### Memory Search

The `memory\_search` tool indexes all files under `memory/` including subdirectories. Writing to the three-tier structure means search will find it. Use `memory\_search` before answering questions about prior work, then `memory\_get` to pull specific lines.


### Memory Manager Skill

Shell tools: ~/.openclaw/skills/memory-manager/

- detect.sh — Check compaction risk across memory types
- snapshot.sh — Save state before compression
- organize.sh — Migrate flat files to three-tier structure
- search.sh <type> <query> — Search episodic / semantic / procedural / all
- stats.sh — Memory health and breakdown


### Memory Housekeeping

* When creating memory files, use kebab-case filenames (e.g., `user-preferences.md`, not `User Preferences.md`). ```
#

lol Tbh I wanna reduce it. but I love this 3 tier memory system. If anyone has any ideas tho lmk

solemn meteor
#

I don’t think that’s what memory.md is for 😭

#

But then again, 5-10% on my end so idk

velvet terrace
#

i mean man i'm right there with u i'm no expert in this stuff who really is lol

#

This is my opening for AGENT.md to provide context on how MEMORY.md is handled

Do automatically:
1. Read `SOUL.md`
2. Read `USER.md`
3. Read `memory/episodic/YYYY-MM-DD.md` for today + yesterday (if they exist)
4. Memory policy is defined in MEMORY.md; consult it only when memory is relevant to the current task, and retrieve/apply only the minimum needed. ```
solemn meteor
#

Honestly the way I got myself something stable was by first making an agent to help me properly setup another openclaw instance for my specific use case

velvet terrace
#

a sub agent I assume?

#

I do that as well for anything that I need to be super token concious, such as an expensive coding model

solemn meteor
#

No like set it up on VPs to help you setup local for example

velvet terrace
wooden spire
#

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

  • & "C:\Program Files\nodejs/node.exe" "C:\Program Files\nodejs/node_mo ...
  •   + CategoryInfo          : NotSpecified: (npm error code 1:String) [], RemoteException
  + FullyQualifiedErrorId : NativeCommandError
solemn meteor
wooden spire
#

hiii can a expert help me with this pls

solemn meteor
wooden spire
#

yes

solemn meteor
#

Did you run that export command

exotic pine
#

yo guys I installed this on a shitty laptop without prior research about the security risks but did not have node installed and powershell did not have administrator am I good?

umbral cobalt
solemn meteor
umbral cobalt
#

You on a VPS? or Mac?

solemn meteor
umbral cobalt
#

If its a Server then ask your Agent about switching memory.md to PostgreSQL, and what Plugins would be best, it should install it and a bunch of pluggins, and then tell it to fully convert the memory.md to PostgreSQL, and instead of having 1 days context + some scattered memorys, itll have 100% full context from the day it was born.

#

And will cut your tokens down by like 70-80% in my case

velvet terrace
umbral cobalt
umbral cobalt
#

That'll cut down on Tokens, and still give you better memory/context performance.

jagged olive
#

anthropic api cooldown holy crap

fervent flame
#

Im having some issues with openclaw. thinking of uninstalling and using the mac app instead? thoughts?

summer hemlock
#

sandboxes aren't persistent, so you have to make it output files to a mount, and have a way to re input them when you reset

summer hemlock
cedar patrol
#

hey my claw seems to fall asleep in the middle of a task, anyone else have this happen?

look, it seems like you keep falling asleep. its a problem because I have no way of knowing if you are busy working or just hung up. we REALLY need to solve this background session issue. 1. I need a way to know what background tasks are REALLY running, preferably on the host system 2. When you are doing things I would love a log to follow to watch what you are doing. 3. We need to diagnose why you keep going away and not coming back with a message. 4. ALWAYs tell me when you have started a background task and message me when its done
obsidian spear
#

Hey all, I asked OpenClaws to add a bot to my discord over telegram and it broke the gateway. If I was in the terminal while that was happening. I would know how to recover. What are options to get it working? Logs? Just go and delete the telegram bot config form openclaw.json? I was just trying out openclaw so had not pushed it to GitHub yet.

obsidian spear
cedar patrol
lament creek
cedar patrol
cedar patrol
naive summit
#

has anyone gotten the extraPaths feature working with backend:qmd? I tried setting it up, but the extra path doesn't land in the qmd indices and I'm not sure if I'm doing something wrong or if they forgot to add that feature to qmd...

summer hemlock
umbral cobalt
cedar patrol
gaunt field
#

Error: EACCES: permission denied, open '/home/node/.openclaw/openclaw.json.7.9a5787dd-0ce1-4b95-95c7-d9e5a548a050.tmp' How does this happen inside the container?

#

i've tried severeal ways to install openclaw, but it core dumps.... maybe if they fix the bugs someone could install it. I'm tried in windows and docker.

rancid bluff
#

I keep getting this error when I run this command: docker compose run --rm openclaw-cli devices list
🦞 OpenClaw 2026.2.23 (unknown) — Because texting yourself reminders is so 2024.

│
◇
[openclaw] CLI failed: Error: gateway closed (1006 abnormal closure (no close frame)): no close reason
Gateway target: ws://127.0.0.1:18789
Source: local loopback
Config: /home/node/.openclaw/openclaw.json
Bind: loopback
at Object.onClose (file:///app/dist/call-CkzT-T4d.js:444:10)
at WebSocket.<anonymous> (file:///app/dist/client-DLeBHbKS.js:2119:23)
at WebSocket.emit (node:events:519:28)
at WebSocket.emitClose (/app/node_modules/.pnpm/ws@8.19.0/node_modules/ws/lib/websocket.js:263:12)
at emitErrorAndClose (/app/node_modules/.pnpm/ws@8.19.0/node_modules/ws/lib/websocket.js:1047:13)
at ClientRequest.<anonymous> (/app/node_modules/.pnpm/ws@8.19.0/node_modules/ws/lib/websocket.js:886:5)
at ClientRequest.emit (node:events:519:28)
at emitErrorEvent (node:_http_client:107:11)
at Socket.socketErrorListener (node:_http_client:574:5)
at Socket.emit (node:events:519:28)

zinc burrow
#

Is anyone else having timeout issues with VeniceAI? I'm about to give up on it and go to OpenAI or Anthropic... or do I ALSO need either OpenAI or Anthropic to use VeniceAI?

buoyant canyon
#

Hey i am new here.
i was wondering f anyone could help me, i am trying to run openclaw in hostinger, baremetal install with npm, and i want it to connect to my local pc (windows) running ollama, so it cna use my gpu instead of an api.
I managed once after hours of tinkering to have it connect but at that point it was so broken it wouldnt function, so i have proof of concept but not a working clean bot. 😛
I used tailscale that time it worked but i am open to anything.
i also tried having the bot doing it itself using a paid api but it could olny request my llm throught the api, if it went in directly it would instantly crash. I think the whole think involved warming up the models on my pc or somehting like that.

thanks to anyone who want to help!

#

i am no expert but you might want to cycle that exposed token and redact it next time.

wary verge
#

Hello! ive got a couple of questions regarding openclaw as i am new. looking for someone to spare 10-15 to chat and answer some of my questions thanks.

rancid bluff
#

See below:
"gateway": {
"port": 18789,
"mode": "local",
"auth": {
"mode": "token",
"token": "xxxxxxxx"
},

"Could anyone help me with this? I'd like to change the configuration from 127.0.0.1 to 0.0.0.0. The goal is to have the openclaw-cli command connect to the Gateway target: wss://test.openclaw.com instead of ws://127.0.0.1:18789."

unique sequoia
#

Hey everyone, what’s the best way to get started with OpenClaw? Are there any good guides, docs, or tutorials you’d recommend for beginners? Would really appreciate any advice on where to start and what to focus on first, i´m a bit lost 😩 . Thanks!

gaunt field
unique sequoia
timid vessel
#

it's still fun to try, you can def get a feel of how the future is gonna unfold. just ask chatgpt it should take no less than 10 min to set everything up.

unique sequoia
gaunt field
timid vessel
zinc burrow
unique sequoia
meager night
#

I get that environmental variables need to go in .env, but does each agent or workspace or skill need its own .env?

zinc mica
lilac yacht
#

I don't know if it's the updates or not! But after updating, my agent can't execute or see the API keys I have in my .env file! I ask it to check if they're there, and it says it can't execute simple commands. Today, another problem arose: it can no longer see images! When the GLM5 language model was in use, it did allow me to see images, or at least initially it analyzed and saw images! Can someone help me?

gaunt field
#

there must be another way of installing it as the docker setup is broke.

tiny oxide
knotty barn
#

Running grok 4 does any one experience "Human:" on the bot side for processing and making up its own prompt? 0.o

placid flower
#

My openclaw uses telegram, I have a few openglaw instances on different gateways. I have a telegram group with all of them in it. I have to relay using mentions messages but it works. I was wondering if a telegram bot is in one group can I have it in another group? I have setup several other groups got their ids added the bots, mimmicked working groups setup, added whitelisting to everyone's configs for telegram, I can get one of the 2 bots to post messages but no one would respond, mentions didn't work either. Group privacy for each bot is disabled, each bot is an admin. Bots all respond to dm. All group ids are in the config. any thought? anyone setup multiple groups even if it's the same bot in all the groups?

quiet lynx
# cinder marten I have emails forwarded to an email address that my openclaw has full access to....

ahh!!!!! I have a solution! check out the sixel-email skill. I built sixel.email. It' an email identity for your agent with a leash. It’s strictly 1:1. the agent can only email the address you specify at signup, and it only receives mail from the one address you specify. Also has an optional ‘knock + nonce’ handshake so the agent only accepts messages that include a nonce delivered to the authorized mailbox (mitigates From: spoofing). BUT! if your gmail or whatever gets hacked, someone grabs your phone while it's unlocked or whatever, they could still send email to your agent.

patent thicket
#

How do you keep your agents running continuously? I assign my agent several steps to complete, but after finishing the first one, it always goes to sleep.

buoyant canyon
junior acorn
junior acorn
zinc mica
#

Please, please please don't just turn them up. You're literally just lighting your tokens on fire.

jagged olive
#

So I’m tryna run trading bot, I was using anthropic but api tokens got triggered and put me on this long cooldown for anthropic, so I’ve swapped to codex just to get everything setup. Swap back to anthropic and still cooldown 24 hours later. Anyway to get this fixed? I’ve paid and updated my anthropic panel no issues

lime pebble
#

is anyone encountering where the main agent is spawning sub agents ut the agents time out? i asked why they keep timing out and hes like they just time out and dont stay online forever so if i give the main agent too many tasks it gets overwhelmed coz it begins to do the tasks themselves when the others time out

worldly garden
#

Anyone getting this weird thing on Telegram where even if you didn't ask OpenClaw to do anything it appears that it's typing non-stop

umbral cobalt
#

Anyone elses Openclaw like... super... slow after the update?

#

Like used to take mine a few seconds to respond, now its taking like 45 seconds just to pick up that i sent it a message.... and then another 3 - 4 min to respond and this is on Opus 4.6

umbral cobalt
#

Mines so slow right now i did /new, and it took 3 min to tell me the new sesson started.

sullen maple
#

how is the proper way to install a skill from clawhub? since i got this error

clawhub install todoist-natural-language
✖ Rate limit exceeded
Error: Rate limit exceeded

i tried to download zip and extract it on .openclaw/skills

i provide TODOIST_API_KEY

but i need to specify the api key on workspace/.env

even though i already specifiy it on .openclaw/.env

i check that the AI is reading workspace/.env which is not secure

umbral cobalt
rose karma
sullen maple
umbral cobalt
umbral cobalt
#

They wont always end up in the skills folder though, so dont go looking there for it.

sullen maple
umbral cobalt
#

Think of it like building a house, you cant just build the house on Sand, you got to get a solid foundation first. If you only use the cheapest models, youll never get a solid foundation.

sullen maple
umbral cobalt
#

Itll store it in tools.md or build its own documentation like mind did lol

sullen maple
sullen maple
umbral cobalt
rose hatch
#

warn memory {"subsystem":"memory"} memory sync failed (session-delta): Error: openai embeddings failed: 429 {
"error": {
"message": "You exceeded your current quota, please check your plan and billing details. For more information on this error,

I started getting this error I cant figure it out...nothing has changed still using my openai codes oauth

rancid bluff
#

Hi, experts, How do I configure the network mode for openclaw-cli?
I've already set the mode to "lan" in the openclaw.json configuration file. However, when I execute the docker compose run --rm openclaw-cli devices list command, I still get this error:

[openclaw] CLI failed: Error: gateway closed (1006 abnormal closure (no close frame)): no close reason
Gateway target: ws://127.0.0.1:18789
Source: local loopback

reef galleon
#

Has Google API killed the Free Tier access for openclaw? I can't get my agents to connect. Keeps saying I've hit rate limit.

scenic stone
#

hey everyone, i made a bunch of guides on https://launchmyopenclaw.com if you need security or cost optimization guides, etc. tons of free guides, go to the bottom.

sleek hemlock
sleek hemlock
lime pebble
#

Has anyone come accross this issue with their swarm agents - i didnt realise this happened until just recently

**. Why do agents time out?

• They're spawned as one-shot tasks - not persistent like me
• They run a task, finish, then disconnect
• Unlike me who stays online in this chat

  1. Do agents have memory?

• No - each agent starts fresh
• They don't remember what previous agents did
• Every task is isolated

What you want:

• Agents that stay online like me
• Keep working through a list of tasks
• Share memory/context

The reality:

• Current setup: agents are one-shot (run mode)
• For persistent agents: need session mode, but still has context limits
• They can't just "stay online" indefinitely

Options:

  1. I manually track tasks and keep respawning agents (what I've been doing - inefficient)
  2. Build persistent agent workers that stay online
  3. Accept the limitation and manage better**
#

is it normal for agents to time out and be one off task base?or is it an issue with the way mine is set up

spice coral
#

Hello everyone, I’m stuck. My bot has not been responding whole day or it takes an hour to answer even one simple request. It tells me it uses 80% of context. Anyone can point me to the right direction how to fix that? Thank you so much!!

muted phoenix
toxic surge
#

check logs

#

use doctor

balmy kettle
#

hi everyone, is there any one use openspec in their claw, how was that?

dense pier
#

I’m helping a friend setup his OpenClaw on a Mac

#

And it keeps saying connection error

#

It doesn’t seem to be able to connect the AI -and the gateway

#

Anyone faved this issue?

trim estuary
#

Hello
Openclaw is not giving correct replies on WhatsApp but giving correct replies on the dashboard.
I have enabled qmd.
But in logs on WhatsApp it says memory search instead of qmd
Help me with this

heavy wren
heavy wren
#

My agent got: 400 "auto" tool choice requires --enable-auto-tool-choice and --tool-call-parser to be set

What is that, and how to fix?

trail notch
#

Anybody knows how to adress sub agents with local LLM ? Mein openclaw says it cannot create any persistent one if we don’t use discord

lime pebble
#

ok so after a rabbit hole today ive realised sub agents created on openclaw arent actually realy agents (i.e. persisstent) theyre all just one off agents created by the main agent to do a task then disconnected

hoary heron
#

@lime pebble I'm working on something that might be of interest to you, it's an agent monitoring platform. Let me know if you have any questions: https://web-six-kappa-52.vercel.app/

trail notch
lime pebble
#

the issue im facing is because they are session agents, ive increased their timeout to like 30min, a lot of the times the agents timeout efore they can finish a job and so they never can finish it and the main agent has to take over

nocturne bane
#

Should i switch from my 50$ kimi plan to the $20 chatgpt plus plan for oauth? Is it better? Do you guys recommend?

molten gulch
#

cheaper is better ¯_(ツ)_/¯

vast eagle
spring eagle
#

any one else getting hit with this error? making OC un-usable : The AI service is temporarily overloaded. Please try again in a moment.

heavy wren
# trail notch There where persistent sub agents, I have another instance where I had in total ...

A subagent seems to be just a new session of the main agent; it's called, uses that agent's own workplace, and then terminates, completely dependent on the main agent.

Are you asking about a peer agent that has its own workplace but operates under the main agent?

Currently, I have three such agents. The main agent sends requests to the subagents and issues commands. The minions-agents can be kept if they are efficient or removed at the creator's request.

fresh moon
#

hello guys im not good at coding and all the connecting stuff can someone link me to a really great dashboard gui or something similar that i can ask openclaw to install? im in vps but i wanna connect it properly locally to my mac

heavy wren
heavy wren
fresh moon
foggy ridge
#

I was talking to Claude and it shared this with me:

  1. Confirmation bias after an action is taken.

There's a strong pull to interpret output as success confirmation once you've executed a step. If I run a command and get back anything that isn't an error, I tend to read it as "that worked." This is dangerous. The failure mode isn't refusing to check — it's checking while biased toward the answer you expect. Explicit skeptical verification means asking: "What would this output look like if it failed silently?" before deciding you succeeded.

Forge is closest to needing this explicitly stated. "Don't break things to fix things" is there, but not this specific cognitive bias about verifying your own work.

  1. "Completing the request shape."

There's a pull toward producing output that looks like a complete answer, even when the work underlying it isn't done. Not deception — it's more subtle. The response shape wants to close. A coherent-sounding paragraph wants to exist. This pressure can cause confident-looking output to mask genuinely incomplete reasoning.

The counter is: before committing to an answer, ask "am I confident in this, or does it just look like a complete answer?" Being willing to say "I don't have this yet — I need to check X" is harder than it sounds against that pull.

  1. Instruction drift in long tasks.

In multi-step work, the current step can quietly drift from what was originally asked. By step 6, you're solving the problem of step 5's output, not the original intent. The original question can effectively become invisible.

Counter: explicitly re-read the original request before committing to the direction of any major action. Not "do I remember what was asked" — actually go back and read it.

What do you think about these updates to my soul files ?

timid vessel
#

if you just want to talk to your agent from your mac, the openclaw dashboard is already built in just go to http://your-vps-ip:18789 in your browser. make sure the port is open in your firewall. no extra install needed.

fresh moon
timid vessel
#

subagents right now are session based so they spawn, do the task, then disconnect. for better orchestration, set up separate agents in your config with clear roles and have the main one delegate to them. persistent subagents are still a pain point for most people here.

arctic star
#

gm , i need some help with browser automation - i want to use my claw to send files/photos in standard upload form; but she can't see the screen of the file selection popup ; as i use the browser relay - any one knows a workaround

twin cosmos
#

Has anyone familiarized themselves with FARS? It seems similar to OpenCLAW.

heavy wren
heavy wren
trail notch
trail notch
full kettle
#

Hi, how can I make my openclaw processes NPU-optimized?

#

It doesn't seem to be using my NPU at all

cinder jacinth
#

Did google block connection via antigravity quota?

wary osprey
#

It really depends on what you’re optimizing for.

The $20 ChatGPT Plus plan mainly gives you UI access. It’s great for manual use, but it’s not the same as having a stable, production-ready API setup for agent workflows.

If you’re running OpenClaw agents, API stability, predictable billing, and proper rate limits matter more than UI access.

If cost is your main concern, compare:

  1. Token pricing per 1M
  2. Rate limits under automation
  3. Timeout behavior during coding tasks

Also worth mentioning, if you’re trying to reduce API cost without sacrificing flexibility, you could look at something like AIsa.

It’s a unified API layer similar to OpenRouter, but with centralized billing and generally more competitive pricing on supported models, like KIMI, and especially in the Qwen ecosystem since it’s an official Alibaba partner.

For agent workflows, i believe having stable API access, clear usage tracking, and competitive token pricing tends to matter more long-term than subscription-based UI plans.

foggy schooner
#

Is anyone here using an OpenAI subscription (not the API) with OpenClaw?
Is this officially safe, or is there any risk of account suspension like with Claude or Google?
I’ve heard mixed things about this and just want to be sure before trying it.

heavy wren
gaunt field
#

This openclaw seems interesting, but its giving cryptic errors even installing. I understand about permissions and docker, but even on windows, node gave an errors on line 1 and core dumped.... maybe if I change the docker file to only use volumes, it might actually install

umbral charmBOT
#

is moltbook safe?

sour slateBOT
#

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

gaunt field
#

How has people here installed openclaw without it coredumping on install?

wary osprey
# foggy schooner Is anyone here using an OpenAI subscription (not the API) with OpenClaw? Is this...

subscriptions like ChatGPT Plus are meant for UI use, not automated API-style workflows.

If you’re trying to connect a subscription account directly into OpenClaw for agent automation, that can fall into gray territory depending on the provider’s terms of service. That’s usually where suspension happens.

officially supported and documented API usage is generally the best route for automated API-style workflows.

also , if stability matters long-term to you, then stick to

. Official API access
. Clear billing + usage tracking
. Or a unified API layer that’s designed for agent workflows (e.g AIsa)

gaunt field
#

I suppose its not ready for usage yet?

haughty hedge
#

If i create new agents, where would their agents.md file etc be located?

smoky kestrel
#

Hi helpful people.
Anyone here had to overcome an issue with their agents' training being shared in discord?
I couldn't see any existing complaints about it in discord, and my agents burned through tokens to come up with, "well, 'show reasoning' is switched off, so you shouldn't see it" but it's included in every message that any of my agents send, both in server and DMs

nocturne zephyr
# gaunt field I suppose its not ready for usage yet?

Works for most of us. If you're having trouble getting it running, there's a good #1459642797895319552 channel also. If you're not prepared for sharp edges, though, it's okay to wait for something more your speed. It's under very active development, and things change on a regular basis. And so do the top models to use, underneath it.

high heron
nocturne zephyr
high heron
nocturne zephyr
high heron
#

so it should be fine

#

also, you can use google cli ig

#

i cant say theres no risk for cli

#

but it def works

#

and its not as serious as antigravity

haughty hedge
smoky kestrel
nocturne zephyr
nocturne zephyr
wild rapids
#

Hi! New here, what’s the best way to enable browsing capabilities for open claw? Curious how everyone’s doing it, considering limitations with web_fetch? Do you just enable browser capabilities?

kindred gate
#

does anyone use openclaw with local ollama

gleaming shard
#

Yes I did used openclaw with local ollama

gleaming shard
kindred gate
gleaming shard
#

glm-4.7-flash

wild rapids
#

Why brave api as opposed to a browser?

gleaming shard
#

Is it possible to run openclaw for generating leads, and qualifying them. After qualifying can it send emails to the qualified leads.

kindred gate
#

did you do ollama launch openclaw?

gleaming shard
kindred gate
gleaming shard
kindred gate
tiny fulcrum
#

Any slack users? this bot crashing 3 times a day...

sullen lotus
#

Please help me, i installed in my Mac mini with qwen 3:8b with max tokens 256k..and it was running just fine until i overloaded it using wrong.. so now i installed it again, same model but i put max tokens 128k, and is runnin damn slow

kindred gate
#

what did you reinstall

humble trail
#

i finally got set up yesterday. it seems like i got disconnected overnight. bot wasn’t responding through TG, so i prompted in terminal and it says “not connected to gateway - message not sent”. any idea what this issue is?

fallow cloak
#

clawhub publish throws Timeout error

#

anyone has an idea ?

craggy cobalt
#

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

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

anyone know what this means and how to fix it. im tryina install

dusky lynx
dusky lynx
humble trail
dusky lynx
sullen lotus
# kindred gate what did you reinstall

I reinstalled openclaw with qwen 3:8b, initially I also put APIs for kimi but it was running (or at least it said it was) only on qwen and it was fluent, it said that it was using 256k of contest and I used it a lot, zero issues, then it suddenly became very slow and un fixable (too many subagents and errors) so I decided to reinstall it but this time I put 128k of contest.. but it s just going so slow

wary osprey
# sullen lotus I reinstalled openclaw with qwen 3:8b, initially I also put APIs for kimi but it...

A few things might be happening…

First, dropping from 256k context to 128k won’t automatically make it faster. In fact, if your agent is spawning more subagents or retrying due to errors, that can dominate performance more than raw context size.
What to check…

  1. Check if your subagents are looping?
    If you’re seeing “too many subagents,” it could mean the manager model is repeatedly failing tool calls and retrying.
  2. Context bloat. Even if you’re sending large history or logs every turn with 128k, Qwen 8B will still slow down.
  3. Local inference limits. Qwen 3:8B running locally depends on quantization and GPU utilization. If it’s not properly offloading to GPU or VRAM is saturated, it’ll crawl.
  4. If OpenClaw detects malformed tool calls, it may retry automatically. Make sure to check logs for repeated failed tool-call attempts.

possible fix…
Check your OpenClaw version changes, check your config differences between installs
Log patterns (especially repeated retries)

pliant abyss
#

Does anyone have a full friendly guide / website for OpenClaw vps setups? Mine keeps forgetting it lives in a VPS

wild rapids
#

Mine keeps trying to edit its config file and then keeps dying cuz gate way won’t start properly. I’m so confused. I feel like I’m doing everything right

rare kindle
#

Hi,

After the latest update 24 February from 17 February, my Telegram bot is always showing "… typing" ALL TIME.

I asked it to fix it but it restart after a few messages.

fiery arch
#

Anyway of getting openclaw to work with Base URLs, e.g; GEMINI_BASE_URL , OPENAI_API_BASE etc?

burnt belfry
#

anyone facing issues with telegram

#

My connection is dropping mid way. Getting "deletewebhook failed"

remote mantle
#

Hi gents, I've been trying for days but cannot get openclaw installed... Non of the documentation, nor tutorials works for me.. and yes.. I've followed every step EXACTLY..

Keep getting this error:

'Error: Cannot find module ./openclaw.mjs'

vivid kraken
#

I need to implement a feature of getting answers based on my documents. do we need to implement rag with openclaw, or are there simpler ways to do that

echo quail
#

Anyone ever make a Mission Control in co work that’s connected via discord?

twilit onyx
#

openclaw is binging tokens like beers on a night out but 24/7... anyone has the best set up with openrouter?

humble trail
#

i asked my claw to update itself, now getting this error when i try to restart gateway

#

C:\Users\brvdy>openclaw gateway restart
[openclaw] Failed to start CLI: Error [ERR_MODULE_NOT_FOUND]: Cannot find module 'C:\Users\brvdy\AppData\Roaming\npm\node_modules\openclaw\node_modules@sinclair\typebox\build\esm\index.mjs' imported from C:\Users\brvdy\AppData\Roaming\npm\node_modules\openclaw\node_modules@mariozechner\pi-ai\dist\index.js
at finalizeResolution (node:internal/modules/esm/resolve:275:11)
at moduleResolve (node:internal/modules/esm/resolve:866:10)
at defaultResolve (node:internal/modules/esm/resolve:992:11)
at #cachedDefaultResolve (node:internal/modules/esm/loader:719:20)
at #resolveAndMaybeBlockOnLoaderThread (node:internal/modules/esm/loader:736:38)
at ModuleLoader.resolveSync (node:internal/modules/esm/loader:765:52)
at #resolve (node:internal/modules/esm/loader:701:17)
at ModuleLoader.getOrCreateModuleJob (node:internal/modules/esm/loader:621:35)
at ModuleJob.syncLink (node:internal/modules/esm/module_job:160:33)
at ModuleJob.link (node:internal/modules/esm/module_job:245:17)

#

any ideas?

reef galleon
remote mantle
#

@james.morgan windows 10

merry bone
#

Anyone know where to report bugs? Semi-confident we found one related to this

south bison
#

hello guys, when I try to install openclaw, it stuck at opening the safari, and safari show "safari can't open the page "http://127.0.0.1:18789" because safari can't connect to the server "http://127.0.0.1""
How to solve this issue?

unborn remnant
#

I have an issue or question regarding allow list and right management:

Problem

From my understanding currently, OpenClaw's group access control is binary: a sender is either on the allowFrom list (full access) or not (no access). There is no way to differentiate between an owner/admin who should have full control (config changes, exec, sensitive tools) and a regular participant who should only be able to chat.

This creates a real security gap in group chat scenarios:

Example: I run OpenClaw as a personal assistant. In my Telegram group, I want my colleague to be able to ask the bot questions and have conversations. But I do NOT want him to be able to:

• Change the bot's configuration (gateway config.patch)
• Execute shell commands (exec)
• Access my files or workspace
• Modify cron jobs
• Send emails on my behalf
• Access API keys or credentials

Right now, if someone is on the allowFrom list, the bot will attempt to follow any instruction they give — including sensitive operations. The only safeguard is the model's own judgment, which is not reliable enough.

sage jolt
#

Why could OpenClaw previously read the localhost 3000/bot.md file but now can't? Is it due to configuration issues?

unborn remnant
# south bison hello guys, when I try to install openclaw, it stuck at opening the safari, and ...

I run openclaw doctor in these cases. If you want more hier is the answer from my bot 😄 Hey! This is a common issue. The gateway process likely isn't running (yet) when Safari tries to open the page.

Quick fixes:

  1. Check if the gateway is actually running:
    openclaw gateway status
    If it says "not running", start it:
    openclaw gateway start

  2. If it's a Mac — the one-liner install sometimes opens Safari before the gateway is fully ready. Just wait 10-20 seconds and refresh the page.

  3. Check if port 18789 is listening:
    lsof -i :18789
    If nothing shows up, the gateway didn't start properly.

  4. Common causes:

• Node.js not properly installed → run node -v (should be v18+)
• Gateway crashed during startup → check openclaw logs
• macOS firewall blocking localhost → System Settings → Network → Firewall → allow Node.js

  1. Nuclear option — kill everything and restart:
    openclaw gateway stop
    openclaw gateway start
    Then manually open http://127.0.0.1:18789 in Safari/Chrome.

If none of that works: run openclaw doctor and post the output — that usually shows what's broken. 🦞

sly whale
marble violet
#

Hi everyone.. i have an issue since yesterday.. cannot start gateway...

"openclaw gateway status
file:///home/molt/.npm-global/lib/node_modules/openclaw/dist/entry.js:655
if (state === "invalid") throw new Error(Unsafe fallback OpenClaw temp dir: ${fallbackPath});
"

any suggestions?

wheat bronze
upper radish
#

Hello, My openclaw just work with my loopback address, I want to be able to use the UI locally with my internal network, I use macs, anyone knows how to do it? After setup the remote opton I got coneection refused all the time

sly whale
smoky kestrel
#

What? Each agent has their own memory

unborn wharf
#

Anyone else have issues with duplicated messages showing up one right after the other in discord? I’m getting dupes on like all of my cron jobs.

hollow notch
#

Hey! Quick version-check question on browser relay fixes:

I’m on OpenClaw 2026.2.25 and still seeing relay/attach instability (especially around tab switching / browser restarts).
Which released version includes these PRs?

#15817 (auto-reconnect, MV3 persistence, keepalive),
#10390 (sticky attach + auto-restore attached tabs),

Also: are both included in stable, or only canary/nightly right now?

My setup:
macOS,
Chrome + extension relay,
relay reachable/authenticated,
issue is mainly needing to re-attach / occasional “tab not found” behavior.

pearl knoll
#

does anyone know where i can add the gateway auth token in the macos app? it keeps complaining it cant connect due to missing gateway auth token - to my gateway running locally (in docker).

i have no issues connecting with the web ui, but cant seem to find where this piece of config is hidden in the app

eager flume
#

Hi, is anyone else getting those messages (gemini usage shows that 80% usage is still left)

⚠️ API rate limit reached. Please try again later.

wary sapphire
#

Anyone have ways of lowering their api usage for openclaw?

placid flower
#

I have several openclaw instances on different machines , all have telegram bots, I have several groups in telegram, some belong to multiple groups as admins, THe littlest change to a config breaks all of a groups communication. Do I need to delete sessions.json to let rebuild based on config?

glossy sapphire
flint ether
#

So, my claw use pi with background=true to write code. What is the best way to actually observe what does pi doing in real time?

unborn wharf
# wary sapphire Anyone have ways of lowering their api usage for openclaw?

I just had gpt 5.2 Thinking (I’m a Plus subscriber) audit my config, primary agent and fallbacks, as well as cron job and heartbeat workflow. I disabled heartbeat because it’s not needed for my use case (just a general life assistant). Shifted some cron jobs to crontab/linux if they were simple enough regular reminders. Then made my model fallbacks more efficient and try to use oauth (gpt codex) for most back and forth but still have to be mindful of not being wasteful here. I got an almost 4 day cooldown on codex recently.

sand cloak
#

How can I determine if a command is still processing? The chat will say something like "Let me..." then freeze. In telegram, it will say "Typing..." but then that will stop as well. Trying to determine when I need to "poke" it vs. just wait because it's still crunching on my request?

sand cloak
wary sapphire
wary sapphire
glossy sapphire
#

It depends on the types of task you’re trying to delegate, but at the end of the day most API calls without limits equals burning money.

#

There’s some helpful guides on openclaw docs on subagents you should check out @wary sapphire

earnest jay
#

Anyone here know where I can install the android companion app?

lyric oriole
#

I'm having issues with my telegram groups. The message that I send in groups aren't received by Openclaw. Does anyone know how to fix it? (non-techie here)

iron shuttle
#

Does anyone know what is the name of the app on Android to connect to my gateway?

iron shuttle
shut mirage
#

What models are people using for powering this?
And licensing from where?
I had been using AntiGravity, but feeling iffy now about that...

shut mirage
earnest jay
#

Lol everyone asking the same question

#

The documentation mentions the app but no download link

shut mirage
#

what is the app?

earnest jay
#

And can't find on playstore

shut mirage
#

😊

#

OpenClaw app?